Description
An ethnic enclave is a physical space with high ethnic concentration; thus these spaces are culturally distinct from the larger receiving society. Ethnic enclaves are found in virtually every country, arising in response to increased immigration of people from the same ethnic background.
ABSTRACT
Title of dissertation:
FIRM OWNERS AND WORKERS: AN ANALYSIS OF IMMIGRANTS AND ETHNIC CONCENTRATION M´ onica Garc´ ?a-P´ erez, Doctor of Philosophy, 2009
Dissertation directed by:
Professor John Haltiwanger Department of Economics
This dissertation consists of three chapters examining the important role of ?rm and coworker characteristics, as well as the use of social networks, in labor markets. The ?rst paper investigates the e?ect of ?rm owners and coworkers on hiring patterns and wages. Immigrant-owned ?rms are more likely to hire immigrant workers. This prevalence is especially strong for Hispanic and Asian workers. We also ?nd that the probability that a new hire is a Hispanic is higher for immigrant ?rms. On wage di?erentials, the results illustrate that much of the di?erence between the log annual wages of immigrants and natives can be explained by immigrants’ propensity to work in non-native owned ?rms, which pay the lowest average wages. Interestingly, though, native workers holding a job in immigrant ?rms are paid less than immigrant workers. The last section examines the potential mechanisms for these ?ndings. It explores the importance of job referral and use of networks for migrants in labor markets. We consider the theoretical implications of social ties between owners and workers in this context. Firms decide whether to ?ll their vacancies by posting their o?ers or by using their current workers’ connections.
Next, we explore the patterns of immigrant concentration relative to native workers at the establishment level in a sample of metropolitan areas. Immigrants are much more likely to have immigrant coworkers than are natives, and are particularly likely to work with others from the same country of origin, even within local markets. The concentration of immigrants is higher for recent immigrants and interestingly for older immigrants. We ?nd large di?erences associated with establishment size that cannot be explained solely by statistical aggregation. Exploring the mechanisms that underlie these patterns, we ?nd that proxies for the role of social networks, as well as the importance of language skills in the production process, are important correlates of immigrant concentration in the workplace.
FIRM OWNERS AND WORKERS: AN ANALYSIS OF IMMIGRANTS AND ETHNIC CONCENTRATION
by M´ onica Isabel Garc´ ?a-P´ erez
Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial ful?llment of the requirements for the degree of Doctor of Philosophy 2009
Advisory Committee: Professor John Haltiwanger, Chair/Advisor Professor John Shea Professor Seth Sanders Professor Judith Hellerstein Professor Howard Leathers
c Copyright by M´ onica Isabel Garc´ ?a-P´ erez 2009
Dedication To Chavela and Nene . . . This thesis is dedicated to my wonderful parents, Isabel y Luis, who have raised me to be the person I am today and sacri?ced a lot to o?er me the means to reach my dreams. You have been with me every step of the way, through good times and bad. Thank you for all the unconditional love, guidance, and support that you have always given me, helping me to succeed and instilling in me the con?dence that I am capable of doing anything I put my mind to. ¡Gracias, los amo!
ii
Acknowledgments
It is a pleasure to thank to many people who made this thesis possible. I owe my gratitude to all of them and because of whom my graduate experience has been one that I will cherish forever. I cannot overemphasize my gratitude and love to my husband Darin, who did more than his share around the house as I sat at the computer. Without his support, and gentle prodding, I would still be trying to write the introduction of my ?rst paper. With his enthusiasm, his patience, his revisions and his inspiration, he helped me to overcome the hardest moments of the creative process. I would have been lost without him. Thank you for your love amor. I would like to express my deep and sincere gratitude to my advisor, Prof. John Haltiwanger for his valuable advice and great encouragement as well as for his excellent guidance and assistance for this research. His wide knowledge and his logical way of thinking have been of great value for me. I’d like to thank him for giving me an invaluable opportunity to work on challenging and extremely interesting projects over the past three years. It has been a pleasure to work with and learn from such an extraordinary individual. I am deeply grateful to Prof. John Shea for o?ering advises in di?cult moments and always have the door open when I needed him. I am also indebted for the amount of time and e?ort he has spent reading and correcting my work. I also wish to express my warm thanks to Professor Seth Sanders who inspired me with his questions and comments. His discussions around my work and initial explorations iii
have been very helpful for this study. My most sincere gratitude to Kristin Sandusky for being next to me and kindly grants me her time for answering my unintelligent questions during the time of this research. I am much indebted to Kristin for her valuable insights on the data details and her emails with great tips. I also thank to Dr. Jose Tessada, Dr. Jeanne LaFortune, Prof. Judy Hellerstein, and the participants in brownbag seminars in the University of Maryland. I am also grateful to my many student and work colleagues for providing a stimulating and fun environment in which to learn and grow. To Helena Schweiger for sharing great times at the beginning of my career and o?ering me a lot of support. I would also like to acknowledge help and support from some of the sta? members, especially to Vickie Fletcher, Elizabeth Martinez, and Terry Davis, who provided me with all the help and advice in all the administrative process during the
A career. I am thankful to Dorothea Brosious for providing the L TEX thesis template.
Finally, I would like to extent my gratitude to Jeremy Wu, Fredrik Andersson and all the LEHD sta? at the Census Bureau for their kind support. I owe my deepest thanks to my family. Words cannot express my gratitude.
iv
Disclaimer
This work is uno?cial and thus has not undergone the review accorded to o?cial Census Bureau publications. The views expressed in the paper are those of the authors and not necessarily those of the U.S. Census Bureau or the U.S. Department of the Treasury. All papers are screened to ensure that they do not disclose con?dential information. Persons who wish to obtain a copy of the paper, submit comments about the paper, or obtain general information about the series should contact Sang V. Nguyen, Editor, Discussion Papers, Center for Economic Studies, Bureau of the Census, 4600 Silver Hill Road, 2K132F, Washington, DC 20233, (301-763-1882) or internet address [email protected].
v
Table of Contents
List of Tables
viii
List of Figures
xi
List of Abbreviations
xiii
1 Introduction
1
2 Does It Matter Who I Work For And Who I Work With? The Impact Of Owners And Coworkers On Wages And Hiring 2.1 2.2 2.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 8
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 On the use of social networks . . . . . . . . . . . . . . . . . . . . . . 18 2.3.1 Small ?rms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4
Data and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Construction of ex post weights . . . . . . . . . . . . . . . . . 30 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Measuring coworker share . . . . . . . . . . . . . . . . . . . . 42
2.5
Analysis of New Hires, Earnings of Workers and Skill Distribution . . 42 2.5.1 2.5.2 New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Earnings of Workers . . . . . . . . . . . . . . . . . . . . . . . 44
vi
2.5.3 2.6
Sorting by Skill . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.6.1 2.6.2 Analysis of ?rms hiring patterns . . . . . . . . . . . . . . . . . 58 Hiring Process by Race/Ethnicity . . . . . . . . . . . . . . . . 63 2.6.2.1 2.6.2.2 2.6.3 Worker Race . . . . . . . . . . . . . . . . . . . . . 63 Worker and Owner Races . . . . . . . . . . . . . 71
Workers’ earnings and analysis of results . . . . . . . . . . . . 78
2.7
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3 Workplace Concentration of Immigrants 3.1 3.2
86
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 3.2.1 3.2.2 Literature on earnings di?erences . . . . . . . . . . . . . . . . 89 Literature on segregation . . . . . . . . . . . . . . . . . . . . . 91
3.3
Methodology and Data . . . . . . . . . . . . . . . . . . . . . . . . . . 97 3.3.1 3.3.2 3.3.3 3.3.4 Measuring immigrant concentration . . . . . . . . . . . . . . . 97 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Regression speci?cations . . . . . . . . . . . . . . . . . . . . . 103 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . 106
3.4
Accounting for immigrant concentration . . . . . . . . . . . . . . . . 112 3.4.1 3.4.2 3.4.3 Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Worker Demographics . . . . . . . . . . . . . . . . . . . . . . 116 Employer characteristics . . . . . . . . . . . . . . . . . . . . . 118
vii
3.4.3.1 3.4.3.2 3.5
Employer size . . . . . . . . . . . . . . . . . . . . . . 119 Industry . . . . . . . . . . . . . . . . . . . . . . . . . 125
Exploring social networks, language skills, and human capital as possible explanations for concentration . . . . . . . . . . . . . . . . . . . 127
3.6 3.7
Country of origin di?erences . . . . . . . . . . . . . . . . . . . . . . . 135 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5 Conclusion
137
Appendix
140
A Matching Rate
140
B De?nitions
141
C Unknown-Owned Firms
142
D Weights and Selection
143
E IPUMS 1990: Descriptive Statistics
147
F Linear Probability Estimates
148
G Simulations of employer size e?ects in a statistical model with segregation 154
Bibliography
164
viii
List of Tables
2.1
Descriptive Statistics - CBO(1992) and Sample/Matched Firms . . . 35
2.2
Descriptive Statistics - Characteristics of Workers . . . . . . . . . . . 38
2.3
Average Race and Ethnic Composition of New Hires by Owner’s Type 43
2.4
Average Race and Ethnic Composition of New Hires by Owner’s Race 44
2.5
Mean Earnings by Owner and Worker Type . . . . . . . . . . . . . . 49
2.6
By Similar Coworker Share: Mean Earnings by Owner and Worker Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.7
Worker types distribution by owner’s skill requirement . . . . . . . . 53
2.8
Linear Estimates of the E?ect of Owner Type on the Probability that a New Hire is an Immigrant . . . . . . . . . . . . . . . . . . . . . . . 61
2.9
Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic . . . . . . . . . . . . . . . . 63
2.10 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian . . . . . . . . . . . . . . . . . . 66
2.11 Multinomial Logit Model: E?ects of Owner Type and Coworkers on Type of New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
viii
2.12 Multinomial Logit Model: Predicted Probability of Covariates . . . . 70
2.13 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic . . . . . . . . . . . . . . . . 71
2.14 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian . . . . . . . . . . . . . . . . . . 74
2.15 Multinomial Logit Model: E?ects of Owner’s Race and Coworkers on Type of New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.16 Multinomial Logit Model: Predicted Probability of Covariates Owner and Worker Races (%) . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.17 OLS Results: E?ect of Owner Type and Coworker Share on Log Real Annual Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.1
Variation in Immigrant Share of Workforce across Sample MSAs . . . 103
3.2
Characteristics of Immigrant and Native Workers, Full Sample
. . . 106
3.3
Contribution of Covariates to Immigrant Concentration (Full Sample) 113
3.4
Characteristics of Matched Sample Workers (Unweighted)
. . . . . . 128
3.5
Characteristics of Immigrant and Native Workers, Matched Sample (weighted) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
ix
3.6
Contribution of Covariates to Immigrant Concentration (Matched Sample) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.7
Network E?ects from Coworker Share Regressions . . . . . . . . . . . 133
A.1 Matching and Non-matching rate of ?rms in CBO and SSEL(singleunit) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
D.1 Descriptive Statistics - CBO(1992) and Sample/Matched Firms . . . 145
E.1 Descriptive Statistics - Characteristics of Workers . . . . . . . . . . . 147
F.1 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Black . . . . . . . . . . . . . . . . . . 148
F.1 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 149
F.2 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is White . . . . . . . . . . . . . . . . . 149
F.2 Linear Probability: E?ect of Owners types on the Probability that a New Hire is White (continued) . . . . . . . . . . . . . . . . . . . . . . 150
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black . . . . . . . . . . . . . . . . . . 150
x
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 151
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 152
F.4 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is White . . . . . . . . . . . . . . . . . 152
F.4 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is White (continued) . . . . . . . . . . . 153
G.1 Characteristics of Weighted Matched Sample . . . . . . . . . . . . . . 162
G.2 Linear Regression of Full Speci?cation . . . . . . . . . . . . . . . . . 163
xi
List of Figures
2.1
Workforce Characteristics of Immigrant, Mix and Native Firms . . . . 45
2.2
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.1
Cumulative Distribution of Coworker Share by Worker Type . . . . . 111
3.2
Coworker share by age of employee . . . . . . . . . . . . . . . . . . . 118
3.3
Coworker share by employer size . . . . . . . . . . . . . . . . . . . . . 120
3.4
Cumulative Distribution of Coworker Share by Worker Type and Employer Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.5
Coworker share by employer sector . . . . . . . . . . . . . . . . . . . 126
G.1 Shape of function d
. . . . . . . . . . . . . . . . . . . . . . . . . . . 156
G.2 Immigrant share distribution with and without segregation
. . . . . 158
xi
G.3 Immigrant coworker mean and employer size (? = 4)
. . . . . . . . . 159
G.4 Native coworker mean and employer size (? = 4)
. . . . . . . . . . . 159
G.5 Immigrant-native di?erence in coworker mean and employer size (? = 4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
G.6 Immigrant-native di?erence in coworker mean and employer size (? = 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
xii
List of Abbreviations
CBO SSEL BR LEHD MSA Characteristics of Business Owners Standard Statistical Establishment List Business Register Longitudinal Employer-Household Dynamics Metropolitan Statistical Area
xiii
Chapter 1 Introduction
Over the last several decades, labor markets in many cities in the US have absorbed large in?ows of new immigrants. During the same period, numerous empirical studies have analyzed the e?ect of immigration in the host economy. In the early 90s the consensus was that there is only a small e?ect of immigration on native economic outcomes (Grossman [1982]). However, since the late 90s, the consensus moved toward a signi?cant e?ect of foreign-born migration on natives (Borjas [1994]). Recent surveys on the economics of immigration [Borjas, 2003, 2005], Friedberg and Hunt,1995; Card, 2001; Card and Lewis,2005; Card, 2006) conclude that the impact of immigration on the wages and employment is still unclear. As of 2007, immigrant workers represented 15% of the U.S. population. The impact of large in?ows of immigrants and their assimilation into the host economy has been a primary objective of analysis in the labor literature. How such large ?ows of workers are incorporated into the labor market and interact with various businesses and workers is of special interest. An alternative literature has focused on how ?rms respond to an in?ow of immigrants. The key question is no longer one of job supply but also one of job distribution. Lewis and Card [2005] and Beaudry et al. [2006] look at an exogenous local unskilled labor supply change and ?nd that areas with higher concentration of immigrants have employed higher number of unskilled
1
workers and increased productivity at the same time. Their ?ndings also suggest a small impact of immigration on natives’ relative wages. In this analysis, the role of business owners in the patterns of hires and earnings in the labor market is relevant. In particular, some studies have found that the type of manager recruiting new workers is a determinant in the workforce composition of the business. In a extensive analysis of race and ethnic segregation across workplaces in the U.S., Hellerstein and Neumark [2007] ?nd that a large degree of segregation remains even after accounting for metropolitan area, education and occupation. In a follow-up paper, they explore the role of residential networks in these patterns, and found preliminary evidence of its relevance for low-educated and low-English-ability workers. On the other hand, many other authors have analyzed the direct e?ect of the type of manager on the type of worker in the ?rm. For instance, Carrington and Troske [1995] and Giuliano and Ransom [2008] have found that females and blacks are disproportionately employed by female and black supervisors respectively. Meanwhile, Stoll et al. [2004] found that black businesses receive more applications from black workers and employ more black workers than other businesses. Giuliano and Ransom [2008] found a causal relation between the race of managers and workers using panel data of a retail store. They control for the unobserved characteristics that can also a?ect the race of the coworkers and hires in a ?rm. Although the primary determinants of the racial composition of new hires are workplace and location characteristics, manager race also stands as a signi?cant component. Nevertheless, this second group of analyses have been mainly focused on black versus white issues 2
and particular industries. In the sociology literature, there have been a limited number of studies that provide some insights on the tendency of immigrants to work for immigrant ?rms. For instance, in Los Angeles in 1989, 30 percent of employed Koreans held jobs in ?rms owned by fellow Koreans even though Koreans composed only one percent of the Los Angeles County population.1 According to Cardenas and Hansen [1988], during the 1980s, Mexican immigrant employers were most likely to hire Mexican, whether legal or undocumented, and were more likely to evaluate their quality favorably. Porter and Wilson [1980] ?nd two relevant patterns in the Cuban immigration to Miami during the 1960s. First, Cubans worked with other Cubans. Second, almost one-third of the Cubans worked for Cuban employers. The phenomenon of immigrants hiring immigrants is not limited to coethnic relationships between employees and employers. Other researchers have found that employers from one immigrant group often hire workers from other ethnic/racial groups.2 Immigrant entrepreneurs can take advantage of their language, cultural background and a?nities to have access to di?erent ethnic groups. Their immigrant status can give them privileged access to sources of labor less available to native entrepreneurs. Immigrant entrepreneurs routinely employ coethnics (including relatives) at rates vastly above chance levels.3 Making use of unique longitudinal and cross-sectional micro level databases, this thesis examines the role of owners, coworkers and networks, focusing on the
1 2
Min [1989]. Light [2006]. 3 Massey [1999], Massey et al. [1987].
3
importance of immigration and race/ethnicity on hiring patters, the scope for segregation and wage di?erentials. The main contributions of the research presented in this document are providing new stylized facts on the immigration issue and evidence on the role of social networks in labor markets4 . The outline of the thesis is as follows. Chapter 2 analyzes the e?ect of the birthplace of ?rm owners and coworkers on hiring patterns and wages. Using a unique matched sample from an employer-employee administrative database and a survey of characteristics of ?rm owners, this chapter studies the impact of the type of employers and individual coworkers (native versus immigrant workers, and ethnic/racial groups) on ?rm hiring patterns and workers’ average log wages. We connect owner and ?rm characteristics (place of birth, size and industry) with workers’ characteristics (wage, age, education, and place of birth) to test di?erent assumptions about ?rm hiring patterns and the wage di?erentials of workers of di?erent types. Given the unique features of the matched database, the data allows asking whether the odds that a worker of a particular group is hired are related to the types of owners and coworkers, and whether there exist wage premia associated with being an immigrant and working for or with other immigrants. Our results suggest that immigrant owners are three percentage points more likely to hire other immigrants than native owners, even after controlling for industry, ?rm size, geographic concentration of immigrants in the population, population density, and the legal form of organization of the ?rm. Looking at ethnic/race groups, immigrant owners are 3 to 4 percentage points more likely than native own4
For an extensive analysis on job information networks see Ioannides and Loury [2004].
4
ers to hire Asians and Hispanics versus blacks. Both types of owners, immigrants and natives, hire white non-Hispanic workers, but native owners have a higher probability of having white workers as new hires. These results are based on linear probability models as well as multinomial logit speci?cation that accounts for the simultaneity of choosing from among di?erent types of workers. Among our strongest ?ndings are the existence of a persistent pattern of hiring similar types and the e?ect of the share of dissimilar coworkers on the likelihood of hiring a particular individual. For instance, the increase of the share of similar coworkers at the time of recruitment by 100 workers increases the probability of hiring a worker of a type by around 60%. The probability is smaller if we look at the e?ect of the fraction of coworkers of other di?erent types. Additionally, this probability depends on whether the employer is immigrant versus native. Immigrant businesses show higher chances of hiring a new immigrant, Hispanic or Asian worker compared to native businesses, even after controlling for whether they have similar workforce distribution at the time of a new recruitment. Later, after controlling for owner’s race, our results are similar. Hispanic and Asian owners are 2.5 percentage points more likely to hire their own type (Hispanic and Asian workers respectively) than white and black owners. Given the lack of representation of native Hispanic and Asian owners in the data, we were not able to control for the cross categories race-birthplace. To the best of our knowledge, no previous study has analyzed the link between employer and coworkers’ birthplaces and employees’ employment opportunities and wages. This research provides initial steps on that branch of analysis. 5
Chapter 3 presents descriptive evidence on immigrant segregation at the workplace and analyzes the mechanisms that drive immigrant concentration. We have unique matched employer-employee data for a large number of states in the US that permits quantifying the extent of and covariates of the workplace concentration of immigrants. A lack of suitable data has limited economists’ ability to address these questions. The paper has two broad objectives. The ?rst is primarily descriptive. The descriptive ?ndings show that immigrants are much more likely to have immigrant coworkers than are natives. This pattern is driven partly by the geographic concentration of immigrants, but the patterns hold true even within local labor markets. At the same time, most immigrants do have native coworkers; only a small share work in immigrant-only workplaces. The concentration of immigrants is higher for recent immigrants and, conditional on recent arrival, for older immigrants. Part of the assimilation process is a movement towards more interaction with natives in the workplace over time, and younger immigrants are more likely to work with natives. We ?nd large di?erences associated with ?rm size: concentration is much higher in smaller ?rms, but is far from zero even in the largest ?rms. We also ?nd substantial variation in the extent of immigrant concentration across industries even after controlling for a detailed set of location, employer and employee characteristics. Second, our ?nding that the allocation of immigrants across workplaces is far from random raises the question of what drives this workplace concentration. Both the existing literature and our descriptive ?ndings suggest that it is important to consider how businesses hire their employees and the choices that businesses 6
make about the skill mix of their workforce. One relevant issue here is the role that language skills play in governing interactions among employees and between employees and customers. A second issue is the role of social networks in the process that matches workers and ?rms. A third issue is human capital - the sorting and concentration of immigrants in the workplace may re?ect sorting by skills. In the second part of the paper, we explore the role of these factors. We ?nd evidence that immigrants with primarily immigrant coworkers are likely to have coworkers who live in the same residential tract. This pattern is robust to inclusion of controls for other closely related factors such as residential segregation. We also ?nd evidence that immigrant workers with poor English speaking ability and low education are more likely to work with immigrant coworkers. Our ?ndings suggest that social connections and social capital may be important for understanding workplace concentration, employment opportunities and wage di?erentials. Continuing this line of thought, Chapter 5 o?ers the conclusions and discusses on the main factors that can explain the previous ?ndings. It is intended to focus on the role on networks in the labor markets, and the connection of our ?ndings with previous empirical and theoretical literature. It also describes the key issues to be considered to develop in appropriate theory.
7
Chapter 2 Does It Matter Who I Work For And Who I Work With? The Impact Of Owners And Coworkers On Wages And Hiring 2.1 Introduction
This paper analyzes the e?ect of the birthplace of ?rm owners and coworkers on hiring patterns and wages. As of 2007, immigrant workers represented 15% of the U.S. population. The impact of large in?ows of immigrants and their assimilation into the host economy has been a primary area of study in the labor literature. How such large ?ows of workers are incorporated into the labor market and interact with various businesses and workers is of special interest. The role of business owners in the patterns of hires and earnings in the labor market has played an important role in this literature. In particular, some studies have found that the type of manager recruiting new workers is a determinant of the ?rm’s workforce composition. For instance, Carrington and Troske [1995] and Giuliano and Ransom [2008] have found that females and blacks are disproportionatly employed by female and black supervisors respectively. Meanwhile, Stoll et al. [2004] found that black businesses receive more applications from black workers and employ more black workers than other businesses.
8
Using a unique matched sample from an employer-employee administrative database and a survey of characteristics of small-?rm owners, this study analyzes the impact of the type of employers and individual coworkers (natives versus immigrants, or ethnic groups) on ?rm hiring patterns and workers’ average log wages. Firm types are de?ned by the type of owner (immigrant-owned versus native-owned), while ‘’coworker” refers to the fraction of same-kind fellow workers holding a job in the same ?rm. The share of immigrant coworkers in the ?rm is called the coworker index.1 We connect owner and ?rm characteristics (place of birth, size and industry) with workers’ characteristics (wage, age, education, and place of birth) to test different assumptions about ?rm hiring patterns and the wage di?erentials of workers of di?erent types. Given the unique features of the matched database, the data allows asking whether there exist wage premia associated with being an immigrant and with working for or with other immigrants. The type of a new hire can be a?ected by the type of employer in di?erent ways. First, social networks, segregated by race or similar background, could be used by job seekers and by employers when looking for new candidates. Ethnic communities provide a network for immigrant entrepreneurs to ?nd workers, to sell ethnic goods, and to obtain credit. Second, matching productivity generated by employer-employee similarity could motivate owners to employ same-kind individuals. In certain industries the use of a common language may be important for productive e?ciency. Third, employer tastes might bias them to employ workers
In this Chapter, the expressions ?rm type and owner type are used to explain that ?rm’s owners correspond to one of the following groups: native-only, immigrant-only, and mix owned ?rms.
1
9
of a similar kind. Employer discrimination could generate scope for segregation.2 However, coworker e?ects could compensate for the presence of employer discrimination. In fact, for all types of owners the share of similar coworkers increases the probability of being hired in the ?rm. We also control for speci?c characteristics in the ?rm, such as the fraction of English speakers, to identify the possible scope for matching productivity. This paper focuses on the importance of social ties in the process of recruitment when ?rms use current employees’ social connections to help ?nd and identify new candidates. However, employers may use this mechanism di?erently for di?erent worker types, depending on their ability to take advantage of their workers’ connections. For instance, given their cultural, linguistic, and social backgrounds, immigrant employers have an advantage, compared to natives, in exploiting their immigrant workers’ social connections. Our results suggest that immigrant owners are three percentage points more likely than native owners to hire other immigrants, even after controlling for industry, ?rm size, geographic concentration of immigrants in the population, population density, and the legal form of organization of the ?rm. Looking at ethnic/race groups, immigrant owners (Hispanic/Asian owned ?rms) are 3 to 4 percentage points more likely than native owners (white and black owned ?rms) to hire Asians and Hispanics versus blacks and whites. Both, native and immigrant owners, hire white non-Hispanic workers, but native owners have a higher probability of having white workers as new hires. These results are based on both linear probability models and a multinomial logit speci?cation that accounts for the simultaneity of choosing from
2
Lang [1986]
10
di?erent types of workers. Among our strongest ?ndings are the existence of a persistent pattern of hiring similar types and the smaller e?ect of the share of dissimilar coworkers on the likelihood of hiring a particular individual. For instance, the share of similar coworkers at the time of recruitment increases the probability of hiring a worker of a type by around 60%. The probability is higher when the owner is similar to the new hired. Additionally, this probability is di?erent whether the employer is immigrant versus native. Immigrant businesses show higher chances of hiring a new immigrant, Hispanic or Asian compared to native businesses, even after looking whether they have similar workforce distribution at the time of a new recruitment. To study the wages of employees, one must understand the role of employers in wage-setting, which necessitates gathering wage data by employer and having detailed information about the employer. Immigrant workers tend to have lower average wages than native workers. Many authors have used a human capital approach to explain that wage gap and have found that skill accounts for almost two thirds of the wage di?erence between Hispanics and white Non-Hispanics.3 Meanwhile, the residual unexplained wage gap has traditionally been used to claim the existence of racial/ethnic discrimination in the labor market. Other authors have found that industry wage-di?erentials are to a very large extent explained by the characteristics of workers and the contribution of industry to wage setting is much smaller after looking at both person and that industry e?ects.4 However, these studies don’t rule
3 4
Borjas [1994], Trejo [1997], Chiswick [1978], Borjas [2003] among others. Abowd et al. [1999]
11
out a signi?cant impact of ?rm-level e?ects on wage formation.5 The results in this paper suggest that much of the di?erence between the log annual wages of immigrants and natives comes from immigrants’ propensity to work in non-native owned ?rms, which pay the lowest average log annual wages. Interestingly, though, native workers holding a job in immigrant ?rms are paid less than immigrant workers. After controlling for typical human capital variables, full-time immigrant workers earn about 8% less than native workers ($3,293 less each year). When working for native employers this di?erence increases to 11%. Meanwhile, immigrant workers earn 10% more than native workers in immigrant owned ?rms ($4,398 more each year). Recent work has used the idea of networks in the labor market to explain labor market inequalities as a function of di?erential social capital (social resources, network structures, network resources). Minority individuals are generally connected to other minority-group workers who cannot provide them with the opportunity to change their employment outcomes. Hispanics and blacks are disadvantaged because they are likely to match with same-kind job contacts, and end up working in lower wage workplaces where other Hispanics and blacks work (Elliot [2001]). To the best of our knowledge, no previous study has analyzed the link between employer and coworkers’ birthplaces and employees’ employment opportunities and wages in a large set of industries and geographic locations. This research provides initial steps on that branch of analysis. These ?ndings suggest that social connections and social capital may be important for understanding employment opportunities
These authors obtained that the average of the di?erence in wages paid to an identical worker employed at two di?erent ?rms in France was 20%-30%.
5
12
and wage di?erentials. The remainder of the paper is organized as follows. Section 2.2 and section 2.3 review previous work on the relation between workers and types of ?rms, ethnic economies and ’ethnic matching’ between supervisors and employees, the usage of networks, and network e?ects on hiring procedures and workers’ wages. It also discusses the importance of analyzing small businesses when looking at the impact of immigration. Section 2.4 examines the data and presents basic descriptive statistics on owners’ and workers’ characteristics. Next, section 2.5 presents preliminary information on workers’ average earnings by worker type and by di?erent levels of coworker shares. Section 2.6 is divided in two sections. The ?rst part analyzes whether the type of employer and coworker characteristics a?ect the composition of new hires in ?rms. The second part evaluates the impact of ?rm owner type on employees’ log annual earnings controlling for worker human capital. Section 2.7 concludes.
2.2 Literature Review
Because no single theory exists to explain the e?ect of ?rm owners and coworkers on hiring patterns and wages, we draw on the literature of several related ?elds to motivate our hypotheses on the subject. Those literatures include ethnic economy theories dealing with ethnic/immigrant concentration, theories of ?rm wage di?erentials and hiring procedures, and network theories. Immigrants tend to work in low-wage/low-productivity ?rms, low-pay occupa-
13
tions, and in ?rms with a high percentage of immigrant workers.6 Some researchers have found occupational and ethnic coworker concentration in the United States (Andersson et al. [2007], Patel and Vella [2007], and Light [2006]) and in other countries (Barr and Oduro [2000] and Andersson and Wadensj´ o [2001]). The literature has attempted to explain workers’ concentration by skill, race, and sex.7 Hellerstein and Neumark [2007] analyzed ethnic segregation in the United States and found a substantial degree of segregation in the workplace. They claim that even though workplace segregation partially results from residential segregation (spatial mismatch explanation) and from ethnically correlated skills, there seem to be other mechanisms that suggest the presence of immigrant social connection e?ects (local residential networking). In an extensive analysis of racial and ethnic segregation across U.S. workplaces, they found that a large degree of segregation remains even after controlling for metropolitan area characteristics, and that very little of this segregation can be explained by observed di?erences in education and occupations. Language, however, seems to be a signi?cant factor for immigrant segregation. Lang [1986]’s theory provides an explanation for worker segregation by language groups. When there are transaction costs associated with employees of di?erent language groups working together, there is scope for segregation. Employers of each language group have incentives to fully segregate to avoid the cost of needing employees who can be the bridge between di?erent language groups. Despite ?ndings on immigrant concentration at di?erent levels, we cannot be
6 7
Borjas [1994], Borjas [2003], Andersson et al. [2007], and Andersson et al. [2008]. Kremer and Maskin [1996], Hellerstein and Neumark [2003].
14
sure that immigrants are more likely to work for immigrant bosses and that such a pattern would a?ect individuals’ labor market outcomes. There is no evidence that immigrant-owned businesses are distributed (or concentrated on) di?erently across speci?c industries, ?rm sizes, or skills, than native businesses, and that this distribution is correlated with the distribution of immigrant workers across industries, sizes, and skills. A recent group of studies analyzes the matching process between managers and workers by racial group. Giuliano et al. [2006] found a signi?cant e?ect of race and ethnicity on hiring procedures. For example, in locations with large Hispanic populations, Hispanic managers tend to hire more Hispanics and fewer whites than white non-Hispanic managers. In a more recent analysis, Giuliano and Ransom [2008] looks at the e?ect of manager ethnicity on hires, separations and promotions across di?erent occupations in a U.S. retail ?rm. Whites were more likely to leave stores where managers were Hispanics than when they were white. Their work is very relevant, although they only focus on a very particular retail ?rm. Their studies do not consider the coworker e?ect. That is, they don’t study the e?ect of the fraction of similar coworkers holding a job in the ?rm on the probability a particular type of worker is hired. There has not yet been a connection established between owner’s birthplace and the type of workers employed at a ?rm or these workers’ earnings. Nevertheless, the literature discusses motivations for supervisor-employee matching. First, ?rm owners could have preferences for employing individuals of their own type or with the same background. Second, the types of goods o?ered by immigrant ?rms may 15
di?er from those o?ered by native ?rms. If immigrants specialize in producing ethnic goods, immigrant workers have a comparative advantage over native workers in these ?rms. The di?erences between products can result in di?erent worker composition.8 However, none of these reasons have obvious predictions of workers’ earnings. That an employer has a preference for a certain group does not necessarily imply higher wages for that group. The distribution of workers and employers in the market also a?ects the labor market equilibrium. In the sociology literature, there have been a limited number of studies that provide some insights on the tendency of immigrants to work for immigrant ?rms. For instance, in Los Angeles in 1989 30 percent of employed Koreans held jobs in ?rms owned by fellow Koreans even though Koreans composed only one percent of the Los Angeles County population.9 According to Cardenas and Hansen [1988], during the 1980s, Mexican immigrant employers were most likely to hire Mexicans, whether legal or undocumented, and to evaluate their quality favorably. Porter and Wilson [1980] ?nd two relevant patterns in the Cuban immigration to Miami during the 1960s. First, Cubans worked with other Cubans. Second, almost one-third of the Cubans worked for Cuban employers. The phenomenon of immigrants hiring immigrants is not limited to coethnic relationships between employees and employers. Other researchers have found that employers from one immigrant group often hire workers from other ethno/racial groups.10 In Los Angeles, during the nineties, 51% of the garment factories were owned by Asians with most of their employees being
Andersson and Wadensj´ o [2001] Min [1989]. 10 Massey [1999], Massey et al. [1987].
9 8
16
Hispanics. Ethnic networks alone cannot expand the supply of coethnic-accessible jobs. Generally, the number of jobs o?ererd by ethnic-speci?c owned ?rms is not equal to the number of possible candidates from the same ethnic group in the local community. Business leaders from ethnic groups whose rates of entrepreneurship are higher than other groups ?nd it di?cult to limit hiring to members of their own groups. Ethnic crossover can expand the economic opportunities provided by immigrant-owned businesses. Immigrant workers often join networks that cross ethnic boundaries. Using the Garment Industry in Los Angeles as an example, Light [2006] analyzes immigrant ownership economies consisting of immigrant employers plus their immigrant but not coethnic employees. He ?nds that this type of economy explains part of the garment industry’s growth during early 1990s in Los Angeles. The cited studies have been limited to small samples from particular geographic areas and speci?c groups of ?rms and immigrants. Most of them also focus on a particular period of time, with a cross-sectional view of the distribution of workers and ?rms. These analyses tended not to look beyond the segregation aspect to analyze the possible causes and consequences of those patterns. Unlike previous studies, this paper uses a representative group of areas, ?rms, industries and workers, and it analyzes the ?ow of hiring and the e?ect of employer-employee type matches on wages. The underlying hypothesis in the analysis is that workers and employers make di?erent use of their social connections in the market, given their speci?c characteristics, such as race/ethnicity and immigration status, which leads to a particular hiring pattern by each ?rm. Immigrant ?rms, for instance, would have an advantage over native ?rms when using their immigrant current workers as 17
a channel to ?nd new workers. On wage e?ects, previous research has suggested that much of the unexplained variation in wages among employees is linked to characteristics of their ?rms, such as size and industry.11 Not only do individual characteristics explain wage di?erentials between immigrants and natives, but potentially so do other characteristics, such as the birthplace or ethnicity of employers and coworkers. Unfortunately, most wage databases come from household surveys of individuals (Decennial Census and CPS), rather than from establishment surveys of wage-paying employers; they provide little employer-speci?c information, except for industry and, in some cases, ?rm size.
2.3 On the use of social networks
Recent work has suggested that supervisor-employee ethnic matching could result from the use of networks.12 On the one hand, according to several sociological studies on the ethnic economy, ethnic solidarity serves to provide entrepreneurs with privileged access to immigrant labor and to legitimize paternalistic work arrangements (Sanders and Nee [1987] and Model [1997]). Di?erent ?rms have di?erent recruitment processes, generating an initial sorting of worker types. On the other hand, networks can also have an impact on wages, providing better matches and more opportunities to the individual. Ethnic networks can generate informal sources
[Groshen, 1990, 1991a,b], Abowd et al. [1999], Abowd et al. [2004] among others. Networks is not a new concept in the literature. For an extensive analysis on job information networks see Ioannides and Loury [2004]. Sociologists have investigated the origins and creation of social networks for more than 40 years. Rees[1966] draws attention to di?erences among workers and their use of available information (formal and informal sources). Job referral is also extensively used in the labor market, as well as family networks (Granovetter [1995]).
12 11
18
of capital formation and captive markets, making these ?rms more self-su?cient and ?exible (Volery [2005]). Social capital becomes another form of capital resource.13 Individual’s social networks are likely to have an impact on labor market outcomes (Simon and Warner [1992]). The di?erential use of social networks does not provide the same access to information and opportunities to all individuals, o?ering a better relative position to those agents with better social connections or better use of their social networks. Recent literature has moved away from spatial mismatch model in explaining inequality across ethnic/race groups towards theories that include how social networks a?ect urban inequality [Hellerstein and Neumark, 2007, Hellerstein et al., 2008a]. Life-chances depend not only on individual resources but also on network characteristics re?ecting the resources of network members. In this context, personal networks are then considered an additional determinant of inequalities (Light [2006]). How do these mechanisms a?ect our groups of analysis? What is di?erent about particular types of workers and ?rms such as immigrant/racial groups? Although the comparison between whites and blacks has been long discussed, immigrant status can be crucial for understanding group di?erences in informal job matching and labor outcomes. Two important characteristics of the immigrant community are relevant for these implications. First, Borjas [1994] pointed out that immigrants tend to be less educated, to have poor English language skills, and to lack domestic experience. Second, immigrants rely heavily on social networks for
Social capital in its simplest form is a social network of strong and weak social ties (Light and Gold [2000]).
13
19
?nding jobs and geographically reallocate (Massey et al. [1987]). Previous literature has also discussed racial and ethnic di?erences in informal job matching (Elliot [2001], Holzer [1987]). These di?erences arise because informal channels permit race and other characteristics in the network to play a more prominent role in the hiring process than it does when formal mechanisms are used. As noted by Elliot [2001], one of the puzzles during 1980s and 1990s was the worsening position of less educated blacks in the labor market while the economy was absorbing thousands of new immigrant workers. Surprisingly, these new workers had, on average, similar characteristics to blacks: low formal education and high geographic segregation. So the question of job distribution became a ?rst order issue, especially in the topics of immigration and immigrant assimilation. Research on this puzzle has focused on the use of social networks by di?erent groups for ?nding employment [Waldinger, 1997], while the role of prospective employers in the use of these mechanisms has been ignored. Our empirical analyses sheds light on the impact of networks on immigrants. Considering the tendency of workers to refer their own, the immediate e?ect of network is the reproduction of the workforce composition across time as shown in this chapter. Our results in the following chapter support the hypothesis that social networks play an important role in workplace concentration. The tendency of social networks to be racially/ethnically homogeneous - exacerbated by individual’s immigration status- increases the probability that workers would refer same-type candidates and that same-type employers would tend to hire from shame-type groups. Immigrant employers can take better advantage of their immigrant employees in 20
hiring than native employers. The di?erential use of job referrals by employers is also evident when we examine who is hired and how the wages are distributed in the ?rm. Immigrants will tend to be hired more by immigrant ?rms with a high share of immigrant workers than by native ?rms with high share of immigrant workers. Immigrant entrepreneurs can take advantage of their language, cultural background and a?nities to have access to di?erent ethnic groups. Their immigrant status can give them privileged access to sources of labor less available to native entrepreneurs. Immigrant entrepreneurs routinely employ coethnics (including relatives) at rates vastly above chance levels. The most important network relationships are based on kinship, friendship, and paisanaje (the feeling of belonging to a common community of origin).14 Immigrant economies rely upon networks to locate jobs. On the one hand, referrals by friends or coworkers remove some of the uncertainty associated with ?nding a job with unfamiliar employers and increase the chance of ?nding a better job match. On the other hand, immigrant entrepreneurs tend to rely on their current employees to help ?ll their vacancies. Workers tend to refer individuals that are ’similar’ to them, from the same group, or with the same characteristics. Referral coworkers could also provide informal training, show the new worker how to perform the job, and have a good interaction with the new hire. Moreover, referral coworkers indirectly accept responsibility for new hires. Employers realize that this practice is bene?cial for them as well. Little cost or e?ort need be expended when new workers are located through employee contacts.
14
Massey[1980].
21
Previous empirical ?ndings show that Hispanic men report more frequent use of friends and relatives for job search than non-Hispanic whites, and are also signi?cantly more likely to have obtained their most recent job through personal contacts. Hispanics use informal contacts 32.8 percent more often than white non-Hispanics and blacks.15 Recent Latino immigrants are more likely than blacks or Latino natives to use personal contacts to ?nd jobs.16 Weak English skills explain much of this di?erence. However, this di?erence comes not only from the use of job networks by workers, but also from a greater reliance on referrals in small workplaces in combination with a concentration of recent immigrants in small ?rms. Employers also have a role in this process given that ?rms’ hiring procedures will a?ect individuals’ likelihood of receiving o?ers from jobs heard about through friends and relatives.
2.3.1 Small ?rms
Our focus on small/medium ?rms17 is motivated by two observations. First, in larger ?rms, the separation between ownership and management could detach the ?rm’s hiring process from owner characteristics. As Haltiwanger [2006] points out, however, in small ?rms the decision process is likely dependent on owner ability and characteristics. When dealing with each worker, small ?rm owners could project their tastes and managerial abilities onto the hiring and production processes of the ?rm. Since it is usually the business owner who makes such choices, the identi?cation of the person responsible for hiring decisions is easier and more relevant for small
Holzer[1987b], Smith [2000]. (Elliot [2001]). 17 We consider small/medium ?rms those with less than 500 employees.
16 15
22
?rms. Second, immigrant workers are more likely than natives to work in small ?rms. In Chapter 3 we ?nd that there is a signi?cant market segmentation that appears in any detailed distribution of workers in ?rms. Immigrants are more likely to be employed in ?rms with less than 10 employees 70% of immigrants work for small ?rms. Meanwhile, more than 60% of native workers are employed at ?rms with more than 100 employees. The labor force changes generated by immigration in?ows are thus borne primarily by smaller, younger ?rms. These ?rms are more sensitive to immigration shocks. If we only look at aggregate numbers (including small and big ?rms), immigration e?ects will be obscured.
2.4 Data and Measures 2.4.1 Sources
In this paper, we use three di?erent databases to match owners’ characteristics to workers’ characteristics. First, we use the Characteristics of Business Owners Survey (CBO) from 1992, and then match this survey with administrative data from the IRS (Business Register) for the years 1992 to 1996. To obtain workers characteristics, we use information from the Longitudinal Household-Employer Dynamics (LEHD) database for the years 1992 to 1996. In this section, we give a brief description of each database and their limitations, and discuss how we construct relevant variables used in the regressions. The Characteristics of Business Owners (CBO), later renamed the Small Busi23
ness Owner (SBO)database, is produced by the Bureau of the Census. The 1992 release of CBO was the ?nal version of this survey, which formerly was conducted every ?ve years. The survey for the 1992 CBO’s release was conducted in 1996, along with the economic census. Therefore, the questions in the survey refer to the business’ and owners’ information for years 1992 and 1994. The CBO is a supplement to the Survey of Minority-Owned Business Enterprises (SMOBE) and Survey of Women-Owned Businesses (WOB). The survey universe considered was ‘’any business which ?les an IRS form 1040, Schedule C (individual proprietors or self-employed persons); form 1065 (partnership); or form 1120S(Subchapter S corporation) in 1992.”18 It considers as business owners those who ?led business tax forms as owners of the ?rm, excluding non-S corporations, with at least 500 dollars in yearly business receipts, and with the largest employment size category equal to ?ve hundred. Note that non-S corporations generally have investors, not decisionmaking owners, and thus this group is not in the CBO survey’s universe. However, excluding non-S corporations often excludes the largest employers, making comparisons of small and large business owners di?cult. The CBO provides details about both business owners and their businesses. The unique ?rm identi?er is the CFN (Census File Number). At the cross-sectional level this number is unique for each ?rm. According to a CBO publication cited in of the Census [1997], almost 62% of the 78,147 ?rms’ surveys
18
19
and 59% of the 116,589 owners’ surveys were returned.
Characteristics of Business Owners 1992:CBO092-1. U.S. Bureau of the Census (September 1997) and Headd [1999]. 19 This is translated into 63% of the 41,297 employer ?rm surveys.
24
One possible reason for this low rate of reply is the di?culty of ?nding owners of exiting ?rms after 3-4 years. Almost 70% of all businesses present in 1992 were still in operation in 1996. This rate is lower for minority-owned ?rms (around 66%). We use employer ?rms in our sample. When sampling weights are used, the survey indicates that in 1992, 20% of owners were in ?rms with employees. According to the minority-?rm surveys, women, Asian, Paci?c Islander, American Indian, black, and Hispanic owners were typically underrepresented in the larger employment size classes. Hispanic-owned ?rms were 3.68% of all employer ?rms, but just 2.04% of ?rms with 100 or more employees. Additionally, 90.6% of business owners were born in the United States, while 9.4% percent were foreign born.
20
The per-
centage of native-owned ?rms was higher in the case of larger ?rms (94.5%). In this paper we focus only on employer ?rms. On average, there exists more than one owner per ?rm. In the CBO(1992), more than 52% of ?rms are employer ?rms, and almost 41% of this group have only one owner. Employer ?rms tend to have more owners than non-employer ?rms. Based on previous research using the CBO,
21
we consider the CBO as a sam-
ple of ?rms even though it is essentially a sample of ?rm owners. The resulting complication is that we need to make assumptions to identify the owner characteristics for multiple-owner ?rms. As a ?rst attempt, we consider three types of ?rms: only-native-owned, only-immigrant-owned, and mix-owned. Using this classi?cation, more than 85% of employer ?rms have 1 or 2 owners for all types.
A foreign born is an individual that was born outside the USA. CBO has a particular question on whether the owner was born in the US or abroad. 21 Carrington and Troske [1995].
20
25
In order to identify the characteristics of the owners of a particular ?rm (particularly immigration status and race), we follow the work of previous research based on the CBO (Carrington and Troske [1996]). For single-owner ?rms, the identi?cation is straightforward. Meanwhile, for multi-owner ?rms the mode is used. The number of hours per week spent at the business was used to break ties. This database has some limitations. First, in the 1992 survey the CBO’s sample universe omits chapter C corporations. This group of corporations corresponds to bigger businesses; therefore, comparison between small and large businesses in the CBO must be done with care. Second, even though we have each ?rm’s average payroll, we know nothing about the inter?rm distribution of payroll between di?erent types of workers. Third, this survey has zero information on human capital or occupational characteristics of workers. We try to overcome some of these limitations by merging CBO with data from Bureau of Labor statistics (UI and ES202) as described below. The second database used in this paper is the Census Bureau’s Standard Statistical Establishment List (SSEL) or Business Register (BR).22 This data has more complete information on ?rms given that the source of the SSEL is at the administrative level. This database works as a register of active employer business
Walker [1997] has an extensive discussion on the Census Bureau’s Business Register. The initial source of information on businesses is the IRS(Parker and Spletzer [2000]). The SSEL receives three main ?les from IRS; the Business Master File (BMF), with information on name, addresses and legal form of organization; the Payroll Tax Return File (Form 941) containing quarterly payroll and ?rst quarter employment (including March 12th employment); and the Annual Business Income Tax Return Files with information on receipts/revenues, industry classi?cation. For all three sources, EIN is the primary business’ id.
22
26
establishments23 in the United States and its territories. The unit of information is an enterprise, which can be associated with one or more establishments and with one or more EIN entities (Employer Identi?cation Number).24 In this paper we concentrate on those businesses organizations associated with only one EIN and one establishment, known as single-establishment enterprises or single-unit ?rms.25 All of the small ?rms in this chapter correspond to single-unit establishments. The assumption that ?rm owners are the ones making the main contracting decisions in a ?rm is more plausible in ?rms with only one establishment than otherwise. In the case of younger and smaller ?rms, this restriction does not exclude many ?rms.26 Additionally, businesses have a CFN (Census File Number) as an identi?er, which is unique for single-unit businesses. To follow the ?rm across time, the longitudinal identi?er for each ?rm is called alpha, and corresponds to the ?rst 6 digits of ?rms’ EIN. In the sample, we only follow ?rms that survived the entire period 1992 to 1996. Because most non-surviving ?rms did not respond to the CBO survey and the weights are constructed such that this pattern is considered, the weighted results are not impacted by this exclusion.27 We take data on industry, legal form of organization and employment from the SSEL ?les. See Appendix B for speci?c description of these variables. Because of the time di?erence between the year of information and the year in which the CBO
Active employer business establishments are those with payroll at anytime during the past three years, or with an indication that the business expects to hire employees in the future. 24 An EIN entity is an administrative unit assigned by IRS for tax purpose. Under the Federal Insurance Contributions Act (FICA) every organization with paid employees has to obtain an EIN. 25 All the matches between CBO(1992) and SSEL(1992) are in this category. 26 Haltiwanger et al. [2005]. 27 Headd [1999].
23
27
survey was conducted, information on employment and sales are from the SSEL dataset.28 We use the common unique ?rm identi?er (CFN) to match CBO with SSEL.29 We then follow the ?rm across time until 1996.30 The second set of information is associated with the characteristics of workers. This information comes from the Longitudinal Employer-Household Dynamics database. Information on workers comes from the Unemployment Insurance wage records for a group of states31 and the ES202 data. Based on availability, we use data from eight states for the years 1992 to 1996. The sample includes states with high immigrant concentration and low immigrant concentration areas. These ?les contain person identi?ers that allow researchers to track a worker’s quarterly earnings within a State across years. We sum over quarters to obtain each worker’s annual earnings. This database also contains ?rm identi?ers that allow for an exact link between the UI ?les and other data sets. The business level identi?ers in UI ?les are State Employer Identi?cation Numbers (SEINs). Therefore, one can match the UI data with the ES202 data, using SEIN to get information on the EIN, and compare it with the data previously matched using CBO(1992) and Business Register. For single-unit ?rms, the units of observation at the ?rm level used for CBO, SSEL and LEHD are generally similar.
The CBO is a retrospective survey. The response rate is a?ected by the survival rate of the ?rm and the extent to which owners can accurately recall past information. 29 We use businesses’ CFN, which are the Census Bureau’s preferred intra-year, cross-dataset link. The CFN contains the EIN ?rm identi?er and is unique for single-unit ?rms. 30 To illustrate the groups of ?rms included in both databases, we include a short discussion on ?rms matching rate in the Appendix A. 31 More detailed analysis on these records is presented in Abowd et al. [2006], and additional information on date of birth, place of birth, and gender are obtained for almost all workers in the sample after linking UI wage records to Census data. 98% of all private, non-agricultural employment is covered by the employer reports.
28
28
The UI wage records contain virtually all business employment for the sample states (for private non-farm ?rms). Earnings reports from these records are more accurate than survey-based earnings data, and one can obtain information for each worker in a speci?c ?rm (or establishment). Using this database, we follow ?rms across time from 1992 to 1996 using the unique identi?er within the state. We end up using only those ?rms that survived the entire period and did not change ownership. This group represents about 67% of the initial set of ?rms in 1992.32 Finally, the data set used in this study is unique in the sense that it contains data from each ?rm on output and inputs used in the production process, as well as data on earnings and some demographic characteristics of each worker in the ?rm. We use the years 1992 to 1996 for the analysis mainly because information about owners’ place of birth (i.e. being born in or outside the US) is only available in the Characteristics of Business Owners Survey in 1992. Our data tracks the total payroll and workforce composition of each ?rm from 1992 to 1996. The drawback of using UI data is its lack of certain demographic information on workers, such as education and occupation. However, the sta? at the LEHD has overcome this limitation by imputing education using administrative data from the Census Bureau containing information such as date of birth, place of birth, geographic area, industry, and sex. In this chapter, we use this imputed information on education,33 which has been used in previous work on the LEHD. This variable is
Few ?rms were dropped because, initially, the survey’s rate of response was highly correlated with the ?rms survival rate, so that most of the ?rms with information in the survey are surviving businesses. 33 See Lengermann et al. [2004] for details on the imputation.
32
29
a proxy for individuals’ human capital. We are aware that the lack of occupational information could be a relevant drawback of the data given that prior research has documented an important role for occupational segregation in creating di?erent workers’ wage gaps. We might think that immigrants tend to concentrate in lowskilled occupations relative to natives. However, as Troske [1999] and Carrington and Troske [1995] point out, occupations and job titles are less likely to be sharply de?ned in small ?rms, and as a result there could be less occupational segregation in small ?rms compared to large ?rms. Despite this limitation, we have to keep in mind that we can account for other workers’ characteristics, such as age, sex and imputed education. Given that workers have varying preferences for place of work depending on the disutility of commuting and amenities of particular areas, the areas where they would be willing to work are better represented by their actual place of work than their place of residence. Therefore, we need data on individuals’ place of work. Location of the ?rm is obtained using the LEHD database.
2.4.2 Construction of ex post weights
A relevant technical issue that arises in the process of using di?erent databases, especially when a survey is included, is the change of sample frame used by the survey database. Additionally, for smaller geographic areas, di?erences in industry and geographic information along with di?erences in the scope of industries covered lead to dissimilarities between the universe considered by the LEHD data and surveys
30
based on the Economic Census.34 In the design of the CBO survey, four panels were created in addition to divisions by employer status (employer versus non-employer), 2-digit industry and state. These panels consider racial categories using the owners’ social security information and the categories: Asian, Asian-American / Paci?c Islander, Hispanic, Black, and White. These groups were created by the Survey on Minority Businesses. Therefore, small ?rms and minority-owned ?rms are over-represented in the survey. The di?erence between the universe and sampling frames used in the CBO survey implies that our matched analysis sample will not be representative. Specifically, the sample frame used in the CBO will over-represent small, minority-owned businesses when linked with the UI database. To deal with this issue, we follow Abowd et al. [2007] and build ex post weights that control for the ?rms’ size, 2-digit industry code, legal form of organization, and employer status. We follow previous research in that we ?rst construct the fractions of ?rms each the category in the universe of ES-202. The universe of ES-202 is single-unit ?rms with more than one employee (coworker shares can be computed only for these ?rms), not in Agriculture, Mining, nor Public Administration, and less than one thousand employees, and are in Economic Census in-scope industries in 1992. This represents the numerator in the ex post weight. Then, we compute the same fractions for the ?nal matched data and use each fraction as the denominator of the ex post weight. This weight has the property that the distribution of employment by each category re?ects the size
LEHD database covers partially agriculture and public administration industries. Surveys based on the Economic Census tend to over-represent businesses in areas with high density population.
34
31
distribution of the ES-202 considered universe. The second section of the adjustment procedure involves the construction of an inverse Mills ratio. We use a probit estimation that considers the probability of being matched as a function of log employment, legal form of organization, owner’s place of birth (in or out the US), and log of sales per employee to generate the propensity scores. This section intends to account for the CBO survey’s sampling frame and the possible selection bias generated by the e?ect of unobservables on ?rms exiting from the universe considered to design the sample of the CBO survey. The ex post weights are included in all regressions. For more details and unweighted summary statistics see appendix D. Before using our approach the matched sample under-represent small, minorityowned businesses (see appendix D.1). After the match, and without considering the re-weighting process, we would be under-representing minority groups in small size ?rms. The sample of ?rms o?ering unemployment bene?ts are relatively of bigger size. After applying our new weight, we try to recover some of the original distribution in the CBO sample. There is a lower representation of sole proprietorship after matching the original sample with the UI database without using the new weights.
2.4.3 Firms
To compare the full CBO sample to the ?nal matched sample used in the analysis, we look at descriptive statistics for a set of variables. The ?nal match uses LEHD information from 8 states,35 which include high and low immigration
35
Those states with available data in 1992 are included.
32
states. For these states we obtain workers’ and ?rms’ information. Firms from the agriculture, mining and public administration sectors are not included. Additionally, only single-unit businesses are considered. The original matched sample in the analysis has 7,200 ?rms, representing 339,040 workers from 1992 to 1996. All results are weighted by the adjusted-weight discussed in section 2.4.2. Table 2.1 shows two blocks of summary statistics. One block (CBO-SSEL) contains the employer ?rms matched from the CBO survey and the BR, while the second block (Sample(CBO-LEHD)) contains the ?nal matched sample, consisting of the subset of CBO-SSEL data matched to the LEHD. For each block, this table presents the distribution of ?rm type across ?rm size categories and sectors, together with the average number of owners, average share of immigrant workers, de-meaned average log sales per employee, average percentage of immigrants in the county in which the ?rm is located and in the counties surrounding this location, and the percentage of each type of owner. Total population and the share of immigrant workers are constructed from the public 1990 Census, and are based on all Census counties surrounding the location of the ?rm. Immigrant ?rms have a higher proportion of immigrants in the local population than native and mixed ?rms. Because immigrants also tend to be geographically segregated, we will use this variable to control for di?erences in ?rms’ local workforce. In the ?nal matched sample, the average immigrant-owned ?rm employs 38% immigrant workers. The distribution of ?rms across sectors and sizes for each type of ?rm by owner birthplace is very similar, except for the tendency of immigrantowned ?rms to be in retail or services, and this distribution is only slightly changed 33
after matching the original database with the LEHD database. From the table we observe that immigrant-owned ?rms’ log sales per employee is slightly higher than native-owned ?rms. Actually, on average, native owned ?rms have the lowest log labor productivity. In general, ?rms are concentrated in size categories with fewer than 50 employees. Meanwhile, regardless their owner type, ?rms are highly concentrated in the sectors Services, Retail, Manufacturing and Construction. Sole proprietorships represent more than 50% of immigrant and native ?rms. Mixed-owned ?rms tend to be larger in size with respect to the other groups. These ?rms are mainly Partnerships and Corporations.
34
Table 2.1: Descriptive Statistics - CBO(1992) and Sample/Matched Firms CBO1 Imm Nat 49.30 20.90 14.68 10.70 2.97 1.46 5.07 10.35 2.83 19.73 29.68 3.19 29.16 42.51 21.35 16.69 12.10 4.56 2.79 12.74 13.89 7.43 17.14 19.73 6.18 22.89 Matched Sample(CBO-LEHD) Mix Imm Nat Unk ALL 16.67 18.75 18.75 28.13 9.90 7.81 5.73 25.00 5.21 18.75 14.58 7.81 22.92 37.27 21.53 18.44 14.25 5.67 2.84 4.71 15.93 2.77 22.18 29.85 3.55 21.02 33.60 20.97 18.02 16.72 6.65 4.03 13.49 17.65 7.58 19.03 16.42 5.30 20.54 33.29 20.33 17.40 16.89 6.78 5.31 9.38 18.36 6.91 22.24 21.34 4.44 17.33 37.18 13.77 49.35 11.63 (1.18) 15.07 81.71 1.83 33.79 20.79 17.88 16.58 6.59 4.37 10.06 17.76 6.34 20.84 20.83 4.70 19.49 50.64 12.99 36.37 11.64 (1.13) 12.80 92.85 1.87
Distribution/Type of ?rm Size (%) 2-4 5-9 10-19 20-49 50-99 100+ Sector (%) Construction Manufacturing Transp. & Utility FIRE Retail Wholesale Services Legal Form (%) Sole Proprietorship Partnership Corporation 2 l(sales/employment) 3 Imm. in the neighborhood4 In MSA Average Number of Owners Continued on next page.
Mix 26.52 18.06 23.48 18.18 6.94 6.82 6.49 20.26 5.45 19.22 17.14 6.10 25.32
Unk 46.43 20.91 15.00 10.95 4.26 2.45 10.41 13.59 6.73 19.24 23.17 5.21 21.65
ALL 44.66 21.05 15.90 11.59 4.26 2.54 10.58 13.38 6.43 18.36 22.47 5.36 23.43
35
- 52.40 51.43 29.15 28.51 12.58 12.26 18.69 71.94 35.10 25.01 52.16 11.64 11.60 11.48 11.54 (1.17) (1.17) (1.05) (1.15) 13.47 22.15 12.43 15.06 92.01 96.50 90.07 80.60 3.58 1.57 1.88 1.80
50.48 - 49.12 56.81 12.97 25.00 12.26 10.46 37.36 75.94 38.61 32.72 11.54 11.73 11.59 11.57 (1.11) (1.08) (1.19) (1.05) 12.90 14.03 21.17 11.34 92.41 91.30 97.37 93.01 1.84 3.78 1.55 1.95
Table 2.1: Descriptive Statistics - CBO(1992) and Sample/Matched Firms (continued) CBO1 Nat . 45.87 78.60 Matched Sample(CBO-LEHD) Mix Imm Nat Unk ALL 33.00 38.05 11.55 28.00 26.00 2.84 18.10 42.54 36.53 100.00 2.84 12.89 77.92 6.40 100.00 10.09 15.50 0.79 73.62 19.20 51.70 0.69 28.41 3.41 2.55 1.48 92.56 7,985 339,040 4.15 3.02 1.20 82.30 5.66 9.41 1.36 83.57
Distribution/Type of ?rm Average Share of imm. Workers Unweighted dist. of ?rms Weighted dist. of ?rms RACE/ETHNICITY Hispanic Asian Black White # of Firms unweighted # of Observations unweighted 36
Mix . 2.03 2.01 10.39 10.60 0.78 78.23
Imm . 14.94 13.50 18.56 35.68 1.56 44.20
Unk ALL . . 37.16 100.00 5.89 100.00 4.69 5.07 2.16 88.08
2.99 4.35 1.34 5.20 2.22 2.60 93.45 87.85 41,297 1,655,750
Note: Statistics based on weighted outcomes unless the contrary is indicated. (1) Single-unit ?rms that matched with SSEL. (2)Only S- Corporation . (3)Source SSEL: Sales (total receipts/sales), and employment (Employment March12th). Numbers in parenthesis are standard deviations. (4)Using Census 1990, computed percentage of immigrant population in the counties including the ?rm and surrounding.
The average number of owners (owner type) is similar in the original and matched samples. The average number of owners by owner birthplace is similar, except, as expected, for mix-owned ?rms which by de?nition have two or more owners. Table 2.1 illustrates that these patterns are similar in the original CBO sample and the ?nal matched CBO-LEHD sample. In the matched sample, Asian-owned ?rms are over-represented, while white immigrant owners are underrepresented. However, as in the original CBO sample, immigrant-owned ?rms are mainly owned by Hispanics and Asians, while most of the native-owned ?rms have white owners. In the original matched data there is a percentage of ?rms with unknown owners’ place of birth. We decide to exclude this group from further analysis. Given that, on average, the characteristics of this unknown group are similar to the rest of the sample (see Appendix(C) for t-tests and a chi-square analysis), we don’t expect this exclusion to a?ect our ?ndings. We drop ?rms with less than two employees. Given that female labor participation is characterize for additional elements di?erent to the ones analyze here we only consider male workers.36 Workers should have at least one coworker, and the analysis of earnings is net of other labor supply factors that could a?ect female workers di?erently. After these restrictions, the ?nal sample is reduced to 4,478 ?rms and 214,398 workers from 1992 to 1996.
The e?ect of networks for female immigrants is also a very important analysis. According to Massey et al. [1987], Mexican female immigrants tended to arrive and go directly to speci?c industries such as babysitter and meat packaging. The variation at such detailed level is not enough in our data, so we cannot disentangle industry e?ect versus owner e?ect. Given the particularity in the way female labor enter the market, there could be additional unobserved elements a?ecting the likelihood of hire an immigrant woman that we cannot consider in this aggregate analysis.
36
37
2.4.4 Workers
Among the relevant workers’ characteristics available in our data are age, immigration status (place of birth), date of entry in the US (date of SSN application), education, quarterly earnings, and race. We sum over quarters to obtain each worker’s annual earnings, and then compute real earnings based on 1992 dollars. The data set used for the analysis includes all male workers with positive earnings. On the distribution of workers, Table (2.2) and Figure (2.1) show the proportions of workers by age, race, sex, education, owner type,size, and sector, as well as, mean age, education and earnings, for all workers and for immigrants and natives. Foreign workers represent almost 24% of the sample. Similar to previous studies, on average, foreign born workers tend to be less educated, younger and tend to have lower income than native workers (Borjas [1994]), although these di?erences are not large in our sample. The fraction of workers across age categories, however, is similar for both types of workers in age categories 40 years and more. Table 2.2: Descriptive Statistics - Characteristics of Workers Individual IM US MEAN (std) Age Education Log(annual earnings) DISTRIBUTION (%) AGE Continued on next page. 38 ALL
34.01 34.14 34.11 (13.33) (12.02) (13.13) 13.04 13.16 13.13 (2.76) (2.94) (2.79) 8.30 8.32 8.33 (1.87) (1.68) (1.84)
Table 2.2: Descriptive Statistics - Characteristics of Workers (continued) Individual IM US 18.09 24.43 51.64 43.03 30.26 32.54 8.89 59.27 30.81 1.04 7.92 37.08 3.81 14.17 19.33 1.36 16.34 2.18 4.70 9.27 19.32 18.96 45.57 17.36 47.21 22.76 1.71 10.88 42.70 8.34 48.96 20.64 0.85 15.58 64.92 43.67 7.41 59.26 32.00 1.33 18.37 26.13 7.11 14.16 16.63 1.58 16.03 1.65 4.23 9.32 21.23 18.91 44.66 75.07 4.85 0.93 11.21 4.67 12.10 5.67 82.23 4.74 0.96 4.94 89.36 37.39 ALL 22.91 45.10 31.99 7.77 59.26 31.71 1.26 15.85 28.76 6.31 14.16 17.28 1.52 16.10 1.78 4.34 9.31 20.77 18.92 44.88 61.18 15.04 6.18 8.92 6.16 19.46 6.31 74.22 7.15 0.95 6.24 85.68 38.90
Under 25 25-39 40+ EDUCATION High School Dropout High School Graduate Some College Education College Graduate SECTOR Construction Manufacturing Transportation and Utilities Wholesale Retail FIRE Services SIZE 2-4 5-9 10-19 20-49 50-99 100+ RACE White Hispanic Asian Black Other TYPE OF OWNER Immigrant Mixed Native RACE OF OWNER Asian Black Hispanic White Part-time Continued on next page. 39
Table 2.2: Descriptive Statistics - Characteristics of Workers (continued) Individual IM US 97.23 83.38 24.06 75.94 ALL 86.71 100
In MSA All
Note: Number of observations equal to 214,398 workers. Statistics based on weighted outcomes. Standard Deviations in parenthesis. Male workers with positive earnings in a year. Log annual wage in 1992 dollars.
We can compare our sample of workers with the distribution and characteristics of workers from IPUMS 1990 (see appendix E.1 we ?nd interesting di?erences. To build the comparable sample, we only look at male workers, older than 16, and not working in Agriculture, Mining nor Public Administration sectors. One main di?erence is the average year of school between our sample and IPUMS. In IPUMS, both immigrant and natives have more year of schooling. Our sample has a very low proportion of college graduate workers (natives and immigrants). This low representation of this group could be driven by the over representation of small ?rms in our sample versus IPUMS database, and the types of workers that these ?rms hire.37 Natives have higher wages, but the wage di?erential between natives and immigrants is higher in IPUMS (around 12%) than in our sample (around 1%). Interestingly, natives in our sample are younger than the national average. By race, the distribution of workers is very similar. The proportion of immigrants in our sample is around 24% versus 13% for the national average. In sum, our sample contains younger male workers with low educational attaintments.
Ipums database does not include ?rm’s size. Therefore, we cannot control for the size of the ?rms.
37
40
The share of workers with a high school diploma or less is over 60% for both immigrants and natives. Immigrants are more concentrated in the high school dropout and high school graduate categories. Looking at sectoral distribution, both foreign and native workers are concentrated in Construction, Manufacturing, Retail and Services, with natives more likely to be in Construction and immigrants in Manufacturing.38 Foreigners are more likely to be working for immigrant owners than native workers. 43% of immigrant workers are employed in immigrant ?rms and 49% are employed in native ?rms. Asian and Hispanic-owned ?rms employ more immigrant workers than the average ?rm. More than forty percent of immigrant employees are hired by immigrant owners (around 43%). Most of the immigrants are Hispanics or Asians, while natives are mainly either white or black. Although there is a fraction of native-Hispanic and native-Asian workers, these proportions are less than 5%. The racial and ethnic categories follow the SSA codes, which form a set of mutually exclusive and collectively exhaustive categories. I also include information on whether the worker is full or part time. A worker is full time if he or she has worked during the full year (worker has positive earnings all four quarters). Most of the survey corresponds to information from ?rms located in MSAs. However, we include a variable that identi?es those ?rms and workers located outside a MSA. Almost 90% of the workers holds jobs in a ?rm located inside a MSA. Looking at place of birth in detail, Mexican, Salvadorian, Indian, and Chinese
One explanation for this pattern is that informal and undocumented immigrants workers are not largely covered by the database.
38
41
workers are the most represented immigrant groups in the data. At the national level, these are also the largest immigrant groups in the US according to Census 1990. In the data, native owners employ almost 75% of the total workforce.
2.4.5 Measuring coworker share
As described further in 3 below, we calculate the immigrant coworker share by considering all workers at the ?rm aside from the sample worker using the following formula:
1 COWij = Ik empj ? 1 emp
j
k=i
(2.1)
Where Ik is one when the worker is an immigrant. Therefore, this measure equals the fraction of immigrant coworkers of an employee in a ?rm. This measure is generally used in concentration analysis.39 Here I use it as an indication of workforce composition in the ?rm.
2.5 Analysis of New Hires, Earnings of Workers and Skill Distribution 2.5.1 New Hires
For the analysis of hiring procedures, we look at the type, race and ethnic composition of new hires by type of owner. During the period of analysis (199239
Hellerstein and Neumark [2007]; Aslund and Skans [2005a], Aslund and Skans [2005b].
42
1996), there were 147,373 new hires. We identify a new hire in the data by following a ?rm and looking at those workers that accessed the sample during the period of analysis. We track information on each new worker. Table (2.3) shows the distribution of new hires by type of owner. While new hires include a large share of natives for every type of owner, the proportions of newly hired immigrants for immigrant and mixed-owned ?rms (more than 30%) are almost three times the proportion of immigrants hired in native-owned ?rms (almost 12%). The second section of Table (2.3) displays the composition of new hires by race and ethnicity. Hispanics and Asians correspond to more than 35% of immigrantowned ?rms’ new hires. Again, this represents almost three times the proportion hired by native ?rms. Both immigrant and native ?rms hired more new workers later in the sample period as the economy recovered from the 1991-1992 recession (see Figures 2.1 to 2.4). The main diagonal shows that immigrant-owned ?rms hire more immigrants (33.33%) than the average ?rm (14.80%)), while native-owned ?rms hire more natives (88.44%) than the average ?rm (85.20%). Table 2.3: Average Race and Ethnic Composition of New Hires by Owner’s Type Owner Type Worker type/ race/ethnicity Immigrant Mixed Native All Immigrant 33.30 37.10 11.56 14.80 Native 66.70 62.90 88.44 85.20 Hispanic Asian White Black 20.14 16.09 48.20 6.49 22.32 12.78 49.80 8.41 10.14 11.50 2.65 4.26 73.61 70.40 8.68 8.46
Note: Number of Observations equal to 147,373. Male workers with positive earnings in a year. The other race/ethnic groups represent 0.5% of the sample. Results are not shown.
43
If we further look at the distribution of new hires across owner’s race, the ?ndings are stronger. Table 2.4 shows a strong correlation between the race of the owner and the racial/ethnic composition of the new hires.40 Asian, Black and Hispanic-owned ?rms hire their own type more than twice as often as the average ?rm. Asian-owned ?rms also hire Hispanic in a large number. Table 2.4: Average Race and Ethnic Composition of New Hires by Owner’s Race Worker / Owner Asian Asian 23.75 Black 6.29 Hispanic 20.10 White 39.03 Black 2.50 38.54 10.52 44.43 Hispanic White 4.27 2.70 10.33 8.04 35.46 9.10 42.24 75.39 All 4.25 8.46 11.50 70.42
Note: Number of Observations equal to 147,373. Male workers with positive earnings in a year. The other race/ethnic groups represent 0.5% of the sample. Results are not shown.
2.5.2 Earnings of Workers
In this section, we look at workers’ earnings. On average, immigrant workers have lower wages than natives. Most of the explanations given by the literature are based on human capital formation. Immigrants have lower host country abilities and generally less education than natives. However, even after controlling for some of these characteristics, immigrants tend to receive lower wages than observationally similar natives (Borjas [1994]). But do workers receive di?erent wages than their counterfactual group regardless of who they work for? To answer this question we undertake two di?erent exercises. First, we look at the average real log annual
Giuliano et al. [2006] ?nd a similar correlation when they look at the race of the hiring manager and the racial composition of the new hires in di?erent establishments of a retail store.
40
44
Figure 2.1: Workforce Characteristics of Immigrant, Mix and Native Firms
Note: Weighted share and percentage. Base on years 1992-1996.
earnings of each worker type across owner types. We also look at these statistics for di?erent groups of ?rms de?ned by the fraction of similar coworkers in the ?rm. This 45
Figure 2.2: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
analysis is a ?rst look at the impact of ?rm owner types on earnings. Second, we estimate owner type wage e?ects after controlling for a number of ?rm and worker characteristics, and evaluate the sources of wage di?erentials. The natural log of real annualized earnings of each worker comes from LEHDUI records.41 Table (2.5) shows how average wages change according to the type
41 When we take the average log annual earnings for each type of ?rm, we ?nd that it is slightly below the log of annual payroll per employee in the SSEL database. According to internal documentation on the ES202/SSEL joint project, annual payroll in SSEL ?les includes non-wage payments, such as bene?t payments, retirement pension funds, annuity funds, supplemental bene?t funds, etc, which are not included in the UI ?les.
46
Figure 2.3: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
47
Figure 2.4: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
48
of owner. The last column of the table shows the t-test computed for worker type wages for each owner type. A t-test can reject the null hypothesis that the mean of immigrant worker wages and the mean of native worker wages are the same at the 90% level. Table 2.5: Mean Earnings by Owner and Worker Type
Variable=log(annual earnings) owner = Immigrant Immigrant Native All owner = Mix Immigrant Native All owner = Native Immigrant Native All (%) 50.30 49.70 100.00 35.94 64.06 100.00 15.87 84.13 100.00 Mean 8.35 8.12 8.23 8.52 9.04 8.71 8.32 8.38 8.37 STD 1.47 1.67 1.64 1.86 1.71 1.82 1.73 1.88 1.73 T-test
24.20
-16.07
-5.83
Note:STD indicates standard deviation. Log annual wage in 1992 dollars. Using workers during the period 1992-1996.(*)T-tests are computed on the di?erence between average wages of immigrant and native workers for each speci?ed owner type.
Looking at Table (2.5) we notice three relevant outcomes for wage di?erential analysis. First, immigrants are paid slightly less by native than by immigrant owners. On average, they are paid the lowest when working for native owners. Second, native workers are paid signi?cantly less in immigrant owned businesses. Third, on average native owned ?rms pay more than immigrant owned ?rms. Fourth, mix-owned ?rms signi?cantly pay less to immigrant workers. However, these ?rms employ a lower proportion of immigrant workers than immigrant-owned ?rms. In sum, immigrant workers end up receiving lower log annual earnings than native workers. If we combine the ?rst three outcomes, we can see that much
49
Table 2.6: By Similar Coworker Share: Mean Earnings by Owner and Worker Type
Coworker Share Below the median Above the median (%) Mean STD (%) Mean STD 33.64 66.69 48.09 26.42 66.36 39.07 6.96 91.48 20.33 7.37 7.98 7.74 7.90 8.67 8.32 7.74 7.80 7.78 1.71 1.68 1.51 1.79 1.63 1.75 1.81 1.77 1.79 66.36 33.31 51.91 73.58 33.64 60.93 93.04 8.52 79.67 7.67 8.19 7.82 8.39 6.91 8.13 8.38 7.68 8.31 1.53 1.34 1.70 2.98 1.21 1.98 1.89 1.96 1.92
Variable=log(annual earnings) owner = imm Native Immigrant all owner = mix Native Immigrant all owner = usa Native Immigrant all
Note: STD indicates standard deviation. Log annual wage in 1992 dollars. Statistics based on estimation sample: all male individuals working between 1992 and 1996.
of the di?erence between the log annual wages of immigrants and natives comes from immigrants’ propensity to work in immigrant owned ?rms. These ?rms pay the lowest wages, and the di?erence in immigrant earnings between immigrant and native ?rms is small. Additionally, native owned ?rms pay immigrant workers less than native workers (see Table(2.5)). It is important to highlight the relevance of having actual earnings of each employee at the ?rm level, so we can exploit these variations to identify the e?ect of owner types on individuals’ wages. Therefore, individual level wages are used in the regressions analyzed in the next sections. Table (2.5) would not be possible if we didn’t have data on both employers and employees’ characteristics. Our unique database allows us to compare average earnings between workers of di?erent types holding a job in the same type of ?rm, and workers of the same type (native or
50
immigrant) working for di?erent types of owners. We now perform a similar exercise, but separating ?rms by the share of coworkers similar to the worker called ”similar coworker share” (see Table 2.6). This measure is di?erent from the measure of immigrant coworker share de?ned previously, in that here we de?ne the similar coworker share as the share of workers that are of a similar type to a particular worker in a speci?c ?rm. For instance, the coworker share of a native worker is the share of native born workers in the ?rm excluding the worker. The second column (%) shows the percentage of workers of each type in the ?rm accordingly below or above the similar coworker share median. We can see in the table that the previous ?ndings in Table 2.5 remain valid. Foreign-born employers pay the lowest wages, on average. However, for businesses with coworker share below the median, immigrant employees working for immigrant employers are paid slightly more than immigrant employees working for native employers. Additionally, workers are paid more when working with similar coworkers. When workers’ similar coworker share is below the median, employers pay lower annual wages. More than 65% of sample businesses have a mixed workforce, that is, the share of immigrant coworkers is neither one nor zero (0 < share < 1). These tables do not control for individuals’ characteristics, so we don’t know the pro?les of native and foreign employees holding jobs in these businesses. Nevertheless, these ?ndings are striking. Immigrant owners pay the lowest on average. Furthermore, they to pay natives less than the rest of the market. This motivates the question of what type of native workers work for immigrant employers.
51
2.5.3 Sorting by Skill
Sorting by skill is a possible cause of sorting by owner type. The incentive to combine workers of identical skills within the same ?rm has been documented previously (Kremer and Maskin [1996]). Job descriptions and skill requirements are also a concern as characteristics of employers and employees are correlated. Additionally, if ?rms of di?erent types have di?erent skill mix productivity, that is, they use a combination of workers’ skills and capital di?erently, then the di?erences in the probability of hiring a speci?c type of worker could be motivated by the capital/labor ?rm’s decisions. For instance, immigrant owners could use labor more intensively than native businesses, or could hire more low-skilled workers than native ?rms. Immigrants, Hispanics, and other minority groups have lower skill on average so they may tend to work in low-skill sectors and low-skill jobs regardless of the owner type. Immigrant owners, on the other hand, may tend to concentrate in lowskill sectors because they also have low skill levels. For both group, the mayority of the ?rms are in the ’Low-Skill Industries’. Almost 30% of immigrant-owned ?rms belong to the ’High-Skill Industries’ group, while more than 45% of native-owned ?rms belong to this group.
52
Table 2.7: Worker types distribution by owner’s skill requirement
Worker / Owner Immigrant Native Race/ethnicity Hispanic* Asian Black White (non-hispanic) All Low-Skill Industries Immigrant Native Mixed 38.60 11.60 27.50 61.40 88.40 72.50 19.23 18.22 5.38 48.71 70.83 10.12 2.63 6.83 76.18 54.16 16.78 8.23 7.62 61.38 62.24 All 15.40 84.60 11.40 4.64 6.68 72.47 100.00 High-Skill Industries Immigrant Native Mixed 33.70 9.20 41.60 66.30 90.80 58.40 17.73 17.09 6.45 52.22 29.17 6.44 2.97 10.84 75.49 45.84 20.05 24.30 4.45 42.07 37.76 All 11.50 88.50 7.43 4.34 10.42 73.28 100.00
53
Note: Using Census 1990 information on workers’ education attainment by industry, industries are separated into High Skill and Low Skill. High skill refers to those industries in which more than 50% of workers have at least a high school diploma. Otherwise we de?ne the industry as low skill. (*) Hispanic refers to all races with ethnic group Hispanic. The group Other includes Native American and otherwise unclassi?ed racial groups. Native-American workers represented only 0.5% of the total sample.
Table (2.7) shows workers’ distribution by owner’s skill requirement. The skill requirement for a ?rm is computed using Census 1990 data after compiling the share of workers by industry at the 2-digit level that have low educational attainment(less than high school) and high educational attainment(more than high school). High skill industries are those in which more than 50% of workers have at least a high school diploma. The remaining industries are low skill. The idea is to illustrate whether speci?c owner and worker types are concentrated in a particular skill group. Not surprisingly, the table shows that ?rms in low-education industries have higher fractions of immigrant workers than ?rms in high-education industries. Immigrant ?rms continue to have a bigger proportion of immigrant workers, except for mix-owned businesses. Results are similar breaking down by workers’ race. However, it is worth mentioning that immigrant-owned ?rms are more than 60% of the group of low-skill ?rms. To account for part of this pattern, in the regressions below we include the share of workers in the ?rm in four education categories: high school dropouts, high school graduate, some college, and college graduate.
2.6 Regression Analysis
The ideal data to analyze the e?ect of owners, coworkers, and social connections on individual labor market outcomes requires information on individuals’ labor market histories, earnings, and, speci?cally, the employer’s source of ex-ante information about the job seekers that apply to its open vacancies. With this information
54
we would be able to measure the actual hiring policies that ?rms use to ?nd new workers. Unfortunately, we don’t have detailed data on hiring procedures used by ?rms. However, we do have a good deal of valuable information on the ?rms and workers. Workers can be divided into di?erent categories by birth place or by race/ethnicity42 to infer workers’ and candidates’ likely social connections. This, together with information on the type of owner, will help us infer the use of social ties in the ?rm’s hiring process and its e?ect on workers’ earnings. More speci?cally, network structure refers to the number of ties an individual has (Smith, 2000). In this paper, we try to identify the impact of networks by using the proportion of coworkers who are potentially tied to a newly hired worker. Besides identifying the type of owner for whom the employee works, I use the proportion of similar employees in the ?rm at the time the new worker is hired as a measure of the network link between coworkers, employers, and the new worker. Following each ?rm from 1992 to 1996, we obtain the number of employees who work for the ?rm and their earnings. We also have the total number of workers possessing any given set of demographic characteristics at each period of time. Following the de?nition of networks used in previous literature, we compute the share of similar coworkers for each new hire at each ?rm in each period, assuming that a similar birthplace or ethnicity implies at least a weak network connection between individuals.43
White, Black, Hispanic, and Asian. At this point, it is worth to mention that even though immigrants are very diverse and it is a group that re?ects a multiple gamma of ethnic/cultural backgrounds, not necessarily captures by the denomination of being foreign-born, it is also true that immigrants tend to have similar
43 42
55
A key challenge in linking owners and employees is that the characteristics of both owners and employees may be correlated with other characteristics of a workplace and its location. Section 2.5.3 above gives preliminary evidence on sorting by skill. The correlation between owner and employee types could also be a result of residential segregation of workers and owners (spatial mismatch). Job descriptions and skill requirements are also a concern, as characteristics of employers and employees are correlated. Immigrants, and in particular Hispanics, tend to be low skilled and therefore are likely to work in low-skilled sectors and low-skilled jobs regardless of the owner type. However, at the same time, immigrant owners could tend to concentrate in low-skill sectors, perhaps because they also have low skill levels. Because the proportions of immigrants are unequally distributed across sectors and regions, we control for the 2-digit industry and geographic location of each ?rm. There exist sectors such as Retail, Services and Construction where immigrants represent a signi?cant proportion of the workforce
44
. We also see this pattern in the
geographic distribution of the immigrant population. For instance, according to Census 2000, Los Angeles and New York represent more than 30% of the total immigrant population in the country. To account for these concerns we need to control for ?xed attributes of the workplace and the local labor market, and also for local trends in labor pool demographics. Therefore, we estimate the model controlling for
strategies to enter into the labor market regardless of their cultural background. Using migrant networks is one common factor among foreign-born workers, especially for new immigrants (Porter and Wilson [1980], Light [2006]). 44 This can be also related to the fact that these sectors are also highly represented by relatively smaller ?rms than in Manufacturing, for instance.
56
characteristics of the ?rm (Fj ) and local community (Zj ). These controls include the immigrant workforce population and population density in the local community, 2-digit industry code dummies, ?rm size (log of reported employment), and legal form of organization. We also include the share of the ?rm’s workers in the four education categories discussed previously. Previous research has noted the impact of English language ability in the use of networks and the level of wages for immigrant workers.45 We capture this feature by interacting the 2-digit industry dummy with an English speaker dummy
46
This
interaction is a proxy that intends to capture whether language is used di?erently in di?erent industries. In the wage regressions, we also control for individual characteristics (Xj ), including worker’s age, education and a dummy for working full time.47 The composition of the labor pool might also be a?ected by changes over time in labor supply and demand. For example, white natives may be more likely to work in low-wage retail jobs when labor markets are weak. Therefore, we also include a dummy variable for each of the years in the sample (Mt ) to control for national ?uctuations in the labor market. The identi?cation strategy exploits variation across owner types for otherwise similar ?rms. By controlling for a rich set of ?rm characteristics we can narrow the possible alternative explanations for any residual correlation between owner type
Hellerstein and Neumark [2007] and Hellerstein et al. [2008a]. For additional analysis, see 3 below. 46 We identify a group of countries where English is the main language, and use this information to identify the worker as English speaker or otherwise. 47 A worker with positive time during the complete year is considered full quarter worker or full year worker.
45
57
and worker outcomes.
2.6.1 Analysis of ?rms hiring patterns
This section starts by looking at the hiring patterns of the ?rm, estimating a model that predicts the probability that a newly hired employee is an immigrant. Firm hiring decisions indirectly re?ect the way owners use current employees to help ?ll their job vacancies. We use a linear probability model to estimate the likelihood that a newly hired worker is of a particular type (immigrant or from a speci?c race/ethnic group).48
Pr(new hire:groupi )kjt = c + B1 ? Oj + ? ? Wjt?1 + B2 ? Oj ? Wjt?1 + ? ? Fj + Z ? Zkj + T ? Mt +
kjt
(2.2)
Where k , j and t designate the worker, ?rm type, and time respectively. Oj is a vector of dummy variables for owner type (de?ned by immigration status or race). If i refers to the group of immigrant workers, we use as the reference group ?rms owned by immigrants. B1 represents the vector of coe?cients associated with the impact of owner type on hiring. The elements of this vector are expected to be negative when the omitted group is the same type as the new hire. For instance, the coe?cient on native owners would be negative if immigrant-owned ?rms are more likely to hire new immigrant workers. Wjt?1 corresponds to the vector of the proportion of workers of
We use a linear probability model over a Probit (Logit) model because we don’t need to restrict the sample to ?rms that hire at least one new worker of each type. This restriction could introduce sample selection bias because ?rms with zero hiring could have a completely di?erent policy than those with a least one new hire.
48
58
type each type i at the ?rm in the previous period. An interaction between owner type and Wjt?1 is included to asses di?erences in use of current employees’ networks across owner types. I also control for ?rm characteristics Fj (a vector of variables measured at the ?rm level), year dummies Mt , and local community information and state dummies Zkj . In a regression with both owner type and coworker share included, the estimated coe?cient on owner type will capture only the direct impact of owner type on hiring, not the total e?ect, which will include both the direct e?ect and the indirect e?ect coming through owner type’s e?ect on coworker share. The use of employee referrals can be correlated with the type of owner and can a?ect hiring patterns if owners have the tendency to hire same-group individuals. When employees tend to refer same-group workers, the owner type’s e?ect may be ampli?ed. If we believe that the share of similar coworkers is a good proxy for social connections, these exercises illustrate the combined result of owner e?ects and hiring patterns. We assume that the error (
kjt )
in equation 2.2 is independent and identi-
cally distributed across ?rms, but not within ?rms. To correct for non spherical disturbances, we estimate Huber-White robust standard errors clustered by ?rm. This procedure is used in all subsequent estimations. We cluster the errors by ?rm since ?rms in the sample may have hired more than one worker and thus may have repeated observations. For purposes of analysis, we estimate di?erent versions of equation (2.2) and look at the impact of the addition of controls on the estimates of B1 and B2 . The ?rst regression includes only year dummies; subsequent speci?cations add controls 59
one by one. Most of the literature on hiring networks argues that current workers’ referrals are more important to ?rm hiring patterns than owners’ personal networks. Owners are likely to hire individuals from their residential area. However, current workers have a larger and more diverse set of connections that can be exploited by the ?rm. We are not able to disentangle these e?ects directly. Nevertheless, by allowing owners of di?erent groups to make use of their workers’ social ties di?erently, the estimated interaction e?ects can measure the ability of owners to use social ties. Table (2.8) shows the probability of a new hire being an immigrant given the characteristics of the ?rm, its community and the share of immigrant coworkers in the ?rm. Controlling only for year dummies, native owners are 25 percentage points less likely to hire a new immigrant worker than immigrant ?rms (column 1). This di?erence is signi?cantly reduced, to 3.5 percentage points, when we include the share of immigrant coworkers (column 2). Controlling for year and industry dummies, the share of immigrant coworker positively a?ects the likelihood of an immigrant being hired. The inclusion of the share of English speaker and its interaction with industry dummies decreases the impart of the share of immigrant coworkers on the probability of being hired. This covariates controls for whether language is used di?erently in di?erent industries (column 3). For instance, a Mexican restaurant would probably hire Mexicans or Spanish speaker because of the type of service they o?er and type of frequent consumers. The use of language can be di?erent in a industry where workers don’t need to communicate with each other, so language di?erences are not obstacle in the production process. Given the results, it seems that ?rms use language in di?erent ways, a?ecting the likelihood of an immigrant
60
Table 2.8: Linear Estimates of the E?ect of Owner Type on the Probability that a New Hire is an Immigrant
Owner Mix Owner Native % Imm. Coworkers % Imn. Coworkers * Owner Mix % Imm. Coworkers * Owner Native Corporation Sole Prop. log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Pop. % immigrant in neighborhood(+) Population in neighborhood(+) In MSA Constant Dummies year Industry Indus*English Spkr R-Square 0.4081*** 0.099 yes yes 0.29 0.0211*** 0.0039 yes yes 0.32 0.0989*** 0.002 yes yes yes 0.34 0.0969*** 0.0016 yes yes yes 0.35 (1) -0.0519*** 0.0057 -0.2358*** 0.0032 (2) -0.041*** 0.0065 -0.0351*** 0.0031 0.9961*** 0.0056 (3) -.0034** 0.001 -0.0342*** 0.0009 0.782*** 0.002 (4) -0.0037** 0.001 -0.033*** 0.0004 0.7724*** 0.0101 -0.0125** 0.005 -0.0711*** 0.003 (5) -0.00313** 0.001 -0.0254*** 0.0014 0.7132*** 0.0234 -0.0094** 0.005 -0.0378*** 0.0041 -0.00085* 0.0033 0.0026 0.003 0.003 0.002 0.0021** 0.0004 -0.0012 0.001 0.005 0.006 0.0162** 0.0068 0.0004*** 0.00 -0.005*** 0.0009 0.0285** 0.0069 yes yes yes 0.38 (FE)
0.6715*** 0.0435
0.1575* 0.781 yes 0.41
Note: Reference group is immigrant ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Neighborhood is de?ned counties adjacent to the county where the ?rm is located. Population in 100,000’s. FE represents the ?rm ?xed-e?ect model. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
being hired and reducing the impact of immigrant coworkers in the ?rm. There is a positive and signi?cant impact on the probability of the new hire
61
being an immigrant when the proportion of workers in the ?rm with low education (high school dropout) increases. The owner e?ect diminishes and the di?erence in the probability of hiring a Hispanic between immigrant and native owners is 2.5 percentage points (column 4). The coworker e?ect is smaller too, although it is still signi?cant. The interaction e?ects between owner type and coworker share decrease slightly when others controls are included, although the results are similar. The e?ect of immigrant coworker share is smaller in mix and native owned ?rms than in immigrant owned ?rms. Immigrant employers can take advantage more e?ciently of their current immigrant workers than other types of employers. The increment of immigrant coworker share by 1 percentage point increases this likelihood by 0.710.67. The inclusion of other characteristics of the ?rm and the local community has a smaller impact on the relative likelihood of native versus immigrant owners hiring a new immigrant worker. We should be cautious when analyzing these results. We include a vast series of covariates to control for all possible observables that can be correlated with employer and employee e?ects. However, the presence of unobservables correlated with ?rm and worker interactions could bias the results. As another exercise, we compute the ?rm ?xed-e?ect version of the model by including ?rm dummies. The last column of Table (2.8) shows the results. The impact of share of immigrant coworkers in the ?rm at the time of the new hire remains positive, high, and signi?cant.
62
2.6.2 Hiring Process by Race/Ethnicity
We next consider the determinants of the probability that a new hire comes from a particular race/ethnic group: white, black, Hispanic and Asian. That is, we estimate equation (2.2), setting i equal to a particular racial category. Tables 2.9 and F.1 show the e?ects of owner types and shares of type i coworker, and other types of coworker, at the time of hiring on the probability that a new hire is Hispanic, Asian, white, or black respectively.
2.6.2.1 Worker Race
The likelihood of a new worker being Hispanic or Asian signi?cantly decreases when the employer is native. This result holds even after including a exhaustive list of controls(Tables 2.9 and 2.10). The direct impact of owner type is reduced, however, once we control for the share of Hispanic coworkers. For instance, having a one percentage point increment of Hispanics as current employees in the ?rm increases the probability that a new hire is Hispanic (by up to 0.88 in immigrant owned ?rms). The impact of Hispanic coworkers is smaller for native owned ?rms. Table 2.9: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic
(1) 0.0214*** 0.0041 -0.0903*** 0.003 (2) 0.012 0.01 -0.0872*** 0.003 Hispanic (3) (4) 0.0077 0.014 0.005 0.0357 -0.0412*** -0.0245** 0.003 0.001 0.9441*** 0.0054 -0.628*** 0.013 (5) -0.0694 0.054 -0.0172** 0.001 FE
Owner Mix Owner Native Hispanic Cowkrs Asian Cowkrs
-0.526*** -0.504*** 0.023 0.029 Continued on next page.
63
Table 2.9: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic (continued)
(1) White Cowkrs Black Cowkrs Hispanic Cow* Owner Mix Asian Cow* Owner Mix White Cow* Owner Mix Black Cow* Owner Mix Hispanic Cow* Owner Native Asian Cow* Owner Native White Cow* Owner Native Black Cow* Owner Native log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Work. Pop. Total.1 Work. Pop. %Hisp 1 Constant year dummies Industry dummies State dummies Other controls(+) p-value R-Square (2) Hispanic (3) (4) -0.703*** 0.0095 -0.879*** 0.0123 -0.0621** 0.031 0.1060** 0.0483 0.0185 0.0384 0.137 0.0845 -0.0869*** 0.043 -0.113*** 0.0295 -0.148** 0.062 -0.094* 0.04 (5) -0.681*** 0.0075 -0.725*** 0.0134 FE -0.596*** 0.0197 -0.616*** 0.0212
0.093** 0.034 0.0175 0.0434 0.105 0.0945
0.1920*** 0.0032 yes 0.0001 0.22
0.1243*** 0.009 yes yes 0.002 0.29
0.9702*** 0.08 yes yes yes 0.0001 0.31
0.8454*** 0.1616 yes yes yes 0.003 0.34
-0.102*** 0.0243 -0.124** 0.056 -0.097* 0.04 0.0013** 0.00 0.0012*** 0.0005 0.0025*** 0.0003 0.0030*** 0.0008 0.0561** 0.0245 0.094*** 0.002 0.9511*** 0.171 yes yes yes yes 0.003 0.42
0.8411*** 0.201 yes 0.01 0.35
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
64
The e?ect of the share of Hispanic coworkers is positive regardless the type of the owner. However, the e?ect is smaller than the baseline e?ect on the Hispanicowned ?rms (column 3). Columns 4 includes the e?ect of all races coworker share on the likelihood of being hired. Other races coworker shares a?ect negatively the probability of a new hire is Hispanic. Interestinly though, Asian coworker share is less negative when the ?rm is mix-owned. Column 5 includes other ?rm and local community characteristics. Their inclusion decreases the average e?ects, but do not change the directions of the results. In section (2.5.3) we discussed the distribution of workers by average industrylevel skill requirement. As a proxy to control for this e?ect, we include the ?rm’s share of workers in four education categories and the fraction of workers of similar type in the local community. The results show that a higher share of low-educated workers in the ?rm increases the probability that the new worker is Hispanic. We also include the share of workers of each racial group in the local labor force. The inclusion of these shares decreases the impact of the coworker shares. Looking at Asian new hires (Table 2.10), we again ?nd that native employers are less likely to hire Asian workers. The inclusion of additional controls reduces the di?erence in probability of hiring an Asian between immigrant and native owned ?rms. Another interesting result is that Asians are less likely to be hired in ?rms with bigger proportion of workers with education attainment below the high school level.
65
Table 2.10: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian
(1) -0.029*** 0.0032 -0.1245*** 0.002 (2) -0.0280*** 0.0032 -0.1114*** 0.002 Asian (3) (4) 0.002 0.001 0.003 0.002 -0.054** -0.052** 0.002 0.002 -0.743*** 0.0079 0.8194*** 0.031 -0.7876*** 0.0071 -0.8795*** 0.0083 -0.0197 0.0303 0.007 0.002 -0.0746*** 0.0022 0.032 0.0446 -0.0064 0.0185 -0.152*** 0.013 -0.0158 0.0154 -0.0076 0.02 (5) 0.007 0.002 -0.06** 0.014 -0.712*** 0.008 FE 0.1948 0.254
Owner Mix Owner Native Hispanic Cowkrs Asian Cowkrs White Cowkrs Black Cowkrs Hispanic Cow* Owner Mix Asian Cow* Owner Mix White Cow* Owner Mix Black Cow* Owner Mix Hispanic Cow* Owner Native Asian Cow* Owner Native White Cow* Owner Native Black Cow* Owner Native log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Work. Pop. % Imm.1 Work. Pop. %Hisp 1 Work. Pop. % Asian 1 Constant year dummies Industry dummies State dummies
-0.622*** 0.008
-0.741*** 0.0072 -0.8214*** 0.0081 -0.0556 0.041
-0.6715*** 0.0074 -0.7631*** 0.0083
-0.076*** 0.0022 0.0404 0.0536 -0.0064 0.0185
-0.031 0.0221 -0.0095 0.02 0.0014** 0.00 -0.0012** 0.0001 -0.000 0.00 -0.0013** 0.00 0.024* 0.001
0.1495*** 0.0018 yes -
0.112** 0.0562 yes yes -
0.065** 0.02 yes yes yes
0.095*** 0.01 yes yes yes
0.043** 0.01 0.094*** 0.097 0.01 0.081 yes yes yes yes Continued on next page.
66
Table 2.10: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian (continued)
(1) 0.0001 0.26 (2) 0.002 0.31 Asian (3) 0.0001 0.34 (4) 0.003 0.35 (5) yes 0.003 0.37 FE 0.01 0.38
Other controls(+) p-value R-Square
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
Whites and blacks are more likely to be hired by native ?rms (See Tables (F.1) and (F.2)). However, the probability that a new hire is black or white depends on the share of blacks or whites in the ?rm at the time of the recruitment process. The signi?cance of the immigrant owner e?ect on black hiring vanishes when I include the black coworker share in the regression. The column FE shows the results of the regression after including ?rm ?xed e?ects. The impact of similar coworkers decreases slightly but is still high and signi?cant. The largest change in coe?cients caused by the inclusion of ?xed e?ects is the drop in the impact of white coworkers on the probability of that a new hire is black. We also experiment with estimating a multinomial logit model to account for the posibility that employers may simultaneously choose among di?erent types of workers. The estimation sample is then restricted to ?rms that hire at least one worker of each race group during the period 1992-1996. This restriction eliminates more homogeneous ?rms. The new sample contains 2,662 ?rms out of the original sample of 4,478 ?rms. We investigate how the owner type and shares of di?erent
67
types of workers at the time of hiring a?ect the type/race of the new hire. We estimate a model49 that aims to reveal whether the birthplace of the employer a?ects the likelihood that a new worker is of the same type as opposed to other types, conditional on having accessed to the ?rm during the period of analysis and controlling for the characteristics of the worker and the ?rm.
Pr(new hire is worker type: i)kjt =
i exp(ci + B1 ? Oj + ? i ? Wjt?1 + ?i ? Fj + Z i ? Zkj + T i ? Mt + 5 s=1 i kjt ) s kjt )
s exp(cs + B1 ? Oj + ? s ? Wjt?1 + ?s ? Fj + Z s ? Zkj + T s ? Mt +
(2.3)
with i = 1, ..., 4 for the four race groups: white, black, Asian, and Hispanic. This procedure makes very strong assumptions with respect to the relevance of other alternatives. The odds ratio of any two options is assumed independent of the other alternatives. This feature is important to consider when more than two alternatives are included. To test the Independence of Irrelevant Alternatives assumption, we conduct a Hausman test by excluding each outcome category in turn. The test indicates that I cannot reject the null hypothesis that the odds of one outcome happening are independent of other alternatives. Additionally, we perform Wald tests for combination of categories. The tests reject the null hypotheses that all coe?cients associated with a given pair of outcomes are zero (except intercepts). We cluster the errors by ?rm since observations within ?rms are not independent. The results for this regression are shown in Tables (2.11) and (2.12).
I speci?cally estimate a mixed logit model that incorporates both characteristics of the individual and the alternatives.
49
68
Table (2.11) shows the change in log odds comparing two alternatives. The share of white coworkers signi?cantly increases in the log odds of a white being hired. We also show the predicted hiring probabilities for each owner type (Table 2.12) computed at the means of all ?rms and dummy variables. The change in log odds between hiring a white worker versus hiring a Hispanic or an Asian decreases when the ?rm is immigrant-owned. Immigrant owners are 3 percentage points more likely to hire Asians and Hispanics than native ?rms. These results support the analysis in the previous section.
69
Table 2.11: Multinomial Logit Model: E?ects of Owner Type and Coworkers on Type of New Hires
Cow. Share White Black Asian Hispanic White to Black 1.97*** 0.646 -5.352*** 0.892 -1.391 0.951 -0.236 1.03 Change in log odds comparing alternative 1 to alternative 2 White to Asian White to Hispanic Black to Hispanic Black to Asian 2.32*** 3.53*** 1.44* 0.53 0.761 0.421 0.71 0.723 1.186 2.145*** 7.456*** 5.456*** 1.086 0.661 0.957 0.968 -7.243*** 0.041 1.433 -5.682*** 1.001 0.591 1.108 0.946 -0.086 -3.675*** -3.127*** 0.15 1.102 0.527 1.09 0.952 Asian to Hispanic 0.92 1.017 1.014 1.131 7.126*** 1.143 -3.654*** 1.361
Note: Other controls include log of employment, percentage of immigrant workers in the surrounding counties, population in the county, legal form of organization, Msa location, 2-digit industry, interaction 2-digit industry and English speaker dummy, state and year dummies. Results from race/ethnicity ’others’ are not shown. Number of observation 135,583 workers, and 2,662 ?rms. Robust standard errors in italic allow for arbitrary correlation within the same ?rm. * signi?cant at 10%,** signi?cant at 5%, *** signi?cant at 1%.
70
Table 2.12: Multinomial Logit Model: Predicted Probability of Covariates
Owner Native Immigrant Mix White 0.740 0.710 0.690 Workers Black Asian 0.120 0.031 0.102 0.060 0.119 0.052 Hispanic 0.100 0.126 0.134
Note: Based on multinomial logit predictions of the race of new hires from previous table.
2.6.2.2 Worker and Owner Races
After looking at the e?ect of owner birthplace on the probability of being hired for each particular worker’s race, the natural question is whether we can detect similar e?ects when we separate owner types by race. As explained in Section 2.4.1, owner’s race is obtained from the Small Minority Owner Business Employers Survey(SMOBE). For multiple-owned ?rms, the median race is used; in the case of ties, the hours worked in the ?rm are also considered to determine the predominant race of the ?rm. The race categories are: white, black, Asian and Hispanic. The likelihood of a new worker being Hispanic or Asian signi?cantly decreases when the employer is White. This result holds even after including a exhaustive list of controls(Tables 2.13 and 2.14). The direct impact of owner type is reduced, however, once we control for the share of Hispanic coworkers. For instance, an increment of one percentage point in the share of Hispanics as current employees in the ?rm increases the probability that a new worker is Hispanic (by up to 0.95 in Hispanic owned ?rms). Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic
(1) -0.2215*** 0.0271 -0.135*** 0.0229 -0.2358*** 0.0154 (2) -0.1514*** 0.0085 -0.1264*** 0.0043 -0.176*** 0.0034 Hispanic (3) -0.024** 0.0112 -0.0257*** 0.0076 -0.0318*** 0.0064 0.9512*** 0.0176 (4) -0.0284** 0.015 0.0868 0.7279 -0.0231*** 0.0015 (5) -0.0165** 0.0013 0.1429 0.1309 -0.0158*** 0.0045
Owner Black Owner Asian Owner White Hispanic Cowkrs Asian Cowkrs
-0.918* -0.9114*** 0.0669 0.0707 Continued on next page.
71
Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic (continued)
(1) White Cowkrs Black Cowkrs Other Cowkrs Owner Black* Hispanic Cowkrs Owner Black* Asian Cowkrs Owner Black* White Cowkrs Owner Black* Black Cowkrs Owner Black* Other Cowkrs Owner Asian* Hispanic Cowkrs Owner Asian* Asian Cowkrs Owner Asian* White Cowkrs Owner Asian* Black Cowkrs Owner Asian* Other Cowkrs Owner White* Hispanic Cowkrs Owner White* Asian Cowkrs Owner White* White Cowkrs Owner White* Black Cowkrs Owner White* Other Cowkrs Share of workers with HSD (?rm) Share of workers with HSG Share of workers with SOG Log employment Work. Pop. Total Work. Pop. -0.0335** 0.0145 0.0367 0.2128 0.0132 0.0678 -0.0382** 0.0109 0.0048 0.1592 -0.0427* 0.0252 -0.0531 0.0711 -0.0869** 0.0326 -0.1764*** 0.0493 -0.2705*** 0.0627 -0.09*** 0.0194 -0.1057* 0.0748 -0.0811** 0.0296 -0.1071*** 0.0385 -0.184*** 0.054 0.0016*** 0.0005 0.0022*** 0.0003 0.0032*** 0.0009 0.0011* 0.0009 0.0422*** 0.024 0.0002** Continued on next page. -0.2077*** 0.0703 -0.1218*** 0.0254 -0.2015*** 0.0342 -0.2358*** 0.0486 -0.1636** 0.0756 -0.1622*** 0.037 -0.144*** 0.0533 -0.33*** 0.0676 0.0922 0.2218 0.0024 0.0748 0.0063 0.0783 0.092 0.1826 (2) Hispanic (3) (4) -0.9086*** 0.0231 -0.8322*** 0.0317 -0.6857*** 0.0434 (5) -0.6494*** 0.0269 -0.6335*** 0.0349 -0.5814*** 0.0478
72
Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic (continued)
(1) % Hisp. Constant Year dumies Industry dummies State dummies Other Controls p-value R-Square 0.3286*** 0.0136 yes 0.01 0.24 (2) 0.0436* 0.0175 yes yes yes 0.01 0.29 Hispanic (3) 0.0211 0.0828 yes yes yes 0.01 0.31 (4) 0.8779 0.0864 yes yes yes yes 0.01 0.33 (5) 0 0.6446 0.1755 yes yes yes yes 0.01 0.45
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
The impact of Hispanic coworkers is smaller for other types of ?rms. The results for other characteristics of the ?rm and its location are similar to previous sections. The results show that a higher share of low-educated workers in the ?rm increases the probability that the new worker is Hispanic. I also include the shares of coworkers in each racial group. Black and White owned ?rms are 2 to 3 percentage points less likely to hire a Hispanic worker compared to Hispanic and Asian owned ?rms, holding constant the worker race distribution. Looking at Asian new hires (Table 2.14), we ?nd that white employers are less likely to hire Asian workers. White owners are mostly natives. The inclusion of additional controls reduces the di?erence in probability of hiring an Asian between Asian and white owned ?rms. Another interesting result is that Asians are less likely to be hired in ?rms with bigger proportion of workers with educational attainment below the high school level.
73
Table 2.14: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian
(1) -0.181*** 0.0067 -0.1977** 0.0215 -0.187*** 0.0049 (2) -0.1771*** 0.0054 -0.1842*** 0.0027 -0.1775*** 0.0021 Asian (3) -0.06*** 0.006 -0.013 0.0034 -0.049*** 0.0026 0.98*** 0.0074 (4) -0.2035* 0.1164 -0.0333*** 0.00419 -0.1257*** 0.0149 (5) -0.1628*** 0.0268 0.0503 0.0462 -0.1749*** 0.0163
Owner Black Owner Hispanic Owner White Asian Cowkrs Hispanic Cowkrs White Cowkrs Black Cowkrs Other Cowkrs Owner Black* Asian Cowkrs Owner Black* Hispanic Cowkrs Owner Black* White Cowkrs Owner Black* Black Cowkrs Owner Black* Other Cowkrs Owner Hispanic* Asian Cowkrs Owner Hispanic* Hispanic Cowkrs Owner Hispanic* White Cowkrs Owner Hispanic* Black Cowkrs Owner Hispanic* Other Cowkrs Owner White* Asian Cowkrs Owner White* Hispanic Cowkrs Owner White* White Cowkrs Owner White* Black Cowkrs Owner White* Other Cowkrs Share of workers
-0.987*** 0.0149 -0.920*** 0.0101 -0.978*** 0.0202 -0.9928*** 0.0237 -0.2765*** 0.1041 0.1801 0.1258 0.1915 0.1186 0.2172 0.1184 0.3027 0.1514 -0.0374 0.0339 0.0019 0.044 -0.0421** 0.0235 -0.0032 0.0506 -0.0684 0.0552 -0.2475*** 0.0131
-0.8883*** 0.017 -0.7414*** 0.0118 -0.7986*** 0.022 -0.8541*** 0.0257
0.1481 0.1367 0.1471 0.1305 0.2031 0.1299 0.2474 0.17
-0.0373 0.0488 -0.0302** 0.0129 -0.0009 0.0559 -0.0754 0.0608
0.2095 0.0229 -0.1092*** 0.0177 -0.1205*** 0.0263 -0.108*** 0.0316 -0.0017*** Continued on next page.
0.1633*** 0.0203 -0.1158*** 0.0158 -0.144*** 0.0242 -0.1223*** 0.0287
74
Table 2.14: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian (continued)
(1) with HSD (?rm) Share of workers with HSG Share of workers with SOG Log employment Work. Pop. Total Work. Pop. % Asian Constant Year dumies Industry dummies State dummies Other Controls p-value R-Square (2) Asian (3) (4) (5) 0.0004 -0.0001 0.0002 -0.0021*** 0.0005 0.0033*** 0.0006 0.0643*** 0.0154 0.0002*** 0 0.1242 0.1129 yes yes yes yes 0.01 0.43
0.2431*** 0.0052 yes 0.01 0.28
0.2131*** 0.0557 yes yes yes 0.01 0.29
0.0373 0.0519 yes yes yes 0.01 0.35
0.0275 0.0528 yes yes yes yes 0.01 0.38
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
Whites are more likely to be hired by white owned ?rms (see Table F.4). However, the probability that a new hire is white depends postively on the share of whites in the ?rm at the time of the recruitment process. The owner’s race e?ect is lower for those owners from a di?erent racial group. Multinomial analysis is also applied to the combination of worker and owner races. The results for this regression are shown in Tables (2.15) and (2.16). Table (2.15) shows the change in log odds comparing two alternatives. The change in log odds between hiring a white worker versus hiring a Hispanic or an Asian decreases when ?rms are Hispanic or Asian owned. A higher share of white coworkers sig75
ni?cantly increases the log odds of a white being hired, and a similar result holds for other races. We also show the predicted hiring probabilities for each owner race (Table 2.12) computed at the means of all ?rms and dummy variables. Hispanic owners are 3 percentage points more likely to hire Asians and Hispanics than White and Black owners. These results support the analysis in previous sections. In sum, Hispanic and Asian workers are generally more likely to be hired by Hispanic or Asian owned ?rms. In this detailed presentation, it seems that Asian owned ?rms tend to employ Asian and Hispanic workers more readily than black and white workers. Almost 70% of Asian and Hispanic owners are immigrants. We would also like to analyse the impact of immigrant/native oner e?ects after controllinf for owner race, including owner birthpalce and race simultaneously. However, the variation across the sample is not enough to identify whether birthplace or owner race is more important. Most of the native owners are either white or black, with a large proportion of them being white. While, our sample has a small representation of black immigrant owners. Given the structure of our sample, white and black owners are mainly natives, while Asian and Hispanic owners are immigrants, and after looking at our by racial groups results, we can see that our previous result. Immigrant owners tend to hire immigrant workers, while native owners tend to hire native workers.
76
Table 2.15: Multinomial Logit Model: E?ects of Owner’s Race and Coworkers on Type of New Hires
Covariates Change in log odds comparing alternative 1 to alternative 2 White White White Black Black Asian to Black to Asian to Hispanic to Hispanic to Asian to Hispanic -0.379*** -0.313* -0.234 0.145*** 0.067 -0.079 0.03 0.15 0.2 0.04 0.1 0.1 -0.09* -0.186*** -0.052** -0.039** -0.096*** -0.134*** 0.06 0.03 0.01 0.01 0.02 0.03 0.11 -0.229*** -0.113 -0.340** -0.223** 0.116*** 0.09 0.03 0.03 0.04 0.03 0.02 2.522*** 1.7560** 6.631*** 4.111*** 0.766 1.875 0.892 0.413 1.121 1.153 0.651 1.034 -7.130*** -0.823 -0.385 6.745*** 6.308*** 0.438 1.203 0.723 0.241 1.412 1.324 0.56 -3.532*** -7.742*** -1.832* 1.700 -4.210*** 5.910*** 0.731 1.202 0.891 1.342 1.154 1.265 -1.522* -1.756*** -6.632*** -4.210** 0.667 -4.875*** 0.641 0.952 1.678 1.023 0.801 1.123
Owner Black Owner Hispanic Owner Asian Cow. Share white Cow. Share black Cow. Share Asian Cow. Share Hispanic
77
Note: Other controls include log of employment, percentage of immigrant workers in the surrounding counties, population in the county, legal form of organization, Msa location, 2-digit industry, interaction 2-digit industry and English speaker dummy, state and year dummies. Results from race/ethnicity ’others’ are not shown. Number of observation 135,583 workers, and 2,662 ?rms. Robust standard errors in italic allow for arbitrary correlation within the same ?rm. * signi?cant at 10%,** signi?cant at 5%, *** signi?cant at 1%.
Table 2.16: Multinomial Logit Model: Predicted Probability of Covariates Owner and Worker Races (%)
Owner White Black Asian Hispanic White 74.11 65.66 60.92 61.00 Workers Black Asian 11.99 3.05 15.33 3.90 10.91 6.84 12.75 6.71 Hispanic 10.64 11.97 12.50 13.87
Note: Based on multinomial logit predictions of the race of new hires from previous table.
2.6.3 Workers’ earnings and analysis of results
We estimate the e?ects of owner type and coworker shares on workers’ compensation using a human capital approach. The dependent variable is the natural logarithm of workers’ real annual wages.50 The regression includes dummy variables for owner type, the share of similar coworkers, worker type, and other ?rm characteristics. Using wage estimates at the individual level, we can evaluate the impact of owners’ characteristics on wage di?erentials by using equation (2.4).
ln(wkjt ) = c + ?1 ? Ik + Xk ? B2 + Oj ? B3 + Ik ? Oj ? B4 +COWkj ? B5 + Ik ? COWkj ? B6 (2.4) +Oj ? COWkj ? B7 + Ik ? Oj ? COWkj ? B8 +Fj ? ? + Zkj ? Z + T ? Mt + µkjt
In order to approximate the individual’s full-year annual wage rate and thus reduce the importance of within-year labor supply decision, we include the additional information of whether the worker is a full quarter employee. That is, full quarter worker is an individual with positive earnings during all the quarters of the year. Controlling for full quarter workers allows us to make UI’s annual earnings comparable to CPS salary and wages. Abowd et al. [2002] have a discussion on the comparison between LEHD and CPS annualized wages. After controlling for dominant employer and full-time status, CPS and LEHD earnings data are more comparable. LEHD annualized wages are slightly higher than CPS’ annualized wages. However, when looking at our analysis we should keep in mind that an individual’s labor supply depends on both the duration and the average number of hours worked at the job.
50
78
where k identi?es information on the worker and j refers to information on the ?rm. wkj stands for worker k ’s log real annual earnings at ?rm j . Ik is a dummy variable for whether the worker is an immigrant. In an e?ort to establish how much the immigrant earnings di?erential is due to di?erences in predetermined personal characteristics, we add a vector Xk of employee characteristics including age, age squared, education, sex, and race. Oj is a vector of dummy variables for owner type (birthplace). COWkj stands for the proportion of immigrant coworkers in the ?rm (explained in section 2.4.5). The expected sign for ?1 is negative, assuming that immigrants earn lower wages, and its signi?cance would indicates whether there is substantial wage variation across the di?erent worker types. With the inclusion of owner type dummies, the estimate of ?1 will represent the di?erence in wages between immigrants and natives in native owned ?rms. The sum of ?1 and the B3 and B4 coe?cients corresponding to an immigrant owned ?rm will be positive if immigrant workers earn higher wages when working for immigrant-owned businesses than native workers in an immigrant ?rm. The coworker share accounts for the potential impact on wages of having better connections to similar types of workers in the ?rm. The interaction between COWkj and the vector of owner types is included to assess whether the e?ect of coworkers di?ers according to the type of employer that is hiring the employee. We explore a 3-way interaction among owner type, worker type, and the immigrant coworker share. In equation (2.4), B2 Xk absorbs the e?ects of variations in personal characteristics. We would expect estimates of ?1 and the vector B3 to change after including workers’ characteristics. We should be aware of the potential presence of omitted variable bias. Unob79
servable characteristics could bias estimated coe?cients in equation (2.4). Ignoring these unobservables could causes us to overestimate the impact of owner type and immigrant coworkers on individual earnings. High ability workers of type k should look for ?rms that pay higher earnings. If native-owned ?rms o?er higher wages and employ these high ability workers, the estimated model would not be capturing the e?ect of owner type on workers’ earnings; rather it would be capturing individuals’ ability to ?nd better jobs. Also, worker preferences and comparative advantage can in?uence the results. Variations in preferences for particular job characteristics across di?erent workers could provide an alternative explanation for both earnings di?erentials and sorting. To account for some of this variation, we include the fraction of workers in the ?rm with education lower than high school, equal to high school, higher than high school with some college, and equal to college or higher. The omitted category is college graduate. Characteristics of ?rms (Fj ) and of the local community (Zj ) are also included. These controls include the population share of each group in the local community, population density, ?rm’s size (log of reported employment), and legal form of organization. Mt are year dummies. The ?rst column of Table (2.17) shows results from a baseline model including immigrant status, individual age, education, and part-time status, but excluding other variables of interest. The table reports the betas estimated by equation (2.4). To make the analysis easier to interpret, we transform these unstandardized ? coef?cients with the usual formula [(e? ? 1) ? 100], so that we can analyze the percentage change in wages associated with a 1-unit change in a continuous independent pre80
dictor variable. In the case of a dichotomous independent variable, we interpret the percentage wage di?erence in the target category compared to the reference category. After controlling for typical human capital variables, full-time immigrant workers earn about 8% less than native workers (3,293 dollars less each year). In the Table (2.17), we progressively include covariates that control for ?rms and coworker shares. Column 1 shows the typical human capital analysis. Column 2 includes owner dummies and their interaction with worker type. Then, in column 3, immigrant coworker share is included and its interaction with worker type to see whether the e?ect of immigrant coworker share di?ers across worker types. Then, the interaction of immigrant coworker share and owner types are added to the regression (column 5). Finally, we include a 3-way interaction among immigrant coworker share, worker type, and owner type. Evaluating the variables at their means and sample distribution, we ?nd that, when working for native employers, the di?erence between native and immigrant wages increases to 11%. Meanwhile, immigrant workers earn 10% more than native workers in immigrant owned ?rms (4,398 dollar more each year). The human capital results in Table (2.17) are consistent with the literature. Age positively a?ects wages but at a decreasing rate. Education is signi?cant and positive. Part-time workers earn less than full-time workers. The inclusion of additional independent variables does not modify these patterns. After controlling for individual characteristics, immigrant workers are paid less than native workers in native ?rms, but they receive a signi?cantly higher wage than native workers when working for immigrant ?rms. The inclusion of the share of immigrant coworkers 81
Table 2.17: OLS Results: E?ect of Owner Type and Coworker Share on Log Real Annual Wages
Immigrant Age Age square (’) Education Partime Owner Mix Owner Immigrant Owner Mix*Immigrant Owner Immigrant*Immigrant Imm.Coworker Imm.Coworker*Immigrant Imm.Coworker*Oimm Imm.Coworker*Omix Imm.Coworker*Oimm* Immigrant Imm.Coworker*Omix* Immigrant Constant Year dummies 2-digit industry dummies Other controls R-Square Adjusted (1) -0.08*** 0.007 0.0806*** 0.0007 -0.080*** 0.000 0.506*** 0.0007 -2.1847*** 0.0044 (2) -0.1503*** 0.0069 0.080*** 0.0007 -0.080*** 0.000 0.506*** 0.0007 -2.1805*** 0.0044 0.1808*** 0.0128 -0.1191*** 0.0097 0.0054 0.0224 0.3205*** 0.0251 (3) -0.1205*** 0.0073 0.0803*** 0.0007 -0.080*** 0.000 0.511*** 0.0007 -2.1792*** 0.0044 0.1615*** 0.013 -0.1443*** 0.0101 -0.0007 0.02 0.3030*** 0.0153 -0.1398*** 0.0163 0.09*** 0.013 (4) -0.1171*** 0.0025 0.0802*** 0.0006 -0.080*** 0.000 0.504*** 0.0007 -2.1783*** 0.0036 0.1407*** 0.0481 -0.1495*** 0.017 0.1866 0.251 0.3174*** 0.017 -0.2457*** 0.0203 0.12*** 0.011 -0.0966*** 0.036 -0.095** 0.0582 (5) -0.1017*** 0.0008 0.0749*** 0.0008 -0.080*** 0.000 0.484*** 0.0007 -2.1240*** 0.0049 0.0936 0.0155 -0.1087*** 0.012 0.1368 0.243 0.3131*** 0.0252 -0.3797*** 0.0276 0.1520*** 0.02 -0.2925*** 0.0734 -0.1359 0.0513 0.6456*** 0.1285 -0.3285 0.641 10.615*** 0.235 yes yes yes 0.35
10.665*** 0.26 yes yes no 0.25
10.7015*** 0.262 yes yes no 0.27
10.686*** 0.263 yes yes no 0.28
10.653*** 0.241 yes yes no 0.31
Note: The number of observations includes 214,398 workers. Standard Errors are Huber-White robust standard errors, corrected by ?rm clustering. Reference group are full time native workers in native ?rms. (+) Neighborhood is de?ned as the contiguous counties to the county where the ?rm is located. Population in 100,000’s. (’) Age ? 102 . ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
produces interesting results. Immigrants earn more when working for immigrant employers and when the immigrant coworker share increases. The opposite is true
82
for native workers. In general, a native worker receives higher wages if he or she works for a native ?rm with a low share of immigrant workers. These results are striking in two senses. One, the ability to look at individual wages and identify the types of ?rm owners is only possible with this database. We have individual earnings for each ?rm. Although immigrants are paid less on average, they ?nd themselves in a better position when working for immigrant ?rms. Second, we can look at the entire workforce and identify each individuals’ types of coworkers in the ?rm. This allows us to make inference on the impact of social ties on worker wages.
2.7 Conclusions
This paper takes advantage of unique employee and employer matched microdata from the U.S. Census Bureau to examine the e?ect of owner types and coworker types on ?rms’ hiring patterns and workers’ earnings. Particular attention was paid to the birthplace of employers and to the share of similar coworkers (by birthplace and ethnicity) at ?rms when new workers are hired. We examined the e?ect of those variables on hiring rates and on the wage di?erential between immigrants and natives. In general, employees’ wages are a?ected by the type of owner of the ?rm. For native employees working for immigrant owners the e?ect is very interesting. Natives are paid lower when working for immigrant employers, and in these ?rms natives have lower average earnings than immigrants. One explanation for these ?ndings
83
is that immigrant bosses have a better understanding of and networking with the immigrant community, and therefore can ?nd and contract immigrant workers more easily than native-owned ?rms. Why can’t native-owned ?rms quickly adjust and ?nd this cheaper labor? Lack of language knowledge and lack of networking make it harder for native bosses to ?nd immigrant workers. These ?ndings justify further analysis of di?erences in contracting ability across employers. The evidence that the type of owner matters for wage di?erentials among workers also implies an important role for owner type on personnel policy. In addition to examining the e?ect of owners and coworkers on di?erences between immigrants and natives, we evaluate the e?ect of owner and coworker types on ethnically(racially) di?erent groups. An individual’s race is an important source of variation across workers and owners. The evidence suggests that employers tend to hire workers from the same ethnic group. A signi?cant impact of similar coworkers in the hiring process is observed across all types of owners, even after controlling for ?rm ?xed e?ects. Immigrant owners tend to hire more Hispanics and Asians, while native owners hire more blacks and whites. By shedding light on the ways workers and employers interact in the labor market to a?ect job and wage outcomes, this research makes a contribution to the sociology, labor economics, and demography literatures. It also opens up numerous avenues for future research. On the microeconomic side, we can further evaluate job ?ows and wage pro?les of workers inside di?erent types of ?rms. The analysis of assimilation can also take advantage of the results presented here, to further our understanding of the adjustment process of new immigrant workers. The empir84
ical analysis in this paper makes some progress toward mitigating biases of skill sorting. This paper controls for a broad number of observable characteristics that try to capture other explanations for segregation. However, if owner unobservable characteristics are correlated with worker characteristics, the results of the analysis would be biased. Di?erent empirical approaches such as instrumental variables or owner ?xed-e?ects could be good options in future research, although this would demand a more exhaustive matched database that follows workers after leaving the ?rm and ?rms after ownership changes. Narrowing the scope of the analysis by looking at one industry could also provide information on the costs and bene?ts of ?rm recruitment processes. For instance, we could examine with more detail the e?ect of worker type concentration on ?rms’ labor productivity. On an aggregate view, we can evaluate the e?ect of large ?ows of immigrants on the economy with the combined analysis of push and pull factors. Immigrant ?rms and immigrant workers seem to match quickly in the labor market. The analysis of the impact of immigration on unemployment and aggregate vacancies in the labor market can be extended to incorporate the ?ndings in this paper.
85
Chapter 3 Workplace Concentration of Immigrants 3.1 Introduction1
Over the last several decades, labor markets in many U.S. cities have absorbed large in?ows of new immigrants. The size of these ?ows has generated intense interest in their e?ects on the employment and wages of natives, as well as in the extent to which new immigrants have assimilated into the U.S. economy. New immigrants ?nd employment and accumulate location-speci?c skills and work experience, gradually becoming integrated into local economies and potentially changing them in substantial ways. While outcomes of this process have been the subject of much research, less is known about the process itself. Which businesses hire immigrants? To what extent do immigrants work with natives? How does these patterns change as immigrants accumulate U.S. speci?c skills? Do the characteristics of di?erent immigrant groups and di?erent geographic labor markets a?ect the way in which assimilation plays out? A lack of suitable data has limited economists’ ability to address these questions. Our contribution is to bring to bear a rich set of matched employer-employee data that allows us to identify immigrants, their coworkers, and their employers. Our unique data permit quantifying the extent of and covariates of the workplace
1 This chapter draws heavily on a joint paper with John Haltiwanger, Kristin McCue, Seth Sanders and Fredrik Andersson with the same title.
86
concentration of immigrants. The paper has two broad objectives. The ?rst is primarily descriptive. The descriptive ?ndings show that immigrants are much more likely to have immigrant coworkers than are natives. This pattern is driven partly by the geographic concentration of immigrants, but the patterns hold true even within local labor markets. At the same time, most immigrants do have native coworkers: only a small share work in immigrant-only workplaces. The concentration of immigrants is higher for recent immigrants and, conditional on recent arrival, for older immigrants: part of the assimilation process is a movement towards more interaction with natives in the workplace, and younger immigrants are more likely to work with natives. We ?nd large di?erences associated with ?rm size: concentration is much higher in smaller ?rms, but is far from zero even in the largest ?rms. We also ?nd substantial variation in the extent of immigrant concentration across industries even after controlling for a detailed set of location, employer and employee characteristics. Second, our ?nding that the allocation of immigrants across workplaces is far from random raises the question: what does drive this workplace concentration? Both the existing literature and our descriptive ?ndings suggest that it is important to consider how businesses hire their employees and the choices that businesses make about the skill mix of their workforce. One relevant issue here is the role that language skills play in governing interactions among employees and between employees and customers. A second issue is the role of social networks in the process that matches workers and ?rms. A third issue is human capital - the sorting and concentration of immigrants in the workplace may re?ect sorting by skills. In the 87
second part of the paper, we explore the role of these factors. We ?nd evidence that immigrants with primarily immigrant coworkers are likely to have coworkers who live in the same residential tract. This pattern is robust to the inclusion of controls for other closely related factors such as residential segregation. We also ?nd evidence that immigrant workers with poor English speaking ability and low education are more likely to work with immigrant coworkers. The paper proceeds as follows. Section 3.2 provides an overview of the relevant theoretical and empirical literature that helps guide our empirical analysis. Section 3.3 describes the measurement of immigrant concentration, the matched employer-employee data we use in our analysis and the methods we use to explore the correlates of immigrant concentration. In section 3.4 we present our main results quantifying the extent and nature of immigrant concentration across workers and businesses. Section 3.5 analyzes the impact of factors such as social networks, language skills and human capital on the patterns of immigrant concentration. Most of the analysis focuses on native born, recent immigrants and established immigrants without speci?c reference to country of origin. Section 3.6 extends the analysis in terms of the basic patterns of concentration by country of origin. Concluding remarks are provided in section 3.7.
88
3.2 Background 3.2.1 Literature on earnings di?erences
Work examining earnings di?erences between whites and other groups in the U.S. has largely focused on netting out di?erences in skill (often captured by education and labor market experience) and geography (often using place of residence and urban residence) to assess the potential role of discrimination in labor market outcomes. This assumes that earnings di?erences are generated either by di?ering worker characteristics or di?ering returns to those characteristics. By extension, closing gaps in earnings requires equalizing worker characteristics and their return across groups. Di?erences in returns to characteristics are assumed to re?ect unobserved ways in which the wage generating process di?ers and is typically viewed as an upper bound on the potential for discrimination to play a role in explaining wage disparities. A huge number of papers use this approach; some classic examples that examine earnings di?erences relative to white men are Smith and Welch [1977] for African American men, Borjas [1982] for Hispanic men, Chiswick [1983] for Asian men, and Corcoran et al. [1983] for women. There is also a large literature assessing the sources of earnings di?erences between immigrants and native born workers (for example, Chiswick [1978], or Butcher and DiNardo [2002]). These papers generally augment the basic human capital framework used in the studies above by allowing for skill di?erences that are speci?cally relevant to immigrants. These include potential di?erences in the value of education and work experience accumulated outside the U.S., and the importance 89
of di?erences in English language skills. Immigrant assimilation into the U.S. labor market is viewed as occurring through a narrowing of the earnings gap, resulting largely from increased U.S.-speci?c skills with time spent in the U.S. While there is debate over the speed at which the earnings gap between immigrant and native born workers closes, most studies ?nd a substantial narrowing with time spent in the U.S. (see Chiswick [1978] and Borjas [1985]). An older literature in sociology and economics stresses that earnings di?erences between groups may be driven by the characteristics of the ?rms that employ the majority and minority groups, rather than solely by human capital characteristics. Usually termed ’dual labor market theory,’ this idea gained considerable attention in the late 1960s and early 1970s (see for example Averitt [1968] or Galbraith [1971]). According to this theory, many ?rms (especially industrial ?rms) are not governed by competitive processes. Instead, these ?rms enjoy market power. They insulate themselves and stabilize their workforce through job training and promotional ladders (Edwards [1972]). Firms that are constrained by competition do not invest in work skills and are characterized by low wages and high turnover, with low returns to human capital including job tenure. The existence of ’good jobs’ and ’bad jobs’ by itself would not imply an earnings disadvantage to minority workers. Sociologists typically rely on a form of employer discrimination to explain why dual labor markets lead to minority disadvantage. Queuing theory suggests that good jobs always have an excess supply of applicants and ?rms then order workers by preferences and hire down the queue until vacancies are ?lled. If race or ethnicity plays a role in this ordering, a higher 90
fraction of minority workers will be employed in the secondary market and have relatively low wages and wage growth. While dual labor market theory per se has largely fallen out of the mainstream literature in economics and sociology, a newer literature that similarly argues that ?rm characteristics may be partially responsible for the level and growth in earnings of workers has gained growing acceptance. Wages appear to be positively correlated with ?rm productivity and ?rm size (Abowd et al. [2005]). While more controversial, there is some evidence that ?rm-level technological adoption also affects workers’ wages (Dunne et al. [2004]). Lengermann [2002] ?nds that coworker characteristics, in addition to ?rm characteristics, may a?ect wages. Speci?cally, he ?nds that having more skilled coworkers independently raises a worker’s wages. If ?rm characteristics play a major role in wage setting, then understanding how race and ethnicity a?ect the matching of workers to ?rms becomes important for understanding wage disparities across groups. Lengermann et al. [2004] explore the issues of sorting of immigrants across ?rms and ?nd that sorting matters for wage di?erences between native born and immigrant workers.2 We now turn to theories of worker segregation with special attention to how immigrants sort into ?rms.
3.2.2 Literature on segregation
Four broad overlapping theories explain segregation of workers into ?rms. These theories focus on sorting based on (a) productive characteristics, (b) prefSome of our basic ?ndings on immigrant concentration are also found in Lengermann et al. [2004]. Using the same data infrastructure that we use in this paper, they ?nd for example di?erences in immigrant concentration by industry and employer size.
2
91
erences of workers or employers, (c) information available to workers or employers, or (d) cost of commuting to jobs. Some, but not all, of these theories imply that segregation results in a disadvantage for one group of workers relative to another. There is substantial evidence of segregation by skill. For example, Kremer and Maskin [1996] look at the sorting of high and low skilled workers into ?rms over time and across three countries, the U.S., Britain and France. They ?nd a high and rising correlation between worker skill levels in ?rms over the 1970s and 1980s. This may occur either because a ?rm demands a particular type of worker (for example skilled workers) or because coordination within a ?rm demands that workers share a common characteristic such as a common language. Cabrales et al. [2008] emphasize a di?erent skill-based mechanism: if a worker’s utility is a function of both absolute wages and their wages relative to those of coworkers, and if movement of workers across ?rms is costless, complete segregation of workers by skill is optimal. A mixed-skill workforce generates wage inequality within a ?rm, reducing worker utility. All workers are made better o? by grouping workers with similar skills and avoiding these reference group costs. Regardless of the mechanism, segregation by skill will cause immigrant-native di?erences in the distribution of skill to contribute to segregation. For example, immigrants are both much more likely than natives to have an 8th grade education or less (23% vs. 5.2% for natives in the 2000 census), and also more likely to have an advanced degree (10.3% vs. 8.6% for natives). Therefore, ?rms that specialize in hiring exclusively low-skilled or exclusively highskilled workers will tend to have a workforce that has a higher fraction of immigrants than the fraction in the population. 92
Language di?erences provide another productivity-based motivation for segregation. If working with someone who does not speak the same language generates transaction costs, employers may increase productivity by hiring only workers who share a common language. In this case, immigrants from non-English speaking countries may be particularly likely to be segregated, and may also be particularly likely to work with their compatriots rather than other immigrants. Lang [1986] develops a formal model of wage di?erences arising because of the costs to ?rms of having to pay a premium for bilingual workers who can bridge the language barrier. One of the results of this model is that complete segregation would occur if both capital and labor were owned by each language group. Hellerstein and Neumark [2003] ?nd evidence that Hispanics with poor English-language skills are particularly likely to work with other Hispanics. Their data do not allow them to examine how much of this is due to Hispanic workers working for Hispanic-owned ?rms as in the Lang model. Becker [1957] is the classic model of preference-based segregation. In this model, segregation of workers by race occurs as the result of discriminatory preferences on the part of co-workers. White workers would demand a premium to work with black workers. In response, ?rms segregate workers into separate facilities, avoiding the need to pay a wage premium to discriminating white workers. Depending on conditions including the relative size of the minority and majority group, the number of ?rms, and returns to scale in production, segregation may be extreme but with limited disadvantage in wages to the minority group. Dual labor market theory, described above, also generates wage di?erences across groups 93
if discriminating employers put minority job candidates lower down the queue. In this case, higher wages in the primary sector ensure that a higher fraction of the majority group works in the primary sector and hence gives a wage advantage to the majority group. Information-based theories concentrate on the mechanisms that workers use to ?nd jobs. For example, ?rm use of employee referrals to ?ll jobs may contribute to workplace segregation. For workers, use of personal contacts to search for jobs is inexpensive and has relatively high rates of success (Holzer [1988]). For employers, employee referrals provide both a low cost recruitment strategy and, on average, new hires with higher productivity and lower turnover rates (Holzer [1987]; Montgomery [1991]). If workers tend to refer others who have similar characteristics, use of referrals can increase rates of segregation. Elliot [2001] ?nds that recent Latino immigrants are more likely than blacks or Latino natives to use personal contacts to ?nd jobs. Weak English skills explain much of this di?erence. A greater reliance on referrals in small workplaces in combination with a concentration of recent immigrants in small ?rms also contributes to the di?erence. Information ?ows may combine with residential segregation to contribute to workplace segregation. Neighborhoods play an important role in who you know and hence may provide important job contacts and references. Several papers have established that workers in the same ?rm are disproportionately from the same neighborhoods. Using data from Boston, Bayer et al. [2008] ?nd that a worker is about one-third more likely to work with someone who lives in the same census block as to work with someone who lives in other blocks in their block group (typically 94
eight or so contiguous blocks). This comparison of blocks to block groups provides important evidence that having coworkers who are neighbors does not stem from unobserved factors such as transportation routes or distance that make a place of employment a natural place to work for those living in a particular location. Many of these unobserved factors would be similar for a block group and block of residence, and so should have similar e?ects on the likelihood of working with more or less immediate neighbors. This paper is limited in that the exact establishment can not be observed, while sample sizes as well as the ethnic make-up of Boston restrict the authors’ investigation to black-white di?erences. Hellerstein et al. [2008a] also present evidence of neighborhood network e?ects. Using matched employer-employee data, they compare how likely an individual is to work in the same establishment as his neighbor, relative to the likelihood that this would result if their employer hired workers randomly from the geographic areas of residence of all individuals who work in the employer’s census tract. Their dataset is large enough to disaggregate the analysis for whites, blacks and Hispanics. They ?nd that another worker living in the same census tract has twice the probability of working in your ?rm than what one would expect from randomness. They do not investigate the importance of other mechanisms for sorting workers into ?rms. A ?nal theory of the sorting of workers into ?rms also works through residential segregation but focuses on the fact that not all jobs are equally accessible from di?erent places of residence. Kain [1968] investigated employment patterns of blacks and whites in Chicago and Detroit. He found that blacks were unlikely to be employed in areas that were predominantly white, that blacks would have higher 95
employment rates if housing segregation was lower, and that the movement of jobs from central cities to suburban areas depressed the employment prospects of blacks. A number of other studies followed that compared employment di?erences between central city and suburban residents within an urban area. These tests often found employment prospects lower for central city residents, but controlling for unmeasured skill di?erences between residents of di?erent locations remained an issue in inference. A recent study by Hellerstein et al. [2008b] questions the interpretation that a lack of jobs near where blacks live is a major source of racial employment di?erences. They ?nd that the employment prospects of black residents are positively correlated with the number of nearby jobs in which blacks work, but not with the number of nearby jobs in which whites work. This indicates that even within close geographic proximity, job markets are racially segregated. They conclude that spatial mismatch has little e?ect on employment prospects of blacks but that what they term racial mismatch—few nearby jobs that employ blacks—has a large e?ect. Clearly, residential segregation could contribute to workplace segregation of immigrants. There is ample evidence that immigrants’ places of residence are spatially concentrated. Iceland [2009] describes the high level of residential segregation in the U.S. among immigrant groups but also shows that immigrants migrate to neighborhoods that are more ethnically integrated as they spend more time in the U.S. However, Porter and Wilson [1980] argue that, unlike for black Americans, residential segregation may aid immigrants—especially new immigrants–while also leading to segregation of workers in ?rms. Studying the post-Castro immigration from Cuba to Miami, Portes and Wilson show that not only do Cubans in the U.S. 96
work together, many work in ?rms owned by other Cubans. Moreover, Cuban employees of Cuban-owned ?rms tended to display the same patterns of wage growth and returns to human capital as workers in ?rms classi?ed as in the ’primary sector,’de?ned as ?rms with a promotion ladder, over 1000 workers, and high average wages. While an impressive source of employment, it is not clear that the example of Cubans generalizes to other foreign-born groups. Capital owners speci?cally were forced to leave Cuba, which may have led to higher levels of capital with which to start businesses and more experience with small businesses among Cubans than among other foreign born groups. Having said this, Wilson and Portes report that much of the capital used to start these businesses was accumulated in the U.S. and not transferred from Cuban concerns.
3.3 Methodology and Data 3.3.1 Measuring immigrant concentration
We follow several recent papers that study workplace segregation (Hellerstein and Neumark [2007]; Aslund and Skans [2005a], Aslund and Skans [2005b]— henceforth HN and AS) by using the share of coworkers in a particular group as a measure of exposure. That is, we exclude the worker himself when measuring the concentration of immigrants in the business he works in. For worker i, employed by business j which has sj employees, the share of immigrants among coworkers is:
sj
1 Cij = sj ? 1 97
Ik
k =i
(3.1)
where Ik is an indicator for whether or not worker k is an immigrant. For the sake of brevity, we will refer to this simply as the coworker share. As pointed out by these authors, excluding the worker’s own characteristic in calculating concentration ensures that in large samples the coworker share for both immigrants and natives should on average equal the share of immigrants in the workforce in the absence of any systematic concentration. Based on this, we use the di?erence between the mean coworker share for immigrants and natives as a measure of immigrant concentration. A positive value indicates that immigrants are more concentrated than would be expected based on random allocation. At the extreme, if immigrants worked only with immigrants and natives with natives, the di?erence in coworker means would equal one. A negative value for this di?erence would indicate that immigrants were more likely to work with natives than would be expected based on random allocation—a pattern that could arise where the two groups provide di?erent but complementary skills. We depart from the approach of these authors in two ways: in the way in which we condition on observable characteristics, and in choice of a normalization to gauge whether the concentration we ?nd is large relative to some alternative. There are two types of questions that can be addressed by conditioning on observable characteristics in studying segregation: to what extent can segregation be explained by di?erences in the characteristics of the two groups, and which characteristics are most associated with segregation. HN and AS both focus more on the ?rst issue, while we explore some aspects of both questions. As an example to provide some context, the immigrant and native education distributions di?er, and particular em98
ployers may hire primarily from one part of the education distribution, leading to concentration of immigrants because of di?erences in skill. HN and AS both use the di?erence between measured concentration and the amount of concentration that would be generated solely by the way in which education is distributed across employers as their conditional measure of concentration. In contrast, we condition on a worker’s own characteristics and on the characteristics of his or her employer (e.g. employer size and industry), but do not directly condition on coworker characteristics. Our measure of concentration is the mean di?erence between immigrants and natives with the same characteristics. We take a di?erent approach in part because the worker characteristics in our data that vary within employer (age, gender) do not di?er dramatically between immigrants and natives, and they also turn out not to have a strong correlation with immigrant concentration. Controlling for a worker’s own characteristics should remove the e?ects of age and gender from the measured di?erence in coworker mean, and the estimated coe?cients allow us to examine the characteristics of immigrants and natives who work in heavily immigrant workplaces. Both HN and AS normalize their measures of concentration, though they choose di?erent references for the normalization. While both of their normalizations have intuitive appeal, we take a di?erent approach. We use the immigrant-native di?erence in coworker shares as our measure of concentration, but in most cases also present information on the coworker share for natives as a point of reference. Our regression approach makes doing so straightforward, and also allows us to more directly illustrate patterns of concentration. For example, using the regressions to 99
predict means for a given set of covariates allows us to illustrate the strong positive relationship between immigrant concentration and immigrant share of the workforce, when looking across groups de?ned by characteristics such as area of residence and employer size. In addition, the regression approach using our coworker index at the person level as the dependent variable permits us to normalize our measure of concentration e?ectively along a number of dimensions. For example, HN normalize to control for between MSA di?erences in various groups (e.g., di?erences in the distribution of blacks and whites across MSAs). We control for such di?erences directly in our regression approach by, for example, including controls for MSAs.
3.3.2 Data
We use the data from the Longitudinal Employer - Household Dynamics (LEHD) database, which draws much of its data from complete sets of unemployment insurance (UI) earnings records for a subset of U.S. states. The database includes records for 1990 to 2004, though some states only have data for a subset of those years. The workers’ earnings records have also been matched to characteristics of their employer gathered in quarterly administrative reports and through Census Bureau business censuses and surveys. Basic demographic data are also available for workers, including place of birth. For those born outside the U.S. (and its territories), we treat the year in which they ?rst applied for a Social Security Number (SSN) as the date of their arrival. While this may not precisely date arrival, preliminary results based on a sample of immigrants for whom both LEHD and decennial
100
population census data are available suggest that the year the individual ?rst applied for a social security number proxies the reported year of arrival fairly well.3 In the current analysis, we use data from selected metropolitan areas in 11 states. While we do not use a large number of states, our sample does include ?ve of the six states that had immigrant populations of 1 million or more. These data give us two unique advantages. First, we have earnings for a group large enough to include millions of immigrants. Second, we can observe the ?rms in which workers are employed, allowing us to measure both employer characteristics and the characteristics of coworkers. These data have other advantages that we do not exploit here but plan to in future work: for example, the data can be used to generate a panle on both employers and employees that would allow us to track earnings of immigrants over time in the U.S. as well as to observe contemporaneous changes for native-born workers. The main disadvantage of these data for studying immigration is that they include only on-the-books employees and so do not cover the self-employed or those working in the informal sector. Thus they likely have poor coverage of undocumented immigrants. Coverage of employment in agriculture is incomplete in the LEHD data, so we exclude employers in that sector. Calculating the share of coworkers who are immigrants requires at least one
Here we use year of arrival only to split immigrants between those arriving very recently (within the last 5 years) and other immigrants. Comparing our classi?cation based on date of SSN application to one based on responses in the 2000 census, 92% of immigrants are classi?ed in the same way according to both sources. Among those for whom the classi?cation di?ers, the most common pattern is that 4% of Mexicans are considered new immigrants in Decennial Census versus 10% in LEHD. The lag in the registration process by immigrants, specially in the case of Mexicans, explains these di?erences. The patterns by age are very similar between LEHD and Decennial Census, however younger immigrant workers are also reporting a small lag in their application for social security.
3
101
coworker, so we restrict our sample to businesses with at least two employees.4 We measure concentration using a cross-section of data based primarily on the second quarter of 2000, but we use LEHD data for the 1995-2000 period to de?ne business age. In computing the coworker share, we use all coworkers, whether or not they hold other jobs. However, the set of observations used in our regressions includes only one job for each individual: the job where they received their highest earnings in that quarter. We draw data from employers in 31 MSAs. We include all MSAs that have substantial foreign-born populations and are in states for which we have the required data, but we also included several smaller MSAs that experienced very rapid growth in foreign-born residents between 1990 and 2000.5 Even in the smallest of our MSAs we have data on more than 30,000 immigrant workers, so small sample sizes are never an issue. Table 3.1 summarizes the across-MSA variation in immigrant shares for our sample of MSAs. In the average MSA in our dataset, 18.9% of workers are immigrants. In what follows, we are interested in deviations in workplace shares from the overall-average. Clearly the substantial variation in immigrant share across MSAs will contribute to ?nding immigrant concentration. The shares of both recent and
Immigrants account for 27% of employment in single-employee businesses, and 16% of employment in businesses with more than one employee. 5 More precisely, we started from the list of MSAs used in Singer [2004], which included all MSAs with at least 1 million residents in 2000, and meeting at least one of the following criterion: (i) at least 200,000 foreign-born residents, (ii) a foreign-born share higher than the 2000 national average (11.1%), (iii) 1990-2000 growth rate of the foreign-born population above the national growth rate (57.4%), or (iv) above national average percentage foreign-born in 1900-1930 (‘’former gateways”). We drop 14 of Singer’s 45 MSAs because we do not currently have access to all of the data we need from the relevant states.
4
102
Table 3.1: Variation in Immigrant Share of Workforce across Sample MSAs Percent Immigrant Recent Established 3.40 15.46 1.85 8.57 1.94 8.52 2.92 13.54 4.37 22.82 6.03 27.23
Mean Standard Deviation P25 Median P75 P90
Total 18.86 10.27 10.57 16.26 26.60 32.58
Source: Authors calculations based on LEHD UI-ES202 database. Note: Unit of observation is an MSA. Immigrant shares are measured as of the second quarter of 2000, and recent immigrants are those arriving between 1995 and 2000. The table presents fuzzed percentile values.
established immigrants vary substantially across MSAs as well. For roughly 10% of workers in our sample, we match in additional information on educational attainment and English language skills from the long form of the 2000 population census. Using propensity score models, we develop weights for the matched sample that allow us to closely replicate our results based on the overall sample.6 We then use weighted estimation with the matched sample to examine the relationship between these measures of skill and immigrant concentration.
3.3.3 Regression speci?cations
Our primary empirical approach is to run a series of regressions with the coworker share as the dependent variable, and individual workers on their primary job as the unit of analysis. As a rough way to capture the way in which immigrant
The variables used in the propensity score procedure were: worker age, sex, country of origin (11 groups=Mexico, China, Cuba, El Salvador, India, Korea, Japan, Vietnam, Phillipines, other country of origin groups, and natives), log earnings, worker status, industry (4 digits), Msa indicator variables and population density, plant age and size, and ?rm’s # of establishments.
6
103
concentration changes with time spent in the U.S., we include indicators for whether an individual is a recent immigrant (RI, de?ned as arriving in the last 5 years), or a more established immigrant (EI, arriving more than 5 years ago). Since we use a cross-section of data, the di?erences between recent and more established immigrants confound the e?ects of time spent in the U.S. with changes in labor markets and in immigrant and native characteristics over time. We would need to exploit the panel aspect of our database to seriously address the a?ects of assimilation, but believe this is useful as a starting point that provides suggestive evidence on whether assimilation e?ects on concentration are likely to be important. Our initial regression speci?cation is:
Cij = ?N + ?EI EIi + ?RI RIi + ?xij +
ij
(3.2)
where (again) i denotes an individual and j denotes a workplace. Here, the constant term (?N ) represents the mean coworker share for the omitted category, which in our simplest speci?cation consists simply of natives. Coe?cients ?EI and ?RI give us estimates of the di?erences between immigrants and natives in how likely they are to have immigrant coworkers. We use controls for MSA and for various worker and employer characteristics to examine the extent to which immigrant concentration can be accounted for by di?erences between natives and immigrants in their geographic distribution and in worker and job characteristics. In section 3.6, we de?ne coworker shares for speci?c countries of origin and look at which immigrants are most likely to work together.
104
Speci?cation (3.2) assumes that the e?ects of covariates are the same for immigrants and natives. To examine whether this in fact holds, we use an alternative speci?cation that includes interactions between our immigrant dummy variables and other covariates:
Cij = ?N + ?EI EIi + ?RI RIi + ?xij + ?EI EIi ? xij + ?RI RIi ? xij +
(3.3)
Once we add interaction terms, the intercept rarely identi?es e?ects for a group of particular interest. To illustrate the e?ects of a particular covariate in speci?cations of form 3.3, we present predicted means for immigrants and natives, by which we evaluate di?erences between immigrants and natives based on the pooled distribution of the variables in x. To ease computations with our 36 million records, we use linear regression models rather than adopting an approach that accounts for the limited range of the dependent variable. In this draft, we also ignore the e?ect of clustering within employer in estimating the standard errors. For most of our speci?cations, the dependent variable mean is not close to either 0 or 1, which mitigates some of the problems inherent in the linear model. The strong positive correlation in the coworker share among employees of the same business will lead to a downward bias in our estimated standard errors in all worker-level regressions. Given the huge size of our sample, the results we present would generally remain signi?cant at standard levels even if the corrected standard errors were 100 times larger. The few exceptions 105
(in Table 3.6) are estimates that are too small to be meaningfully di?erent from zero anyway.
7
3.3.4 Descriptive statistics
Table 3.2 presents summary statistics for immigrant and native workers in our full sample. The ?rst row gives coworker shares for the three groups. For the average native, about 15% of coworkers are immigrants, while 42% of the coworkers of recent immigrants are fellow immigrants, and 36% of the coworkers of established immigrants are immigrants. The immigrant-native di?erence in coworker means—our measure of concentration—is .272 for recent immigrants and .214 for more established immigrants, indicating substantial concentration. Table 3.2: Characteristics of Immigrant and Native Workers, Full Sample
Immigrants Recent Established 42.1 36.3 43.6 35.6 20.8 56.8 1.1 36.2 37.0 25.7 8.5 23.6 19.7 33.2 47.0 56.4 14.7 49.6 24.8 10.9 9.0 22.6 Natives 14.9 29.3 30.0 40.7 51.7 . . . . 8.0 23.5
Coworker share Age Age
An ethnic enclave is a physical space with high ethnic concentration; thus these spaces are culturally distinct from the larger receiving society. Ethnic enclaves are found in virtually every country, arising in response to increased immigration of people from the same ethnic background.
ABSTRACT
Title of dissertation:
FIRM OWNERS AND WORKERS: AN ANALYSIS OF IMMIGRANTS AND ETHNIC CONCENTRATION M´ onica Garc´ ?a-P´ erez, Doctor of Philosophy, 2009
Dissertation directed by:
Professor John Haltiwanger Department of Economics
This dissertation consists of three chapters examining the important role of ?rm and coworker characteristics, as well as the use of social networks, in labor markets. The ?rst paper investigates the e?ect of ?rm owners and coworkers on hiring patterns and wages. Immigrant-owned ?rms are more likely to hire immigrant workers. This prevalence is especially strong for Hispanic and Asian workers. We also ?nd that the probability that a new hire is a Hispanic is higher for immigrant ?rms. On wage di?erentials, the results illustrate that much of the di?erence between the log annual wages of immigrants and natives can be explained by immigrants’ propensity to work in non-native owned ?rms, which pay the lowest average wages. Interestingly, though, native workers holding a job in immigrant ?rms are paid less than immigrant workers. The last section examines the potential mechanisms for these ?ndings. It explores the importance of job referral and use of networks for migrants in labor markets. We consider the theoretical implications of social ties between owners and workers in this context. Firms decide whether to ?ll their vacancies by posting their o?ers or by using their current workers’ connections.
Next, we explore the patterns of immigrant concentration relative to native workers at the establishment level in a sample of metropolitan areas. Immigrants are much more likely to have immigrant coworkers than are natives, and are particularly likely to work with others from the same country of origin, even within local markets. The concentration of immigrants is higher for recent immigrants and interestingly for older immigrants. We ?nd large di?erences associated with establishment size that cannot be explained solely by statistical aggregation. Exploring the mechanisms that underlie these patterns, we ?nd that proxies for the role of social networks, as well as the importance of language skills in the production process, are important correlates of immigrant concentration in the workplace.
FIRM OWNERS AND WORKERS: AN ANALYSIS OF IMMIGRANTS AND ETHNIC CONCENTRATION
by M´ onica Isabel Garc´ ?a-P´ erez
Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial ful?llment of the requirements for the degree of Doctor of Philosophy 2009
Advisory Committee: Professor John Haltiwanger, Chair/Advisor Professor John Shea Professor Seth Sanders Professor Judith Hellerstein Professor Howard Leathers
c Copyright by M´ onica Isabel Garc´ ?a-P´ erez 2009
Dedication To Chavela and Nene . . . This thesis is dedicated to my wonderful parents, Isabel y Luis, who have raised me to be the person I am today and sacri?ced a lot to o?er me the means to reach my dreams. You have been with me every step of the way, through good times and bad. Thank you for all the unconditional love, guidance, and support that you have always given me, helping me to succeed and instilling in me the con?dence that I am capable of doing anything I put my mind to. ¡Gracias, los amo!
ii
Acknowledgments
It is a pleasure to thank to many people who made this thesis possible. I owe my gratitude to all of them and because of whom my graduate experience has been one that I will cherish forever. I cannot overemphasize my gratitude and love to my husband Darin, who did more than his share around the house as I sat at the computer. Without his support, and gentle prodding, I would still be trying to write the introduction of my ?rst paper. With his enthusiasm, his patience, his revisions and his inspiration, he helped me to overcome the hardest moments of the creative process. I would have been lost without him. Thank you for your love amor. I would like to express my deep and sincere gratitude to my advisor, Prof. John Haltiwanger for his valuable advice and great encouragement as well as for his excellent guidance and assistance for this research. His wide knowledge and his logical way of thinking have been of great value for me. I’d like to thank him for giving me an invaluable opportunity to work on challenging and extremely interesting projects over the past three years. It has been a pleasure to work with and learn from such an extraordinary individual. I am deeply grateful to Prof. John Shea for o?ering advises in di?cult moments and always have the door open when I needed him. I am also indebted for the amount of time and e?ort he has spent reading and correcting my work. I also wish to express my warm thanks to Professor Seth Sanders who inspired me with his questions and comments. His discussions around my work and initial explorations iii
have been very helpful for this study. My most sincere gratitude to Kristin Sandusky for being next to me and kindly grants me her time for answering my unintelligent questions during the time of this research. I am much indebted to Kristin for her valuable insights on the data details and her emails with great tips. I also thank to Dr. Jose Tessada, Dr. Jeanne LaFortune, Prof. Judy Hellerstein, and the participants in brownbag seminars in the University of Maryland. I am also grateful to my many student and work colleagues for providing a stimulating and fun environment in which to learn and grow. To Helena Schweiger for sharing great times at the beginning of my career and o?ering me a lot of support. I would also like to acknowledge help and support from some of the sta? members, especially to Vickie Fletcher, Elizabeth Martinez, and Terry Davis, who provided me with all the help and advice in all the administrative process during the
A career. I am thankful to Dorothea Brosious for providing the L TEX thesis template.
Finally, I would like to extent my gratitude to Jeremy Wu, Fredrik Andersson and all the LEHD sta? at the Census Bureau for their kind support. I owe my deepest thanks to my family. Words cannot express my gratitude.
iv
Disclaimer
This work is uno?cial and thus has not undergone the review accorded to o?cial Census Bureau publications. The views expressed in the paper are those of the authors and not necessarily those of the U.S. Census Bureau or the U.S. Department of the Treasury. All papers are screened to ensure that they do not disclose con?dential information. Persons who wish to obtain a copy of the paper, submit comments about the paper, or obtain general information about the series should contact Sang V. Nguyen, Editor, Discussion Papers, Center for Economic Studies, Bureau of the Census, 4600 Silver Hill Road, 2K132F, Washington, DC 20233, (301-763-1882) or internet address [email protected].
v
Table of Contents
List of Tables
viii
List of Figures
xi
List of Abbreviations
xiii
1 Introduction
1
2 Does It Matter Who I Work For And Who I Work With? The Impact Of Owners And Coworkers On Wages And Hiring 2.1 2.2 2.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 8
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 On the use of social networks . . . . . . . . . . . . . . . . . . . . . . 18 2.3.1 Small ?rms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4
Data and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Construction of ex post weights . . . . . . . . . . . . . . . . . 30 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Measuring coworker share . . . . . . . . . . . . . . . . . . . . 42
2.5
Analysis of New Hires, Earnings of Workers and Skill Distribution . . 42 2.5.1 2.5.2 New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Earnings of Workers . . . . . . . . . . . . . . . . . . . . . . . 44
vi
2.5.3 2.6
Sorting by Skill . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.6.1 2.6.2 Analysis of ?rms hiring patterns . . . . . . . . . . . . . . . . . 58 Hiring Process by Race/Ethnicity . . . . . . . . . . . . . . . . 63 2.6.2.1 2.6.2.2 2.6.3 Worker Race . . . . . . . . . . . . . . . . . . . . . 63 Worker and Owner Races . . . . . . . . . . . . . 71
Workers’ earnings and analysis of results . . . . . . . . . . . . 78
2.7
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3 Workplace Concentration of Immigrants 3.1 3.2
86
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 3.2.1 3.2.2 Literature on earnings di?erences . . . . . . . . . . . . . . . . 89 Literature on segregation . . . . . . . . . . . . . . . . . . . . . 91
3.3
Methodology and Data . . . . . . . . . . . . . . . . . . . . . . . . . . 97 3.3.1 3.3.2 3.3.3 3.3.4 Measuring immigrant concentration . . . . . . . . . . . . . . . 97 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Regression speci?cations . . . . . . . . . . . . . . . . . . . . . 103 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . 106
3.4
Accounting for immigrant concentration . . . . . . . . . . . . . . . . 112 3.4.1 3.4.2 3.4.3 Location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Worker Demographics . . . . . . . . . . . . . . . . . . . . . . 116 Employer characteristics . . . . . . . . . . . . . . . . . . . . . 118
vii
3.4.3.1 3.4.3.2 3.5
Employer size . . . . . . . . . . . . . . . . . . . . . . 119 Industry . . . . . . . . . . . . . . . . . . . . . . . . . 125
Exploring social networks, language skills, and human capital as possible explanations for concentration . . . . . . . . . . . . . . . . . . . 127
3.6 3.7
Country of origin di?erences . . . . . . . . . . . . . . . . . . . . . . . 135 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5 Conclusion
137
Appendix
140
A Matching Rate
140
B De?nitions
141
C Unknown-Owned Firms
142
D Weights and Selection
143
E IPUMS 1990: Descriptive Statistics
147
F Linear Probability Estimates
148
G Simulations of employer size e?ects in a statistical model with segregation 154
Bibliography
164
viii
List of Tables
2.1
Descriptive Statistics - CBO(1992) and Sample/Matched Firms . . . 35
2.2
Descriptive Statistics - Characteristics of Workers . . . . . . . . . . . 38
2.3
Average Race and Ethnic Composition of New Hires by Owner’s Type 43
2.4
Average Race and Ethnic Composition of New Hires by Owner’s Race 44
2.5
Mean Earnings by Owner and Worker Type . . . . . . . . . . . . . . 49
2.6
By Similar Coworker Share: Mean Earnings by Owner and Worker Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.7
Worker types distribution by owner’s skill requirement . . . . . . . . 53
2.8
Linear Estimates of the E?ect of Owner Type on the Probability that a New Hire is an Immigrant . . . . . . . . . . . . . . . . . . . . . . . 61
2.9
Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic . . . . . . . . . . . . . . . . 63
2.10 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian . . . . . . . . . . . . . . . . . . 66
2.11 Multinomial Logit Model: E?ects of Owner Type and Coworkers on Type of New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
viii
2.12 Multinomial Logit Model: Predicted Probability of Covariates . . . . 70
2.13 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic . . . . . . . . . . . . . . . . 71
2.14 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian . . . . . . . . . . . . . . . . . . 74
2.15 Multinomial Logit Model: E?ects of Owner’s Race and Coworkers on Type of New Hires . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.16 Multinomial Logit Model: Predicted Probability of Covariates Owner and Worker Races (%) . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.17 OLS Results: E?ect of Owner Type and Coworker Share on Log Real Annual Wages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.1
Variation in Immigrant Share of Workforce across Sample MSAs . . . 103
3.2
Characteristics of Immigrant and Native Workers, Full Sample
. . . 106
3.3
Contribution of Covariates to Immigrant Concentration (Full Sample) 113
3.4
Characteristics of Matched Sample Workers (Unweighted)
. . . . . . 128
3.5
Characteristics of Immigrant and Native Workers, Matched Sample (weighted) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
ix
3.6
Contribution of Covariates to Immigrant Concentration (Matched Sample) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.7
Network E?ects from Coworker Share Regressions . . . . . . . . . . . 133
A.1 Matching and Non-matching rate of ?rms in CBO and SSEL(singleunit) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
D.1 Descriptive Statistics - CBO(1992) and Sample/Matched Firms . . . 145
E.1 Descriptive Statistics - Characteristics of Workers . . . . . . . . . . . 147
F.1 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Black . . . . . . . . . . . . . . . . . . 148
F.1 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 149
F.2 Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is White . . . . . . . . . . . . . . . . . 149
F.2 Linear Probability: E?ect of Owners types on the Probability that a New Hire is White (continued) . . . . . . . . . . . . . . . . . . . . . . 150
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black . . . . . . . . . . . . . . . . . . 150
x
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 151
F.3 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Black (continued) . . . . . . . . . . . 152
F.4 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is White . . . . . . . . . . . . . . . . . 152
F.4 Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is White (continued) . . . . . . . . . . . 153
G.1 Characteristics of Weighted Matched Sample . . . . . . . . . . . . . . 162
G.2 Linear Regression of Full Speci?cation . . . . . . . . . . . . . . . . . 163
xi
List of Figures
2.1
Workforce Characteristics of Immigrant, Mix and Native Firms . . . . 45
2.2
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4
Workforce Characteristics of Immigrant, Mix and Native Firms Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.1
Cumulative Distribution of Coworker Share by Worker Type . . . . . 111
3.2
Coworker share by age of employee . . . . . . . . . . . . . . . . . . . 118
3.3
Coworker share by employer size . . . . . . . . . . . . . . . . . . . . . 120
3.4
Cumulative Distribution of Coworker Share by Worker Type and Employer Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.5
Coworker share by employer sector . . . . . . . . . . . . . . . . . . . 126
G.1 Shape of function d
. . . . . . . . . . . . . . . . . . . . . . . . . . . 156
G.2 Immigrant share distribution with and without segregation
. . . . . 158
xi
G.3 Immigrant coworker mean and employer size (? = 4)
. . . . . . . . . 159
G.4 Native coworker mean and employer size (? = 4)
. . . . . . . . . . . 159
G.5 Immigrant-native di?erence in coworker mean and employer size (? = 4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
G.6 Immigrant-native di?erence in coworker mean and employer size (? = 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
xii
List of Abbreviations
CBO SSEL BR LEHD MSA Characteristics of Business Owners Standard Statistical Establishment List Business Register Longitudinal Employer-Household Dynamics Metropolitan Statistical Area
xiii
Chapter 1 Introduction
Over the last several decades, labor markets in many cities in the US have absorbed large in?ows of new immigrants. During the same period, numerous empirical studies have analyzed the e?ect of immigration in the host economy. In the early 90s the consensus was that there is only a small e?ect of immigration on native economic outcomes (Grossman [1982]). However, since the late 90s, the consensus moved toward a signi?cant e?ect of foreign-born migration on natives (Borjas [1994]). Recent surveys on the economics of immigration [Borjas, 2003, 2005], Friedberg and Hunt,1995; Card, 2001; Card and Lewis,2005; Card, 2006) conclude that the impact of immigration on the wages and employment is still unclear. As of 2007, immigrant workers represented 15% of the U.S. population. The impact of large in?ows of immigrants and their assimilation into the host economy has been a primary objective of analysis in the labor literature. How such large ?ows of workers are incorporated into the labor market and interact with various businesses and workers is of special interest. An alternative literature has focused on how ?rms respond to an in?ow of immigrants. The key question is no longer one of job supply but also one of job distribution. Lewis and Card [2005] and Beaudry et al. [2006] look at an exogenous local unskilled labor supply change and ?nd that areas with higher concentration of immigrants have employed higher number of unskilled
1
workers and increased productivity at the same time. Their ?ndings also suggest a small impact of immigration on natives’ relative wages. In this analysis, the role of business owners in the patterns of hires and earnings in the labor market is relevant. In particular, some studies have found that the type of manager recruiting new workers is a determinant in the workforce composition of the business. In a extensive analysis of race and ethnic segregation across workplaces in the U.S., Hellerstein and Neumark [2007] ?nd that a large degree of segregation remains even after accounting for metropolitan area, education and occupation. In a follow-up paper, they explore the role of residential networks in these patterns, and found preliminary evidence of its relevance for low-educated and low-English-ability workers. On the other hand, many other authors have analyzed the direct e?ect of the type of manager on the type of worker in the ?rm. For instance, Carrington and Troske [1995] and Giuliano and Ransom [2008] have found that females and blacks are disproportionately employed by female and black supervisors respectively. Meanwhile, Stoll et al. [2004] found that black businesses receive more applications from black workers and employ more black workers than other businesses. Giuliano and Ransom [2008] found a causal relation between the race of managers and workers using panel data of a retail store. They control for the unobserved characteristics that can also a?ect the race of the coworkers and hires in a ?rm. Although the primary determinants of the racial composition of new hires are workplace and location characteristics, manager race also stands as a signi?cant component. Nevertheless, this second group of analyses have been mainly focused on black versus white issues 2
and particular industries. In the sociology literature, there have been a limited number of studies that provide some insights on the tendency of immigrants to work for immigrant ?rms. For instance, in Los Angeles in 1989, 30 percent of employed Koreans held jobs in ?rms owned by fellow Koreans even though Koreans composed only one percent of the Los Angeles County population.1 According to Cardenas and Hansen [1988], during the 1980s, Mexican immigrant employers were most likely to hire Mexican, whether legal or undocumented, and were more likely to evaluate their quality favorably. Porter and Wilson [1980] ?nd two relevant patterns in the Cuban immigration to Miami during the 1960s. First, Cubans worked with other Cubans. Second, almost one-third of the Cubans worked for Cuban employers. The phenomenon of immigrants hiring immigrants is not limited to coethnic relationships between employees and employers. Other researchers have found that employers from one immigrant group often hire workers from other ethnic/racial groups.2 Immigrant entrepreneurs can take advantage of their language, cultural background and a?nities to have access to di?erent ethnic groups. Their immigrant status can give them privileged access to sources of labor less available to native entrepreneurs. Immigrant entrepreneurs routinely employ coethnics (including relatives) at rates vastly above chance levels.3 Making use of unique longitudinal and cross-sectional micro level databases, this thesis examines the role of owners, coworkers and networks, focusing on the
1 2
Min [1989]. Light [2006]. 3 Massey [1999], Massey et al. [1987].
3
importance of immigration and race/ethnicity on hiring patters, the scope for segregation and wage di?erentials. The main contributions of the research presented in this document are providing new stylized facts on the immigration issue and evidence on the role of social networks in labor markets4 . The outline of the thesis is as follows. Chapter 2 analyzes the e?ect of the birthplace of ?rm owners and coworkers on hiring patterns and wages. Using a unique matched sample from an employer-employee administrative database and a survey of characteristics of ?rm owners, this chapter studies the impact of the type of employers and individual coworkers (native versus immigrant workers, and ethnic/racial groups) on ?rm hiring patterns and workers’ average log wages. We connect owner and ?rm characteristics (place of birth, size and industry) with workers’ characteristics (wage, age, education, and place of birth) to test di?erent assumptions about ?rm hiring patterns and the wage di?erentials of workers of di?erent types. Given the unique features of the matched database, the data allows asking whether the odds that a worker of a particular group is hired are related to the types of owners and coworkers, and whether there exist wage premia associated with being an immigrant and working for or with other immigrants. Our results suggest that immigrant owners are three percentage points more likely to hire other immigrants than native owners, even after controlling for industry, ?rm size, geographic concentration of immigrants in the population, population density, and the legal form of organization of the ?rm. Looking at ethnic/race groups, immigrant owners are 3 to 4 percentage points more likely than native own4
For an extensive analysis on job information networks see Ioannides and Loury [2004].
4
ers to hire Asians and Hispanics versus blacks. Both types of owners, immigrants and natives, hire white non-Hispanic workers, but native owners have a higher probability of having white workers as new hires. These results are based on linear probability models as well as multinomial logit speci?cation that accounts for the simultaneity of choosing from among di?erent types of workers. Among our strongest ?ndings are the existence of a persistent pattern of hiring similar types and the e?ect of the share of dissimilar coworkers on the likelihood of hiring a particular individual. For instance, the increase of the share of similar coworkers at the time of recruitment by 100 workers increases the probability of hiring a worker of a type by around 60%. The probability is smaller if we look at the e?ect of the fraction of coworkers of other di?erent types. Additionally, this probability depends on whether the employer is immigrant versus native. Immigrant businesses show higher chances of hiring a new immigrant, Hispanic or Asian worker compared to native businesses, even after controlling for whether they have similar workforce distribution at the time of a new recruitment. Later, after controlling for owner’s race, our results are similar. Hispanic and Asian owners are 2.5 percentage points more likely to hire their own type (Hispanic and Asian workers respectively) than white and black owners. Given the lack of representation of native Hispanic and Asian owners in the data, we were not able to control for the cross categories race-birthplace. To the best of our knowledge, no previous study has analyzed the link between employer and coworkers’ birthplaces and employees’ employment opportunities and wages. This research provides initial steps on that branch of analysis. 5
Chapter 3 presents descriptive evidence on immigrant segregation at the workplace and analyzes the mechanisms that drive immigrant concentration. We have unique matched employer-employee data for a large number of states in the US that permits quantifying the extent of and covariates of the workplace concentration of immigrants. A lack of suitable data has limited economists’ ability to address these questions. The paper has two broad objectives. The ?rst is primarily descriptive. The descriptive ?ndings show that immigrants are much more likely to have immigrant coworkers than are natives. This pattern is driven partly by the geographic concentration of immigrants, but the patterns hold true even within local labor markets. At the same time, most immigrants do have native coworkers; only a small share work in immigrant-only workplaces. The concentration of immigrants is higher for recent immigrants and, conditional on recent arrival, for older immigrants. Part of the assimilation process is a movement towards more interaction with natives in the workplace over time, and younger immigrants are more likely to work with natives. We ?nd large di?erences associated with ?rm size: concentration is much higher in smaller ?rms, but is far from zero even in the largest ?rms. We also ?nd substantial variation in the extent of immigrant concentration across industries even after controlling for a detailed set of location, employer and employee characteristics. Second, our ?nding that the allocation of immigrants across workplaces is far from random raises the question of what drives this workplace concentration. Both the existing literature and our descriptive ?ndings suggest that it is important to consider how businesses hire their employees and the choices that businesses 6
make about the skill mix of their workforce. One relevant issue here is the role that language skills play in governing interactions among employees and between employees and customers. A second issue is the role of social networks in the process that matches workers and ?rms. A third issue is human capital - the sorting and concentration of immigrants in the workplace may re?ect sorting by skills. In the second part of the paper, we explore the role of these factors. We ?nd evidence that immigrants with primarily immigrant coworkers are likely to have coworkers who live in the same residential tract. This pattern is robust to inclusion of controls for other closely related factors such as residential segregation. We also ?nd evidence that immigrant workers with poor English speaking ability and low education are more likely to work with immigrant coworkers. Our ?ndings suggest that social connections and social capital may be important for understanding workplace concentration, employment opportunities and wage di?erentials. Continuing this line of thought, Chapter 5 o?ers the conclusions and discusses on the main factors that can explain the previous ?ndings. It is intended to focus on the role on networks in the labor markets, and the connection of our ?ndings with previous empirical and theoretical literature. It also describes the key issues to be considered to develop in appropriate theory.
7
Chapter 2 Does It Matter Who I Work For And Who I Work With? The Impact Of Owners And Coworkers On Wages And Hiring 2.1 Introduction
This paper analyzes the e?ect of the birthplace of ?rm owners and coworkers on hiring patterns and wages. As of 2007, immigrant workers represented 15% of the U.S. population. The impact of large in?ows of immigrants and their assimilation into the host economy has been a primary area of study in the labor literature. How such large ?ows of workers are incorporated into the labor market and interact with various businesses and workers is of special interest. The role of business owners in the patterns of hires and earnings in the labor market has played an important role in this literature. In particular, some studies have found that the type of manager recruiting new workers is a determinant of the ?rm’s workforce composition. For instance, Carrington and Troske [1995] and Giuliano and Ransom [2008] have found that females and blacks are disproportionatly employed by female and black supervisors respectively. Meanwhile, Stoll et al. [2004] found that black businesses receive more applications from black workers and employ more black workers than other businesses.
8
Using a unique matched sample from an employer-employee administrative database and a survey of characteristics of small-?rm owners, this study analyzes the impact of the type of employers and individual coworkers (natives versus immigrants, or ethnic groups) on ?rm hiring patterns and workers’ average log wages. Firm types are de?ned by the type of owner (immigrant-owned versus native-owned), while ‘’coworker” refers to the fraction of same-kind fellow workers holding a job in the same ?rm. The share of immigrant coworkers in the ?rm is called the coworker index.1 We connect owner and ?rm characteristics (place of birth, size and industry) with workers’ characteristics (wage, age, education, and place of birth) to test different assumptions about ?rm hiring patterns and the wage di?erentials of workers of di?erent types. Given the unique features of the matched database, the data allows asking whether there exist wage premia associated with being an immigrant and with working for or with other immigrants. The type of a new hire can be a?ected by the type of employer in di?erent ways. First, social networks, segregated by race or similar background, could be used by job seekers and by employers when looking for new candidates. Ethnic communities provide a network for immigrant entrepreneurs to ?nd workers, to sell ethnic goods, and to obtain credit. Second, matching productivity generated by employer-employee similarity could motivate owners to employ same-kind individuals. In certain industries the use of a common language may be important for productive e?ciency. Third, employer tastes might bias them to employ workers
In this Chapter, the expressions ?rm type and owner type are used to explain that ?rm’s owners correspond to one of the following groups: native-only, immigrant-only, and mix owned ?rms.
1
9
of a similar kind. Employer discrimination could generate scope for segregation.2 However, coworker e?ects could compensate for the presence of employer discrimination. In fact, for all types of owners the share of similar coworkers increases the probability of being hired in the ?rm. We also control for speci?c characteristics in the ?rm, such as the fraction of English speakers, to identify the possible scope for matching productivity. This paper focuses on the importance of social ties in the process of recruitment when ?rms use current employees’ social connections to help ?nd and identify new candidates. However, employers may use this mechanism di?erently for di?erent worker types, depending on their ability to take advantage of their workers’ connections. For instance, given their cultural, linguistic, and social backgrounds, immigrant employers have an advantage, compared to natives, in exploiting their immigrant workers’ social connections. Our results suggest that immigrant owners are three percentage points more likely than native owners to hire other immigrants, even after controlling for industry, ?rm size, geographic concentration of immigrants in the population, population density, and the legal form of organization of the ?rm. Looking at ethnic/race groups, immigrant owners (Hispanic/Asian owned ?rms) are 3 to 4 percentage points more likely than native owners (white and black owned ?rms) to hire Asians and Hispanics versus blacks and whites. Both, native and immigrant owners, hire white non-Hispanic workers, but native owners have a higher probability of having white workers as new hires. These results are based on both linear probability models and a multinomial logit speci?cation that accounts for the simultaneity of choosing from
2
Lang [1986]
10
di?erent types of workers. Among our strongest ?ndings are the existence of a persistent pattern of hiring similar types and the smaller e?ect of the share of dissimilar coworkers on the likelihood of hiring a particular individual. For instance, the share of similar coworkers at the time of recruitment increases the probability of hiring a worker of a type by around 60%. The probability is higher when the owner is similar to the new hired. Additionally, this probability is di?erent whether the employer is immigrant versus native. Immigrant businesses show higher chances of hiring a new immigrant, Hispanic or Asian compared to native businesses, even after looking whether they have similar workforce distribution at the time of a new recruitment. To study the wages of employees, one must understand the role of employers in wage-setting, which necessitates gathering wage data by employer and having detailed information about the employer. Immigrant workers tend to have lower average wages than native workers. Many authors have used a human capital approach to explain that wage gap and have found that skill accounts for almost two thirds of the wage di?erence between Hispanics and white Non-Hispanics.3 Meanwhile, the residual unexplained wage gap has traditionally been used to claim the existence of racial/ethnic discrimination in the labor market. Other authors have found that industry wage-di?erentials are to a very large extent explained by the characteristics of workers and the contribution of industry to wage setting is much smaller after looking at both person and that industry e?ects.4 However, these studies don’t rule
3 4
Borjas [1994], Trejo [1997], Chiswick [1978], Borjas [2003] among others. Abowd et al. [1999]
11
out a signi?cant impact of ?rm-level e?ects on wage formation.5 The results in this paper suggest that much of the di?erence between the log annual wages of immigrants and natives comes from immigrants’ propensity to work in non-native owned ?rms, which pay the lowest average log annual wages. Interestingly, though, native workers holding a job in immigrant ?rms are paid less than immigrant workers. After controlling for typical human capital variables, full-time immigrant workers earn about 8% less than native workers ($3,293 less each year). When working for native employers this di?erence increases to 11%. Meanwhile, immigrant workers earn 10% more than native workers in immigrant owned ?rms ($4,398 more each year). Recent work has used the idea of networks in the labor market to explain labor market inequalities as a function of di?erential social capital (social resources, network structures, network resources). Minority individuals are generally connected to other minority-group workers who cannot provide them with the opportunity to change their employment outcomes. Hispanics and blacks are disadvantaged because they are likely to match with same-kind job contacts, and end up working in lower wage workplaces where other Hispanics and blacks work (Elliot [2001]). To the best of our knowledge, no previous study has analyzed the link between employer and coworkers’ birthplaces and employees’ employment opportunities and wages in a large set of industries and geographic locations. This research provides initial steps on that branch of analysis. These ?ndings suggest that social connections and social capital may be important for understanding employment opportunities
These authors obtained that the average of the di?erence in wages paid to an identical worker employed at two di?erent ?rms in France was 20%-30%.
5
12
and wage di?erentials. The remainder of the paper is organized as follows. Section 2.2 and section 2.3 review previous work on the relation between workers and types of ?rms, ethnic economies and ’ethnic matching’ between supervisors and employees, the usage of networks, and network e?ects on hiring procedures and workers’ wages. It also discusses the importance of analyzing small businesses when looking at the impact of immigration. Section 2.4 examines the data and presents basic descriptive statistics on owners’ and workers’ characteristics. Next, section 2.5 presents preliminary information on workers’ average earnings by worker type and by di?erent levels of coworker shares. Section 2.6 is divided in two sections. The ?rst part analyzes whether the type of employer and coworker characteristics a?ect the composition of new hires in ?rms. The second part evaluates the impact of ?rm owner type on employees’ log annual earnings controlling for worker human capital. Section 2.7 concludes.
2.2 Literature Review
Because no single theory exists to explain the e?ect of ?rm owners and coworkers on hiring patterns and wages, we draw on the literature of several related ?elds to motivate our hypotheses on the subject. Those literatures include ethnic economy theories dealing with ethnic/immigrant concentration, theories of ?rm wage di?erentials and hiring procedures, and network theories. Immigrants tend to work in low-wage/low-productivity ?rms, low-pay occupa-
13
tions, and in ?rms with a high percentage of immigrant workers.6 Some researchers have found occupational and ethnic coworker concentration in the United States (Andersson et al. [2007], Patel and Vella [2007], and Light [2006]) and in other countries (Barr and Oduro [2000] and Andersson and Wadensj´ o [2001]). The literature has attempted to explain workers’ concentration by skill, race, and sex.7 Hellerstein and Neumark [2007] analyzed ethnic segregation in the United States and found a substantial degree of segregation in the workplace. They claim that even though workplace segregation partially results from residential segregation (spatial mismatch explanation) and from ethnically correlated skills, there seem to be other mechanisms that suggest the presence of immigrant social connection e?ects (local residential networking). In an extensive analysis of racial and ethnic segregation across U.S. workplaces, they found that a large degree of segregation remains even after controlling for metropolitan area characteristics, and that very little of this segregation can be explained by observed di?erences in education and occupations. Language, however, seems to be a signi?cant factor for immigrant segregation. Lang [1986]’s theory provides an explanation for worker segregation by language groups. When there are transaction costs associated with employees of di?erent language groups working together, there is scope for segregation. Employers of each language group have incentives to fully segregate to avoid the cost of needing employees who can be the bridge between di?erent language groups. Despite ?ndings on immigrant concentration at di?erent levels, we cannot be
6 7
Borjas [1994], Borjas [2003], Andersson et al. [2007], and Andersson et al. [2008]. Kremer and Maskin [1996], Hellerstein and Neumark [2003].
14
sure that immigrants are more likely to work for immigrant bosses and that such a pattern would a?ect individuals’ labor market outcomes. There is no evidence that immigrant-owned businesses are distributed (or concentrated on) di?erently across speci?c industries, ?rm sizes, or skills, than native businesses, and that this distribution is correlated with the distribution of immigrant workers across industries, sizes, and skills. A recent group of studies analyzes the matching process between managers and workers by racial group. Giuliano et al. [2006] found a signi?cant e?ect of race and ethnicity on hiring procedures. For example, in locations with large Hispanic populations, Hispanic managers tend to hire more Hispanics and fewer whites than white non-Hispanic managers. In a more recent analysis, Giuliano and Ransom [2008] looks at the e?ect of manager ethnicity on hires, separations and promotions across di?erent occupations in a U.S. retail ?rm. Whites were more likely to leave stores where managers were Hispanics than when they were white. Their work is very relevant, although they only focus on a very particular retail ?rm. Their studies do not consider the coworker e?ect. That is, they don’t study the e?ect of the fraction of similar coworkers holding a job in the ?rm on the probability a particular type of worker is hired. There has not yet been a connection established between owner’s birthplace and the type of workers employed at a ?rm or these workers’ earnings. Nevertheless, the literature discusses motivations for supervisor-employee matching. First, ?rm owners could have preferences for employing individuals of their own type or with the same background. Second, the types of goods o?ered by immigrant ?rms may 15
di?er from those o?ered by native ?rms. If immigrants specialize in producing ethnic goods, immigrant workers have a comparative advantage over native workers in these ?rms. The di?erences between products can result in di?erent worker composition.8 However, none of these reasons have obvious predictions of workers’ earnings. That an employer has a preference for a certain group does not necessarily imply higher wages for that group. The distribution of workers and employers in the market also a?ects the labor market equilibrium. In the sociology literature, there have been a limited number of studies that provide some insights on the tendency of immigrants to work for immigrant ?rms. For instance, in Los Angeles in 1989 30 percent of employed Koreans held jobs in ?rms owned by fellow Koreans even though Koreans composed only one percent of the Los Angeles County population.9 According to Cardenas and Hansen [1988], during the 1980s, Mexican immigrant employers were most likely to hire Mexicans, whether legal or undocumented, and to evaluate their quality favorably. Porter and Wilson [1980] ?nd two relevant patterns in the Cuban immigration to Miami during the 1960s. First, Cubans worked with other Cubans. Second, almost one-third of the Cubans worked for Cuban employers. The phenomenon of immigrants hiring immigrants is not limited to coethnic relationships between employees and employers. Other researchers have found that employers from one immigrant group often hire workers from other ethno/racial groups.10 In Los Angeles, during the nineties, 51% of the garment factories were owned by Asians with most of their employees being
Andersson and Wadensj´ o [2001] Min [1989]. 10 Massey [1999], Massey et al. [1987].
9 8
16
Hispanics. Ethnic networks alone cannot expand the supply of coethnic-accessible jobs. Generally, the number of jobs o?ererd by ethnic-speci?c owned ?rms is not equal to the number of possible candidates from the same ethnic group in the local community. Business leaders from ethnic groups whose rates of entrepreneurship are higher than other groups ?nd it di?cult to limit hiring to members of their own groups. Ethnic crossover can expand the economic opportunities provided by immigrant-owned businesses. Immigrant workers often join networks that cross ethnic boundaries. Using the Garment Industry in Los Angeles as an example, Light [2006] analyzes immigrant ownership economies consisting of immigrant employers plus their immigrant but not coethnic employees. He ?nds that this type of economy explains part of the garment industry’s growth during early 1990s in Los Angeles. The cited studies have been limited to small samples from particular geographic areas and speci?c groups of ?rms and immigrants. Most of them also focus on a particular period of time, with a cross-sectional view of the distribution of workers and ?rms. These analyses tended not to look beyond the segregation aspect to analyze the possible causes and consequences of those patterns. Unlike previous studies, this paper uses a representative group of areas, ?rms, industries and workers, and it analyzes the ?ow of hiring and the e?ect of employer-employee type matches on wages. The underlying hypothesis in the analysis is that workers and employers make di?erent use of their social connections in the market, given their speci?c characteristics, such as race/ethnicity and immigration status, which leads to a particular hiring pattern by each ?rm. Immigrant ?rms, for instance, would have an advantage over native ?rms when using their immigrant current workers as 17
a channel to ?nd new workers. On wage e?ects, previous research has suggested that much of the unexplained variation in wages among employees is linked to characteristics of their ?rms, such as size and industry.11 Not only do individual characteristics explain wage di?erentials between immigrants and natives, but potentially so do other characteristics, such as the birthplace or ethnicity of employers and coworkers. Unfortunately, most wage databases come from household surveys of individuals (Decennial Census and CPS), rather than from establishment surveys of wage-paying employers; they provide little employer-speci?c information, except for industry and, in some cases, ?rm size.
2.3 On the use of social networks
Recent work has suggested that supervisor-employee ethnic matching could result from the use of networks.12 On the one hand, according to several sociological studies on the ethnic economy, ethnic solidarity serves to provide entrepreneurs with privileged access to immigrant labor and to legitimize paternalistic work arrangements (Sanders and Nee [1987] and Model [1997]). Di?erent ?rms have di?erent recruitment processes, generating an initial sorting of worker types. On the other hand, networks can also have an impact on wages, providing better matches and more opportunities to the individual. Ethnic networks can generate informal sources
[Groshen, 1990, 1991a,b], Abowd et al. [1999], Abowd et al. [2004] among others. Networks is not a new concept in the literature. For an extensive analysis on job information networks see Ioannides and Loury [2004]. Sociologists have investigated the origins and creation of social networks for more than 40 years. Rees[1966] draws attention to di?erences among workers and their use of available information (formal and informal sources). Job referral is also extensively used in the labor market, as well as family networks (Granovetter [1995]).
12 11
18
of capital formation and captive markets, making these ?rms more self-su?cient and ?exible (Volery [2005]). Social capital becomes another form of capital resource.13 Individual’s social networks are likely to have an impact on labor market outcomes (Simon and Warner [1992]). The di?erential use of social networks does not provide the same access to information and opportunities to all individuals, o?ering a better relative position to those agents with better social connections or better use of their social networks. Recent literature has moved away from spatial mismatch model in explaining inequality across ethnic/race groups towards theories that include how social networks a?ect urban inequality [Hellerstein and Neumark, 2007, Hellerstein et al., 2008a]. Life-chances depend not only on individual resources but also on network characteristics re?ecting the resources of network members. In this context, personal networks are then considered an additional determinant of inequalities (Light [2006]). How do these mechanisms a?ect our groups of analysis? What is di?erent about particular types of workers and ?rms such as immigrant/racial groups? Although the comparison between whites and blacks has been long discussed, immigrant status can be crucial for understanding group di?erences in informal job matching and labor outcomes. Two important characteristics of the immigrant community are relevant for these implications. First, Borjas [1994] pointed out that immigrants tend to be less educated, to have poor English language skills, and to lack domestic experience. Second, immigrants rely heavily on social networks for
Social capital in its simplest form is a social network of strong and weak social ties (Light and Gold [2000]).
13
19
?nding jobs and geographically reallocate (Massey et al. [1987]). Previous literature has also discussed racial and ethnic di?erences in informal job matching (Elliot [2001], Holzer [1987]). These di?erences arise because informal channels permit race and other characteristics in the network to play a more prominent role in the hiring process than it does when formal mechanisms are used. As noted by Elliot [2001], one of the puzzles during 1980s and 1990s was the worsening position of less educated blacks in the labor market while the economy was absorbing thousands of new immigrant workers. Surprisingly, these new workers had, on average, similar characteristics to blacks: low formal education and high geographic segregation. So the question of job distribution became a ?rst order issue, especially in the topics of immigration and immigrant assimilation. Research on this puzzle has focused on the use of social networks by di?erent groups for ?nding employment [Waldinger, 1997], while the role of prospective employers in the use of these mechanisms has been ignored. Our empirical analyses sheds light on the impact of networks on immigrants. Considering the tendency of workers to refer their own, the immediate e?ect of network is the reproduction of the workforce composition across time as shown in this chapter. Our results in the following chapter support the hypothesis that social networks play an important role in workplace concentration. The tendency of social networks to be racially/ethnically homogeneous - exacerbated by individual’s immigration status- increases the probability that workers would refer same-type candidates and that same-type employers would tend to hire from shame-type groups. Immigrant employers can take better advantage of their immigrant employees in 20
hiring than native employers. The di?erential use of job referrals by employers is also evident when we examine who is hired and how the wages are distributed in the ?rm. Immigrants will tend to be hired more by immigrant ?rms with a high share of immigrant workers than by native ?rms with high share of immigrant workers. Immigrant entrepreneurs can take advantage of their language, cultural background and a?nities to have access to di?erent ethnic groups. Their immigrant status can give them privileged access to sources of labor less available to native entrepreneurs. Immigrant entrepreneurs routinely employ coethnics (including relatives) at rates vastly above chance levels. The most important network relationships are based on kinship, friendship, and paisanaje (the feeling of belonging to a common community of origin).14 Immigrant economies rely upon networks to locate jobs. On the one hand, referrals by friends or coworkers remove some of the uncertainty associated with ?nding a job with unfamiliar employers and increase the chance of ?nding a better job match. On the other hand, immigrant entrepreneurs tend to rely on their current employees to help ?ll their vacancies. Workers tend to refer individuals that are ’similar’ to them, from the same group, or with the same characteristics. Referral coworkers could also provide informal training, show the new worker how to perform the job, and have a good interaction with the new hire. Moreover, referral coworkers indirectly accept responsibility for new hires. Employers realize that this practice is bene?cial for them as well. Little cost or e?ort need be expended when new workers are located through employee contacts.
14
Massey[1980].
21
Previous empirical ?ndings show that Hispanic men report more frequent use of friends and relatives for job search than non-Hispanic whites, and are also signi?cantly more likely to have obtained their most recent job through personal contacts. Hispanics use informal contacts 32.8 percent more often than white non-Hispanics and blacks.15 Recent Latino immigrants are more likely than blacks or Latino natives to use personal contacts to ?nd jobs.16 Weak English skills explain much of this di?erence. However, this di?erence comes not only from the use of job networks by workers, but also from a greater reliance on referrals in small workplaces in combination with a concentration of recent immigrants in small ?rms. Employers also have a role in this process given that ?rms’ hiring procedures will a?ect individuals’ likelihood of receiving o?ers from jobs heard about through friends and relatives.
2.3.1 Small ?rms
Our focus on small/medium ?rms17 is motivated by two observations. First, in larger ?rms, the separation between ownership and management could detach the ?rm’s hiring process from owner characteristics. As Haltiwanger [2006] points out, however, in small ?rms the decision process is likely dependent on owner ability and characteristics. When dealing with each worker, small ?rm owners could project their tastes and managerial abilities onto the hiring and production processes of the ?rm. Since it is usually the business owner who makes such choices, the identi?cation of the person responsible for hiring decisions is easier and more relevant for small
Holzer[1987b], Smith [2000]. (Elliot [2001]). 17 We consider small/medium ?rms those with less than 500 employees.
16 15
22
?rms. Second, immigrant workers are more likely than natives to work in small ?rms. In Chapter 3 we ?nd that there is a signi?cant market segmentation that appears in any detailed distribution of workers in ?rms. Immigrants are more likely to be employed in ?rms with less than 10 employees 70% of immigrants work for small ?rms. Meanwhile, more than 60% of native workers are employed at ?rms with more than 100 employees. The labor force changes generated by immigration in?ows are thus borne primarily by smaller, younger ?rms. These ?rms are more sensitive to immigration shocks. If we only look at aggregate numbers (including small and big ?rms), immigration e?ects will be obscured.
2.4 Data and Measures 2.4.1 Sources
In this paper, we use three di?erent databases to match owners’ characteristics to workers’ characteristics. First, we use the Characteristics of Business Owners Survey (CBO) from 1992, and then match this survey with administrative data from the IRS (Business Register) for the years 1992 to 1996. To obtain workers characteristics, we use information from the Longitudinal Household-Employer Dynamics (LEHD) database for the years 1992 to 1996. In this section, we give a brief description of each database and their limitations, and discuss how we construct relevant variables used in the regressions. The Characteristics of Business Owners (CBO), later renamed the Small Busi23
ness Owner (SBO)database, is produced by the Bureau of the Census. The 1992 release of CBO was the ?nal version of this survey, which formerly was conducted every ?ve years. The survey for the 1992 CBO’s release was conducted in 1996, along with the economic census. Therefore, the questions in the survey refer to the business’ and owners’ information for years 1992 and 1994. The CBO is a supplement to the Survey of Minority-Owned Business Enterprises (SMOBE) and Survey of Women-Owned Businesses (WOB). The survey universe considered was ‘’any business which ?les an IRS form 1040, Schedule C (individual proprietors or self-employed persons); form 1065 (partnership); or form 1120S(Subchapter S corporation) in 1992.”18 It considers as business owners those who ?led business tax forms as owners of the ?rm, excluding non-S corporations, with at least 500 dollars in yearly business receipts, and with the largest employment size category equal to ?ve hundred. Note that non-S corporations generally have investors, not decisionmaking owners, and thus this group is not in the CBO survey’s universe. However, excluding non-S corporations often excludes the largest employers, making comparisons of small and large business owners di?cult. The CBO provides details about both business owners and their businesses. The unique ?rm identi?er is the CFN (Census File Number). At the cross-sectional level this number is unique for each ?rm. According to a CBO publication cited in of the Census [1997], almost 62% of the 78,147 ?rms’ surveys
18
19
and 59% of the 116,589 owners’ surveys were returned.
Characteristics of Business Owners 1992:CBO092-1. U.S. Bureau of the Census (September 1997) and Headd [1999]. 19 This is translated into 63% of the 41,297 employer ?rm surveys.
24
One possible reason for this low rate of reply is the di?culty of ?nding owners of exiting ?rms after 3-4 years. Almost 70% of all businesses present in 1992 were still in operation in 1996. This rate is lower for minority-owned ?rms (around 66%). We use employer ?rms in our sample. When sampling weights are used, the survey indicates that in 1992, 20% of owners were in ?rms with employees. According to the minority-?rm surveys, women, Asian, Paci?c Islander, American Indian, black, and Hispanic owners were typically underrepresented in the larger employment size classes. Hispanic-owned ?rms were 3.68% of all employer ?rms, but just 2.04% of ?rms with 100 or more employees. Additionally, 90.6% of business owners were born in the United States, while 9.4% percent were foreign born.
20
The per-
centage of native-owned ?rms was higher in the case of larger ?rms (94.5%). In this paper we focus only on employer ?rms. On average, there exists more than one owner per ?rm. In the CBO(1992), more than 52% of ?rms are employer ?rms, and almost 41% of this group have only one owner. Employer ?rms tend to have more owners than non-employer ?rms. Based on previous research using the CBO,
21
we consider the CBO as a sam-
ple of ?rms even though it is essentially a sample of ?rm owners. The resulting complication is that we need to make assumptions to identify the owner characteristics for multiple-owner ?rms. As a ?rst attempt, we consider three types of ?rms: only-native-owned, only-immigrant-owned, and mix-owned. Using this classi?cation, more than 85% of employer ?rms have 1 or 2 owners for all types.
A foreign born is an individual that was born outside the USA. CBO has a particular question on whether the owner was born in the US or abroad. 21 Carrington and Troske [1995].
20
25
In order to identify the characteristics of the owners of a particular ?rm (particularly immigration status and race), we follow the work of previous research based on the CBO (Carrington and Troske [1996]). For single-owner ?rms, the identi?cation is straightforward. Meanwhile, for multi-owner ?rms the mode is used. The number of hours per week spent at the business was used to break ties. This database has some limitations. First, in the 1992 survey the CBO’s sample universe omits chapter C corporations. This group of corporations corresponds to bigger businesses; therefore, comparison between small and large businesses in the CBO must be done with care. Second, even though we have each ?rm’s average payroll, we know nothing about the inter?rm distribution of payroll between di?erent types of workers. Third, this survey has zero information on human capital or occupational characteristics of workers. We try to overcome some of these limitations by merging CBO with data from Bureau of Labor statistics (UI and ES202) as described below. The second database used in this paper is the Census Bureau’s Standard Statistical Establishment List (SSEL) or Business Register (BR).22 This data has more complete information on ?rms given that the source of the SSEL is at the administrative level. This database works as a register of active employer business
Walker [1997] has an extensive discussion on the Census Bureau’s Business Register. The initial source of information on businesses is the IRS(Parker and Spletzer [2000]). The SSEL receives three main ?les from IRS; the Business Master File (BMF), with information on name, addresses and legal form of organization; the Payroll Tax Return File (Form 941) containing quarterly payroll and ?rst quarter employment (including March 12th employment); and the Annual Business Income Tax Return Files with information on receipts/revenues, industry classi?cation. For all three sources, EIN is the primary business’ id.
22
26
establishments23 in the United States and its territories. The unit of information is an enterprise, which can be associated with one or more establishments and with one or more EIN entities (Employer Identi?cation Number).24 In this paper we concentrate on those businesses organizations associated with only one EIN and one establishment, known as single-establishment enterprises or single-unit ?rms.25 All of the small ?rms in this chapter correspond to single-unit establishments. The assumption that ?rm owners are the ones making the main contracting decisions in a ?rm is more plausible in ?rms with only one establishment than otherwise. In the case of younger and smaller ?rms, this restriction does not exclude many ?rms.26 Additionally, businesses have a CFN (Census File Number) as an identi?er, which is unique for single-unit businesses. To follow the ?rm across time, the longitudinal identi?er for each ?rm is called alpha, and corresponds to the ?rst 6 digits of ?rms’ EIN. In the sample, we only follow ?rms that survived the entire period 1992 to 1996. Because most non-surviving ?rms did not respond to the CBO survey and the weights are constructed such that this pattern is considered, the weighted results are not impacted by this exclusion.27 We take data on industry, legal form of organization and employment from the SSEL ?les. See Appendix B for speci?c description of these variables. Because of the time di?erence between the year of information and the year in which the CBO
Active employer business establishments are those with payroll at anytime during the past three years, or with an indication that the business expects to hire employees in the future. 24 An EIN entity is an administrative unit assigned by IRS for tax purpose. Under the Federal Insurance Contributions Act (FICA) every organization with paid employees has to obtain an EIN. 25 All the matches between CBO(1992) and SSEL(1992) are in this category. 26 Haltiwanger et al. [2005]. 27 Headd [1999].
23
27
survey was conducted, information on employment and sales are from the SSEL dataset.28 We use the common unique ?rm identi?er (CFN) to match CBO with SSEL.29 We then follow the ?rm across time until 1996.30 The second set of information is associated with the characteristics of workers. This information comes from the Longitudinal Employer-Household Dynamics database. Information on workers comes from the Unemployment Insurance wage records for a group of states31 and the ES202 data. Based on availability, we use data from eight states for the years 1992 to 1996. The sample includes states with high immigrant concentration and low immigrant concentration areas. These ?les contain person identi?ers that allow researchers to track a worker’s quarterly earnings within a State across years. We sum over quarters to obtain each worker’s annual earnings. This database also contains ?rm identi?ers that allow for an exact link between the UI ?les and other data sets. The business level identi?ers in UI ?les are State Employer Identi?cation Numbers (SEINs). Therefore, one can match the UI data with the ES202 data, using SEIN to get information on the EIN, and compare it with the data previously matched using CBO(1992) and Business Register. For single-unit ?rms, the units of observation at the ?rm level used for CBO, SSEL and LEHD are generally similar.
The CBO is a retrospective survey. The response rate is a?ected by the survival rate of the ?rm and the extent to which owners can accurately recall past information. 29 We use businesses’ CFN, which are the Census Bureau’s preferred intra-year, cross-dataset link. The CFN contains the EIN ?rm identi?er and is unique for single-unit ?rms. 30 To illustrate the groups of ?rms included in both databases, we include a short discussion on ?rms matching rate in the Appendix A. 31 More detailed analysis on these records is presented in Abowd et al. [2006], and additional information on date of birth, place of birth, and gender are obtained for almost all workers in the sample after linking UI wage records to Census data. 98% of all private, non-agricultural employment is covered by the employer reports.
28
28
The UI wage records contain virtually all business employment for the sample states (for private non-farm ?rms). Earnings reports from these records are more accurate than survey-based earnings data, and one can obtain information for each worker in a speci?c ?rm (or establishment). Using this database, we follow ?rms across time from 1992 to 1996 using the unique identi?er within the state. We end up using only those ?rms that survived the entire period and did not change ownership. This group represents about 67% of the initial set of ?rms in 1992.32 Finally, the data set used in this study is unique in the sense that it contains data from each ?rm on output and inputs used in the production process, as well as data on earnings and some demographic characteristics of each worker in the ?rm. We use the years 1992 to 1996 for the analysis mainly because information about owners’ place of birth (i.e. being born in or outside the US) is only available in the Characteristics of Business Owners Survey in 1992. Our data tracks the total payroll and workforce composition of each ?rm from 1992 to 1996. The drawback of using UI data is its lack of certain demographic information on workers, such as education and occupation. However, the sta? at the LEHD has overcome this limitation by imputing education using administrative data from the Census Bureau containing information such as date of birth, place of birth, geographic area, industry, and sex. In this chapter, we use this imputed information on education,33 which has been used in previous work on the LEHD. This variable is
Few ?rms were dropped because, initially, the survey’s rate of response was highly correlated with the ?rms survival rate, so that most of the ?rms with information in the survey are surviving businesses. 33 See Lengermann et al. [2004] for details on the imputation.
32
29
a proxy for individuals’ human capital. We are aware that the lack of occupational information could be a relevant drawback of the data given that prior research has documented an important role for occupational segregation in creating di?erent workers’ wage gaps. We might think that immigrants tend to concentrate in lowskilled occupations relative to natives. However, as Troske [1999] and Carrington and Troske [1995] point out, occupations and job titles are less likely to be sharply de?ned in small ?rms, and as a result there could be less occupational segregation in small ?rms compared to large ?rms. Despite this limitation, we have to keep in mind that we can account for other workers’ characteristics, such as age, sex and imputed education. Given that workers have varying preferences for place of work depending on the disutility of commuting and amenities of particular areas, the areas where they would be willing to work are better represented by their actual place of work than their place of residence. Therefore, we need data on individuals’ place of work. Location of the ?rm is obtained using the LEHD database.
2.4.2 Construction of ex post weights
A relevant technical issue that arises in the process of using di?erent databases, especially when a survey is included, is the change of sample frame used by the survey database. Additionally, for smaller geographic areas, di?erences in industry and geographic information along with di?erences in the scope of industries covered lead to dissimilarities between the universe considered by the LEHD data and surveys
30
based on the Economic Census.34 In the design of the CBO survey, four panels were created in addition to divisions by employer status (employer versus non-employer), 2-digit industry and state. These panels consider racial categories using the owners’ social security information and the categories: Asian, Asian-American / Paci?c Islander, Hispanic, Black, and White. These groups were created by the Survey on Minority Businesses. Therefore, small ?rms and minority-owned ?rms are over-represented in the survey. The di?erence between the universe and sampling frames used in the CBO survey implies that our matched analysis sample will not be representative. Specifically, the sample frame used in the CBO will over-represent small, minority-owned businesses when linked with the UI database. To deal with this issue, we follow Abowd et al. [2007] and build ex post weights that control for the ?rms’ size, 2-digit industry code, legal form of organization, and employer status. We follow previous research in that we ?rst construct the fractions of ?rms each the category in the universe of ES-202. The universe of ES-202 is single-unit ?rms with more than one employee (coworker shares can be computed only for these ?rms), not in Agriculture, Mining, nor Public Administration, and less than one thousand employees, and are in Economic Census in-scope industries in 1992. This represents the numerator in the ex post weight. Then, we compute the same fractions for the ?nal matched data and use each fraction as the denominator of the ex post weight. This weight has the property that the distribution of employment by each category re?ects the size
LEHD database covers partially agriculture and public administration industries. Surveys based on the Economic Census tend to over-represent businesses in areas with high density population.
34
31
distribution of the ES-202 considered universe. The second section of the adjustment procedure involves the construction of an inverse Mills ratio. We use a probit estimation that considers the probability of being matched as a function of log employment, legal form of organization, owner’s place of birth (in or out the US), and log of sales per employee to generate the propensity scores. This section intends to account for the CBO survey’s sampling frame and the possible selection bias generated by the e?ect of unobservables on ?rms exiting from the universe considered to design the sample of the CBO survey. The ex post weights are included in all regressions. For more details and unweighted summary statistics see appendix D. Before using our approach the matched sample under-represent small, minorityowned businesses (see appendix D.1). After the match, and without considering the re-weighting process, we would be under-representing minority groups in small size ?rms. The sample of ?rms o?ering unemployment bene?ts are relatively of bigger size. After applying our new weight, we try to recover some of the original distribution in the CBO sample. There is a lower representation of sole proprietorship after matching the original sample with the UI database without using the new weights.
2.4.3 Firms
To compare the full CBO sample to the ?nal matched sample used in the analysis, we look at descriptive statistics for a set of variables. The ?nal match uses LEHD information from 8 states,35 which include high and low immigration
35
Those states with available data in 1992 are included.
32
states. For these states we obtain workers’ and ?rms’ information. Firms from the agriculture, mining and public administration sectors are not included. Additionally, only single-unit businesses are considered. The original matched sample in the analysis has 7,200 ?rms, representing 339,040 workers from 1992 to 1996. All results are weighted by the adjusted-weight discussed in section 2.4.2. Table 2.1 shows two blocks of summary statistics. One block (CBO-SSEL) contains the employer ?rms matched from the CBO survey and the BR, while the second block (Sample(CBO-LEHD)) contains the ?nal matched sample, consisting of the subset of CBO-SSEL data matched to the LEHD. For each block, this table presents the distribution of ?rm type across ?rm size categories and sectors, together with the average number of owners, average share of immigrant workers, de-meaned average log sales per employee, average percentage of immigrants in the county in which the ?rm is located and in the counties surrounding this location, and the percentage of each type of owner. Total population and the share of immigrant workers are constructed from the public 1990 Census, and are based on all Census counties surrounding the location of the ?rm. Immigrant ?rms have a higher proportion of immigrants in the local population than native and mixed ?rms. Because immigrants also tend to be geographically segregated, we will use this variable to control for di?erences in ?rms’ local workforce. In the ?nal matched sample, the average immigrant-owned ?rm employs 38% immigrant workers. The distribution of ?rms across sectors and sizes for each type of ?rm by owner birthplace is very similar, except for the tendency of immigrantowned ?rms to be in retail or services, and this distribution is only slightly changed 33
after matching the original database with the LEHD database. From the table we observe that immigrant-owned ?rms’ log sales per employee is slightly higher than native-owned ?rms. Actually, on average, native owned ?rms have the lowest log labor productivity. In general, ?rms are concentrated in size categories with fewer than 50 employees. Meanwhile, regardless their owner type, ?rms are highly concentrated in the sectors Services, Retail, Manufacturing and Construction. Sole proprietorships represent more than 50% of immigrant and native ?rms. Mixed-owned ?rms tend to be larger in size with respect to the other groups. These ?rms are mainly Partnerships and Corporations.
34
Table 2.1: Descriptive Statistics - CBO(1992) and Sample/Matched Firms CBO1 Imm Nat 49.30 20.90 14.68 10.70 2.97 1.46 5.07 10.35 2.83 19.73 29.68 3.19 29.16 42.51 21.35 16.69 12.10 4.56 2.79 12.74 13.89 7.43 17.14 19.73 6.18 22.89 Matched Sample(CBO-LEHD) Mix Imm Nat Unk ALL 16.67 18.75 18.75 28.13 9.90 7.81 5.73 25.00 5.21 18.75 14.58 7.81 22.92 37.27 21.53 18.44 14.25 5.67 2.84 4.71 15.93 2.77 22.18 29.85 3.55 21.02 33.60 20.97 18.02 16.72 6.65 4.03 13.49 17.65 7.58 19.03 16.42 5.30 20.54 33.29 20.33 17.40 16.89 6.78 5.31 9.38 18.36 6.91 22.24 21.34 4.44 17.33 37.18 13.77 49.35 11.63 (1.18) 15.07 81.71 1.83 33.79 20.79 17.88 16.58 6.59 4.37 10.06 17.76 6.34 20.84 20.83 4.70 19.49 50.64 12.99 36.37 11.64 (1.13) 12.80 92.85 1.87
Distribution/Type of ?rm Size (%) 2-4 5-9 10-19 20-49 50-99 100+ Sector (%) Construction Manufacturing Transp. & Utility FIRE Retail Wholesale Services Legal Form (%) Sole Proprietorship Partnership Corporation 2 l(sales/employment) 3 Imm. in the neighborhood4 In MSA Average Number of Owners Continued on next page.
Mix 26.52 18.06 23.48 18.18 6.94 6.82 6.49 20.26 5.45 19.22 17.14 6.10 25.32
Unk 46.43 20.91 15.00 10.95 4.26 2.45 10.41 13.59 6.73 19.24 23.17 5.21 21.65
ALL 44.66 21.05 15.90 11.59 4.26 2.54 10.58 13.38 6.43 18.36 22.47 5.36 23.43
35
- 52.40 51.43 29.15 28.51 12.58 12.26 18.69 71.94 35.10 25.01 52.16 11.64 11.60 11.48 11.54 (1.17) (1.17) (1.05) (1.15) 13.47 22.15 12.43 15.06 92.01 96.50 90.07 80.60 3.58 1.57 1.88 1.80
50.48 - 49.12 56.81 12.97 25.00 12.26 10.46 37.36 75.94 38.61 32.72 11.54 11.73 11.59 11.57 (1.11) (1.08) (1.19) (1.05) 12.90 14.03 21.17 11.34 92.41 91.30 97.37 93.01 1.84 3.78 1.55 1.95
Table 2.1: Descriptive Statistics - CBO(1992) and Sample/Matched Firms (continued) CBO1 Nat . 45.87 78.60 Matched Sample(CBO-LEHD) Mix Imm Nat Unk ALL 33.00 38.05 11.55 28.00 26.00 2.84 18.10 42.54 36.53 100.00 2.84 12.89 77.92 6.40 100.00 10.09 15.50 0.79 73.62 19.20 51.70 0.69 28.41 3.41 2.55 1.48 92.56 7,985 339,040 4.15 3.02 1.20 82.30 5.66 9.41 1.36 83.57
Distribution/Type of ?rm Average Share of imm. Workers Unweighted dist. of ?rms Weighted dist. of ?rms RACE/ETHNICITY Hispanic Asian Black White # of Firms unweighted # of Observations unweighted 36
Mix . 2.03 2.01 10.39 10.60 0.78 78.23
Imm . 14.94 13.50 18.56 35.68 1.56 44.20
Unk ALL . . 37.16 100.00 5.89 100.00 4.69 5.07 2.16 88.08
2.99 4.35 1.34 5.20 2.22 2.60 93.45 87.85 41,297 1,655,750
Note: Statistics based on weighted outcomes unless the contrary is indicated. (1) Single-unit ?rms that matched with SSEL. (2)Only S- Corporation . (3)Source SSEL: Sales (total receipts/sales), and employment (Employment March12th). Numbers in parenthesis are standard deviations. (4)Using Census 1990, computed percentage of immigrant population in the counties including the ?rm and surrounding.
The average number of owners (owner type) is similar in the original and matched samples. The average number of owners by owner birthplace is similar, except, as expected, for mix-owned ?rms which by de?nition have two or more owners. Table 2.1 illustrates that these patterns are similar in the original CBO sample and the ?nal matched CBO-LEHD sample. In the matched sample, Asian-owned ?rms are over-represented, while white immigrant owners are underrepresented. However, as in the original CBO sample, immigrant-owned ?rms are mainly owned by Hispanics and Asians, while most of the native-owned ?rms have white owners. In the original matched data there is a percentage of ?rms with unknown owners’ place of birth. We decide to exclude this group from further analysis. Given that, on average, the characteristics of this unknown group are similar to the rest of the sample (see Appendix(C) for t-tests and a chi-square analysis), we don’t expect this exclusion to a?ect our ?ndings. We drop ?rms with less than two employees. Given that female labor participation is characterize for additional elements di?erent to the ones analyze here we only consider male workers.36 Workers should have at least one coworker, and the analysis of earnings is net of other labor supply factors that could a?ect female workers di?erently. After these restrictions, the ?nal sample is reduced to 4,478 ?rms and 214,398 workers from 1992 to 1996.
The e?ect of networks for female immigrants is also a very important analysis. According to Massey et al. [1987], Mexican female immigrants tended to arrive and go directly to speci?c industries such as babysitter and meat packaging. The variation at such detailed level is not enough in our data, so we cannot disentangle industry e?ect versus owner e?ect. Given the particularity in the way female labor enter the market, there could be additional unobserved elements a?ecting the likelihood of hire an immigrant woman that we cannot consider in this aggregate analysis.
36
37
2.4.4 Workers
Among the relevant workers’ characteristics available in our data are age, immigration status (place of birth), date of entry in the US (date of SSN application), education, quarterly earnings, and race. We sum over quarters to obtain each worker’s annual earnings, and then compute real earnings based on 1992 dollars. The data set used for the analysis includes all male workers with positive earnings. On the distribution of workers, Table (2.2) and Figure (2.1) show the proportions of workers by age, race, sex, education, owner type,size, and sector, as well as, mean age, education and earnings, for all workers and for immigrants and natives. Foreign workers represent almost 24% of the sample. Similar to previous studies, on average, foreign born workers tend to be less educated, younger and tend to have lower income than native workers (Borjas [1994]), although these di?erences are not large in our sample. The fraction of workers across age categories, however, is similar for both types of workers in age categories 40 years and more. Table 2.2: Descriptive Statistics - Characteristics of Workers Individual IM US MEAN (std) Age Education Log(annual earnings) DISTRIBUTION (%) AGE Continued on next page. 38 ALL
34.01 34.14 34.11 (13.33) (12.02) (13.13) 13.04 13.16 13.13 (2.76) (2.94) (2.79) 8.30 8.32 8.33 (1.87) (1.68) (1.84)
Table 2.2: Descriptive Statistics - Characteristics of Workers (continued) Individual IM US 18.09 24.43 51.64 43.03 30.26 32.54 8.89 59.27 30.81 1.04 7.92 37.08 3.81 14.17 19.33 1.36 16.34 2.18 4.70 9.27 19.32 18.96 45.57 17.36 47.21 22.76 1.71 10.88 42.70 8.34 48.96 20.64 0.85 15.58 64.92 43.67 7.41 59.26 32.00 1.33 18.37 26.13 7.11 14.16 16.63 1.58 16.03 1.65 4.23 9.32 21.23 18.91 44.66 75.07 4.85 0.93 11.21 4.67 12.10 5.67 82.23 4.74 0.96 4.94 89.36 37.39 ALL 22.91 45.10 31.99 7.77 59.26 31.71 1.26 15.85 28.76 6.31 14.16 17.28 1.52 16.10 1.78 4.34 9.31 20.77 18.92 44.88 61.18 15.04 6.18 8.92 6.16 19.46 6.31 74.22 7.15 0.95 6.24 85.68 38.90
Under 25 25-39 40+ EDUCATION High School Dropout High School Graduate Some College Education College Graduate SECTOR Construction Manufacturing Transportation and Utilities Wholesale Retail FIRE Services SIZE 2-4 5-9 10-19 20-49 50-99 100+ RACE White Hispanic Asian Black Other TYPE OF OWNER Immigrant Mixed Native RACE OF OWNER Asian Black Hispanic White Part-time Continued on next page. 39
Table 2.2: Descriptive Statistics - Characteristics of Workers (continued) Individual IM US 97.23 83.38 24.06 75.94 ALL 86.71 100
In MSA All
Note: Number of observations equal to 214,398 workers. Statistics based on weighted outcomes. Standard Deviations in parenthesis. Male workers with positive earnings in a year. Log annual wage in 1992 dollars.
We can compare our sample of workers with the distribution and characteristics of workers from IPUMS 1990 (see appendix E.1 we ?nd interesting di?erences. To build the comparable sample, we only look at male workers, older than 16, and not working in Agriculture, Mining nor Public Administration sectors. One main di?erence is the average year of school between our sample and IPUMS. In IPUMS, both immigrant and natives have more year of schooling. Our sample has a very low proportion of college graduate workers (natives and immigrants). This low representation of this group could be driven by the over representation of small ?rms in our sample versus IPUMS database, and the types of workers that these ?rms hire.37 Natives have higher wages, but the wage di?erential between natives and immigrants is higher in IPUMS (around 12%) than in our sample (around 1%). Interestingly, natives in our sample are younger than the national average. By race, the distribution of workers is very similar. The proportion of immigrants in our sample is around 24% versus 13% for the national average. In sum, our sample contains younger male workers with low educational attaintments.
Ipums database does not include ?rm’s size. Therefore, we cannot control for the size of the ?rms.
37
40
The share of workers with a high school diploma or less is over 60% for both immigrants and natives. Immigrants are more concentrated in the high school dropout and high school graduate categories. Looking at sectoral distribution, both foreign and native workers are concentrated in Construction, Manufacturing, Retail and Services, with natives more likely to be in Construction and immigrants in Manufacturing.38 Foreigners are more likely to be working for immigrant owners than native workers. 43% of immigrant workers are employed in immigrant ?rms and 49% are employed in native ?rms. Asian and Hispanic-owned ?rms employ more immigrant workers than the average ?rm. More than forty percent of immigrant employees are hired by immigrant owners (around 43%). Most of the immigrants are Hispanics or Asians, while natives are mainly either white or black. Although there is a fraction of native-Hispanic and native-Asian workers, these proportions are less than 5%. The racial and ethnic categories follow the SSA codes, which form a set of mutually exclusive and collectively exhaustive categories. I also include information on whether the worker is full or part time. A worker is full time if he or she has worked during the full year (worker has positive earnings all four quarters). Most of the survey corresponds to information from ?rms located in MSAs. However, we include a variable that identi?es those ?rms and workers located outside a MSA. Almost 90% of the workers holds jobs in a ?rm located inside a MSA. Looking at place of birth in detail, Mexican, Salvadorian, Indian, and Chinese
One explanation for this pattern is that informal and undocumented immigrants workers are not largely covered by the database.
38
41
workers are the most represented immigrant groups in the data. At the national level, these are also the largest immigrant groups in the US according to Census 1990. In the data, native owners employ almost 75% of the total workforce.
2.4.5 Measuring coworker share
As described further in 3 below, we calculate the immigrant coworker share by considering all workers at the ?rm aside from the sample worker using the following formula:
1 COWij = Ik empj ? 1 emp
j
k=i
(2.1)
Where Ik is one when the worker is an immigrant. Therefore, this measure equals the fraction of immigrant coworkers of an employee in a ?rm. This measure is generally used in concentration analysis.39 Here I use it as an indication of workforce composition in the ?rm.
2.5 Analysis of New Hires, Earnings of Workers and Skill Distribution 2.5.1 New Hires
For the analysis of hiring procedures, we look at the type, race and ethnic composition of new hires by type of owner. During the period of analysis (199239
Hellerstein and Neumark [2007]; Aslund and Skans [2005a], Aslund and Skans [2005b].
42
1996), there were 147,373 new hires. We identify a new hire in the data by following a ?rm and looking at those workers that accessed the sample during the period of analysis. We track information on each new worker. Table (2.3) shows the distribution of new hires by type of owner. While new hires include a large share of natives for every type of owner, the proportions of newly hired immigrants for immigrant and mixed-owned ?rms (more than 30%) are almost three times the proportion of immigrants hired in native-owned ?rms (almost 12%). The second section of Table (2.3) displays the composition of new hires by race and ethnicity. Hispanics and Asians correspond to more than 35% of immigrantowned ?rms’ new hires. Again, this represents almost three times the proportion hired by native ?rms. Both immigrant and native ?rms hired more new workers later in the sample period as the economy recovered from the 1991-1992 recession (see Figures 2.1 to 2.4). The main diagonal shows that immigrant-owned ?rms hire more immigrants (33.33%) than the average ?rm (14.80%)), while native-owned ?rms hire more natives (88.44%) than the average ?rm (85.20%). Table 2.3: Average Race and Ethnic Composition of New Hires by Owner’s Type Owner Type Worker type/ race/ethnicity Immigrant Mixed Native All Immigrant 33.30 37.10 11.56 14.80 Native 66.70 62.90 88.44 85.20 Hispanic Asian White Black 20.14 16.09 48.20 6.49 22.32 12.78 49.80 8.41 10.14 11.50 2.65 4.26 73.61 70.40 8.68 8.46
Note: Number of Observations equal to 147,373. Male workers with positive earnings in a year. The other race/ethnic groups represent 0.5% of the sample. Results are not shown.
43
If we further look at the distribution of new hires across owner’s race, the ?ndings are stronger. Table 2.4 shows a strong correlation between the race of the owner and the racial/ethnic composition of the new hires.40 Asian, Black and Hispanic-owned ?rms hire their own type more than twice as often as the average ?rm. Asian-owned ?rms also hire Hispanic in a large number. Table 2.4: Average Race and Ethnic Composition of New Hires by Owner’s Race Worker / Owner Asian Asian 23.75 Black 6.29 Hispanic 20.10 White 39.03 Black 2.50 38.54 10.52 44.43 Hispanic White 4.27 2.70 10.33 8.04 35.46 9.10 42.24 75.39 All 4.25 8.46 11.50 70.42
Note: Number of Observations equal to 147,373. Male workers with positive earnings in a year. The other race/ethnic groups represent 0.5% of the sample. Results are not shown.
2.5.2 Earnings of Workers
In this section, we look at workers’ earnings. On average, immigrant workers have lower wages than natives. Most of the explanations given by the literature are based on human capital formation. Immigrants have lower host country abilities and generally less education than natives. However, even after controlling for some of these characteristics, immigrants tend to receive lower wages than observationally similar natives (Borjas [1994]). But do workers receive di?erent wages than their counterfactual group regardless of who they work for? To answer this question we undertake two di?erent exercises. First, we look at the average real log annual
Giuliano et al. [2006] ?nd a similar correlation when they look at the race of the hiring manager and the racial composition of the new hires in di?erent establishments of a retail store.
40
44
Figure 2.1: Workforce Characteristics of Immigrant, Mix and Native Firms
Note: Weighted share and percentage. Base on years 1992-1996.
earnings of each worker type across owner types. We also look at these statistics for di?erent groups of ?rms de?ned by the fraction of similar coworkers in the ?rm. This 45
Figure 2.2: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
analysis is a ?rst look at the impact of ?rm owner types on earnings. Second, we estimate owner type wage e?ects after controlling for a number of ?rm and worker characteristics, and evaluate the sources of wage di?erentials. The natural log of real annualized earnings of each worker comes from LEHDUI records.41 Table (2.5) shows how average wages change according to the type
41 When we take the average log annual earnings for each type of ?rm, we ?nd that it is slightly below the log of annual payroll per employee in the SSEL database. According to internal documentation on the ES202/SSEL joint project, annual payroll in SSEL ?les includes non-wage payments, such as bene?t payments, retirement pension funds, annuity funds, supplemental bene?t funds, etc, which are not included in the UI ?les.
46
Figure 2.3: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
47
Figure 2.4: Workforce Characteristics of Immigrant, Mix and Native Firms Continuation
Note: Weighted share and percentage. Base on years 1992-1996.
48
of owner. The last column of the table shows the t-test computed for worker type wages for each owner type. A t-test can reject the null hypothesis that the mean of immigrant worker wages and the mean of native worker wages are the same at the 90% level. Table 2.5: Mean Earnings by Owner and Worker Type
Variable=log(annual earnings) owner = Immigrant Immigrant Native All owner = Mix Immigrant Native All owner = Native Immigrant Native All (%) 50.30 49.70 100.00 35.94 64.06 100.00 15.87 84.13 100.00 Mean 8.35 8.12 8.23 8.52 9.04 8.71 8.32 8.38 8.37 STD 1.47 1.67 1.64 1.86 1.71 1.82 1.73 1.88 1.73 T-test
24.20
-16.07
-5.83
Note:STD indicates standard deviation. Log annual wage in 1992 dollars. Using workers during the period 1992-1996.(*)T-tests are computed on the di?erence between average wages of immigrant and native workers for each speci?ed owner type.
Looking at Table (2.5) we notice three relevant outcomes for wage di?erential analysis. First, immigrants are paid slightly less by native than by immigrant owners. On average, they are paid the lowest when working for native owners. Second, native workers are paid signi?cantly less in immigrant owned businesses. Third, on average native owned ?rms pay more than immigrant owned ?rms. Fourth, mix-owned ?rms signi?cantly pay less to immigrant workers. However, these ?rms employ a lower proportion of immigrant workers than immigrant-owned ?rms. In sum, immigrant workers end up receiving lower log annual earnings than native workers. If we combine the ?rst three outcomes, we can see that much
49
Table 2.6: By Similar Coworker Share: Mean Earnings by Owner and Worker Type
Coworker Share Below the median Above the median (%) Mean STD (%) Mean STD 33.64 66.69 48.09 26.42 66.36 39.07 6.96 91.48 20.33 7.37 7.98 7.74 7.90 8.67 8.32 7.74 7.80 7.78 1.71 1.68 1.51 1.79 1.63 1.75 1.81 1.77 1.79 66.36 33.31 51.91 73.58 33.64 60.93 93.04 8.52 79.67 7.67 8.19 7.82 8.39 6.91 8.13 8.38 7.68 8.31 1.53 1.34 1.70 2.98 1.21 1.98 1.89 1.96 1.92
Variable=log(annual earnings) owner = imm Native Immigrant all owner = mix Native Immigrant all owner = usa Native Immigrant all
Note: STD indicates standard deviation. Log annual wage in 1992 dollars. Statistics based on estimation sample: all male individuals working between 1992 and 1996.
of the di?erence between the log annual wages of immigrants and natives comes from immigrants’ propensity to work in immigrant owned ?rms. These ?rms pay the lowest wages, and the di?erence in immigrant earnings between immigrant and native ?rms is small. Additionally, native owned ?rms pay immigrant workers less than native workers (see Table(2.5)). It is important to highlight the relevance of having actual earnings of each employee at the ?rm level, so we can exploit these variations to identify the e?ect of owner types on individuals’ wages. Therefore, individual level wages are used in the regressions analyzed in the next sections. Table (2.5) would not be possible if we didn’t have data on both employers and employees’ characteristics. Our unique database allows us to compare average earnings between workers of di?erent types holding a job in the same type of ?rm, and workers of the same type (native or
50
immigrant) working for di?erent types of owners. We now perform a similar exercise, but separating ?rms by the share of coworkers similar to the worker called ”similar coworker share” (see Table 2.6). This measure is di?erent from the measure of immigrant coworker share de?ned previously, in that here we de?ne the similar coworker share as the share of workers that are of a similar type to a particular worker in a speci?c ?rm. For instance, the coworker share of a native worker is the share of native born workers in the ?rm excluding the worker. The second column (%) shows the percentage of workers of each type in the ?rm accordingly below or above the similar coworker share median. We can see in the table that the previous ?ndings in Table 2.5 remain valid. Foreign-born employers pay the lowest wages, on average. However, for businesses with coworker share below the median, immigrant employees working for immigrant employers are paid slightly more than immigrant employees working for native employers. Additionally, workers are paid more when working with similar coworkers. When workers’ similar coworker share is below the median, employers pay lower annual wages. More than 65% of sample businesses have a mixed workforce, that is, the share of immigrant coworkers is neither one nor zero (0 < share < 1). These tables do not control for individuals’ characteristics, so we don’t know the pro?les of native and foreign employees holding jobs in these businesses. Nevertheless, these ?ndings are striking. Immigrant owners pay the lowest on average. Furthermore, they to pay natives less than the rest of the market. This motivates the question of what type of native workers work for immigrant employers.
51
2.5.3 Sorting by Skill
Sorting by skill is a possible cause of sorting by owner type. The incentive to combine workers of identical skills within the same ?rm has been documented previously (Kremer and Maskin [1996]). Job descriptions and skill requirements are also a concern as characteristics of employers and employees are correlated. Additionally, if ?rms of di?erent types have di?erent skill mix productivity, that is, they use a combination of workers’ skills and capital di?erently, then the di?erences in the probability of hiring a speci?c type of worker could be motivated by the capital/labor ?rm’s decisions. For instance, immigrant owners could use labor more intensively than native businesses, or could hire more low-skilled workers than native ?rms. Immigrants, Hispanics, and other minority groups have lower skill on average so they may tend to work in low-skill sectors and low-skill jobs regardless of the owner type. Immigrant owners, on the other hand, may tend to concentrate in lowskill sectors because they also have low skill levels. For both group, the mayority of the ?rms are in the ’Low-Skill Industries’. Almost 30% of immigrant-owned ?rms belong to the ’High-Skill Industries’ group, while more than 45% of native-owned ?rms belong to this group.
52
Table 2.7: Worker types distribution by owner’s skill requirement
Worker / Owner Immigrant Native Race/ethnicity Hispanic* Asian Black White (non-hispanic) All Low-Skill Industries Immigrant Native Mixed 38.60 11.60 27.50 61.40 88.40 72.50 19.23 18.22 5.38 48.71 70.83 10.12 2.63 6.83 76.18 54.16 16.78 8.23 7.62 61.38 62.24 All 15.40 84.60 11.40 4.64 6.68 72.47 100.00 High-Skill Industries Immigrant Native Mixed 33.70 9.20 41.60 66.30 90.80 58.40 17.73 17.09 6.45 52.22 29.17 6.44 2.97 10.84 75.49 45.84 20.05 24.30 4.45 42.07 37.76 All 11.50 88.50 7.43 4.34 10.42 73.28 100.00
53
Note: Using Census 1990 information on workers’ education attainment by industry, industries are separated into High Skill and Low Skill. High skill refers to those industries in which more than 50% of workers have at least a high school diploma. Otherwise we de?ne the industry as low skill. (*) Hispanic refers to all races with ethnic group Hispanic. The group Other includes Native American and otherwise unclassi?ed racial groups. Native-American workers represented only 0.5% of the total sample.
Table (2.7) shows workers’ distribution by owner’s skill requirement. The skill requirement for a ?rm is computed using Census 1990 data after compiling the share of workers by industry at the 2-digit level that have low educational attainment(less than high school) and high educational attainment(more than high school). High skill industries are those in which more than 50% of workers have at least a high school diploma. The remaining industries are low skill. The idea is to illustrate whether speci?c owner and worker types are concentrated in a particular skill group. Not surprisingly, the table shows that ?rms in low-education industries have higher fractions of immigrant workers than ?rms in high-education industries. Immigrant ?rms continue to have a bigger proportion of immigrant workers, except for mix-owned businesses. Results are similar breaking down by workers’ race. However, it is worth mentioning that immigrant-owned ?rms are more than 60% of the group of low-skill ?rms. To account for part of this pattern, in the regressions below we include the share of workers in the ?rm in four education categories: high school dropouts, high school graduate, some college, and college graduate.
2.6 Regression Analysis
The ideal data to analyze the e?ect of owners, coworkers, and social connections on individual labor market outcomes requires information on individuals’ labor market histories, earnings, and, speci?cally, the employer’s source of ex-ante information about the job seekers that apply to its open vacancies. With this information
54
we would be able to measure the actual hiring policies that ?rms use to ?nd new workers. Unfortunately, we don’t have detailed data on hiring procedures used by ?rms. However, we do have a good deal of valuable information on the ?rms and workers. Workers can be divided into di?erent categories by birth place or by race/ethnicity42 to infer workers’ and candidates’ likely social connections. This, together with information on the type of owner, will help us infer the use of social ties in the ?rm’s hiring process and its e?ect on workers’ earnings. More speci?cally, network structure refers to the number of ties an individual has (Smith, 2000). In this paper, we try to identify the impact of networks by using the proportion of coworkers who are potentially tied to a newly hired worker. Besides identifying the type of owner for whom the employee works, I use the proportion of similar employees in the ?rm at the time the new worker is hired as a measure of the network link between coworkers, employers, and the new worker. Following each ?rm from 1992 to 1996, we obtain the number of employees who work for the ?rm and their earnings. We also have the total number of workers possessing any given set of demographic characteristics at each period of time. Following the de?nition of networks used in previous literature, we compute the share of similar coworkers for each new hire at each ?rm in each period, assuming that a similar birthplace or ethnicity implies at least a weak network connection between individuals.43
White, Black, Hispanic, and Asian. At this point, it is worth to mention that even though immigrants are very diverse and it is a group that re?ects a multiple gamma of ethnic/cultural backgrounds, not necessarily captures by the denomination of being foreign-born, it is also true that immigrants tend to have similar
43 42
55
A key challenge in linking owners and employees is that the characteristics of both owners and employees may be correlated with other characteristics of a workplace and its location. Section 2.5.3 above gives preliminary evidence on sorting by skill. The correlation between owner and employee types could also be a result of residential segregation of workers and owners (spatial mismatch). Job descriptions and skill requirements are also a concern, as characteristics of employers and employees are correlated. Immigrants, and in particular Hispanics, tend to be low skilled and therefore are likely to work in low-skilled sectors and low-skilled jobs regardless of the owner type. However, at the same time, immigrant owners could tend to concentrate in low-skill sectors, perhaps because they also have low skill levels. Because the proportions of immigrants are unequally distributed across sectors and regions, we control for the 2-digit industry and geographic location of each ?rm. There exist sectors such as Retail, Services and Construction where immigrants represent a signi?cant proportion of the workforce
44
. We also see this pattern in the
geographic distribution of the immigrant population. For instance, according to Census 2000, Los Angeles and New York represent more than 30% of the total immigrant population in the country. To account for these concerns we need to control for ?xed attributes of the workplace and the local labor market, and also for local trends in labor pool demographics. Therefore, we estimate the model controlling for
strategies to enter into the labor market regardless of their cultural background. Using migrant networks is one common factor among foreign-born workers, especially for new immigrants (Porter and Wilson [1980], Light [2006]). 44 This can be also related to the fact that these sectors are also highly represented by relatively smaller ?rms than in Manufacturing, for instance.
56
characteristics of the ?rm (Fj ) and local community (Zj ). These controls include the immigrant workforce population and population density in the local community, 2-digit industry code dummies, ?rm size (log of reported employment), and legal form of organization. We also include the share of the ?rm’s workers in the four education categories discussed previously. Previous research has noted the impact of English language ability in the use of networks and the level of wages for immigrant workers.45 We capture this feature by interacting the 2-digit industry dummy with an English speaker dummy
46
This
interaction is a proxy that intends to capture whether language is used di?erently in di?erent industries. In the wage regressions, we also control for individual characteristics (Xj ), including worker’s age, education and a dummy for working full time.47 The composition of the labor pool might also be a?ected by changes over time in labor supply and demand. For example, white natives may be more likely to work in low-wage retail jobs when labor markets are weak. Therefore, we also include a dummy variable for each of the years in the sample (Mt ) to control for national ?uctuations in the labor market. The identi?cation strategy exploits variation across owner types for otherwise similar ?rms. By controlling for a rich set of ?rm characteristics we can narrow the possible alternative explanations for any residual correlation between owner type
Hellerstein and Neumark [2007] and Hellerstein et al. [2008a]. For additional analysis, see 3 below. 46 We identify a group of countries where English is the main language, and use this information to identify the worker as English speaker or otherwise. 47 A worker with positive time during the complete year is considered full quarter worker or full year worker.
45
57
and worker outcomes.
2.6.1 Analysis of ?rms hiring patterns
This section starts by looking at the hiring patterns of the ?rm, estimating a model that predicts the probability that a newly hired employee is an immigrant. Firm hiring decisions indirectly re?ect the way owners use current employees to help ?ll their job vacancies. We use a linear probability model to estimate the likelihood that a newly hired worker is of a particular type (immigrant or from a speci?c race/ethnic group).48
Pr(new hire:groupi )kjt = c + B1 ? Oj + ? ? Wjt?1 + B2 ? Oj ? Wjt?1 + ? ? Fj + Z ? Zkj + T ? Mt +
kjt
(2.2)
Where k , j and t designate the worker, ?rm type, and time respectively. Oj is a vector of dummy variables for owner type (de?ned by immigration status or race). If i refers to the group of immigrant workers, we use as the reference group ?rms owned by immigrants. B1 represents the vector of coe?cients associated with the impact of owner type on hiring. The elements of this vector are expected to be negative when the omitted group is the same type as the new hire. For instance, the coe?cient on native owners would be negative if immigrant-owned ?rms are more likely to hire new immigrant workers. Wjt?1 corresponds to the vector of the proportion of workers of
We use a linear probability model over a Probit (Logit) model because we don’t need to restrict the sample to ?rms that hire at least one new worker of each type. This restriction could introduce sample selection bias because ?rms with zero hiring could have a completely di?erent policy than those with a least one new hire.
48
58
type each type i at the ?rm in the previous period. An interaction between owner type and Wjt?1 is included to asses di?erences in use of current employees’ networks across owner types. I also control for ?rm characteristics Fj (a vector of variables measured at the ?rm level), year dummies Mt , and local community information and state dummies Zkj . In a regression with both owner type and coworker share included, the estimated coe?cient on owner type will capture only the direct impact of owner type on hiring, not the total e?ect, which will include both the direct e?ect and the indirect e?ect coming through owner type’s e?ect on coworker share. The use of employee referrals can be correlated with the type of owner and can a?ect hiring patterns if owners have the tendency to hire same-group individuals. When employees tend to refer same-group workers, the owner type’s e?ect may be ampli?ed. If we believe that the share of similar coworkers is a good proxy for social connections, these exercises illustrate the combined result of owner e?ects and hiring patterns. We assume that the error (
kjt )
in equation 2.2 is independent and identi-
cally distributed across ?rms, but not within ?rms. To correct for non spherical disturbances, we estimate Huber-White robust standard errors clustered by ?rm. This procedure is used in all subsequent estimations. We cluster the errors by ?rm since ?rms in the sample may have hired more than one worker and thus may have repeated observations. For purposes of analysis, we estimate di?erent versions of equation (2.2) and look at the impact of the addition of controls on the estimates of B1 and B2 . The ?rst regression includes only year dummies; subsequent speci?cations add controls 59
one by one. Most of the literature on hiring networks argues that current workers’ referrals are more important to ?rm hiring patterns than owners’ personal networks. Owners are likely to hire individuals from their residential area. However, current workers have a larger and more diverse set of connections that can be exploited by the ?rm. We are not able to disentangle these e?ects directly. Nevertheless, by allowing owners of di?erent groups to make use of their workers’ social ties di?erently, the estimated interaction e?ects can measure the ability of owners to use social ties. Table (2.8) shows the probability of a new hire being an immigrant given the characteristics of the ?rm, its community and the share of immigrant coworkers in the ?rm. Controlling only for year dummies, native owners are 25 percentage points less likely to hire a new immigrant worker than immigrant ?rms (column 1). This di?erence is signi?cantly reduced, to 3.5 percentage points, when we include the share of immigrant coworkers (column 2). Controlling for year and industry dummies, the share of immigrant coworker positively a?ects the likelihood of an immigrant being hired. The inclusion of the share of English speaker and its interaction with industry dummies decreases the impart of the share of immigrant coworkers on the probability of being hired. This covariates controls for whether language is used di?erently in di?erent industries (column 3). For instance, a Mexican restaurant would probably hire Mexicans or Spanish speaker because of the type of service they o?er and type of frequent consumers. The use of language can be di?erent in a industry where workers don’t need to communicate with each other, so language di?erences are not obstacle in the production process. Given the results, it seems that ?rms use language in di?erent ways, a?ecting the likelihood of an immigrant
60
Table 2.8: Linear Estimates of the E?ect of Owner Type on the Probability that a New Hire is an Immigrant
Owner Mix Owner Native % Imm. Coworkers % Imn. Coworkers * Owner Mix % Imm. Coworkers * Owner Native Corporation Sole Prop. log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Pop. % immigrant in neighborhood(+) Population in neighborhood(+) In MSA Constant Dummies year Industry Indus*English Spkr R-Square 0.4081*** 0.099 yes yes 0.29 0.0211*** 0.0039 yes yes 0.32 0.0989*** 0.002 yes yes yes 0.34 0.0969*** 0.0016 yes yes yes 0.35 (1) -0.0519*** 0.0057 -0.2358*** 0.0032 (2) -0.041*** 0.0065 -0.0351*** 0.0031 0.9961*** 0.0056 (3) -.0034** 0.001 -0.0342*** 0.0009 0.782*** 0.002 (4) -0.0037** 0.001 -0.033*** 0.0004 0.7724*** 0.0101 -0.0125** 0.005 -0.0711*** 0.003 (5) -0.00313** 0.001 -0.0254*** 0.0014 0.7132*** 0.0234 -0.0094** 0.005 -0.0378*** 0.0041 -0.00085* 0.0033 0.0026 0.003 0.003 0.002 0.0021** 0.0004 -0.0012 0.001 0.005 0.006 0.0162** 0.0068 0.0004*** 0.00 -0.005*** 0.0009 0.0285** 0.0069 yes yes yes 0.38 (FE)
0.6715*** 0.0435
0.1575* 0.781 yes 0.41
Note: Reference group is immigrant ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Neighborhood is de?ned counties adjacent to the county where the ?rm is located. Population in 100,000’s. FE represents the ?rm ?xed-e?ect model. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
being hired and reducing the impact of immigrant coworkers in the ?rm. There is a positive and signi?cant impact on the probability of the new hire
61
being an immigrant when the proportion of workers in the ?rm with low education (high school dropout) increases. The owner e?ect diminishes and the di?erence in the probability of hiring a Hispanic between immigrant and native owners is 2.5 percentage points (column 4). The coworker e?ect is smaller too, although it is still signi?cant. The interaction e?ects between owner type and coworker share decrease slightly when others controls are included, although the results are similar. The e?ect of immigrant coworker share is smaller in mix and native owned ?rms than in immigrant owned ?rms. Immigrant employers can take advantage more e?ciently of their current immigrant workers than other types of employers. The increment of immigrant coworker share by 1 percentage point increases this likelihood by 0.710.67. The inclusion of other characteristics of the ?rm and the local community has a smaller impact on the relative likelihood of native versus immigrant owners hiring a new immigrant worker. We should be cautious when analyzing these results. We include a vast series of covariates to control for all possible observables that can be correlated with employer and employee e?ects. However, the presence of unobservables correlated with ?rm and worker interactions could bias the results. As another exercise, we compute the ?rm ?xed-e?ect version of the model by including ?rm dummies. The last column of Table (2.8) shows the results. The impact of share of immigrant coworkers in the ?rm at the time of the new hire remains positive, high, and signi?cant.
62
2.6.2 Hiring Process by Race/Ethnicity
We next consider the determinants of the probability that a new hire comes from a particular race/ethnic group: white, black, Hispanic and Asian. That is, we estimate equation (2.2), setting i equal to a particular racial category. Tables 2.9 and F.1 show the e?ects of owner types and shares of type i coworker, and other types of coworker, at the time of hiring on the probability that a new hire is Hispanic, Asian, white, or black respectively.
2.6.2.1 Worker Race
The likelihood of a new worker being Hispanic or Asian signi?cantly decreases when the employer is native. This result holds even after including a exhaustive list of controls(Tables 2.9 and 2.10). The direct impact of owner type is reduced, however, once we control for the share of Hispanic coworkers. For instance, having a one percentage point increment of Hispanics as current employees in the ?rm increases the probability that a new hire is Hispanic (by up to 0.88 in immigrant owned ?rms). The impact of Hispanic coworkers is smaller for native owned ?rms. Table 2.9: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic
(1) 0.0214*** 0.0041 -0.0903*** 0.003 (2) 0.012 0.01 -0.0872*** 0.003 Hispanic (3) (4) 0.0077 0.014 0.005 0.0357 -0.0412*** -0.0245** 0.003 0.001 0.9441*** 0.0054 -0.628*** 0.013 (5) -0.0694 0.054 -0.0172** 0.001 FE
Owner Mix Owner Native Hispanic Cowkrs Asian Cowkrs
-0.526*** -0.504*** 0.023 0.029 Continued on next page.
63
Table 2.9: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Hispanic (continued)
(1) White Cowkrs Black Cowkrs Hispanic Cow* Owner Mix Asian Cow* Owner Mix White Cow* Owner Mix Black Cow* Owner Mix Hispanic Cow* Owner Native Asian Cow* Owner Native White Cow* Owner Native Black Cow* Owner Native log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Work. Pop. Total.1 Work. Pop. %Hisp 1 Constant year dummies Industry dummies State dummies Other controls(+) p-value R-Square (2) Hispanic (3) (4) -0.703*** 0.0095 -0.879*** 0.0123 -0.0621** 0.031 0.1060** 0.0483 0.0185 0.0384 0.137 0.0845 -0.0869*** 0.043 -0.113*** 0.0295 -0.148** 0.062 -0.094* 0.04 (5) -0.681*** 0.0075 -0.725*** 0.0134 FE -0.596*** 0.0197 -0.616*** 0.0212
0.093** 0.034 0.0175 0.0434 0.105 0.0945
0.1920*** 0.0032 yes 0.0001 0.22
0.1243*** 0.009 yes yes 0.002 0.29
0.9702*** 0.08 yes yes yes 0.0001 0.31
0.8454*** 0.1616 yes yes yes 0.003 0.34
-0.102*** 0.0243 -0.124** 0.056 -0.097* 0.04 0.0013** 0.00 0.0012*** 0.0005 0.0025*** 0.0003 0.0030*** 0.0008 0.0561** 0.0245 0.094*** 0.002 0.9511*** 0.171 yes yes yes yes 0.003 0.42
0.8411*** 0.201 yes 0.01 0.35
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
64
The e?ect of the share of Hispanic coworkers is positive regardless the type of the owner. However, the e?ect is smaller than the baseline e?ect on the Hispanicowned ?rms (column 3). Columns 4 includes the e?ect of all races coworker share on the likelihood of being hired. Other races coworker shares a?ect negatively the probability of a new hire is Hispanic. Interestinly though, Asian coworker share is less negative when the ?rm is mix-owned. Column 5 includes other ?rm and local community characteristics. Their inclusion decreases the average e?ects, but do not change the directions of the results. In section (2.5.3) we discussed the distribution of workers by average industrylevel skill requirement. As a proxy to control for this e?ect, we include the ?rm’s share of workers in four education categories and the fraction of workers of similar type in the local community. The results show that a higher share of low-educated workers in the ?rm increases the probability that the new worker is Hispanic. We also include the share of workers of each racial group in the local labor force. The inclusion of these shares decreases the impact of the coworker shares. Looking at Asian new hires (Table 2.10), we again ?nd that native employers are less likely to hire Asian workers. The inclusion of additional controls reduces the di?erence in probability of hiring an Asian between immigrant and native owned ?rms. Another interesting result is that Asians are less likely to be hired in ?rms with bigger proportion of workers with education attainment below the high school level.
65
Table 2.10: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian
(1) -0.029*** 0.0032 -0.1245*** 0.002 (2) -0.0280*** 0.0032 -0.1114*** 0.002 Asian (3) (4) 0.002 0.001 0.003 0.002 -0.054** -0.052** 0.002 0.002 -0.743*** 0.0079 0.8194*** 0.031 -0.7876*** 0.0071 -0.8795*** 0.0083 -0.0197 0.0303 0.007 0.002 -0.0746*** 0.0022 0.032 0.0446 -0.0064 0.0185 -0.152*** 0.013 -0.0158 0.0154 -0.0076 0.02 (5) 0.007 0.002 -0.06** 0.014 -0.712*** 0.008 FE 0.1948 0.254
Owner Mix Owner Native Hispanic Cowkrs Asian Cowkrs White Cowkrs Black Cowkrs Hispanic Cow* Owner Mix Asian Cow* Owner Mix White Cow* Owner Mix Black Cow* Owner Mix Hispanic Cow* Owner Native Asian Cow* Owner Native White Cow* Owner Native Black Cow* Owner Native log(employment) Share of workers with HSD (?rm) Share of workers with HSG (?rm) Share of workers with SCG (?rm) Work. Pop. % Imm.1 Work. Pop. %Hisp 1 Work. Pop. % Asian 1 Constant year dummies Industry dummies State dummies
-0.622*** 0.008
-0.741*** 0.0072 -0.8214*** 0.0081 -0.0556 0.041
-0.6715*** 0.0074 -0.7631*** 0.0083
-0.076*** 0.0022 0.0404 0.0536 -0.0064 0.0185
-0.031 0.0221 -0.0095 0.02 0.0014** 0.00 -0.0012** 0.0001 -0.000 0.00 -0.0013** 0.00 0.024* 0.001
0.1495*** 0.0018 yes -
0.112** 0.0562 yes yes -
0.065** 0.02 yes yes yes
0.095*** 0.01 yes yes yes
0.043** 0.01 0.094*** 0.097 0.01 0.081 yes yes yes yes Continued on next page.
66
Table 2.10: Linear Probability Estimates of the E?ect of Owner Type on the Probability that a New Hire is Asian (continued)
(1) 0.0001 0.26 (2) 0.002 0.31 Asian (3) 0.0001 0.34 (4) 0.003 0.35 (5) yes 0.003 0.37 FE 0.01 0.38
Other controls(+) p-value R-Square
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
Whites and blacks are more likely to be hired by native ?rms (See Tables (F.1) and (F.2)). However, the probability that a new hire is black or white depends on the share of blacks or whites in the ?rm at the time of the recruitment process. The signi?cance of the immigrant owner e?ect on black hiring vanishes when I include the black coworker share in the regression. The column FE shows the results of the regression after including ?rm ?xed e?ects. The impact of similar coworkers decreases slightly but is still high and signi?cant. The largest change in coe?cients caused by the inclusion of ?xed e?ects is the drop in the impact of white coworkers on the probability of that a new hire is black. We also experiment with estimating a multinomial logit model to account for the posibility that employers may simultaneously choose among di?erent types of workers. The estimation sample is then restricted to ?rms that hire at least one worker of each race group during the period 1992-1996. This restriction eliminates more homogeneous ?rms. The new sample contains 2,662 ?rms out of the original sample of 4,478 ?rms. We investigate how the owner type and shares of di?erent
67
types of workers at the time of hiring a?ect the type/race of the new hire. We estimate a model49 that aims to reveal whether the birthplace of the employer a?ects the likelihood that a new worker is of the same type as opposed to other types, conditional on having accessed to the ?rm during the period of analysis and controlling for the characteristics of the worker and the ?rm.
Pr(new hire is worker type: i)kjt =
i exp(ci + B1 ? Oj + ? i ? Wjt?1 + ?i ? Fj + Z i ? Zkj + T i ? Mt + 5 s=1 i kjt ) s kjt )
s exp(cs + B1 ? Oj + ? s ? Wjt?1 + ?s ? Fj + Z s ? Zkj + T s ? Mt +
(2.3)
with i = 1, ..., 4 for the four race groups: white, black, Asian, and Hispanic. This procedure makes very strong assumptions with respect to the relevance of other alternatives. The odds ratio of any two options is assumed independent of the other alternatives. This feature is important to consider when more than two alternatives are included. To test the Independence of Irrelevant Alternatives assumption, we conduct a Hausman test by excluding each outcome category in turn. The test indicates that I cannot reject the null hypothesis that the odds of one outcome happening are independent of other alternatives. Additionally, we perform Wald tests for combination of categories. The tests reject the null hypotheses that all coe?cients associated with a given pair of outcomes are zero (except intercepts). We cluster the errors by ?rm since observations within ?rms are not independent. The results for this regression are shown in Tables (2.11) and (2.12).
I speci?cally estimate a mixed logit model that incorporates both characteristics of the individual and the alternatives.
49
68
Table (2.11) shows the change in log odds comparing two alternatives. The share of white coworkers signi?cantly increases in the log odds of a white being hired. We also show the predicted hiring probabilities for each owner type (Table 2.12) computed at the means of all ?rms and dummy variables. The change in log odds between hiring a white worker versus hiring a Hispanic or an Asian decreases when the ?rm is immigrant-owned. Immigrant owners are 3 percentage points more likely to hire Asians and Hispanics than native ?rms. These results support the analysis in the previous section.
69
Table 2.11: Multinomial Logit Model: E?ects of Owner Type and Coworkers on Type of New Hires
Cow. Share White Black Asian Hispanic White to Black 1.97*** 0.646 -5.352*** 0.892 -1.391 0.951 -0.236 1.03 Change in log odds comparing alternative 1 to alternative 2 White to Asian White to Hispanic Black to Hispanic Black to Asian 2.32*** 3.53*** 1.44* 0.53 0.761 0.421 0.71 0.723 1.186 2.145*** 7.456*** 5.456*** 1.086 0.661 0.957 0.968 -7.243*** 0.041 1.433 -5.682*** 1.001 0.591 1.108 0.946 -0.086 -3.675*** -3.127*** 0.15 1.102 0.527 1.09 0.952 Asian to Hispanic 0.92 1.017 1.014 1.131 7.126*** 1.143 -3.654*** 1.361
Note: Other controls include log of employment, percentage of immigrant workers in the surrounding counties, population in the county, legal form of organization, Msa location, 2-digit industry, interaction 2-digit industry and English speaker dummy, state and year dummies. Results from race/ethnicity ’others’ are not shown. Number of observation 135,583 workers, and 2,662 ?rms. Robust standard errors in italic allow for arbitrary correlation within the same ?rm. * signi?cant at 10%,** signi?cant at 5%, *** signi?cant at 1%.
70
Table 2.12: Multinomial Logit Model: Predicted Probability of Covariates
Owner Native Immigrant Mix White 0.740 0.710 0.690 Workers Black Asian 0.120 0.031 0.102 0.060 0.119 0.052 Hispanic 0.100 0.126 0.134
Note: Based on multinomial logit predictions of the race of new hires from previous table.
2.6.2.2 Worker and Owner Races
After looking at the e?ect of owner birthplace on the probability of being hired for each particular worker’s race, the natural question is whether we can detect similar e?ects when we separate owner types by race. As explained in Section 2.4.1, owner’s race is obtained from the Small Minority Owner Business Employers Survey(SMOBE). For multiple-owned ?rms, the median race is used; in the case of ties, the hours worked in the ?rm are also considered to determine the predominant race of the ?rm. The race categories are: white, black, Asian and Hispanic. The likelihood of a new worker being Hispanic or Asian signi?cantly decreases when the employer is White. This result holds even after including a exhaustive list of controls(Tables 2.13 and 2.14). The direct impact of owner type is reduced, however, once we control for the share of Hispanic coworkers. For instance, an increment of one percentage point in the share of Hispanics as current employees in the ?rm increases the probability that a new worker is Hispanic (by up to 0.95 in Hispanic owned ?rms). Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic
(1) -0.2215*** 0.0271 -0.135*** 0.0229 -0.2358*** 0.0154 (2) -0.1514*** 0.0085 -0.1264*** 0.0043 -0.176*** 0.0034 Hispanic (3) -0.024** 0.0112 -0.0257*** 0.0076 -0.0318*** 0.0064 0.9512*** 0.0176 (4) -0.0284** 0.015 0.0868 0.7279 -0.0231*** 0.0015 (5) -0.0165** 0.0013 0.1429 0.1309 -0.0158*** 0.0045
Owner Black Owner Asian Owner White Hispanic Cowkrs Asian Cowkrs
-0.918* -0.9114*** 0.0669 0.0707 Continued on next page.
71
Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic (continued)
(1) White Cowkrs Black Cowkrs Other Cowkrs Owner Black* Hispanic Cowkrs Owner Black* Asian Cowkrs Owner Black* White Cowkrs Owner Black* Black Cowkrs Owner Black* Other Cowkrs Owner Asian* Hispanic Cowkrs Owner Asian* Asian Cowkrs Owner Asian* White Cowkrs Owner Asian* Black Cowkrs Owner Asian* Other Cowkrs Owner White* Hispanic Cowkrs Owner White* Asian Cowkrs Owner White* White Cowkrs Owner White* Black Cowkrs Owner White* Other Cowkrs Share of workers with HSD (?rm) Share of workers with HSG Share of workers with SOG Log employment Work. Pop. Total Work. Pop. -0.0335** 0.0145 0.0367 0.2128 0.0132 0.0678 -0.0382** 0.0109 0.0048 0.1592 -0.0427* 0.0252 -0.0531 0.0711 -0.0869** 0.0326 -0.1764*** 0.0493 -0.2705*** 0.0627 -0.09*** 0.0194 -0.1057* 0.0748 -0.0811** 0.0296 -0.1071*** 0.0385 -0.184*** 0.054 0.0016*** 0.0005 0.0022*** 0.0003 0.0032*** 0.0009 0.0011* 0.0009 0.0422*** 0.024 0.0002** Continued on next page. -0.2077*** 0.0703 -0.1218*** 0.0254 -0.2015*** 0.0342 -0.2358*** 0.0486 -0.1636** 0.0756 -0.1622*** 0.037 -0.144*** 0.0533 -0.33*** 0.0676 0.0922 0.2218 0.0024 0.0748 0.0063 0.0783 0.092 0.1826 (2) Hispanic (3) (4) -0.9086*** 0.0231 -0.8322*** 0.0317 -0.6857*** 0.0434 (5) -0.6494*** 0.0269 -0.6335*** 0.0349 -0.5814*** 0.0478
72
Table 2.13: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Hispanic (continued)
(1) % Hisp. Constant Year dumies Industry dummies State dummies Other Controls p-value R-Square 0.3286*** 0.0136 yes 0.01 0.24 (2) 0.0436* 0.0175 yes yes yes 0.01 0.29 Hispanic (3) 0.0211 0.0828 yes yes yes 0.01 0.31 (4) 0.8779 0.0864 yes yes yes yes 0.01 0.33 (5) 0 0.6446 0.1755 yes yes yes yes 0.01 0.45
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
The impact of Hispanic coworkers is smaller for other types of ?rms. The results for other characteristics of the ?rm and its location are similar to previous sections. The results show that a higher share of low-educated workers in the ?rm increases the probability that the new worker is Hispanic. I also include the shares of coworkers in each racial group. Black and White owned ?rms are 2 to 3 percentage points less likely to hire a Hispanic worker compared to Hispanic and Asian owned ?rms, holding constant the worker race distribution. Looking at Asian new hires (Table 2.14), we ?nd that white employers are less likely to hire Asian workers. White owners are mostly natives. The inclusion of additional controls reduces the di?erence in probability of hiring an Asian between Asian and white owned ?rms. Another interesting result is that Asians are less likely to be hired in ?rms with bigger proportion of workers with educational attainment below the high school level.
73
Table 2.14: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian
(1) -0.181*** 0.0067 -0.1977** 0.0215 -0.187*** 0.0049 (2) -0.1771*** 0.0054 -0.1842*** 0.0027 -0.1775*** 0.0021 Asian (3) -0.06*** 0.006 -0.013 0.0034 -0.049*** 0.0026 0.98*** 0.0074 (4) -0.2035* 0.1164 -0.0333*** 0.00419 -0.1257*** 0.0149 (5) -0.1628*** 0.0268 0.0503 0.0462 -0.1749*** 0.0163
Owner Black Owner Hispanic Owner White Asian Cowkrs Hispanic Cowkrs White Cowkrs Black Cowkrs Other Cowkrs Owner Black* Asian Cowkrs Owner Black* Hispanic Cowkrs Owner Black* White Cowkrs Owner Black* Black Cowkrs Owner Black* Other Cowkrs Owner Hispanic* Asian Cowkrs Owner Hispanic* Hispanic Cowkrs Owner Hispanic* White Cowkrs Owner Hispanic* Black Cowkrs Owner Hispanic* Other Cowkrs Owner White* Asian Cowkrs Owner White* Hispanic Cowkrs Owner White* White Cowkrs Owner White* Black Cowkrs Owner White* Other Cowkrs Share of workers
-0.987*** 0.0149 -0.920*** 0.0101 -0.978*** 0.0202 -0.9928*** 0.0237 -0.2765*** 0.1041 0.1801 0.1258 0.1915 0.1186 0.2172 0.1184 0.3027 0.1514 -0.0374 0.0339 0.0019 0.044 -0.0421** 0.0235 -0.0032 0.0506 -0.0684 0.0552 -0.2475*** 0.0131
-0.8883*** 0.017 -0.7414*** 0.0118 -0.7986*** 0.022 -0.8541*** 0.0257
0.1481 0.1367 0.1471 0.1305 0.2031 0.1299 0.2474 0.17
-0.0373 0.0488 -0.0302** 0.0129 -0.0009 0.0559 -0.0754 0.0608
0.2095 0.0229 -0.1092*** 0.0177 -0.1205*** 0.0263 -0.108*** 0.0316 -0.0017*** Continued on next page.
0.1633*** 0.0203 -0.1158*** 0.0158 -0.144*** 0.0242 -0.1223*** 0.0287
74
Table 2.14: Linear Probability Estimates of the E?ect of Owner Race on the Probability that a New Hire is Asian (continued)
(1) with HSD (?rm) Share of workers with HSG Share of workers with SOG Log employment Work. Pop. Total Work. Pop. % Asian Constant Year dumies Industry dummies State dummies Other Controls p-value R-Square (2) Asian (3) (4) (5) 0.0004 -0.0001 0.0002 -0.0021*** 0.0005 0.0033*** 0.0006 0.0643*** 0.0154 0.0002*** 0 0.1242 0.1129 yes yes yes yes 0.01 0.43
0.2431*** 0.0052 yes 0.01 0.28
0.2131*** 0.0557 yes yes yes 0.01 0.29
0.0373 0.0519 yes yes yes 0.01 0.35
0.0275 0.0528 yes yes yes yes 0.01 0.38
Note: Reference group is native ?rms.Reference Sector is Services. The number of observations is 147,373. Standard Errors are Huber-White robust standard errors, corrected for ?rm clustering. (+) Other controls include: location in a MSA dummy, legal form of organization, population in thousands in the neighborhood, interaction between 2-digit industry dummy and English speaker dummy. Neighborhood is de?ned as the adjacent counties to the county where the ?rm is located. Population in 100,000’s. ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
Whites are more likely to be hired by white owned ?rms (see Table F.4). However, the probability that a new hire is white depends postively on the share of whites in the ?rm at the time of the recruitment process. The owner’s race e?ect is lower for those owners from a di?erent racial group. Multinomial analysis is also applied to the combination of worker and owner races. The results for this regression are shown in Tables (2.15) and (2.16). Table (2.15) shows the change in log odds comparing two alternatives. The change in log odds between hiring a white worker versus hiring a Hispanic or an Asian decreases when ?rms are Hispanic or Asian owned. A higher share of white coworkers sig75
ni?cantly increases the log odds of a white being hired, and a similar result holds for other races. We also show the predicted hiring probabilities for each owner race (Table 2.12) computed at the means of all ?rms and dummy variables. Hispanic owners are 3 percentage points more likely to hire Asians and Hispanics than White and Black owners. These results support the analysis in previous sections. In sum, Hispanic and Asian workers are generally more likely to be hired by Hispanic or Asian owned ?rms. In this detailed presentation, it seems that Asian owned ?rms tend to employ Asian and Hispanic workers more readily than black and white workers. Almost 70% of Asian and Hispanic owners are immigrants. We would also like to analyse the impact of immigrant/native oner e?ects after controllinf for owner race, including owner birthpalce and race simultaneously. However, the variation across the sample is not enough to identify whether birthplace or owner race is more important. Most of the native owners are either white or black, with a large proportion of them being white. While, our sample has a small representation of black immigrant owners. Given the structure of our sample, white and black owners are mainly natives, while Asian and Hispanic owners are immigrants, and after looking at our by racial groups results, we can see that our previous result. Immigrant owners tend to hire immigrant workers, while native owners tend to hire native workers.
76
Table 2.15: Multinomial Logit Model: E?ects of Owner’s Race and Coworkers on Type of New Hires
Covariates Change in log odds comparing alternative 1 to alternative 2 White White White Black Black Asian to Black to Asian to Hispanic to Hispanic to Asian to Hispanic -0.379*** -0.313* -0.234 0.145*** 0.067 -0.079 0.03 0.15 0.2 0.04 0.1 0.1 -0.09* -0.186*** -0.052** -0.039** -0.096*** -0.134*** 0.06 0.03 0.01 0.01 0.02 0.03 0.11 -0.229*** -0.113 -0.340** -0.223** 0.116*** 0.09 0.03 0.03 0.04 0.03 0.02 2.522*** 1.7560** 6.631*** 4.111*** 0.766 1.875 0.892 0.413 1.121 1.153 0.651 1.034 -7.130*** -0.823 -0.385 6.745*** 6.308*** 0.438 1.203 0.723 0.241 1.412 1.324 0.56 -3.532*** -7.742*** -1.832* 1.700 -4.210*** 5.910*** 0.731 1.202 0.891 1.342 1.154 1.265 -1.522* -1.756*** -6.632*** -4.210** 0.667 -4.875*** 0.641 0.952 1.678 1.023 0.801 1.123
Owner Black Owner Hispanic Owner Asian Cow. Share white Cow. Share black Cow. Share Asian Cow. Share Hispanic
77
Note: Other controls include log of employment, percentage of immigrant workers in the surrounding counties, population in the county, legal form of organization, Msa location, 2-digit industry, interaction 2-digit industry and English speaker dummy, state and year dummies. Results from race/ethnicity ’others’ are not shown. Number of observation 135,583 workers, and 2,662 ?rms. Robust standard errors in italic allow for arbitrary correlation within the same ?rm. * signi?cant at 10%,** signi?cant at 5%, *** signi?cant at 1%.
Table 2.16: Multinomial Logit Model: Predicted Probability of Covariates Owner and Worker Races (%)
Owner White Black Asian Hispanic White 74.11 65.66 60.92 61.00 Workers Black Asian 11.99 3.05 15.33 3.90 10.91 6.84 12.75 6.71 Hispanic 10.64 11.97 12.50 13.87
Note: Based on multinomial logit predictions of the race of new hires from previous table.
2.6.3 Workers’ earnings and analysis of results
We estimate the e?ects of owner type and coworker shares on workers’ compensation using a human capital approach. The dependent variable is the natural logarithm of workers’ real annual wages.50 The regression includes dummy variables for owner type, the share of similar coworkers, worker type, and other ?rm characteristics. Using wage estimates at the individual level, we can evaluate the impact of owners’ characteristics on wage di?erentials by using equation (2.4).
ln(wkjt ) = c + ?1 ? Ik + Xk ? B2 + Oj ? B3 + Ik ? Oj ? B4 +COWkj ? B5 + Ik ? COWkj ? B6 (2.4) +Oj ? COWkj ? B7 + Ik ? Oj ? COWkj ? B8 +Fj ? ? + Zkj ? Z + T ? Mt + µkjt
In order to approximate the individual’s full-year annual wage rate and thus reduce the importance of within-year labor supply decision, we include the additional information of whether the worker is a full quarter employee. That is, full quarter worker is an individual with positive earnings during all the quarters of the year. Controlling for full quarter workers allows us to make UI’s annual earnings comparable to CPS salary and wages. Abowd et al. [2002] have a discussion on the comparison between LEHD and CPS annualized wages. After controlling for dominant employer and full-time status, CPS and LEHD earnings data are more comparable. LEHD annualized wages are slightly higher than CPS’ annualized wages. However, when looking at our analysis we should keep in mind that an individual’s labor supply depends on both the duration and the average number of hours worked at the job.
50
78
where k identi?es information on the worker and j refers to information on the ?rm. wkj stands for worker k ’s log real annual earnings at ?rm j . Ik is a dummy variable for whether the worker is an immigrant. In an e?ort to establish how much the immigrant earnings di?erential is due to di?erences in predetermined personal characteristics, we add a vector Xk of employee characteristics including age, age squared, education, sex, and race. Oj is a vector of dummy variables for owner type (birthplace). COWkj stands for the proportion of immigrant coworkers in the ?rm (explained in section 2.4.5). The expected sign for ?1 is negative, assuming that immigrants earn lower wages, and its signi?cance would indicates whether there is substantial wage variation across the di?erent worker types. With the inclusion of owner type dummies, the estimate of ?1 will represent the di?erence in wages between immigrants and natives in native owned ?rms. The sum of ?1 and the B3 and B4 coe?cients corresponding to an immigrant owned ?rm will be positive if immigrant workers earn higher wages when working for immigrant-owned businesses than native workers in an immigrant ?rm. The coworker share accounts for the potential impact on wages of having better connections to similar types of workers in the ?rm. The interaction between COWkj and the vector of owner types is included to assess whether the e?ect of coworkers di?ers according to the type of employer that is hiring the employee. We explore a 3-way interaction among owner type, worker type, and the immigrant coworker share. In equation (2.4), B2 Xk absorbs the e?ects of variations in personal characteristics. We would expect estimates of ?1 and the vector B3 to change after including workers’ characteristics. We should be aware of the potential presence of omitted variable bias. Unob79
servable characteristics could bias estimated coe?cients in equation (2.4). Ignoring these unobservables could causes us to overestimate the impact of owner type and immigrant coworkers on individual earnings. High ability workers of type k should look for ?rms that pay higher earnings. If native-owned ?rms o?er higher wages and employ these high ability workers, the estimated model would not be capturing the e?ect of owner type on workers’ earnings; rather it would be capturing individuals’ ability to ?nd better jobs. Also, worker preferences and comparative advantage can in?uence the results. Variations in preferences for particular job characteristics across di?erent workers could provide an alternative explanation for both earnings di?erentials and sorting. To account for some of this variation, we include the fraction of workers in the ?rm with education lower than high school, equal to high school, higher than high school with some college, and equal to college or higher. The omitted category is college graduate. Characteristics of ?rms (Fj ) and of the local community (Zj ) are also included. These controls include the population share of each group in the local community, population density, ?rm’s size (log of reported employment), and legal form of organization. Mt are year dummies. The ?rst column of Table (2.17) shows results from a baseline model including immigrant status, individual age, education, and part-time status, but excluding other variables of interest. The table reports the betas estimated by equation (2.4). To make the analysis easier to interpret, we transform these unstandardized ? coef?cients with the usual formula [(e? ? 1) ? 100], so that we can analyze the percentage change in wages associated with a 1-unit change in a continuous independent pre80
dictor variable. In the case of a dichotomous independent variable, we interpret the percentage wage di?erence in the target category compared to the reference category. After controlling for typical human capital variables, full-time immigrant workers earn about 8% less than native workers (3,293 dollars less each year). In the Table (2.17), we progressively include covariates that control for ?rms and coworker shares. Column 1 shows the typical human capital analysis. Column 2 includes owner dummies and their interaction with worker type. Then, in column 3, immigrant coworker share is included and its interaction with worker type to see whether the e?ect of immigrant coworker share di?ers across worker types. Then, the interaction of immigrant coworker share and owner types are added to the regression (column 5). Finally, we include a 3-way interaction among immigrant coworker share, worker type, and owner type. Evaluating the variables at their means and sample distribution, we ?nd that, when working for native employers, the di?erence between native and immigrant wages increases to 11%. Meanwhile, immigrant workers earn 10% more than native workers in immigrant owned ?rms (4,398 dollar more each year). The human capital results in Table (2.17) are consistent with the literature. Age positively a?ects wages but at a decreasing rate. Education is signi?cant and positive. Part-time workers earn less than full-time workers. The inclusion of additional independent variables does not modify these patterns. After controlling for individual characteristics, immigrant workers are paid less than native workers in native ?rms, but they receive a signi?cantly higher wage than native workers when working for immigrant ?rms. The inclusion of the share of immigrant coworkers 81
Table 2.17: OLS Results: E?ect of Owner Type and Coworker Share on Log Real Annual Wages
Immigrant Age Age square (’) Education Partime Owner Mix Owner Immigrant Owner Mix*Immigrant Owner Immigrant*Immigrant Imm.Coworker Imm.Coworker*Immigrant Imm.Coworker*Oimm Imm.Coworker*Omix Imm.Coworker*Oimm* Immigrant Imm.Coworker*Omix* Immigrant Constant Year dummies 2-digit industry dummies Other controls R-Square Adjusted (1) -0.08*** 0.007 0.0806*** 0.0007 -0.080*** 0.000 0.506*** 0.0007 -2.1847*** 0.0044 (2) -0.1503*** 0.0069 0.080*** 0.0007 -0.080*** 0.000 0.506*** 0.0007 -2.1805*** 0.0044 0.1808*** 0.0128 -0.1191*** 0.0097 0.0054 0.0224 0.3205*** 0.0251 (3) -0.1205*** 0.0073 0.0803*** 0.0007 -0.080*** 0.000 0.511*** 0.0007 -2.1792*** 0.0044 0.1615*** 0.013 -0.1443*** 0.0101 -0.0007 0.02 0.3030*** 0.0153 -0.1398*** 0.0163 0.09*** 0.013 (4) -0.1171*** 0.0025 0.0802*** 0.0006 -0.080*** 0.000 0.504*** 0.0007 -2.1783*** 0.0036 0.1407*** 0.0481 -0.1495*** 0.017 0.1866 0.251 0.3174*** 0.017 -0.2457*** 0.0203 0.12*** 0.011 -0.0966*** 0.036 -0.095** 0.0582 (5) -0.1017*** 0.0008 0.0749*** 0.0008 -0.080*** 0.000 0.484*** 0.0007 -2.1240*** 0.0049 0.0936 0.0155 -0.1087*** 0.012 0.1368 0.243 0.3131*** 0.0252 -0.3797*** 0.0276 0.1520*** 0.02 -0.2925*** 0.0734 -0.1359 0.0513 0.6456*** 0.1285 -0.3285 0.641 10.615*** 0.235 yes yes yes 0.35
10.665*** 0.26 yes yes no 0.25
10.7015*** 0.262 yes yes no 0.27
10.686*** 0.263 yes yes no 0.28
10.653*** 0.241 yes yes no 0.31
Note: The number of observations includes 214,398 workers. Standard Errors are Huber-White robust standard errors, corrected by ?rm clustering. Reference group are full time native workers in native ?rms. (+) Neighborhood is de?ned as the contiguous counties to the county where the ?rm is located. Population in 100,000’s. (’) Age ? 102 . ***signi?cant at 1%, ** signi?cant at 5%, * signi?cant at 10%.
produces interesting results. Immigrants earn more when working for immigrant employers and when the immigrant coworker share increases. The opposite is true
82
for native workers. In general, a native worker receives higher wages if he or she works for a native ?rm with a low share of immigrant workers. These results are striking in two senses. One, the ability to look at individual wages and identify the types of ?rm owners is only possible with this database. We have individual earnings for each ?rm. Although immigrants are paid less on average, they ?nd themselves in a better position when working for immigrant ?rms. Second, we can look at the entire workforce and identify each individuals’ types of coworkers in the ?rm. This allows us to make inference on the impact of social ties on worker wages.
2.7 Conclusions
This paper takes advantage of unique employee and employer matched microdata from the U.S. Census Bureau to examine the e?ect of owner types and coworker types on ?rms’ hiring patterns and workers’ earnings. Particular attention was paid to the birthplace of employers and to the share of similar coworkers (by birthplace and ethnicity) at ?rms when new workers are hired. We examined the e?ect of those variables on hiring rates and on the wage di?erential between immigrants and natives. In general, employees’ wages are a?ected by the type of owner of the ?rm. For native employees working for immigrant owners the e?ect is very interesting. Natives are paid lower when working for immigrant employers, and in these ?rms natives have lower average earnings than immigrants. One explanation for these ?ndings
83
is that immigrant bosses have a better understanding of and networking with the immigrant community, and therefore can ?nd and contract immigrant workers more easily than native-owned ?rms. Why can’t native-owned ?rms quickly adjust and ?nd this cheaper labor? Lack of language knowledge and lack of networking make it harder for native bosses to ?nd immigrant workers. These ?ndings justify further analysis of di?erences in contracting ability across employers. The evidence that the type of owner matters for wage di?erentials among workers also implies an important role for owner type on personnel policy. In addition to examining the e?ect of owners and coworkers on di?erences between immigrants and natives, we evaluate the e?ect of owner and coworker types on ethnically(racially) di?erent groups. An individual’s race is an important source of variation across workers and owners. The evidence suggests that employers tend to hire workers from the same ethnic group. A signi?cant impact of similar coworkers in the hiring process is observed across all types of owners, even after controlling for ?rm ?xed e?ects. Immigrant owners tend to hire more Hispanics and Asians, while native owners hire more blacks and whites. By shedding light on the ways workers and employers interact in the labor market to a?ect job and wage outcomes, this research makes a contribution to the sociology, labor economics, and demography literatures. It also opens up numerous avenues for future research. On the microeconomic side, we can further evaluate job ?ows and wage pro?les of workers inside di?erent types of ?rms. The analysis of assimilation can also take advantage of the results presented here, to further our understanding of the adjustment process of new immigrant workers. The empir84
ical analysis in this paper makes some progress toward mitigating biases of skill sorting. This paper controls for a broad number of observable characteristics that try to capture other explanations for segregation. However, if owner unobservable characteristics are correlated with worker characteristics, the results of the analysis would be biased. Di?erent empirical approaches such as instrumental variables or owner ?xed-e?ects could be good options in future research, although this would demand a more exhaustive matched database that follows workers after leaving the ?rm and ?rms after ownership changes. Narrowing the scope of the analysis by looking at one industry could also provide information on the costs and bene?ts of ?rm recruitment processes. For instance, we could examine with more detail the e?ect of worker type concentration on ?rms’ labor productivity. On an aggregate view, we can evaluate the e?ect of large ?ows of immigrants on the economy with the combined analysis of push and pull factors. Immigrant ?rms and immigrant workers seem to match quickly in the labor market. The analysis of the impact of immigration on unemployment and aggregate vacancies in the labor market can be extended to incorporate the ?ndings in this paper.
85
Chapter 3 Workplace Concentration of Immigrants 3.1 Introduction1
Over the last several decades, labor markets in many U.S. cities have absorbed large in?ows of new immigrants. The size of these ?ows has generated intense interest in their e?ects on the employment and wages of natives, as well as in the extent to which new immigrants have assimilated into the U.S. economy. New immigrants ?nd employment and accumulate location-speci?c skills and work experience, gradually becoming integrated into local economies and potentially changing them in substantial ways. While outcomes of this process have been the subject of much research, less is known about the process itself. Which businesses hire immigrants? To what extent do immigrants work with natives? How does these patterns change as immigrants accumulate U.S. speci?c skills? Do the characteristics of di?erent immigrant groups and di?erent geographic labor markets a?ect the way in which assimilation plays out? A lack of suitable data has limited economists’ ability to address these questions. Our contribution is to bring to bear a rich set of matched employer-employee data that allows us to identify immigrants, their coworkers, and their employers. Our unique data permit quantifying the extent of and covariates of the workplace
1 This chapter draws heavily on a joint paper with John Haltiwanger, Kristin McCue, Seth Sanders and Fredrik Andersson with the same title.
86
concentration of immigrants. The paper has two broad objectives. The ?rst is primarily descriptive. The descriptive ?ndings show that immigrants are much more likely to have immigrant coworkers than are natives. This pattern is driven partly by the geographic concentration of immigrants, but the patterns hold true even within local labor markets. At the same time, most immigrants do have native coworkers: only a small share work in immigrant-only workplaces. The concentration of immigrants is higher for recent immigrants and, conditional on recent arrival, for older immigrants: part of the assimilation process is a movement towards more interaction with natives in the workplace, and younger immigrants are more likely to work with natives. We ?nd large di?erences associated with ?rm size: concentration is much higher in smaller ?rms, but is far from zero even in the largest ?rms. We also ?nd substantial variation in the extent of immigrant concentration across industries even after controlling for a detailed set of location, employer and employee characteristics. Second, our ?nding that the allocation of immigrants across workplaces is far from random raises the question: what does drive this workplace concentration? Both the existing literature and our descriptive ?ndings suggest that it is important to consider how businesses hire their employees and the choices that businesses make about the skill mix of their workforce. One relevant issue here is the role that language skills play in governing interactions among employees and between employees and customers. A second issue is the role of social networks in the process that matches workers and ?rms. A third issue is human capital - the sorting and concentration of immigrants in the workplace may re?ect sorting by skills. In the 87
second part of the paper, we explore the role of these factors. We ?nd evidence that immigrants with primarily immigrant coworkers are likely to have coworkers who live in the same residential tract. This pattern is robust to the inclusion of controls for other closely related factors such as residential segregation. We also ?nd evidence that immigrant workers with poor English speaking ability and low education are more likely to work with immigrant coworkers. The paper proceeds as follows. Section 3.2 provides an overview of the relevant theoretical and empirical literature that helps guide our empirical analysis. Section 3.3 describes the measurement of immigrant concentration, the matched employer-employee data we use in our analysis and the methods we use to explore the correlates of immigrant concentration. In section 3.4 we present our main results quantifying the extent and nature of immigrant concentration across workers and businesses. Section 3.5 analyzes the impact of factors such as social networks, language skills and human capital on the patterns of immigrant concentration. Most of the analysis focuses on native born, recent immigrants and established immigrants without speci?c reference to country of origin. Section 3.6 extends the analysis in terms of the basic patterns of concentration by country of origin. Concluding remarks are provided in section 3.7.
88
3.2 Background 3.2.1 Literature on earnings di?erences
Work examining earnings di?erences between whites and other groups in the U.S. has largely focused on netting out di?erences in skill (often captured by education and labor market experience) and geography (often using place of residence and urban residence) to assess the potential role of discrimination in labor market outcomes. This assumes that earnings di?erences are generated either by di?ering worker characteristics or di?ering returns to those characteristics. By extension, closing gaps in earnings requires equalizing worker characteristics and their return across groups. Di?erences in returns to characteristics are assumed to re?ect unobserved ways in which the wage generating process di?ers and is typically viewed as an upper bound on the potential for discrimination to play a role in explaining wage disparities. A huge number of papers use this approach; some classic examples that examine earnings di?erences relative to white men are Smith and Welch [1977] for African American men, Borjas [1982] for Hispanic men, Chiswick [1983] for Asian men, and Corcoran et al. [1983] for women. There is also a large literature assessing the sources of earnings di?erences between immigrants and native born workers (for example, Chiswick [1978], or Butcher and DiNardo [2002]). These papers generally augment the basic human capital framework used in the studies above by allowing for skill di?erences that are speci?cally relevant to immigrants. These include potential di?erences in the value of education and work experience accumulated outside the U.S., and the importance 89
of di?erences in English language skills. Immigrant assimilation into the U.S. labor market is viewed as occurring through a narrowing of the earnings gap, resulting largely from increased U.S.-speci?c skills with time spent in the U.S. While there is debate over the speed at which the earnings gap between immigrant and native born workers closes, most studies ?nd a substantial narrowing with time spent in the U.S. (see Chiswick [1978] and Borjas [1985]). An older literature in sociology and economics stresses that earnings di?erences between groups may be driven by the characteristics of the ?rms that employ the majority and minority groups, rather than solely by human capital characteristics. Usually termed ’dual labor market theory,’ this idea gained considerable attention in the late 1960s and early 1970s (see for example Averitt [1968] or Galbraith [1971]). According to this theory, many ?rms (especially industrial ?rms) are not governed by competitive processes. Instead, these ?rms enjoy market power. They insulate themselves and stabilize their workforce through job training and promotional ladders (Edwards [1972]). Firms that are constrained by competition do not invest in work skills and are characterized by low wages and high turnover, with low returns to human capital including job tenure. The existence of ’good jobs’ and ’bad jobs’ by itself would not imply an earnings disadvantage to minority workers. Sociologists typically rely on a form of employer discrimination to explain why dual labor markets lead to minority disadvantage. Queuing theory suggests that good jobs always have an excess supply of applicants and ?rms then order workers by preferences and hire down the queue until vacancies are ?lled. If race or ethnicity plays a role in this ordering, a higher 90
fraction of minority workers will be employed in the secondary market and have relatively low wages and wage growth. While dual labor market theory per se has largely fallen out of the mainstream literature in economics and sociology, a newer literature that similarly argues that ?rm characteristics may be partially responsible for the level and growth in earnings of workers has gained growing acceptance. Wages appear to be positively correlated with ?rm productivity and ?rm size (Abowd et al. [2005]). While more controversial, there is some evidence that ?rm-level technological adoption also affects workers’ wages (Dunne et al. [2004]). Lengermann [2002] ?nds that coworker characteristics, in addition to ?rm characteristics, may a?ect wages. Speci?cally, he ?nds that having more skilled coworkers independently raises a worker’s wages. If ?rm characteristics play a major role in wage setting, then understanding how race and ethnicity a?ect the matching of workers to ?rms becomes important for understanding wage disparities across groups. Lengermann et al. [2004] explore the issues of sorting of immigrants across ?rms and ?nd that sorting matters for wage di?erences between native born and immigrant workers.2 We now turn to theories of worker segregation with special attention to how immigrants sort into ?rms.
3.2.2 Literature on segregation
Four broad overlapping theories explain segregation of workers into ?rms. These theories focus on sorting based on (a) productive characteristics, (b) prefSome of our basic ?ndings on immigrant concentration are also found in Lengermann et al. [2004]. Using the same data infrastructure that we use in this paper, they ?nd for example di?erences in immigrant concentration by industry and employer size.
2
91
erences of workers or employers, (c) information available to workers or employers, or (d) cost of commuting to jobs. Some, but not all, of these theories imply that segregation results in a disadvantage for one group of workers relative to another. There is substantial evidence of segregation by skill. For example, Kremer and Maskin [1996] look at the sorting of high and low skilled workers into ?rms over time and across three countries, the U.S., Britain and France. They ?nd a high and rising correlation between worker skill levels in ?rms over the 1970s and 1980s. This may occur either because a ?rm demands a particular type of worker (for example skilled workers) or because coordination within a ?rm demands that workers share a common characteristic such as a common language. Cabrales et al. [2008] emphasize a di?erent skill-based mechanism: if a worker’s utility is a function of both absolute wages and their wages relative to those of coworkers, and if movement of workers across ?rms is costless, complete segregation of workers by skill is optimal. A mixed-skill workforce generates wage inequality within a ?rm, reducing worker utility. All workers are made better o? by grouping workers with similar skills and avoiding these reference group costs. Regardless of the mechanism, segregation by skill will cause immigrant-native di?erences in the distribution of skill to contribute to segregation. For example, immigrants are both much more likely than natives to have an 8th grade education or less (23% vs. 5.2% for natives in the 2000 census), and also more likely to have an advanced degree (10.3% vs. 8.6% for natives). Therefore, ?rms that specialize in hiring exclusively low-skilled or exclusively highskilled workers will tend to have a workforce that has a higher fraction of immigrants than the fraction in the population. 92
Language di?erences provide another productivity-based motivation for segregation. If working with someone who does not speak the same language generates transaction costs, employers may increase productivity by hiring only workers who share a common language. In this case, immigrants from non-English speaking countries may be particularly likely to be segregated, and may also be particularly likely to work with their compatriots rather than other immigrants. Lang [1986] develops a formal model of wage di?erences arising because of the costs to ?rms of having to pay a premium for bilingual workers who can bridge the language barrier. One of the results of this model is that complete segregation would occur if both capital and labor were owned by each language group. Hellerstein and Neumark [2003] ?nd evidence that Hispanics with poor English-language skills are particularly likely to work with other Hispanics. Their data do not allow them to examine how much of this is due to Hispanic workers working for Hispanic-owned ?rms as in the Lang model. Becker [1957] is the classic model of preference-based segregation. In this model, segregation of workers by race occurs as the result of discriminatory preferences on the part of co-workers. White workers would demand a premium to work with black workers. In response, ?rms segregate workers into separate facilities, avoiding the need to pay a wage premium to discriminating white workers. Depending on conditions including the relative size of the minority and majority group, the number of ?rms, and returns to scale in production, segregation may be extreme but with limited disadvantage in wages to the minority group. Dual labor market theory, described above, also generates wage di?erences across groups 93
if discriminating employers put minority job candidates lower down the queue. In this case, higher wages in the primary sector ensure that a higher fraction of the majority group works in the primary sector and hence gives a wage advantage to the majority group. Information-based theories concentrate on the mechanisms that workers use to ?nd jobs. For example, ?rm use of employee referrals to ?ll jobs may contribute to workplace segregation. For workers, use of personal contacts to search for jobs is inexpensive and has relatively high rates of success (Holzer [1988]). For employers, employee referrals provide both a low cost recruitment strategy and, on average, new hires with higher productivity and lower turnover rates (Holzer [1987]; Montgomery [1991]). If workers tend to refer others who have similar characteristics, use of referrals can increase rates of segregation. Elliot [2001] ?nds that recent Latino immigrants are more likely than blacks or Latino natives to use personal contacts to ?nd jobs. Weak English skills explain much of this di?erence. A greater reliance on referrals in small workplaces in combination with a concentration of recent immigrants in small ?rms also contributes to the di?erence. Information ?ows may combine with residential segregation to contribute to workplace segregation. Neighborhoods play an important role in who you know and hence may provide important job contacts and references. Several papers have established that workers in the same ?rm are disproportionately from the same neighborhoods. Using data from Boston, Bayer et al. [2008] ?nd that a worker is about one-third more likely to work with someone who lives in the same census block as to work with someone who lives in other blocks in their block group (typically 94
eight or so contiguous blocks). This comparison of blocks to block groups provides important evidence that having coworkers who are neighbors does not stem from unobserved factors such as transportation routes or distance that make a place of employment a natural place to work for those living in a particular location. Many of these unobserved factors would be similar for a block group and block of residence, and so should have similar e?ects on the likelihood of working with more or less immediate neighbors. This paper is limited in that the exact establishment can not be observed, while sample sizes as well as the ethnic make-up of Boston restrict the authors’ investigation to black-white di?erences. Hellerstein et al. [2008a] also present evidence of neighborhood network e?ects. Using matched employer-employee data, they compare how likely an individual is to work in the same establishment as his neighbor, relative to the likelihood that this would result if their employer hired workers randomly from the geographic areas of residence of all individuals who work in the employer’s census tract. Their dataset is large enough to disaggregate the analysis for whites, blacks and Hispanics. They ?nd that another worker living in the same census tract has twice the probability of working in your ?rm than what one would expect from randomness. They do not investigate the importance of other mechanisms for sorting workers into ?rms. A ?nal theory of the sorting of workers into ?rms also works through residential segregation but focuses on the fact that not all jobs are equally accessible from di?erent places of residence. Kain [1968] investigated employment patterns of blacks and whites in Chicago and Detroit. He found that blacks were unlikely to be employed in areas that were predominantly white, that blacks would have higher 95
employment rates if housing segregation was lower, and that the movement of jobs from central cities to suburban areas depressed the employment prospects of blacks. A number of other studies followed that compared employment di?erences between central city and suburban residents within an urban area. These tests often found employment prospects lower for central city residents, but controlling for unmeasured skill di?erences between residents of di?erent locations remained an issue in inference. A recent study by Hellerstein et al. [2008b] questions the interpretation that a lack of jobs near where blacks live is a major source of racial employment di?erences. They ?nd that the employment prospects of black residents are positively correlated with the number of nearby jobs in which blacks work, but not with the number of nearby jobs in which whites work. This indicates that even within close geographic proximity, job markets are racially segregated. They conclude that spatial mismatch has little e?ect on employment prospects of blacks but that what they term racial mismatch—few nearby jobs that employ blacks—has a large e?ect. Clearly, residential segregation could contribute to workplace segregation of immigrants. There is ample evidence that immigrants’ places of residence are spatially concentrated. Iceland [2009] describes the high level of residential segregation in the U.S. among immigrant groups but also shows that immigrants migrate to neighborhoods that are more ethnically integrated as they spend more time in the U.S. However, Porter and Wilson [1980] argue that, unlike for black Americans, residential segregation may aid immigrants—especially new immigrants–while also leading to segregation of workers in ?rms. Studying the post-Castro immigration from Cuba to Miami, Portes and Wilson show that not only do Cubans in the U.S. 96
work together, many work in ?rms owned by other Cubans. Moreover, Cuban employees of Cuban-owned ?rms tended to display the same patterns of wage growth and returns to human capital as workers in ?rms classi?ed as in the ’primary sector,’de?ned as ?rms with a promotion ladder, over 1000 workers, and high average wages. While an impressive source of employment, it is not clear that the example of Cubans generalizes to other foreign-born groups. Capital owners speci?cally were forced to leave Cuba, which may have led to higher levels of capital with which to start businesses and more experience with small businesses among Cubans than among other foreign born groups. Having said this, Wilson and Portes report that much of the capital used to start these businesses was accumulated in the U.S. and not transferred from Cuban concerns.
3.3 Methodology and Data 3.3.1 Measuring immigrant concentration
We follow several recent papers that study workplace segregation (Hellerstein and Neumark [2007]; Aslund and Skans [2005a], Aslund and Skans [2005b]— henceforth HN and AS) by using the share of coworkers in a particular group as a measure of exposure. That is, we exclude the worker himself when measuring the concentration of immigrants in the business he works in. For worker i, employed by business j which has sj employees, the share of immigrants among coworkers is:
sj
1 Cij = sj ? 1 97
Ik
k =i
(3.1)
where Ik is an indicator for whether or not worker k is an immigrant. For the sake of brevity, we will refer to this simply as the coworker share. As pointed out by these authors, excluding the worker’s own characteristic in calculating concentration ensures that in large samples the coworker share for both immigrants and natives should on average equal the share of immigrants in the workforce in the absence of any systematic concentration. Based on this, we use the di?erence between the mean coworker share for immigrants and natives as a measure of immigrant concentration. A positive value indicates that immigrants are more concentrated than would be expected based on random allocation. At the extreme, if immigrants worked only with immigrants and natives with natives, the di?erence in coworker means would equal one. A negative value for this di?erence would indicate that immigrants were more likely to work with natives than would be expected based on random allocation—a pattern that could arise where the two groups provide di?erent but complementary skills. We depart from the approach of these authors in two ways: in the way in which we condition on observable characteristics, and in choice of a normalization to gauge whether the concentration we ?nd is large relative to some alternative. There are two types of questions that can be addressed by conditioning on observable characteristics in studying segregation: to what extent can segregation be explained by di?erences in the characteristics of the two groups, and which characteristics are most associated with segregation. HN and AS both focus more on the ?rst issue, while we explore some aspects of both questions. As an example to provide some context, the immigrant and native education distributions di?er, and particular em98
ployers may hire primarily from one part of the education distribution, leading to concentration of immigrants because of di?erences in skill. HN and AS both use the di?erence between measured concentration and the amount of concentration that would be generated solely by the way in which education is distributed across employers as their conditional measure of concentration. In contrast, we condition on a worker’s own characteristics and on the characteristics of his or her employer (e.g. employer size and industry), but do not directly condition on coworker characteristics. Our measure of concentration is the mean di?erence between immigrants and natives with the same characteristics. We take a di?erent approach in part because the worker characteristics in our data that vary within employer (age, gender) do not di?er dramatically between immigrants and natives, and they also turn out not to have a strong correlation with immigrant concentration. Controlling for a worker’s own characteristics should remove the e?ects of age and gender from the measured di?erence in coworker mean, and the estimated coe?cients allow us to examine the characteristics of immigrants and natives who work in heavily immigrant workplaces. Both HN and AS normalize their measures of concentration, though they choose di?erent references for the normalization. While both of their normalizations have intuitive appeal, we take a di?erent approach. We use the immigrant-native di?erence in coworker shares as our measure of concentration, but in most cases also present information on the coworker share for natives as a point of reference. Our regression approach makes doing so straightforward, and also allows us to more directly illustrate patterns of concentration. For example, using the regressions to 99
predict means for a given set of covariates allows us to illustrate the strong positive relationship between immigrant concentration and immigrant share of the workforce, when looking across groups de?ned by characteristics such as area of residence and employer size. In addition, the regression approach using our coworker index at the person level as the dependent variable permits us to normalize our measure of concentration e?ectively along a number of dimensions. For example, HN normalize to control for between MSA di?erences in various groups (e.g., di?erences in the distribution of blacks and whites across MSAs). We control for such di?erences directly in our regression approach by, for example, including controls for MSAs.
3.3.2 Data
We use the data from the Longitudinal Employer - Household Dynamics (LEHD) database, which draws much of its data from complete sets of unemployment insurance (UI) earnings records for a subset of U.S. states. The database includes records for 1990 to 2004, though some states only have data for a subset of those years. The workers’ earnings records have also been matched to characteristics of their employer gathered in quarterly administrative reports and through Census Bureau business censuses and surveys. Basic demographic data are also available for workers, including place of birth. For those born outside the U.S. (and its territories), we treat the year in which they ?rst applied for a Social Security Number (SSN) as the date of their arrival. While this may not precisely date arrival, preliminary results based on a sample of immigrants for whom both LEHD and decennial
100
population census data are available suggest that the year the individual ?rst applied for a social security number proxies the reported year of arrival fairly well.3 In the current analysis, we use data from selected metropolitan areas in 11 states. While we do not use a large number of states, our sample does include ?ve of the six states that had immigrant populations of 1 million or more. These data give us two unique advantages. First, we have earnings for a group large enough to include millions of immigrants. Second, we can observe the ?rms in which workers are employed, allowing us to measure both employer characteristics and the characteristics of coworkers. These data have other advantages that we do not exploit here but plan to in future work: for example, the data can be used to generate a panle on both employers and employees that would allow us to track earnings of immigrants over time in the U.S. as well as to observe contemporaneous changes for native-born workers. The main disadvantage of these data for studying immigration is that they include only on-the-books employees and so do not cover the self-employed or those working in the informal sector. Thus they likely have poor coverage of undocumented immigrants. Coverage of employment in agriculture is incomplete in the LEHD data, so we exclude employers in that sector. Calculating the share of coworkers who are immigrants requires at least one
Here we use year of arrival only to split immigrants between those arriving very recently (within the last 5 years) and other immigrants. Comparing our classi?cation based on date of SSN application to one based on responses in the 2000 census, 92% of immigrants are classi?ed in the same way according to both sources. Among those for whom the classi?cation di?ers, the most common pattern is that 4% of Mexicans are considered new immigrants in Decennial Census versus 10% in LEHD. The lag in the registration process by immigrants, specially in the case of Mexicans, explains these di?erences. The patterns by age are very similar between LEHD and Decennial Census, however younger immigrant workers are also reporting a small lag in their application for social security.
3
101
coworker, so we restrict our sample to businesses with at least two employees.4 We measure concentration using a cross-section of data based primarily on the second quarter of 2000, but we use LEHD data for the 1995-2000 period to de?ne business age. In computing the coworker share, we use all coworkers, whether or not they hold other jobs. However, the set of observations used in our regressions includes only one job for each individual: the job where they received their highest earnings in that quarter. We draw data from employers in 31 MSAs. We include all MSAs that have substantial foreign-born populations and are in states for which we have the required data, but we also included several smaller MSAs that experienced very rapid growth in foreign-born residents between 1990 and 2000.5 Even in the smallest of our MSAs we have data on more than 30,000 immigrant workers, so small sample sizes are never an issue. Table 3.1 summarizes the across-MSA variation in immigrant shares for our sample of MSAs. In the average MSA in our dataset, 18.9% of workers are immigrants. In what follows, we are interested in deviations in workplace shares from the overall-average. Clearly the substantial variation in immigrant share across MSAs will contribute to ?nding immigrant concentration. The shares of both recent and
Immigrants account for 27% of employment in single-employee businesses, and 16% of employment in businesses with more than one employee. 5 More precisely, we started from the list of MSAs used in Singer [2004], which included all MSAs with at least 1 million residents in 2000, and meeting at least one of the following criterion: (i) at least 200,000 foreign-born residents, (ii) a foreign-born share higher than the 2000 national average (11.1%), (iii) 1990-2000 growth rate of the foreign-born population above the national growth rate (57.4%), or (iv) above national average percentage foreign-born in 1900-1930 (‘’former gateways”). We drop 14 of Singer’s 45 MSAs because we do not currently have access to all of the data we need from the relevant states.
4
102
Table 3.1: Variation in Immigrant Share of Workforce across Sample MSAs Percent Immigrant Recent Established 3.40 15.46 1.85 8.57 1.94 8.52 2.92 13.54 4.37 22.82 6.03 27.23
Mean Standard Deviation P25 Median P75 P90
Total 18.86 10.27 10.57 16.26 26.60 32.58
Source: Authors calculations based on LEHD UI-ES202 database. Note: Unit of observation is an MSA. Immigrant shares are measured as of the second quarter of 2000, and recent immigrants are those arriving between 1995 and 2000. The table presents fuzzed percentile values.
established immigrants vary substantially across MSAs as well. For roughly 10% of workers in our sample, we match in additional information on educational attainment and English language skills from the long form of the 2000 population census. Using propensity score models, we develop weights for the matched sample that allow us to closely replicate our results based on the overall sample.6 We then use weighted estimation with the matched sample to examine the relationship between these measures of skill and immigrant concentration.
3.3.3 Regression speci?cations
Our primary empirical approach is to run a series of regressions with the coworker share as the dependent variable, and individual workers on their primary job as the unit of analysis. As a rough way to capture the way in which immigrant
The variables used in the propensity score procedure were: worker age, sex, country of origin (11 groups=Mexico, China, Cuba, El Salvador, India, Korea, Japan, Vietnam, Phillipines, other country of origin groups, and natives), log earnings, worker status, industry (4 digits), Msa indicator variables and population density, plant age and size, and ?rm’s # of establishments.
6
103
concentration changes with time spent in the U.S., we include indicators for whether an individual is a recent immigrant (RI, de?ned as arriving in the last 5 years), or a more established immigrant (EI, arriving more than 5 years ago). Since we use a cross-section of data, the di?erences between recent and more established immigrants confound the e?ects of time spent in the U.S. with changes in labor markets and in immigrant and native characteristics over time. We would need to exploit the panel aspect of our database to seriously address the a?ects of assimilation, but believe this is useful as a starting point that provides suggestive evidence on whether assimilation e?ects on concentration are likely to be important. Our initial regression speci?cation is:
Cij = ?N + ?EI EIi + ?RI RIi + ?xij +
ij
(3.2)
where (again) i denotes an individual and j denotes a workplace. Here, the constant term (?N ) represents the mean coworker share for the omitted category, which in our simplest speci?cation consists simply of natives. Coe?cients ?EI and ?RI give us estimates of the di?erences between immigrants and natives in how likely they are to have immigrant coworkers. We use controls for MSA and for various worker and employer characteristics to examine the extent to which immigrant concentration can be accounted for by di?erences between natives and immigrants in their geographic distribution and in worker and job characteristics. In section 3.6, we de?ne coworker shares for speci?c countries of origin and look at which immigrants are most likely to work together.
104
Speci?cation (3.2) assumes that the e?ects of covariates are the same for immigrants and natives. To examine whether this in fact holds, we use an alternative speci?cation that includes interactions between our immigrant dummy variables and other covariates:
Cij = ?N + ?EI EIi + ?RI RIi + ?xij + ?EI EIi ? xij + ?RI RIi ? xij +
(3.3)
Once we add interaction terms, the intercept rarely identi?es e?ects for a group of particular interest. To illustrate the e?ects of a particular covariate in speci?cations of form 3.3, we present predicted means for immigrants and natives, by which we evaluate di?erences between immigrants and natives based on the pooled distribution of the variables in x. To ease computations with our 36 million records, we use linear regression models rather than adopting an approach that accounts for the limited range of the dependent variable. In this draft, we also ignore the e?ect of clustering within employer in estimating the standard errors. For most of our speci?cations, the dependent variable mean is not close to either 0 or 1, which mitigates some of the problems inherent in the linear model. The strong positive correlation in the coworker share among employees of the same business will lead to a downward bias in our estimated standard errors in all worker-level regressions. Given the huge size of our sample, the results we present would generally remain signi?cant at standard levels even if the corrected standard errors were 100 times larger. The few exceptions 105
(in Table 3.6) are estimates that are too small to be meaningfully di?erent from zero anyway.
7
3.3.4 Descriptive statistics
Table 3.2 presents summary statistics for immigrant and native workers in our full sample. The ?rst row gives coworker shares for the three groups. For the average native, about 15% of coworkers are immigrants, while 42% of the coworkers of recent immigrants are fellow immigrants, and 36% of the coworkers of established immigrants are immigrants. The immigrant-native di?erence in coworker means—our measure of concentration—is .272 for recent immigrants and .214 for more established immigrants, indicating substantial concentration. Table 3.2: Characteristics of Immigrant and Native Workers, Full Sample
Immigrants Recent Established 42.1 36.3 43.6 35.6 20.8 56.8 1.1 36.2 37.0 25.7 8.5 23.6 19.7 33.2 47.0 56.4 14.7 49.6 24.8 10.9 9.0 22.6 Natives 14.9 29.3 30.0 40.7 51.7 . . . . 8.0 23.5
Coworker share Age Age