Description
In Sub-Saharan Africa, manufacturers operating in spontaneously developed industrial clusters are very small in size, have low productivity, and stagnate except when they are young. The literature has related the preponderance of such enterprises to their socio-economic surroundings.
Entrepreneurial Skills and Industrial Development: The
Case of a Car Repair and Metalworking Cluster in Ghana
February 27, 2009
Alhassan Iddrisu
Ministry of Finance and Economic Planning, Ghana
Yukichi Mano
National Graduate Institute for Policy Studies, Japan
Tetsushi Sonobe
Foundation for Advanced Studies on International Development, Japan
Email: [email protected]
Preliminary. Please do not cite without author permission
1
Entrepreneurial Skills and Industrial Development: The
Case of a Car Repair and Metalworking Cluster in Ghana
Abstract
In Sub-Saharan Africa, manufacturers operating in spontaneously developed industrial
clusters are very small in size, have low productivity, and stagnate except when they
are young. The literature has related the preponderance of such enterprises to their
socio-economic surroundings. This paper reconsiders the issue by looking at the way
small entrepreneurs engage in business in a car repair and metalworking industrial
cluster in Ghana. We hypothesize that these entrepreneurs are unaware of or unskilled
in basic techniques in marketing, management, and accounting, which is necessary for
enterprise growth. Evidence suggests that small entrepreneurs in the cluster are thirsty
for such techniques.
Keywords: Africa, Ghana, industrial development, industrial cluster, entrepreneurial
skills, training, propensity score matching, impact evaluation
JFL classification: O14, O33, O55
1
1. Introduction
In developing countries, a large number of manufacturers are operating in
spontaneously developed industrial clusters as Schmitz and Nadvi (1999), McCormick
(1999), and Sonobe and Otsuka (2006) attest regarding South America and South Asia,
Africa, and East Asia, respectively. While other regions have witnessed substantial
growth in enterprise size and productivity, Sub-Saharan Africa has had only a small
number of success stories. Enterprises operating in clusters in Africa are often
informal, have low productivity, and grow only when they are very young, according
to Ramachandran and Shah (1999), Sleuwaegen and Goedhuys (2002), Mazumbdar
and Mazaheri (2003), Frazer (2005), Bigsten and Söderbom (2006), Van Biesebroeck
(2005a), Mengistae (2006), and Bigsten and Gebreeyesus (2007). These studies find
that major determinants of enterprise growth and survival include enterprise age and
size and entrepreneurial human and social capital, such as years of schooling, years of
business experience, and access to informal network. Bigsten et al. (1999), Van
Biesebroeck (2005b), Rankin, Söderbom, and Teal (2006) among others explore what
make investment and exporting difficult for African enterprises.
The literature has also explored major constraints facing enterprises, such as
credit constraints, high risks, corruption, limited contract enforcement, labor costs
which tend to increase with enterprise sizes, and high costs of transportation and
electricity due to poor infrastructure (e.g., Bigsten et al. 2003; Gunning and Mengistae
2001; Fafchamps 2004; Söderbom and Teal 2004; Collier and Gunning 1999; Eifert,
Gelb, and Ramachandran 2008). Compared with these constraints, little attention has
been paid to inadequate skills of small entrepreneurs in marketing, production and
quality management, and accounting. Such entrepreneurial skills are part of the
human capital of entrepreneurs but not necessarily captured by the number of years of
2
schooling or the number of years of business experience. While it is fair to say that
many entrepreneurs are deficient in entrepreneurial skills, questions arise as to whether
the skill level is a major determinant of enterprise performance. If important, why do
small entrepreneurs remain deficient in such skills even after many years in business?
Is it possible to teach such skills to entrepreneurs? Karlan and Valdivia (2009) present
evidence that a program of teaching entrepreneurship to small entrepreneurs in Peru
improved knowledge, practices and revenues. Does the same apply to Africa?
This paper attempts to answer some of these questions by using enterprise data
collected from an industrial cluster consisting of garage mechanics and metalworking
enterprises in Ghana. To obtain tight evidence for the importance of entrepreneurial
skills as a determinant of enterprise development and for the usefulness of
entrepreneurial training, it would be necessary to carry out a randomized experiment in
which such training is provided to randomly selected participants. This paper is not a
report of such an experiment, but an assessment of demand for entrepreneurial skill
training. We examine the associations among entrepreneurs’ characteristics, their
participation in such training in the past, and their current performance, in order to
infer how beneficial it will be if we provide a training program to them.
The industrial cluster under study is very large in terms of the number of
enterprises and the number of workers and apprentices, but the way in which small
entrepreneurs run their businesses is no different from that found in other clusters of
small businesses in Sub-Saharan Africa, such as the metalworking clusters in Nairobi,
Kenya, and a suburb of Kampara, Uganda, and the leather-shoe cluster in Addis Ababa,
Ethiopia.
1
That is, small entrepreneurs seldom keep records, seldom tout their
potential customers, and seldom take the initiative in making efficient use of materials,
1
For information on the metalwork clusters at Kariobangi and Kamukunji, see Sonobe, Akoten, and Otsuka
(2009a) and Kinyanjui (2007), respectively. For the leather-shoe cluster in Addis Ababa, see Sonobe, Akoten, and
Otsuka (2009b).
3
energy, and time.
Frazer’s (2005) account of apprenticeship applies perfectly to this cluster. The
majority of these entrepreneurs learnt production and business operation from their
masters through apprenticeship. The business model they were taught would be
suitable for self-employed masters working with several apprentices, but not for
owners and managers ambitious to expand their businesses. It is no wonder their
enterprises seldom grow beyond certain small sizes. Moreover, their enterprises may
become less profitable gradually since apprenticeship reproduces competitors who
produce or provide exactly the same products or services. In fact, profitability has
declined substantially in recent years in the cluster.
In response, many entrepreneurs have attempted to change the way of running
their businesses. We hypothesize that knowledge and skills that they did not learn
from apprentice training have assumed importance for their businesses, and that those
who have such knowledge and skills have better business results. Schools do not teach
such knowledge and skills, but education will help entrepreneurs search for useful
information, knowledge, and skills. In Sub-Saharan Africa, however, such intangible
inputs useful for a managerial reform are difficult to obtain not only for elementary
school dropouts but for polytechnic graduates. We conjecture that managerial training
plays an important role, and find that the current business results are strongly
associated with years of schooling and participation in managerial training in the
metalwork sector. Participation in technical training programs has no effects. These
results are obtained from OLS regressions and the propensity score matching (PSM)
estimation. By contrast, we find no such effects in the garage sector.
The rest of the paper is organized as follows. Section 2 describes the formation
of the large cluster under study. Section 3 presents the basic data of the sample
4
enterprises. Section 4 advances some testable hypotheses and explains empirical
strategy. Section 5 presents the results of OLS regressions and PSM estimations.
Section 6 tries to draw for future research and policies.
2. Brief History of the Cluster
The industrial cluster under study is located in the Suame area in Kumasi, the
second largest city in Ghana and the center of Ashanti Region. The cluster is called
Suame Magazine and dates from the 1930s when the dispersed craftsmen set up
workshops at the site of the present Kumasi Zoo, which used to be the site of an army
depot called Magazine during the colonial times. When the workshops resettled in the
current location, they kept this name (Institute of African Studies 1992). It has
expanded tremendously ever since in terms of the number of enterprise, employment,
and area. Table 1 presents the number of member enterprises of the Suame branch of
the Ghana National Association of Garages (GNAG), which comprises not only
garages but also blacksmiths, machinists, and manufactures and is generally believed
to cover 80 percent or more of the enterprise population in the cluster. The garage
sector is by far the largest and continues to grow rapidly in terms of the number of
enterprises.
In developed countries, garages are scattered far and wide to serve dispersed car
owners. In developing countries, most vehicles are business fleets. In Ghana, trailers
and trucks are concentrated on the artery roads connecting the major cities in the south,
such as the capital city, Accra, and port cities, and the major cities in the north, such as
Tamale, and the capital city of Burkina Faso. Kumasi is the most important junction
of these arteries. The number of the vehicles going back and forth on these arteries has
rapidly increased. The demand for garage services has increased accordingly. While
5
garages are clustered not only in Kumasi but also in Accra and other cities, Suame
Magazine is said to be larger and have higher technical skills and better equipment
than any other clusters in West Africa. The division of labor among specialists is
highly developed in this cluster. Each master specializes in a particular type of service
(such as automotive electricians and engine re-borers) and in a particular type of
vehicle (such as large trucks) of a particular brand (such as Mercedes-Benz).
Collaboration among specialists is coordinated by generalist mechanics called “fitters,”
who receive orders from car owners, determine the cause of the trouble, decide who
should be involved in the repair work and how much they should be paid, and collect
and distribute the money. Such transactions are active probably because the
geographical proximity among transacting parties discourages opportunistic behaviors
and reduces transaction costs. Suame Magazine is equipped with a large number of
machine tools, such as lathes and milling machines, and specialized machines. Skilled
machinists operating these machines overhaul engines, gears, and crankshafts. Such
services are more expensive or unavailable at smaller garage clusters.
The number of these machining shops has increased since the 1980s, when the
Intermediate Technology Transfer Unit (ITTU), a training institution established in
1980 by the Kumasi Nkrumah University of Science and Technology, assisted
promising enterprises in acquiring machine tools. Besides working with fitters,
machinists produce simple auto parts, such as center bolts, U-bolts, and nuts, which
traders buy in bulk. Machinists also repair worn gears and other machine parts for
large firms located outside the cluster, such as lumber mills and mining companies.
Moreover, they process parts for metal products, such as flour mixing machine, water
pumps, and cash safes, which manufacturers fabricate using scrap metal.
Manufacturers are skilled welders. Using welding machines, they could
6
fabricate anything, but they usually specialize in one type of metal products. They are
unskilled in using machine tools or do not own machine tools. Thus, they contract out
machine processing to machinists nearby. This is a reason why they are located in the
cluster. Probably more important reason is that scrap metal is readily available in the
cluster. Because of their increasing demand for scrap metal as raw material, the
number of scrap dealers and scrap collectors has also increased in the cluster. Other
important users of scrap metal are foundries casting iron and other metal products and
blacksmiths forging metal to make farm implements, simple hand tools, and car parts.
Thus, the forward and backward linkages with the garage sector have attracted a
variety of new metalworking workshops to the cluster.
3. Data
According to our informants, profitability has been gradually declining in almost
all kinds of business in the cluster. The initial purpose of our empirical study of this
cluster was to find out the reason for the declining profitability. In 2004, we
conducted preliminary unstructured interviews with entrepreneurs in March,
September, and December and a formal enterprise survey for three months from
January to March, 2005. The sample consists of 100 garage mechanics, 92 machinists,
and 45 manufacturers, as shown in the first row of Table 2. These sample enterprises
were randomly selected within the respective sectors. The garage sector is
underrepresented in the sample because we are interested in the distributions of
variables in each sector but not in the whole cluster and because the garage mechanics
are more homogeneous in behaviors and characteristics than entrepreneur in the other
sectors.
2
2
For example, the sample of car mechanics relative to their population is much smaller than that of
machinists or manufacturers relative to their population.
7
Table 2 shows the data on characteristics of the sample entrepreneurs by sector.
The footnotes attached to the table list up the products and services of each sector.
The entrepreneurs are about forty years old on average. Most of them were born in the
Ashanti Region, where the cluster is located, and more than 80 percent of them are
Akan tribesmen.
3
All the sample entrepreneurs are males. The difference in the mean
of the years of education is statistically significant between the garages and the rest but
insignificant between the machinists and the manufacturers. Some entrepreneurs went
to school for more than 12 years. They went to advanced courses of polytechnic and
professional schools and received technical trainings. None in the sample went to a
university.
More than 90 percent of the entrepreneurs were former apprentices for 4.6 years
on average in the case of garages and for a little less than three years on average in the
case of metalworking.
4
They learned from their masters how to produce metal
products or how to repair machines. They learned how to operate a business as well,
but it is important to note that their masters taught the self-employment type business
but not the management of a large organization with, say twenty workers or more. It
should also be noted that apprenticeship training in Africa does not teach knowledge of
marketing and bookkeeping unlike apprenticeship in many developed countries.
Table 2 also shows the percentage of the sample entrepreneurs who have
participated in short-term training programs teaching management (including
bookkeeping) and production techniques. These programs have been provided to the
artisans in the cluster mainly but not exclusively by two training institutions in the
3
Though not shown, the number of young entrepreneurs originally from the outside of the Ashanti
Region has been increasing as the cluster has become widely known.
4
The percentage of workers who are apprentices and the duration of apprenticeship shown in Tables 2
and 3, respectively, are consistent with the data analyzed by Frazer (2006).
8
cluster, i.e., ITTU and the National Vocational Training Institute (NVTI). More than
twenty percent of the sample entrepreneurs have received formal technical training.
In the machining sector, many entrepreneurs have sponsors, even though they
make all decisions except for fixed capital investment. Since they run businesses, we
refer to them as entrepreneurs. They and their sponsors split the profits. The common
rule is that the sponsor takes two thirds and the entrepreneur takes the rest. Some of
such sponsors are expatriates. As is shown in the table, about one third of the
entrepreneurs in the machining sector have sponsors living abroad, who are hereafter
referred to as “abroad-based owners.”
Table 3 presents the data on enterprise size in terms of employment, sales
revenues, variable costs, and producer surplus.
5
None of the sample entrepreneurs had
kept financial book systematically.
6
To obtain reasonably accurate data, we checked
the consistency of each respondent’s answers to our questions about different aspects
and by revising estimates of sales and costs, in front of the respondent until estimates
converge to the one that made much sense to both the respondent and us. If the
respondent kept any fragmentary records, we used them as well.
A typical enterprise in the cluster has less than ten workers including several
apprentices. The number of workers is not a good indicator of labor input since
apprentices are considerably heterogeneous in skill level, even though it is commonly
used in the cluster as an indicator of enterprise size. While the machinists have a
smaller number of workers, they have greater revenues, variable costs, and producer
surpluses than the manufacturers and garages. The manufacturers have greater sales
revenues but smaller producer surplus than the garages, because the former have to
5
Variable costs are measured as the sum of costs of materials, labor, subcontracting, and electricity.
Producer surplus is sales revenue minus variable cost.
6
See de Mel, McKenzie, and Woodruff (2009) for the difficulty in obtaining accurate data from micro
enterprises.
9
spend more on materials.
Each growth rates shown toward the bottom of the table is the difference of the
levels in 2004 and 2000 divided by four, which is intended to approximate the annual
growth rate. The garages had large negative growth in producer surplus because of the
declining sales and the soaring variable cost, especially labor costs. For the machinists,
the median growth rate was negative for the variable cost and positive for the producer
surplus, but the mean growth rate of producer surplus was negative. The
manufacturers spent more and sold more in 2004 than in 2004, and their producer
surplus declined. In the cluster, competitors who produce or provide the same
products or services are always produced by the apprenticeship. Since the
entrepreneurs do not know how to find new markets for their products and services, the
increased number of competitors implies smaller sales per enterprise. Masters would
like to prevent graduates, who finished apprentice training, from leaving and starting
their own businesses. To do so, they have to raise salaries to the graduates. The
manufacturers suffered from the soaring price of scrap metal, due to the increased
demand from China and India as well as the increased demand within the cluster.
Thus, profitability has tended to decline in the cluster.
Table 4 offers additional information on material procurement and marketing, as
well as data on capital stock, of the machinists and manufacturers in 2000 and 2004.
The garage sector is not included in this table because material procurement is not an
important activity in this sector, because this sector lacks marketing activities more
completely than the other sectors, and because garages have very little fixed capital.
The first two rows of the table demonstrate how material procurement has become
difficult for the machinists and manufacturers in recent years. The entrepreneurs have
to spend longer time on material procurement. Since decisions must be made quickly
10
at auction, the entrepreneurs are in charge of material procurement. Many
entrepreneurs go to all the way to Accra in search of good and inexpensive material
more than once a month.
The third row of the table presents the percentage of sales to traders, who buy
products in bulk, and to companies outside the cluster, which tend to place relatively
lucrative orders with them. Other customers of the machinists are mostly fitters, and
those of the manufacturers are individuals, whom few manufacturers can characterize.
Some entrepreneurs have the conception of market research and promotion, but few
practice it. The typical way of selling products or services in the cluster is simply to
wait for customers to come to their workshops. Ordinary consumers do not have a
favorable impression of the cluster. It looks like a big junkyard with wreckage of
vehicles abandoned here and there and the ground smeared with engine oil. Neither
street names nor street numbers exist in the cluster. However, only a small number of
workshops have signboards. Moreover, the entrepreneurs seldom give their customers
detailed explanations of how their products work. They behave just as their masters
did decades ago, when there were few competitors. The near constancy of the
percentages of sales to traders and companies in Table 4 is probably a reflection of the
lack of progress in marketing.
The bottom of the table shows the mean of capital stock measured by the
entrepreneur’s assessment of the replacement cost of their equipment. A substantial
increase in capital stock is found in the machining sector. This may have a bearing on
the negative median growth rate of variable cost in this sector.
4. Hypotheses and Empirical Framework
In the previous section, we saw that enterprise growth is stagnant and
11
profitability tends to decline in the cluster. To restore the high profitability, they need
to upgrade product lines, improve production efficiency, and find new markets.
Upgrading product lines, however, is often too difficult without using better equipment
than they have used, and credit constraints facing them often prohibit them from using
better equipment. Thus, it is important to devise more efficient methods of production,
practice economy, and adopt a marketing method suitable to their products or services.
These reforms require managerial knowledge or skill that the entrepreneurs did not
learn from their apprentice training or from their own experiences of running
businesses in the past. How can they obtain such knowledge or skill?
In the literature on industrial clusters in developing countries, a number of case
studies report that a serious decline in profits can induce entrepreneurs in a cluster to
improve their products, production and quality management, marketing methods, and
financial management (e.g., Schmitz and Nadvi 1999; Sonobe and Otsuka 2006). In
garment clusters in China, Vietnam, and Kenya, entrepreneurs responded to crisis by
improving changing marketing channels (Sonobe, Hu, and Otsuka 2002; Nam, Sonobe,
and Otsuka 2009; Akoten and Otsuka 2006). In a surgical instrument cluster in
Pakistan, entrepreneurs took collective actions in order to obtain from abroad the
information necessary to upgrade the quality of their products (Nadvi 1999). In a
machine tool cluster in Taiwan, entrepreneurs reduced production cost for high quality
products by taking full advantage of the already developed division of labor among
enterprises (Sonobe, Kawakami, and Otsuka 2003). In a electric fitting cluster in
China, entrepreneurs achieved a set of improvements in product quality, marketing
method, and production organization (Sonobe, Hu, and Otsuka, 2004). With these
induced improvements in marketing and management, enterprises could accumulate
funds quickly for investments in better equipment, which in turn allowed them to
12
upgrade product lines.
Most of these case studies find that the induced upgrading is led by highly
educated entrepreneurs, even though schools may not teach knowledge or skills
directly useful for the upgrading of management. Probably education helps
entrepreneurs search for useful information. The human capital literature maintains
that the ability to respond to changing opportunities is “one of the major benefits of
education accruing to people personally in a modernizing economy” (Schultz 1975, p.
843). Moreover, educated persons tend to keep records and make plans better than the
uneducated. Such ability is directly useful for business management. Paulson and
Townsend (2004) use years of schooling as a proxy for entrepreneurial talent in their
empirical analysis of credit constraints facing small entrepreneurs in Thailand.
In the context of Suame Magazine, high education means education at
polytechnic and professional schools. These schools do not teach managerial
bookkeeping, marketing, or anything directly useful for business administration. Still,
high education may give entrepreneurs the higher ability to calculate, search for useful
information, and adjust to changing opportunities. Besides, it may well be that highly
educated entrepreneurs are from relatively wealthy families, faced with less severe
credit constraints, and thus able to use greater working and fixed capital. Our data set
does not contain information of working capital itself, but it can be roughly captured
by variable costs. Based on these considerations, we hypothesize as follows:
Hypothesis 1: Capital stock, variable costs, and sales revenues are positively
associated with years of schooling.
It may be more difficult for small entrepreneurs in Sub-Saharan Africa to search
13
for useful managerial and technical information than their East Asian counterparts. In
China, for example, small entrepreneurs could obtain such information by
subcontracting with state-owned enterprises (SOEs) or by hiring managers and
engineers who quit SOEs (Otsuka, Liu, and Murakami 1998; Sonobe, Hu, Otsuka
2006). In Sub-Saharan Africa, such a source of information does not seem to exist
within countries. Some leather-shoe makers in Addis Ababa, Ethiopia, who have
recently succeeded in finding markets in Europe, have frequently visited Italy for many
years in search for new ideas. Such business trips are too expensive for most
entrepreneurs of small enterprises in Sub-Saharan Africa. Thus, training programs
offered in their clusters or nearby cities must be very important opportunities to learn
useful knowledge and skills.
While both managerial and technical knowledge and skills are important for
small entrepreneurs, the first step to cope with declining profits will be managerial
reforms for the reason discussed above. Thus, we expect that managerial training will
have stronger impacts on enterprise performance than technical training at the current
stage of development in Suame Magazine. By the same token, we also expect that
talented entrepreneurs are more willing to participate in managerial training than in
technical training if they find managerial training more useful. Thus, enterprise
performance will be more strongly associated with participation in managerial training
than technical training.
Capital stock, however, may be more closely associated with technical training
participation. Since technical training tends to teach how to make better use of
machines, those entrepreneurs who have relatively large capital stock may have
stronger incentive to participate in technical training, or technical training participants
may have stronger incentive to use machinery after training. Thus, it seems reasonable
14
to postulate the following hypothesis:
Hypothesis 2: Sales revenues and producer surplus are more strongly associated with
participation in managerial training than participation in technical training, whereas
capital stock is not associated with managerial training but with technical training.
These hypotheses are not about causal effects but about associations between
variables. This is because, given our data set, it is impossible to cope with the
endogeneity problem due to the correlation of education and training participation with
unobservable entrepreneurial talents. As to the effects of training, the problem may be
better referred to as selection bias since more talented entrepreneurs may be more
likely to obtain information about invitation to training or more willing to seize the
opportunity of participating in training. Identifying causal effects would require a
randomized experiment or the application of the difference-in-differences (DID)
propensity score matching (PSM) estimator proposed by Heckman, Ichimura, and
Todd (1997, 1998) to non-experimental but panel data containing information on
enterprise performances before and after training. Since such data are unavailable, the
second choice for us is to apply the PSM estimator developed by Rosenbaum and
Rubin (1983) and extended by Heckman, Ichimura, and Todd (1997, 1998). After
running OLS regressions of sales revenues, variable costs, producer surplus on the
training participation dummies, educational backgrounds, and other characteristics of
entrepreneurs, we attempt the PSM estimation to see whether the positive association
between training participation and enterprise performance remains after mitigating the
selection bias.
In our PSM estimation, the effect of a training program is evaluated by
15
comparing the participants’ actual post-program performance and their estimated
counterfactual performance that they would achieve if they did not participate in the
program. More precisely, what is estimated is the average effect of treatment on the
treated (ATT) defined as E(Y
1
–
Y
0
|the entrepreneur has ever participated in a training
program), where Y
1
is the performance of a participant and Y
0
is the counterfactual
performance that the same entrepreneur would have achieved if he had not participated
in any training program. The matching estimator tries to obtain an estimate of this
counterfactual performance by matching each participant with one or more non-
participants who have similar characteristics. Let X denote a set of observed
characteristics of entrepreneurs and their enterprises. As shown by Rosenbaum and
Rubin (1983), a consistent matching estimate of ATT can be obtained by matching a
participant with non-participants whose propensity scores are similar. Propensity
score is defined as the conditional probability of participating in training given X.
The PSM estimator of ATT can be generally expressed as
( ) ( ) ( )
? ?
? ?
?
?
?
?
?
?
? =
1 0
0 1
,
1
I i I j
j j i i
Y X p X p W Y
N
PSM ,
where N is the number of the participants, I
1
and I
0
are respectively the treatment (or
participant) group and the matched control (or matched non-participant) group, p(X) is
the propensity score, and W is a weight determined by the distance between propensity
scores of participant i and the matched non-participants j. This estimator has been
widely applied to non-experimental data from developing economies. For example,
Rosholm, Nielsen, and Dabalen (2007) use this estimator to evaluate the impacts of
technical training programs for workers on labor productivity in Kenya and Zambia.
We estimate the propensity score using a probit model of training participation
with X being covariates. We impose the common support condition (Heckman,
16
Ichimura, and Todd 1998), and drop from the sample the non-participants whose
propensity scores are higher than the maximum or less than the minimum propensity
scores among the participants. The literature has proposed several types of matching
methods, including stratification matching, nearest neighborhood matching,
Epanechnikov (or quadratic) kernel, normal (or Gaussian) kernel, bi-weight kernel, and
local linear matching. We use these matching methods to see if the estimation results
are robust to the choice of matching methods. We obtain the standard errors of PSM
estimates by bootstrapping with 1000 replications, following the lead of Smith and
Todd (2005).
According to Rosenbaum and Rubin (1983), there should be no significant
differences in X between the matched participants and non-participants. The so-called
balancing test is a specification test using this property. We perform two types of
balancing test. The first is the t test of equality in the mean of each covariate between
the two groups. Second, we estimate the same probit model using the samples with
and without matching to obtain two values of pseudo-R squared and compare them.
We also perform a likelihood-ratio test. If matching is successful, the after-matching
probit should have no explanatory power so that the pseudo-R squared should be low
and the estimated coefficients should be closed to zero.
5. Estimation Results
OLS results
Tables 5 to 7 present the OLS estimates of the functions explaining sales
revenue (S), variable cost (C), producer surplus (P), capital stock (K), the proportion of
sales to traders and companies, and employment size. Table 5 reports the results for
the machinists. The number of years of schooling has positive and highly significant
17
coefficients in all the equations, which is consistent with Hypothesis 1. These
coefficients may overstate the effects of education for some reasons, including the ones
that we discussed in the previous section. One such reason is that the education level
of an entrepreneur may be positively correlated with his family’s wealth, which is
likely to influence enterprise sizes in the presence of credit constraints. We hope that
the inclusion of years of father’s schooling mitigates the bias due to this correlation.
Participation in managerial training has positive and significant coefficients in
equations explaining sales revenues, variable costs, producer surplus, and labor
employment, but not the fixed capital equation. By contrast, the coefficients of
technical training participation are not significantly different from zero in any equation.
These results are consistent with Hypothesis 2, except that capital stock is not closely
associated with technical training participation. Capital stock is associated closely
with the abroad-based owner dummy, which is a natural result since the role of such
owners is to finance investment for skilled artisans faced with credit constraints.
According to the estimates, the enterprises with abroad-based owners have high-
proportions of sale to traders and companies. This is probably because they are
equipped better and able to attract repair work orders from large companies outside the
cluster.
Interestingly, apprentice training has a negative and marginally significant
coefficient in the worker equation. This result suggests that the traditional
apprenticeship is becoming less popular among the youth. As shown by Frazer (2006),
apprentice training pays poorly unless a graduate from apprenticeship starts an own
business. In the machining sector, starting business without relying on investment by
sponsors is becoming difficult. Since entrepreneurs have to pay two thirds of profits to
sponsors, it takes long time for entrepreneurs to reap returns to apprenticeship training.
18
It may well be that apprenticeship is becoming less sustainable in this sector. Another
major finding is that enterprises with abroad-based owners have neither larger sales
revenues nor greater producer surplus despite their greater capital stock. This calls for
future investigation into the issue of moral hazard. The estimation results concerning
age and years of operation are somewhat unique. They differ from the results of
preceding studies of enterprise growth in that entrepreneurial age as well as enterprise
age is correlated negatively with enterprise sizes.
Table 6, which is about the manufacturers, shows qualitatively similar results,
even though statistical significance levels tend to be lower than in Table 5, probably
because the sample size is smaller. A major difference is that the coefficient of
technical training participation is insignificant in the capital stock equation in Table 5
whereas that is significant in Table 6, consistent with Hypothesis 2. The positive
coefficient of managerial training participation in the producer surplus equation has a
higher significance level in Table 6 than in Table 5. Overall, the results shown in
Tables 5 and 6 are consistent with Hypotheses 1 and 2.
Table 7, which is about the garages, looks very different from the previous two
tables. In this table, only a few estimated coefficients are significantly different from
zero. Moreover, no indicators of enterprise performance we employed are correlated
with the level of education and participation in managerial training. These results are
inconsistent with Hypotheses 1 and 2 and also with the results of the preceding studies,
such as the one by Mengistae (2006), who finds positive effects of entrepreneurial
human capital on enterprise performance. The contrast between Table 7 and the
preceding two tables suggests that managerial training as well as schooling is not
useful for all types of enterprises but for only some types. Machinists and
manufacturers may be able to improve enterprise performance more easily than
19
garages, if equipped with a little better knowledge of management.
PSM Results
We now turn to the PSM estimation of the ATT of technical training and
managerial training. The results of the OLS regressions indicate clearly that there is no
effect of training on enterprise performance in the garage sector and that the machinists
and manufacturers share qualitatively similar associations between training
participation and enterprise performance. Thus, we will focus on the machinists and
manufacturers, that is, the metalwork sector of the cluster, and not apply the PSM
method to the data of the garage sector.
The PSM estimation begins by estimating probit models that explain
participation in technical and managerial training. Table 8 reports the estimation
results. The probit models include all the characteristics of the entrepreneurs and two
additional dummy variables: one indicating whether the enterprise is located in the part
of the cluster called the “old site”, and the other indicating whether the entrepreneur is
a manufacturer. In both columns, the entrepreneur’s age and years of schooling have
positive and significant effects on participation. Older entrepreneurs are more likely to
have participated in training, probably because they have encountered a greater
number of opportunities to attend training programs. More highly educated
entrepreneurs may be better informed of learning opportunities. While the significance
of the effect of years of schooling on managerial training participation is of the five
percent level, that on technical training participation is of the one percent level. The
difference in significance level may be linked with a possible correlation between
education and wealth. If technical training may teach techniques suitable to well-
equipped workshops, technical training participants may be relatively wealthy
20
entrepreneurs, who tend to be highly educated. We have no idea of why the negative
correlation between the old site dummy and managerial training participation is highly
significant. The pseudo R squared obtained from each estimate of the probit model is
sufficiently high for the matching purpose.
These results of probit estimation indicate that the participants and non-
participants differ substantially in terms of entrepreneur characteristics. Differences
between them are also apparent in Figures 1 and 2, which show the histograms of the
propensity scores calculated from the estimated probit models. Thus, if we simply
compared the average performance of the participants with that of the non-participants,
we would fail to isolate the effects of training participation from the effects of
entrepreneur characteristics. However, if they differed completely, it would be
impossible to estimate the counterfactual performance based on the performance of the
matched non-participants, or matching would be impossible. Thus, the distribution of
propensity score for the participants and that for the non-participants must have a
common range of support, in order for the PSM estimation to be feasible. As shown
clearly in Figures 1 and 2, there exists such a common support.
Using the estimated propensity score, we match the participant (treatment) group
and the non-participant (control) group so that the two groups after matching have
similar characteristics. We applied six matching methods, and for each method, we
performed two types of balancing test. Table A1 presents the results of the two
balancing tests in the case in which the local linear matching method is applied to the
technical training participants and non-participants. Balancing test 1 is a set of t-tests
for each of covariates X. We find no significant difference in any covariate between
the treated and control groups. Balancing test 2 is based on the fitting measures of the
probit model. The pseudo-R squared is very low. According to the result of the
21
likelihood ratio test, the hypothesis that the coefficients are all zero is not rejected.
The p-value for this test is very high. These test results indicate clearly that the local
linear matching is successful. Table A2 shows similar test results for managerial
training in the case of local linear matching. Although not shown in this paper, we
obtained similar results of the balancing tests in the cases of the five other matching
methods.
Table 9 presents the estimated ATT of formal technical training on sales revenue,
variable cost, producer surplus, capital stock, the proportion of sales to traders and
companies, and the number of workers. Consistent with the result of OLS regression
shown in Tables 5 and 6, we find little evidence that participation in technical training
helps entrepreneurs achieve higher performances. Exceptions are positive and
significant ATT of training on capital stock, which is obtained when stratification
matching is applied, and the positive and significant ATT on the number of workers,
which is found when stratification matching and nearest neighborhood matching are
applied. Participation in technical training has a positive and significant ATT on
capital stock probably because technical training tends to teach how to make better use
of machinery. The positive and significant ATT on the number of workers may be
obtained because technical training participants have more to teach apprentices and can
attract a greater number of apprentices than non-participants.
Table 10 presents the estimated ATT of managerial training. Regardless of
matching method, the estimated ATT’s on sales revenues and producer surplus are
positive and highly significant, whereas the estimated ATT on capital stock is
insignificant. These results, together with the results shown in Table 9, are highly
consistent with Hypothesis 2. Among the different estimates based on the different
matching methods, the smallest estimate of the ATT on the logarithm of producer
22
surplus is 0.932. That is, the smallest estimate of the ATT on producer surplus is
exp(0.932) or 2.5. Thus, compared with the counterfactual producer surplus that the
treated group (i.e., managerial training participants) would have earned if they had not
participated in any managerial training, their actual producer surplus is 2.5 times as
large. These results of the PSM estimation lend support to our view that managerial
training is useful in the metalwork sector. Yet it is important to note that these results
are not enough to establish causal effects since the selection bias arising from the
correlation between unobservable factors and training participation may remain.
6. Conclusions
Industrial clusters have recently attracted much attention from development
economists because clusters are said to increase productivity and enlarge growth
opportunities of the industry and because they are ubiquitous even in Sub-Saharan
Africa. However, enterprises in this region remain extremely small in size. Many of
them suffer from declining profits. This paper has asked if a reason for the stagnancy
of enterprise development in Africa may be found in inadequate managerial skills of
entrepreneurs themselves. Our analysis suggests that managerial skills of small
entrepreneurs in a large cluster in Ghana are indeed poor, and that managerial training
is helpful for some types of small entrepreneurs.
The findings of this paper, however, raise several questions. For example, we
should explore why managerial training works in some industries but not in others.
But more fundamentally, we need tight evidence. In order to obtain tight evidence, it
is desirable to carry out randomized experiments in which managerial training is
provided to randomly selected entrepreneurs. Experiments and post experiment
surveys may be desired so that we can investigate the issues like which aspect of
23
managerial training is more useful and how long it takes until the effects of training
make themselves felt. We may be able to investigate the extent of information
spillovers within industrial clusters. The results of our analysis in this paper warrant
such experiments.
24
References
Adeya, C. N. 2003. “Sources of Training in African Clusters and Awareness of ICTs:
A Study of Kenya and Ghana,” Discussion Paper Series, No.2003-6, United
Nations University, Institute for New Technologies (INTECH).
Akoten, J.E., and Otsuka, K. 2006. “From Tailors to Mini-manufacturers: The Role of
Traders in the Transformation of Garment Enterprises in Nairobi,” Journal of
African Economies, forthcoming.
Bigsten, A., P. Collier, S. Dercon, B. Gauthier, J. W. Gunning, A. Isaksson, A. Oduro,
R. Oostendorp, C. Pattillo, M. Söderbom, M. Sylvain, F. Teal, and A. Zeufack.
1999. “Investment in Africa’s Manufacturing Sector: A Four Country Panel Data
Analysis.” Oxford Bulletin of Economics and Statistics 61(4):489–512.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J. W. Gunning, A.
Oduro, R. Oostendorp, C. Pattillo, M. Söderbom, F. Teal, and A. Zeufack. 2003.
“Credit Constraints in Manufacturing Enterprises in Africa.” Journal of African
Economies 12(1): 104–25.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J. W. Gunning, A.
Oduro, R. Oostendorp, C. Pattillo, M. Söderbom, F. Teal, and A. Zeufack. 2004.
“Do African Manufacturing Firms Learn from Exporting?” Journal of
Development Studies 40(3): 115–71.
Bigsten, A., and Söderbom, M. 2006. “What Have We Learned From a Decade of
Manufacturing Enterprise Survey in Africa?” The World Bank Research Observer,
21 (2), 241-65.
Bigsten, Arne, and Mulu Gebreeyesus. 2007. “The Small, the Young, and the
Productive: Determinants of Manufacturing Firm Growth in Ethiopia.” Economic
Development and Cultural Change 55 (4): 813-840.
25
Collier, P., and Gunning, J.W. 1999. “Explaining African Economic Performance,”
Journal of Economic Literature, 37, 64-111.
Davidson, R., and MacKinnon, J.G. 1993. Estimation and Inference in Econometrics.
Oxford: Oxford University Press.
de Mel, Suresh, McKenzie, David, and Woodruff, Christopher 2009. “Measuring
Microenterprise Profits: Must We Ask How the Sausage is Made?” Journal of
Development Economics 88 (1), 19-31.
Fafchamps, M. 2004. Market Institutions and Sub-Saharan Africa: Theory and
Evidence. Cambridge, MA: MIT Press.
Fafchamps, M. and Söderbom, M. (2006). Wages and labor management in African
manufacturing. Journal of Human Resources, 41(2), 346-379.
Frazer, Garth. 2005. “Which Firms Die? A Look at Manufacturing Firm Exit in
Ghana.” Economic Development and Cultural Change 53(3):585–617.
Frazer, Garth. 2006. “Learning the Master's Trade: Apprenticeship and Human Capital
in Ghana.” Journal of Development Economics 81 (2): 259-98.
Goedhuys, M. and Sleuwaegen, L. (2000). “Entrepreneurship and Growth of
Entrepreneurial Firms in Côte d’Ivoire.” Journal of Development Studies: 36(3),
122-145.
Gunning, Jan Willem, and Mengistae, Taye. 2001 “Determinants of African
Manufacturing Investment: the Microeconomic Evidence.” Journal of African
Economies 10 (1): 48-80.
Heckman, James J., Hidehiko Ichimura, and Petra Todd, 1997. “Matching as an
Econometric Evaluation Estimator: Evidence from Evaluating a Job Training
Programme,” Review of Economic Studies 64: 605-654.
26
Heckman, James J., Hidehiko Ichimura, and Petra Todd, 1998. “Matching as an
Econometric Evaluation Estimator,” Review of Economic Studies 65: 261-294.
Institute of African Studies. 1992. The City of Kumasi, Handbook – Past, Present and
Future. The Institute of African Studies, University of Ghana in collaboration with
the Kumasi Metropolitan Assembly.
Kinyanjui, N. (2007). “Tha Kamkunji metalwork cluster in Kenya.” In D. Z. Zeng
(Ed.), Knowledge, Technology, and Cluster-based Growth in Africa, WBI
Development Studies, Washington, DC: The World Bank.
McCormick, Dorothy. 1999 “African Enterprise Clusters and Industrialization: Theory
and Reality,” World Development 27(9): 1531-1551.
Mengistae, T. 2006. “Competition and entrepreneurs’ human capital in small business
longetivity and growth.” Journal of Development Studies 42 (5), 812-836.
Nam, V. H, Sonobe, T, and Otsuka, K. 2009. “An Inquiry into the Development
Process of Village Industries: The Case of a Knitwear Cluster in Northern
Vietnam,” Journal of Development Studies, forthcoming.
Obeng, G.Y. 2002. “Kumasi Suame Magazine: A Background Paper,” Technology
Consultancy Centre, Kwame Nkrumah University of Science & Technology,
Kumasi, Ghana.
Otsuka, K., Liu, D., and Murakami, N. 1998. Industrial Reform in China: Past
Performance and Future Prospects. Oxford: Clarendon Press.
Paulson, Anna. L., and Townsend, Robert 2004. “Entrepreneurship and Financial
Constraints in Thailand.” Journal of Corporate Finance 10 (2): 229-262.
Ramachandran, V. and Shah, M. K. 1999. “Minority Entrepreneurs and Firm
Performance in Sub-Saharan Africa,” Journal of Development Studies 36 (2): 71-
87.
27
Rosenbaum, P. R. and Rubin, D. B. 1983. “The Central Role of the Propensity Score
in Observational Studies for Causal Effects,” Biometrika 70 (1): 41-55.
Rosholm, Michael; Nielsen, Helena, Skyt; and Dabalen, Andrew 2007. “Evaluation of
Training in African Enterprises.” Journal of Development Economics 84 (1): 310-
329.
Schmitz, H., and Nadvi, K. 1999. “Clustering and Industrialization,” World
Development, 27, 1503-1514.
Schultz, T.W. 1975. “The Value of the Ability to Deal with Disequilibria,” Journal of
Economic Literature 13 (3), 827-46.
Sluwaegen, L. and Goedhuys, M. 2002. “Growth of firms in developing countries:
Evidence form Côte d’Ivoire.” Journal of Development Economics, 68 (1): 117-
135.
Smith, J.A. and Todd, P. E. 2005. “Does Matching Overcome LaLonde’s Critique of
Nonexperimental Estimators? Journal of Econometrics 125 (1-2): 305-353.
Söderbom, M. and Teal, F. 2004. “Size and efficiency in African manufacturing firms:
Evidence from firm-level panel data.” Journal of Development Economics 73 (2):
369-394
Sonobe, T. and Otsuka, K. 2006. Cluster-Based Industrial Development: An East
Asian Model. Basingstoke: Palgrave Macmillan.
Sonobe, T., Akoten, J.E., and Otsuka, K. 2006. “The Development of the Footwear
Industry in Ethiopia: How Different Is It from the East Asian Experience.” FASID
Discussion paper 2006-09-003.
Sonobe, T., Hu, D., and Otsuka, K. 2004. “Process of Cluster Formation in China: A
Case Study of a Garment Town,” Journal of Development Studies, 39 (1), 118-39.
Sonobe, T., Hu, D., and Otsuka, K. 2004. “From Inferior to Superior Products: An
28
Inquiry into the Wenzhou Model of Industrial Development in China,” Journal of
Comparative Economics, 32 (3), 542-63.
Sonobe, T., Hu, D., and Otsuka, K. 2006. “Industrial Development in the Inland
Region of China: A Case Study of the Motorcycle Industry,” Journal of
Comparative Economics, 34 (4), 818-38.
Sonobe, T., Kawakami, M., and Otsuka, K. 2003. “Changing Roles of Innovation and
Imitation in Industrial Development: The Case of the Machine Tool Industry in
Taiwan,” Economic Development and Cultural Change, 52 (1), 103-28.
Van Biesebroeck, J. 2005a. “Exporting Raises Productivity in Sub-Saharan African
Manufacturing Firms.” Journal of International Economics 67(2):373–91.
Van Biesebroeck, J. 2005b. “Firm Size Matters: Growth and Productivity Growth in
African Manufacturing.” Economic Development and Cultural Change 53(3):545–
83.
29
Table 1. Estimates of the Enterprise Population by Sector in the Suame Magazine
Cluster in 2000, 2002, and 2003
a
Garages Metalworking
enterprises
Others
b
Total
2000 4,958 807 2,204 7,969
2002 6,222 990 2,618 9,830
2003 7,847 1139 2,844 11,830
Notes:
a. These estimates are taken from the data base of the Suame branch of the Ghana National
Association of Garages (GNAG). The estimates do not include ancillary trades such as
restaurants and telecommunication shops.
b. “Others” include car body builders, pot makers, sign writers, and some types of welders.
30
Table 2. Characteristics of the Entrepreneurs by Sector
Garages
a
Machinists
b
Manufacturers
c
No. sample enterprises 100 92 45
Age 40.3 40.3 41.7
% from Ashanti 77.0 84.9 80.0
Years of schooling 9.49 11.41 10.53
Apprentice training 99.0 89.1 91.1
Years of apprenticeship 5.61 3.05 4.16
% technical training 24.0 28.3 24.4
% managerial training 4.0 16.3 11.1
% owner abroad 1.0 32.6 15.6
Years of operation 16.4 12.4 14.2
Notes:
a. Garage mechanics include fitters, vehicle body strainghteners, sprayers, and auto
electricians. Fitters diagnose trouble in engine and other auto parts and assign repair work
to specialized mechanics.
b. Machinists are defined here as machining enterprises which do not specialize in engine
block re-boring. They operate lathes, milling machines, screw thread cutters, gear-cutting
machines, and other machine tools, to provide customers with machining services such as
shaft lathing, long shaft turning, gear cutting, and screw thread cutting. They also produce
bolts and nuts, wheel bolts, center bolts, bushings and other fabricated metal parts.
c. Manufacturers produce coffers, carbide welding pots, food processing machinery and
equipment, such as palm oil press, cassava press, rice huller, rice thresher, flour mixing
machine, and other items including cooking stoves for households and large ovens for
industrial use.
31
Table 3. Enterprise Size in 2004 and Growth from 2000 to 2004 by Sector
Garages Machinists Manufacturers
Mean Median Mean Median Mean Median
No of workers 10.2 7.0 5.2 4.5 6.0 6.0
% apprentices 75.9 80.0 61.4 66.7 67.3 75.0
Sales revenue (1,000USD) 14.1 8.0 44.6 30.6 22.6 8.8
Variable cost (1,000USD) 6.0 2.3 20.1 11.7 15.5 8.3
Producer surplus (1,000USD) 8.1 4.7 24.5 16.3 7.1 2.3
Growth in sales revenue (%) -5.0 -5.7 -0.4 -0.9 1.9 0.6
Growth in variable cost (%) 27.1 26.0 0.5 -2.1 5.9 4.8
Growth in producer surplus(%) -13.4 -12.3 -1.0 0.8 -2.6 -1.6
32
Table 4. Material Procurement, Marketing Channels, Capital Stock, 2000, 2004
Machinists Manufacturers
2000 2004 2000 2004
% work time of
entrepreneur allocated
to material procurement
11.3 27.1 12.0 24.1
Material cost-sales
revenue ratio
0.16 0.20 0.32 0.40
% sales to traders and
companies
31.9 35.3 34.6 34.8
Capital stock (USD)
7,364 10,091 3,855 4,026
33
Table 5. OLS Estimates for Machinists, 2004
lnS lnC lnP lnK
Proportion of
sale to
traders and
companies
Workers
Age
-0.018*
(0.010)
-0.030**
(0.013)
-0.011
(0.010)
0.012
(0.008)
-0.006***
(0.002)
-0.023
(0.026)
From Ashanti
0.405
(0.289)
0.192
(0.361)
0.493*
(0.257)
0.359
(0.219)
0.115*
(0.064)
0.500
(0.708)
Years of schooling
0.119***
(0.034)
0.070*
(0.042)
0.150***
(0.030)
0.121***
(0.026)
0.025***
(0.007)
0.198**
(0.082)
Apprentice training
0.399
(0.358)
0.412
(0.447)
0.473
(0.318)
-0.179
(0.272)
0.078
(0.080)
-1.465*
(0.876)
Technical training
-0.074
(0.247)
0.080
(0.308)
-0.157
(0.220)
0.0753
(0.188)
0.077
(0.055)
-0.408
(0.605)
Managerial training
0.566*
(0.301)
0.784**
(0.376)
0.465*
(0.268)
0.268
(0.228)
0.106
(0.067)
2.136***
(0.737)
Abroad-based
owner
0.003
(0.240)
-0.205
(0.299)
0.087
(0.213)
0.521***
(0.182)
0.237***
(0.053)
0.123
(0.586)
Years of operation
-0.022*
(0.011)
-0.016
(0.013)
-0.017*
(0.009)
-0.027**
(0.008)
0.001
(0.002)
0.001
(0.026)
Years of father’s
schooling
0.020
(0.018)
0.002
(0.023)
0.041**
(0.017)
0.020
(0.014)
0.002
(0.004)
-0.0003
(0.045)
Intercept
9.067***
(0.760)
9.339**
(0.948)
7.539***
(0.676)
6.784***
(0.576)
0.002
(0.169)
4.074**
(1.858)
Adj. R-squared 0.183 0.090 0.290 0.399 0.275 0.116
Notes:
Number of observations is 92. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
34
Table 6. OLS Estimates for Manufacturers, 2004
lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Age
-0.001
(0.019)
-0.002
(0.020)
-0.002
(0.024)
-0.033
(0.028)
-0.009*
(0.005)
-0.011
(0.052)
From Ashanti
0.363
(0.440)
0.346
(0.463)
0.520
(0.571)
0.376
(0.647)
0.152
(0.124)
1.559
(1.206)
Years of schooling
0.147*
(0.079)
0.159*
(0.083)
0.160
(0.102)
-0.003
(0.118)
0.036
(0.022)
-0.139
(0.216)
Apprentice training
0.752
(1.007)
1.023
(1.059)
0.980
(1.305)
-1.413
(1.457)
0.393
(0.284)
-2.191
(2.760)
Technical training
0.048
(0.551)
-0.116
(0.580)
0.041
(0.717)
1.697**
(0.814)
-0.240
(0.156)
3.189**
(1.510)
Managerial training
1.276
(0.817)
1.137
(0.859)
2.105**
(1.058)
0.133
(1.218)
0.445*
(0.231)
1.243
(2.239)
Abroad-based
owner
0.364
(0.497)
0.696
(0.523)
0.012
(0.647)
2.447***
(0.733)
0.141
(0.140)
2.612*
(1.363)
Years of operation
0.024
(0.019)
0.023
(0.020)
0.017
(0.025)
-0.004
(0.028)
0.004
(0.022)
0.022
(0.052)
Years of father’s
schooling
-0.008
(0.030)
-0.018
(0.032)
0.011
(0.039)
-0.075
(0.045)
0.002
(0.008)
0.077
(0.083)
Intercept
6.298**
(1.744)
5.819**
(1.834)
4.302*
(2.257)
8.934**
(2.602)
-0.209
(0.492)
6.112
(4.779)
Adj. R-squared 0.219 0.159
0.199
0.354 0.111
0.263
Notes:
Number of observations is 45. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
35
Table 7. OLS Estimates for Garages, 2004
lnS lnC lnP Workers
Age
-0.011
(0.017)
-0.029
(0.020)
-0.006
(0.020)
0.116*
(0.055)
From Ashanti
0.177
(0.311)
-0.023
(0.363)
0.307
(0.351)
-0.662
(0.996)
Years of schooling
-0.022
(0.045)
0.007
(0.052)
-0.045
(0.051)
-0.055
(0.144)
Apprentice training
-0.758
(1.236)
1.090
(1.446)
-1.630
(1.399)
1.872
(3.966)
Technical training
-0.238
(0.297)
0.001
(0.348)
-0.468
(0.336)
1.777*
(0.953)
Managerial training
-0.427
(0.634)
-0.508
(0.742)
-0.126
(0.718)
-3.760*
(2.034)
Years of operation
0.013
(0.015)
0.021
(0.017)
0.010
(0.017)
-0.028
(0.048)
Years of father’s
schooling
0.010
(0.023)
0.012
(0.026)
0.001
(0.026)
0.151**
(0.072)
Intercept
10.13**
(1.34)
7.53**
(1.57)
10.36**
(1.52)
0.405
(4.304)
Adj. R-squared -0.041 -0.033
-0.032
0.043
Notes
Number of observations is 45. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
36
Table 8. Estimated Probit Models of Training Participation
Technical training
Managerial training
Age
0.032**
(0.013)
0.046**
(0.018)
From Ashanti
0.416
(0.376)
0.504
(0.485)
Apprentice training
-0.118
(0.438)
-0.696
(0.489)
Years of schooling
0.125***
(0.042)
0.111**
(0.052)
Years of operation
0.001
(0.014)
0.018
(0.019)
Abroad-based owner
-0.091
(0.306)
-0.401
(0.443)
Years of father’s schooling
0.009
(0.022)
-0.048
(0.030)
Old site
0.067
(0.287)
-1.187***
(0.391)
Manufacturing sector
-0.095
(0.279)
-0.186
(0.367)
Intercept
-3.695***
(1.033)
-3.380**
(1.307)
LR chi2 (4) 23.73***
32.12***
Log Likelihood -68.05
-40.89
Pseudo R-squared 0.149
0.282
Notes
The number of observations is 137. Marginal effects for dummy variables are calculated as
the effect of a change from 0 to 1. Standard errors are in parentheses. ***, **, and * indicate
the 1, 5, 10 percent levels of statistical significance, respectively.
37
Table 9. PSM estimate of ATT of Technical Training
Matching methods lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Stratification
0.377
0.401
0.398
0.700**
0.048
1.976***
(0.358) (0.337) (0.424) (0.310) (0.059) (0.694)
Nearest Neighborhood
0.206
0.439
0.156
0.626
0.095
2.135**
(0.436) (0.457) (0.519) (0.491) (0.075) (0.996)
Epanechnikov Kernel
0.111
0.285
0.034
0.427
0.050
1.526
(0.361) (0.345) (0.407) (0.322) (0.054) (0.838)
Normal Kernel
0.182
0.322
0.140
0.491*
0.045
1.660*
(0.320) (0.332) (0.408) (0.290) (0.057) (0.879)
Biweight Kernel
0.080
0.274
-0.016
0.447
0.056
1.496*
(0.352) (0.350) (0.407) (0.327) (0.055) (0.839)
Local Linear
-8.81e-05
0.232
-0.060
0.569
0.093
1.902*
(0.451) (0.428) (0.509) (0.441) (0.069) (0.986)
Note. ***, **, * denote the statistical significance at the 1%, 5%, and 10% levels, respectively. In
parentheses are standard errors obtained by bootstraps with 1000 replications and with re-sampling.
The common support condition was imposed.
38
Table 10. PSM estimate of the ATT of Managerial Training
Matching methods lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Stratification
0.832**
0.806***
1.143***
0.627
0.099
2.004*
(0.368) (0.350) (0.414) (0.422) (0.076) (1.102)
Nearest Neighborhood
1.469***
1.652***
1.679***
0.751
0.073
2.750**
(0.476) (0.529) (0.560) (0.597) (0.094) (1.299)
Epanechnikov Kernel
0.792**
0.703*
1.142***
0.749
0.089
2.038*
(0.385) (0.425) (0.426) (0.486) (0.076) (1.126)
Normal Kernel
0.778**
1.021***
0.932**
0.501
0.140*
2.047**
(0.327) (0.393) (0.385) (0.395) (0.069) (1.037)
Biweight Kernel
0.818**
0.740
1.161**
0.735
0.081
2.064*
(0.409) (0.434) (0.413) (0.472) (0.080) (1.125)
Local Linear
0.954**
1.211**
1.110*
0.603
0.084
2.225*
(0.404) (0.474) (0.460) (0.484) (0.083) (1.242)
Note. ***, **, * denote the statistical significance at the 1%, 5%, and 10% levels, respectively. In
parentheses are standard errors obtained by bootstraps with 1000 replications and with re-sampling.
The common support condition was imposed.
39
Table A-1. Balancing Test for Impact Evaluation of Technical Training using Local
Linear Matching
Balancing test 1
Mean
t-test
Variable Sample Treated Control % bias
% reduction
|bias|
t p-value
Age
Unmatched
Matched
41.916
45.622
37.783
46.081
37.8
-4.2
88.9
4.62
-0.19
0.000
0.852
From Ashanti
Unmatched
Matched
0.774
0.865
0.788
0.892
-3.3
-6.5
-95.6
-0.41
-0.35
0.681
0.727
Apprentice
training
Unmatched
Matched
0.847
0.811
0.965
0.919
-41.2
-37.8
8.3
-6.24
-1.36
0.000
0.178
Years of
schooling
Unmatched
Matched
11.526
12.892
9.431
12.838
58.4
1.5
97.4
7.20
0.06
0.000
0.949
Years of
operation
Unmatched
Matched
13.709
14.432
13.624
15.730
0.7
-11.4
-1421.5
0.09
-0.45
0.929
0.654
Abroad-based
owner
Unmatched
Matched
0.153
0.297
0.121
0.270
9.2
7.9
15.0
1.17
0.25
0.243
0.800
Years of father’s
schooling
Unmatched
Matched
5.537
6.514
5.233
5.189
5.1
22.1
-336.3
0.63
0.95
0.529
0.343
Old site
Unmatched
Matched
0.674
0.730
0.676
0.649
-0.6
17.3
2897.8
-0.07
0.75
0.944
0.458
Manufacturing
sector
Unmatched
Matched
0.268
0.297
0.229
0.270
9.1
6.2
31.1
1.13
0.25
0.258
0.800
Balancing test 2
Pseudo
R-squared
LR chi2
p-value
Matched 0.041
4.25
0.894
40
Table A-2. Balancing Test for Impact Evaluation of Managerial Training using Local
Linear Matching
Balancing test 1
Mean
t-test
Variable Sample Treated Control % bias
% reduction
|bias|
t p-value
Age
Unmatched
Matched
45.986
48.250
38.006
49.900
75.9
-15.7
79.3
6.00
-0.54
0.000
0.591
From Ashanti
Unmatched
Matched
0.739
0.800
0.789
0.800
-11.5
0.0
100.0
-0.97
0.00
0.331
1.000
Apprentice
training
Unmatched
Matched
0.822
0.750
0.951
0.850
-41.3
-32.0
22.5
-4.52
-0.78
0.000
0.442
Years of
schooling
Unmatched
Matched
12.247
12.850
9.661
12.050
68.4
21.2
69.1
5.89
0.72
0.000
0.476
Years of
operation
Unmatched
Matched
15.644
16.250
13.456
20.000
20.2
-34.6
-71.4
1.55
-0.88
0.121
0.382
Abroad-based
owner
Unmatched
Matched
0.164
0.250
0.124
0.350
11.4
-28.4
-149.2
0.99
-0.68
0.325
0.503
Years of Father’s
schooling
Unmatched
Matched
4.096
4.350
5.401
4.100
-21.6
4.1
80.9
-1.81
0.13
0.070
0.894
Old site
Unmatched
Matched
0.534
0.450
0.688
0.450
-31.8
0.0
100.0
-2.70
0.00
0.007
1.000
Manufacturing
sector
Unmatched
Matched
0.274
0.250
0.234
0.150
9.1
22.9
-151.2
0.77
0.78
0.444
0.442
Balancing test 2
Pseudo
R-squared
LR chi2
p-value
Matched 0.062
3.41
0.946
41
0
1
2
3
4
0 .5 1 0 .5 1
0 1
D
e
n
s
i
t
y
Pr(dtechtrain)
Graphs by dtechtrain
Figure 1. Histogram for estimated propensity score for technical training
0
5
1
0
1
5
0 .2 .4 .6 .8 0 .2 .4 .6 .8
0 1
D
e
n
s
i
t
y
Pr(dnontechtrain)
Graphs by dnontechtrain
Figure 2. Histogram for estimated propensity score for managerial training
doc_873751266.pdf
In Sub-Saharan Africa, manufacturers operating in spontaneously developed industrial clusters are very small in size, have low productivity, and stagnate except when they are young. The literature has related the preponderance of such enterprises to their socio-economic surroundings.
Entrepreneurial Skills and Industrial Development: The
Case of a Car Repair and Metalworking Cluster in Ghana
February 27, 2009
Alhassan Iddrisu
Ministry of Finance and Economic Planning, Ghana
Yukichi Mano
National Graduate Institute for Policy Studies, Japan
Tetsushi Sonobe
Foundation for Advanced Studies on International Development, Japan
Email: [email protected]
Preliminary. Please do not cite without author permission
1
Entrepreneurial Skills and Industrial Development: The
Case of a Car Repair and Metalworking Cluster in Ghana
Abstract
In Sub-Saharan Africa, manufacturers operating in spontaneously developed industrial
clusters are very small in size, have low productivity, and stagnate except when they
are young. The literature has related the preponderance of such enterprises to their
socio-economic surroundings. This paper reconsiders the issue by looking at the way
small entrepreneurs engage in business in a car repair and metalworking industrial
cluster in Ghana. We hypothesize that these entrepreneurs are unaware of or unskilled
in basic techniques in marketing, management, and accounting, which is necessary for
enterprise growth. Evidence suggests that small entrepreneurs in the cluster are thirsty
for such techniques.
Keywords: Africa, Ghana, industrial development, industrial cluster, entrepreneurial
skills, training, propensity score matching, impact evaluation
JFL classification: O14, O33, O55
1
1. Introduction
In developing countries, a large number of manufacturers are operating in
spontaneously developed industrial clusters as Schmitz and Nadvi (1999), McCormick
(1999), and Sonobe and Otsuka (2006) attest regarding South America and South Asia,
Africa, and East Asia, respectively. While other regions have witnessed substantial
growth in enterprise size and productivity, Sub-Saharan Africa has had only a small
number of success stories. Enterprises operating in clusters in Africa are often
informal, have low productivity, and grow only when they are very young, according
to Ramachandran and Shah (1999), Sleuwaegen and Goedhuys (2002), Mazumbdar
and Mazaheri (2003), Frazer (2005), Bigsten and Söderbom (2006), Van Biesebroeck
(2005a), Mengistae (2006), and Bigsten and Gebreeyesus (2007). These studies find
that major determinants of enterprise growth and survival include enterprise age and
size and entrepreneurial human and social capital, such as years of schooling, years of
business experience, and access to informal network. Bigsten et al. (1999), Van
Biesebroeck (2005b), Rankin, Söderbom, and Teal (2006) among others explore what
make investment and exporting difficult for African enterprises.
The literature has also explored major constraints facing enterprises, such as
credit constraints, high risks, corruption, limited contract enforcement, labor costs
which tend to increase with enterprise sizes, and high costs of transportation and
electricity due to poor infrastructure (e.g., Bigsten et al. 2003; Gunning and Mengistae
2001; Fafchamps 2004; Söderbom and Teal 2004; Collier and Gunning 1999; Eifert,
Gelb, and Ramachandran 2008). Compared with these constraints, little attention has
been paid to inadequate skills of small entrepreneurs in marketing, production and
quality management, and accounting. Such entrepreneurial skills are part of the
human capital of entrepreneurs but not necessarily captured by the number of years of
2
schooling or the number of years of business experience. While it is fair to say that
many entrepreneurs are deficient in entrepreneurial skills, questions arise as to whether
the skill level is a major determinant of enterprise performance. If important, why do
small entrepreneurs remain deficient in such skills even after many years in business?
Is it possible to teach such skills to entrepreneurs? Karlan and Valdivia (2009) present
evidence that a program of teaching entrepreneurship to small entrepreneurs in Peru
improved knowledge, practices and revenues. Does the same apply to Africa?
This paper attempts to answer some of these questions by using enterprise data
collected from an industrial cluster consisting of garage mechanics and metalworking
enterprises in Ghana. To obtain tight evidence for the importance of entrepreneurial
skills as a determinant of enterprise development and for the usefulness of
entrepreneurial training, it would be necessary to carry out a randomized experiment in
which such training is provided to randomly selected participants. This paper is not a
report of such an experiment, but an assessment of demand for entrepreneurial skill
training. We examine the associations among entrepreneurs’ characteristics, their
participation in such training in the past, and their current performance, in order to
infer how beneficial it will be if we provide a training program to them.
The industrial cluster under study is very large in terms of the number of
enterprises and the number of workers and apprentices, but the way in which small
entrepreneurs run their businesses is no different from that found in other clusters of
small businesses in Sub-Saharan Africa, such as the metalworking clusters in Nairobi,
Kenya, and a suburb of Kampara, Uganda, and the leather-shoe cluster in Addis Ababa,
Ethiopia.
1
That is, small entrepreneurs seldom keep records, seldom tout their
potential customers, and seldom take the initiative in making efficient use of materials,
1
For information on the metalwork clusters at Kariobangi and Kamukunji, see Sonobe, Akoten, and Otsuka
(2009a) and Kinyanjui (2007), respectively. For the leather-shoe cluster in Addis Ababa, see Sonobe, Akoten, and
Otsuka (2009b).
3
energy, and time.
Frazer’s (2005) account of apprenticeship applies perfectly to this cluster. The
majority of these entrepreneurs learnt production and business operation from their
masters through apprenticeship. The business model they were taught would be
suitable for self-employed masters working with several apprentices, but not for
owners and managers ambitious to expand their businesses. It is no wonder their
enterprises seldom grow beyond certain small sizes. Moreover, their enterprises may
become less profitable gradually since apprenticeship reproduces competitors who
produce or provide exactly the same products or services. In fact, profitability has
declined substantially in recent years in the cluster.
In response, many entrepreneurs have attempted to change the way of running
their businesses. We hypothesize that knowledge and skills that they did not learn
from apprentice training have assumed importance for their businesses, and that those
who have such knowledge and skills have better business results. Schools do not teach
such knowledge and skills, but education will help entrepreneurs search for useful
information, knowledge, and skills. In Sub-Saharan Africa, however, such intangible
inputs useful for a managerial reform are difficult to obtain not only for elementary
school dropouts but for polytechnic graduates. We conjecture that managerial training
plays an important role, and find that the current business results are strongly
associated with years of schooling and participation in managerial training in the
metalwork sector. Participation in technical training programs has no effects. These
results are obtained from OLS regressions and the propensity score matching (PSM)
estimation. By contrast, we find no such effects in the garage sector.
The rest of the paper is organized as follows. Section 2 describes the formation
of the large cluster under study. Section 3 presents the basic data of the sample
4
enterprises. Section 4 advances some testable hypotheses and explains empirical
strategy. Section 5 presents the results of OLS regressions and PSM estimations.
Section 6 tries to draw for future research and policies.
2. Brief History of the Cluster
The industrial cluster under study is located in the Suame area in Kumasi, the
second largest city in Ghana and the center of Ashanti Region. The cluster is called
Suame Magazine and dates from the 1930s when the dispersed craftsmen set up
workshops at the site of the present Kumasi Zoo, which used to be the site of an army
depot called Magazine during the colonial times. When the workshops resettled in the
current location, they kept this name (Institute of African Studies 1992). It has
expanded tremendously ever since in terms of the number of enterprise, employment,
and area. Table 1 presents the number of member enterprises of the Suame branch of
the Ghana National Association of Garages (GNAG), which comprises not only
garages but also blacksmiths, machinists, and manufactures and is generally believed
to cover 80 percent or more of the enterprise population in the cluster. The garage
sector is by far the largest and continues to grow rapidly in terms of the number of
enterprises.
In developed countries, garages are scattered far and wide to serve dispersed car
owners. In developing countries, most vehicles are business fleets. In Ghana, trailers
and trucks are concentrated on the artery roads connecting the major cities in the south,
such as the capital city, Accra, and port cities, and the major cities in the north, such as
Tamale, and the capital city of Burkina Faso. Kumasi is the most important junction
of these arteries. The number of the vehicles going back and forth on these arteries has
rapidly increased. The demand for garage services has increased accordingly. While
5
garages are clustered not only in Kumasi but also in Accra and other cities, Suame
Magazine is said to be larger and have higher technical skills and better equipment
than any other clusters in West Africa. The division of labor among specialists is
highly developed in this cluster. Each master specializes in a particular type of service
(such as automotive electricians and engine re-borers) and in a particular type of
vehicle (such as large trucks) of a particular brand (such as Mercedes-Benz).
Collaboration among specialists is coordinated by generalist mechanics called “fitters,”
who receive orders from car owners, determine the cause of the trouble, decide who
should be involved in the repair work and how much they should be paid, and collect
and distribute the money. Such transactions are active probably because the
geographical proximity among transacting parties discourages opportunistic behaviors
and reduces transaction costs. Suame Magazine is equipped with a large number of
machine tools, such as lathes and milling machines, and specialized machines. Skilled
machinists operating these machines overhaul engines, gears, and crankshafts. Such
services are more expensive or unavailable at smaller garage clusters.
The number of these machining shops has increased since the 1980s, when the
Intermediate Technology Transfer Unit (ITTU), a training institution established in
1980 by the Kumasi Nkrumah University of Science and Technology, assisted
promising enterprises in acquiring machine tools. Besides working with fitters,
machinists produce simple auto parts, such as center bolts, U-bolts, and nuts, which
traders buy in bulk. Machinists also repair worn gears and other machine parts for
large firms located outside the cluster, such as lumber mills and mining companies.
Moreover, they process parts for metal products, such as flour mixing machine, water
pumps, and cash safes, which manufacturers fabricate using scrap metal.
Manufacturers are skilled welders. Using welding machines, they could
6
fabricate anything, but they usually specialize in one type of metal products. They are
unskilled in using machine tools or do not own machine tools. Thus, they contract out
machine processing to machinists nearby. This is a reason why they are located in the
cluster. Probably more important reason is that scrap metal is readily available in the
cluster. Because of their increasing demand for scrap metal as raw material, the
number of scrap dealers and scrap collectors has also increased in the cluster. Other
important users of scrap metal are foundries casting iron and other metal products and
blacksmiths forging metal to make farm implements, simple hand tools, and car parts.
Thus, the forward and backward linkages with the garage sector have attracted a
variety of new metalworking workshops to the cluster.
3. Data
According to our informants, profitability has been gradually declining in almost
all kinds of business in the cluster. The initial purpose of our empirical study of this
cluster was to find out the reason for the declining profitability. In 2004, we
conducted preliminary unstructured interviews with entrepreneurs in March,
September, and December and a formal enterprise survey for three months from
January to March, 2005. The sample consists of 100 garage mechanics, 92 machinists,
and 45 manufacturers, as shown in the first row of Table 2. These sample enterprises
were randomly selected within the respective sectors. The garage sector is
underrepresented in the sample because we are interested in the distributions of
variables in each sector but not in the whole cluster and because the garage mechanics
are more homogeneous in behaviors and characteristics than entrepreneur in the other
sectors.
2
2
For example, the sample of car mechanics relative to their population is much smaller than that of
machinists or manufacturers relative to their population.
7
Table 2 shows the data on characteristics of the sample entrepreneurs by sector.
The footnotes attached to the table list up the products and services of each sector.
The entrepreneurs are about forty years old on average. Most of them were born in the
Ashanti Region, where the cluster is located, and more than 80 percent of them are
Akan tribesmen.
3
All the sample entrepreneurs are males. The difference in the mean
of the years of education is statistically significant between the garages and the rest but
insignificant between the machinists and the manufacturers. Some entrepreneurs went
to school for more than 12 years. They went to advanced courses of polytechnic and
professional schools and received technical trainings. None in the sample went to a
university.
More than 90 percent of the entrepreneurs were former apprentices for 4.6 years
on average in the case of garages and for a little less than three years on average in the
case of metalworking.
4
They learned from their masters how to produce metal
products or how to repair machines. They learned how to operate a business as well,
but it is important to note that their masters taught the self-employment type business
but not the management of a large organization with, say twenty workers or more. It
should also be noted that apprenticeship training in Africa does not teach knowledge of
marketing and bookkeeping unlike apprenticeship in many developed countries.
Table 2 also shows the percentage of the sample entrepreneurs who have
participated in short-term training programs teaching management (including
bookkeeping) and production techniques. These programs have been provided to the
artisans in the cluster mainly but not exclusively by two training institutions in the
3
Though not shown, the number of young entrepreneurs originally from the outside of the Ashanti
Region has been increasing as the cluster has become widely known.
4
The percentage of workers who are apprentices and the duration of apprenticeship shown in Tables 2
and 3, respectively, are consistent with the data analyzed by Frazer (2006).
8
cluster, i.e., ITTU and the National Vocational Training Institute (NVTI). More than
twenty percent of the sample entrepreneurs have received formal technical training.
In the machining sector, many entrepreneurs have sponsors, even though they
make all decisions except for fixed capital investment. Since they run businesses, we
refer to them as entrepreneurs. They and their sponsors split the profits. The common
rule is that the sponsor takes two thirds and the entrepreneur takes the rest. Some of
such sponsors are expatriates. As is shown in the table, about one third of the
entrepreneurs in the machining sector have sponsors living abroad, who are hereafter
referred to as “abroad-based owners.”
Table 3 presents the data on enterprise size in terms of employment, sales
revenues, variable costs, and producer surplus.
5
None of the sample entrepreneurs had
kept financial book systematically.
6
To obtain reasonably accurate data, we checked
the consistency of each respondent’s answers to our questions about different aspects
and by revising estimates of sales and costs, in front of the respondent until estimates
converge to the one that made much sense to both the respondent and us. If the
respondent kept any fragmentary records, we used them as well.
A typical enterprise in the cluster has less than ten workers including several
apprentices. The number of workers is not a good indicator of labor input since
apprentices are considerably heterogeneous in skill level, even though it is commonly
used in the cluster as an indicator of enterprise size. While the machinists have a
smaller number of workers, they have greater revenues, variable costs, and producer
surpluses than the manufacturers and garages. The manufacturers have greater sales
revenues but smaller producer surplus than the garages, because the former have to
5
Variable costs are measured as the sum of costs of materials, labor, subcontracting, and electricity.
Producer surplus is sales revenue minus variable cost.
6
See de Mel, McKenzie, and Woodruff (2009) for the difficulty in obtaining accurate data from micro
enterprises.
9
spend more on materials.
Each growth rates shown toward the bottom of the table is the difference of the
levels in 2004 and 2000 divided by four, which is intended to approximate the annual
growth rate. The garages had large negative growth in producer surplus because of the
declining sales and the soaring variable cost, especially labor costs. For the machinists,
the median growth rate was negative for the variable cost and positive for the producer
surplus, but the mean growth rate of producer surplus was negative. The
manufacturers spent more and sold more in 2004 than in 2004, and their producer
surplus declined. In the cluster, competitors who produce or provide the same
products or services are always produced by the apprenticeship. Since the
entrepreneurs do not know how to find new markets for their products and services, the
increased number of competitors implies smaller sales per enterprise. Masters would
like to prevent graduates, who finished apprentice training, from leaving and starting
their own businesses. To do so, they have to raise salaries to the graduates. The
manufacturers suffered from the soaring price of scrap metal, due to the increased
demand from China and India as well as the increased demand within the cluster.
Thus, profitability has tended to decline in the cluster.
Table 4 offers additional information on material procurement and marketing, as
well as data on capital stock, of the machinists and manufacturers in 2000 and 2004.
The garage sector is not included in this table because material procurement is not an
important activity in this sector, because this sector lacks marketing activities more
completely than the other sectors, and because garages have very little fixed capital.
The first two rows of the table demonstrate how material procurement has become
difficult for the machinists and manufacturers in recent years. The entrepreneurs have
to spend longer time on material procurement. Since decisions must be made quickly
10
at auction, the entrepreneurs are in charge of material procurement. Many
entrepreneurs go to all the way to Accra in search of good and inexpensive material
more than once a month.
The third row of the table presents the percentage of sales to traders, who buy
products in bulk, and to companies outside the cluster, which tend to place relatively
lucrative orders with them. Other customers of the machinists are mostly fitters, and
those of the manufacturers are individuals, whom few manufacturers can characterize.
Some entrepreneurs have the conception of market research and promotion, but few
practice it. The typical way of selling products or services in the cluster is simply to
wait for customers to come to their workshops. Ordinary consumers do not have a
favorable impression of the cluster. It looks like a big junkyard with wreckage of
vehicles abandoned here and there and the ground smeared with engine oil. Neither
street names nor street numbers exist in the cluster. However, only a small number of
workshops have signboards. Moreover, the entrepreneurs seldom give their customers
detailed explanations of how their products work. They behave just as their masters
did decades ago, when there were few competitors. The near constancy of the
percentages of sales to traders and companies in Table 4 is probably a reflection of the
lack of progress in marketing.
The bottom of the table shows the mean of capital stock measured by the
entrepreneur’s assessment of the replacement cost of their equipment. A substantial
increase in capital stock is found in the machining sector. This may have a bearing on
the negative median growth rate of variable cost in this sector.
4. Hypotheses and Empirical Framework
In the previous section, we saw that enterprise growth is stagnant and
11
profitability tends to decline in the cluster. To restore the high profitability, they need
to upgrade product lines, improve production efficiency, and find new markets.
Upgrading product lines, however, is often too difficult without using better equipment
than they have used, and credit constraints facing them often prohibit them from using
better equipment. Thus, it is important to devise more efficient methods of production,
practice economy, and adopt a marketing method suitable to their products or services.
These reforms require managerial knowledge or skill that the entrepreneurs did not
learn from their apprentice training or from their own experiences of running
businesses in the past. How can they obtain such knowledge or skill?
In the literature on industrial clusters in developing countries, a number of case
studies report that a serious decline in profits can induce entrepreneurs in a cluster to
improve their products, production and quality management, marketing methods, and
financial management (e.g., Schmitz and Nadvi 1999; Sonobe and Otsuka 2006). In
garment clusters in China, Vietnam, and Kenya, entrepreneurs responded to crisis by
improving changing marketing channels (Sonobe, Hu, and Otsuka 2002; Nam, Sonobe,
and Otsuka 2009; Akoten and Otsuka 2006). In a surgical instrument cluster in
Pakistan, entrepreneurs took collective actions in order to obtain from abroad the
information necessary to upgrade the quality of their products (Nadvi 1999). In a
machine tool cluster in Taiwan, entrepreneurs reduced production cost for high quality
products by taking full advantage of the already developed division of labor among
enterprises (Sonobe, Kawakami, and Otsuka 2003). In a electric fitting cluster in
China, entrepreneurs achieved a set of improvements in product quality, marketing
method, and production organization (Sonobe, Hu, and Otsuka, 2004). With these
induced improvements in marketing and management, enterprises could accumulate
funds quickly for investments in better equipment, which in turn allowed them to
12
upgrade product lines.
Most of these case studies find that the induced upgrading is led by highly
educated entrepreneurs, even though schools may not teach knowledge or skills
directly useful for the upgrading of management. Probably education helps
entrepreneurs search for useful information. The human capital literature maintains
that the ability to respond to changing opportunities is “one of the major benefits of
education accruing to people personally in a modernizing economy” (Schultz 1975, p.
843). Moreover, educated persons tend to keep records and make plans better than the
uneducated. Such ability is directly useful for business management. Paulson and
Townsend (2004) use years of schooling as a proxy for entrepreneurial talent in their
empirical analysis of credit constraints facing small entrepreneurs in Thailand.
In the context of Suame Magazine, high education means education at
polytechnic and professional schools. These schools do not teach managerial
bookkeeping, marketing, or anything directly useful for business administration. Still,
high education may give entrepreneurs the higher ability to calculate, search for useful
information, and adjust to changing opportunities. Besides, it may well be that highly
educated entrepreneurs are from relatively wealthy families, faced with less severe
credit constraints, and thus able to use greater working and fixed capital. Our data set
does not contain information of working capital itself, but it can be roughly captured
by variable costs. Based on these considerations, we hypothesize as follows:
Hypothesis 1: Capital stock, variable costs, and sales revenues are positively
associated with years of schooling.
It may be more difficult for small entrepreneurs in Sub-Saharan Africa to search
13
for useful managerial and technical information than their East Asian counterparts. In
China, for example, small entrepreneurs could obtain such information by
subcontracting with state-owned enterprises (SOEs) or by hiring managers and
engineers who quit SOEs (Otsuka, Liu, and Murakami 1998; Sonobe, Hu, Otsuka
2006). In Sub-Saharan Africa, such a source of information does not seem to exist
within countries. Some leather-shoe makers in Addis Ababa, Ethiopia, who have
recently succeeded in finding markets in Europe, have frequently visited Italy for many
years in search for new ideas. Such business trips are too expensive for most
entrepreneurs of small enterprises in Sub-Saharan Africa. Thus, training programs
offered in their clusters or nearby cities must be very important opportunities to learn
useful knowledge and skills.
While both managerial and technical knowledge and skills are important for
small entrepreneurs, the first step to cope with declining profits will be managerial
reforms for the reason discussed above. Thus, we expect that managerial training will
have stronger impacts on enterprise performance than technical training at the current
stage of development in Suame Magazine. By the same token, we also expect that
talented entrepreneurs are more willing to participate in managerial training than in
technical training if they find managerial training more useful. Thus, enterprise
performance will be more strongly associated with participation in managerial training
than technical training.
Capital stock, however, may be more closely associated with technical training
participation. Since technical training tends to teach how to make better use of
machines, those entrepreneurs who have relatively large capital stock may have
stronger incentive to participate in technical training, or technical training participants
may have stronger incentive to use machinery after training. Thus, it seems reasonable
14
to postulate the following hypothesis:
Hypothesis 2: Sales revenues and producer surplus are more strongly associated with
participation in managerial training than participation in technical training, whereas
capital stock is not associated with managerial training but with technical training.
These hypotheses are not about causal effects but about associations between
variables. This is because, given our data set, it is impossible to cope with the
endogeneity problem due to the correlation of education and training participation with
unobservable entrepreneurial talents. As to the effects of training, the problem may be
better referred to as selection bias since more talented entrepreneurs may be more
likely to obtain information about invitation to training or more willing to seize the
opportunity of participating in training. Identifying causal effects would require a
randomized experiment or the application of the difference-in-differences (DID)
propensity score matching (PSM) estimator proposed by Heckman, Ichimura, and
Todd (1997, 1998) to non-experimental but panel data containing information on
enterprise performances before and after training. Since such data are unavailable, the
second choice for us is to apply the PSM estimator developed by Rosenbaum and
Rubin (1983) and extended by Heckman, Ichimura, and Todd (1997, 1998). After
running OLS regressions of sales revenues, variable costs, producer surplus on the
training participation dummies, educational backgrounds, and other characteristics of
entrepreneurs, we attempt the PSM estimation to see whether the positive association
between training participation and enterprise performance remains after mitigating the
selection bias.
In our PSM estimation, the effect of a training program is evaluated by
15
comparing the participants’ actual post-program performance and their estimated
counterfactual performance that they would achieve if they did not participate in the
program. More precisely, what is estimated is the average effect of treatment on the
treated (ATT) defined as E(Y
1
–
Y
0
|the entrepreneur has ever participated in a training
program), where Y
1
is the performance of a participant and Y
0
is the counterfactual
performance that the same entrepreneur would have achieved if he had not participated
in any training program. The matching estimator tries to obtain an estimate of this
counterfactual performance by matching each participant with one or more non-
participants who have similar characteristics. Let X denote a set of observed
characteristics of entrepreneurs and their enterprises. As shown by Rosenbaum and
Rubin (1983), a consistent matching estimate of ATT can be obtained by matching a
participant with non-participants whose propensity scores are similar. Propensity
score is defined as the conditional probability of participating in training given X.
The PSM estimator of ATT can be generally expressed as
( ) ( ) ( )
? ?
? ?
?
?
?
?
?
?
? =
1 0
0 1
,
1
I i I j
j j i i
Y X p X p W Y
N
PSM ,
where N is the number of the participants, I
1
and I
0
are respectively the treatment (or
participant) group and the matched control (or matched non-participant) group, p(X) is
the propensity score, and W is a weight determined by the distance between propensity
scores of participant i and the matched non-participants j. This estimator has been
widely applied to non-experimental data from developing economies. For example,
Rosholm, Nielsen, and Dabalen (2007) use this estimator to evaluate the impacts of
technical training programs for workers on labor productivity in Kenya and Zambia.
We estimate the propensity score using a probit model of training participation
with X being covariates. We impose the common support condition (Heckman,
16
Ichimura, and Todd 1998), and drop from the sample the non-participants whose
propensity scores are higher than the maximum or less than the minimum propensity
scores among the participants. The literature has proposed several types of matching
methods, including stratification matching, nearest neighborhood matching,
Epanechnikov (or quadratic) kernel, normal (or Gaussian) kernel, bi-weight kernel, and
local linear matching. We use these matching methods to see if the estimation results
are robust to the choice of matching methods. We obtain the standard errors of PSM
estimates by bootstrapping with 1000 replications, following the lead of Smith and
Todd (2005).
According to Rosenbaum and Rubin (1983), there should be no significant
differences in X between the matched participants and non-participants. The so-called
balancing test is a specification test using this property. We perform two types of
balancing test. The first is the t test of equality in the mean of each covariate between
the two groups. Second, we estimate the same probit model using the samples with
and without matching to obtain two values of pseudo-R squared and compare them.
We also perform a likelihood-ratio test. If matching is successful, the after-matching
probit should have no explanatory power so that the pseudo-R squared should be low
and the estimated coefficients should be closed to zero.
5. Estimation Results
OLS results
Tables 5 to 7 present the OLS estimates of the functions explaining sales
revenue (S), variable cost (C), producer surplus (P), capital stock (K), the proportion of
sales to traders and companies, and employment size. Table 5 reports the results for
the machinists. The number of years of schooling has positive and highly significant
17
coefficients in all the equations, which is consistent with Hypothesis 1. These
coefficients may overstate the effects of education for some reasons, including the ones
that we discussed in the previous section. One such reason is that the education level
of an entrepreneur may be positively correlated with his family’s wealth, which is
likely to influence enterprise sizes in the presence of credit constraints. We hope that
the inclusion of years of father’s schooling mitigates the bias due to this correlation.
Participation in managerial training has positive and significant coefficients in
equations explaining sales revenues, variable costs, producer surplus, and labor
employment, but not the fixed capital equation. By contrast, the coefficients of
technical training participation are not significantly different from zero in any equation.
These results are consistent with Hypothesis 2, except that capital stock is not closely
associated with technical training participation. Capital stock is associated closely
with the abroad-based owner dummy, which is a natural result since the role of such
owners is to finance investment for skilled artisans faced with credit constraints.
According to the estimates, the enterprises with abroad-based owners have high-
proportions of sale to traders and companies. This is probably because they are
equipped better and able to attract repair work orders from large companies outside the
cluster.
Interestingly, apprentice training has a negative and marginally significant
coefficient in the worker equation. This result suggests that the traditional
apprenticeship is becoming less popular among the youth. As shown by Frazer (2006),
apprentice training pays poorly unless a graduate from apprenticeship starts an own
business. In the machining sector, starting business without relying on investment by
sponsors is becoming difficult. Since entrepreneurs have to pay two thirds of profits to
sponsors, it takes long time for entrepreneurs to reap returns to apprenticeship training.
18
It may well be that apprenticeship is becoming less sustainable in this sector. Another
major finding is that enterprises with abroad-based owners have neither larger sales
revenues nor greater producer surplus despite their greater capital stock. This calls for
future investigation into the issue of moral hazard. The estimation results concerning
age and years of operation are somewhat unique. They differ from the results of
preceding studies of enterprise growth in that entrepreneurial age as well as enterprise
age is correlated negatively with enterprise sizes.
Table 6, which is about the manufacturers, shows qualitatively similar results,
even though statistical significance levels tend to be lower than in Table 5, probably
because the sample size is smaller. A major difference is that the coefficient of
technical training participation is insignificant in the capital stock equation in Table 5
whereas that is significant in Table 6, consistent with Hypothesis 2. The positive
coefficient of managerial training participation in the producer surplus equation has a
higher significance level in Table 6 than in Table 5. Overall, the results shown in
Tables 5 and 6 are consistent with Hypotheses 1 and 2.
Table 7, which is about the garages, looks very different from the previous two
tables. In this table, only a few estimated coefficients are significantly different from
zero. Moreover, no indicators of enterprise performance we employed are correlated
with the level of education and participation in managerial training. These results are
inconsistent with Hypotheses 1 and 2 and also with the results of the preceding studies,
such as the one by Mengistae (2006), who finds positive effects of entrepreneurial
human capital on enterprise performance. The contrast between Table 7 and the
preceding two tables suggests that managerial training as well as schooling is not
useful for all types of enterprises but for only some types. Machinists and
manufacturers may be able to improve enterprise performance more easily than
19
garages, if equipped with a little better knowledge of management.
PSM Results
We now turn to the PSM estimation of the ATT of technical training and
managerial training. The results of the OLS regressions indicate clearly that there is no
effect of training on enterprise performance in the garage sector and that the machinists
and manufacturers share qualitatively similar associations between training
participation and enterprise performance. Thus, we will focus on the machinists and
manufacturers, that is, the metalwork sector of the cluster, and not apply the PSM
method to the data of the garage sector.
The PSM estimation begins by estimating probit models that explain
participation in technical and managerial training. Table 8 reports the estimation
results. The probit models include all the characteristics of the entrepreneurs and two
additional dummy variables: one indicating whether the enterprise is located in the part
of the cluster called the “old site”, and the other indicating whether the entrepreneur is
a manufacturer. In both columns, the entrepreneur’s age and years of schooling have
positive and significant effects on participation. Older entrepreneurs are more likely to
have participated in training, probably because they have encountered a greater
number of opportunities to attend training programs. More highly educated
entrepreneurs may be better informed of learning opportunities. While the significance
of the effect of years of schooling on managerial training participation is of the five
percent level, that on technical training participation is of the one percent level. The
difference in significance level may be linked with a possible correlation between
education and wealth. If technical training may teach techniques suitable to well-
equipped workshops, technical training participants may be relatively wealthy
20
entrepreneurs, who tend to be highly educated. We have no idea of why the negative
correlation between the old site dummy and managerial training participation is highly
significant. The pseudo R squared obtained from each estimate of the probit model is
sufficiently high for the matching purpose.
These results of probit estimation indicate that the participants and non-
participants differ substantially in terms of entrepreneur characteristics. Differences
between them are also apparent in Figures 1 and 2, which show the histograms of the
propensity scores calculated from the estimated probit models. Thus, if we simply
compared the average performance of the participants with that of the non-participants,
we would fail to isolate the effects of training participation from the effects of
entrepreneur characteristics. However, if they differed completely, it would be
impossible to estimate the counterfactual performance based on the performance of the
matched non-participants, or matching would be impossible. Thus, the distribution of
propensity score for the participants and that for the non-participants must have a
common range of support, in order for the PSM estimation to be feasible. As shown
clearly in Figures 1 and 2, there exists such a common support.
Using the estimated propensity score, we match the participant (treatment) group
and the non-participant (control) group so that the two groups after matching have
similar characteristics. We applied six matching methods, and for each method, we
performed two types of balancing test. Table A1 presents the results of the two
balancing tests in the case in which the local linear matching method is applied to the
technical training participants and non-participants. Balancing test 1 is a set of t-tests
for each of covariates X. We find no significant difference in any covariate between
the treated and control groups. Balancing test 2 is based on the fitting measures of the
probit model. The pseudo-R squared is very low. According to the result of the
21
likelihood ratio test, the hypothesis that the coefficients are all zero is not rejected.
The p-value for this test is very high. These test results indicate clearly that the local
linear matching is successful. Table A2 shows similar test results for managerial
training in the case of local linear matching. Although not shown in this paper, we
obtained similar results of the balancing tests in the cases of the five other matching
methods.
Table 9 presents the estimated ATT of formal technical training on sales revenue,
variable cost, producer surplus, capital stock, the proportion of sales to traders and
companies, and the number of workers. Consistent with the result of OLS regression
shown in Tables 5 and 6, we find little evidence that participation in technical training
helps entrepreneurs achieve higher performances. Exceptions are positive and
significant ATT of training on capital stock, which is obtained when stratification
matching is applied, and the positive and significant ATT on the number of workers,
which is found when stratification matching and nearest neighborhood matching are
applied. Participation in technical training has a positive and significant ATT on
capital stock probably because technical training tends to teach how to make better use
of machinery. The positive and significant ATT on the number of workers may be
obtained because technical training participants have more to teach apprentices and can
attract a greater number of apprentices than non-participants.
Table 10 presents the estimated ATT of managerial training. Regardless of
matching method, the estimated ATT’s on sales revenues and producer surplus are
positive and highly significant, whereas the estimated ATT on capital stock is
insignificant. These results, together with the results shown in Table 9, are highly
consistent with Hypothesis 2. Among the different estimates based on the different
matching methods, the smallest estimate of the ATT on the logarithm of producer
22
surplus is 0.932. That is, the smallest estimate of the ATT on producer surplus is
exp(0.932) or 2.5. Thus, compared with the counterfactual producer surplus that the
treated group (i.e., managerial training participants) would have earned if they had not
participated in any managerial training, their actual producer surplus is 2.5 times as
large. These results of the PSM estimation lend support to our view that managerial
training is useful in the metalwork sector. Yet it is important to note that these results
are not enough to establish causal effects since the selection bias arising from the
correlation between unobservable factors and training participation may remain.
6. Conclusions
Industrial clusters have recently attracted much attention from development
economists because clusters are said to increase productivity and enlarge growth
opportunities of the industry and because they are ubiquitous even in Sub-Saharan
Africa. However, enterprises in this region remain extremely small in size. Many of
them suffer from declining profits. This paper has asked if a reason for the stagnancy
of enterprise development in Africa may be found in inadequate managerial skills of
entrepreneurs themselves. Our analysis suggests that managerial skills of small
entrepreneurs in a large cluster in Ghana are indeed poor, and that managerial training
is helpful for some types of small entrepreneurs.
The findings of this paper, however, raise several questions. For example, we
should explore why managerial training works in some industries but not in others.
But more fundamentally, we need tight evidence. In order to obtain tight evidence, it
is desirable to carry out randomized experiments in which managerial training is
provided to randomly selected entrepreneurs. Experiments and post experiment
surveys may be desired so that we can investigate the issues like which aspect of
23
managerial training is more useful and how long it takes until the effects of training
make themselves felt. We may be able to investigate the extent of information
spillovers within industrial clusters. The results of our analysis in this paper warrant
such experiments.
24
References
Adeya, C. N. 2003. “Sources of Training in African Clusters and Awareness of ICTs:
A Study of Kenya and Ghana,” Discussion Paper Series, No.2003-6, United
Nations University, Institute for New Technologies (INTECH).
Akoten, J.E., and Otsuka, K. 2006. “From Tailors to Mini-manufacturers: The Role of
Traders in the Transformation of Garment Enterprises in Nairobi,” Journal of
African Economies, forthcoming.
Bigsten, A., P. Collier, S. Dercon, B. Gauthier, J. W. Gunning, A. Isaksson, A. Oduro,
R. Oostendorp, C. Pattillo, M. Söderbom, M. Sylvain, F. Teal, and A. Zeufack.
1999. “Investment in Africa’s Manufacturing Sector: A Four Country Panel Data
Analysis.” Oxford Bulletin of Economics and Statistics 61(4):489–512.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J. W. Gunning, A.
Oduro, R. Oostendorp, C. Pattillo, M. Söderbom, F. Teal, and A. Zeufack. 2003.
“Credit Constraints in Manufacturing Enterprises in Africa.” Journal of African
Economies 12(1): 104–25.
Bigsten, A., P. Collier, S. Dercon, M. Fafchamps, B. Gauthier, J. W. Gunning, A.
Oduro, R. Oostendorp, C. Pattillo, M. Söderbom, F. Teal, and A. Zeufack. 2004.
“Do African Manufacturing Firms Learn from Exporting?” Journal of
Development Studies 40(3): 115–71.
Bigsten, A., and Söderbom, M. 2006. “What Have We Learned From a Decade of
Manufacturing Enterprise Survey in Africa?” The World Bank Research Observer,
21 (2), 241-65.
Bigsten, Arne, and Mulu Gebreeyesus. 2007. “The Small, the Young, and the
Productive: Determinants of Manufacturing Firm Growth in Ethiopia.” Economic
Development and Cultural Change 55 (4): 813-840.
25
Collier, P., and Gunning, J.W. 1999. “Explaining African Economic Performance,”
Journal of Economic Literature, 37, 64-111.
Davidson, R., and MacKinnon, J.G. 1993. Estimation and Inference in Econometrics.
Oxford: Oxford University Press.
de Mel, Suresh, McKenzie, David, and Woodruff, Christopher 2009. “Measuring
Microenterprise Profits: Must We Ask How the Sausage is Made?” Journal of
Development Economics 88 (1), 19-31.
Fafchamps, M. 2004. Market Institutions and Sub-Saharan Africa: Theory and
Evidence. Cambridge, MA: MIT Press.
Fafchamps, M. and Söderbom, M. (2006). Wages and labor management in African
manufacturing. Journal of Human Resources, 41(2), 346-379.
Frazer, Garth. 2005. “Which Firms Die? A Look at Manufacturing Firm Exit in
Ghana.” Economic Development and Cultural Change 53(3):585–617.
Frazer, Garth. 2006. “Learning the Master's Trade: Apprenticeship and Human Capital
in Ghana.” Journal of Development Economics 81 (2): 259-98.
Goedhuys, M. and Sleuwaegen, L. (2000). “Entrepreneurship and Growth of
Entrepreneurial Firms in Côte d’Ivoire.” Journal of Development Studies: 36(3),
122-145.
Gunning, Jan Willem, and Mengistae, Taye. 2001 “Determinants of African
Manufacturing Investment: the Microeconomic Evidence.” Journal of African
Economies 10 (1): 48-80.
Heckman, James J., Hidehiko Ichimura, and Petra Todd, 1997. “Matching as an
Econometric Evaluation Estimator: Evidence from Evaluating a Job Training
Programme,” Review of Economic Studies 64: 605-654.
26
Heckman, James J., Hidehiko Ichimura, and Petra Todd, 1998. “Matching as an
Econometric Evaluation Estimator,” Review of Economic Studies 65: 261-294.
Institute of African Studies. 1992. The City of Kumasi, Handbook – Past, Present and
Future. The Institute of African Studies, University of Ghana in collaboration with
the Kumasi Metropolitan Assembly.
Kinyanjui, N. (2007). “Tha Kamkunji metalwork cluster in Kenya.” In D. Z. Zeng
(Ed.), Knowledge, Technology, and Cluster-based Growth in Africa, WBI
Development Studies, Washington, DC: The World Bank.
McCormick, Dorothy. 1999 “African Enterprise Clusters and Industrialization: Theory
and Reality,” World Development 27(9): 1531-1551.
Mengistae, T. 2006. “Competition and entrepreneurs’ human capital in small business
longetivity and growth.” Journal of Development Studies 42 (5), 812-836.
Nam, V. H, Sonobe, T, and Otsuka, K. 2009. “An Inquiry into the Development
Process of Village Industries: The Case of a Knitwear Cluster in Northern
Vietnam,” Journal of Development Studies, forthcoming.
Obeng, G.Y. 2002. “Kumasi Suame Magazine: A Background Paper,” Technology
Consultancy Centre, Kwame Nkrumah University of Science & Technology,
Kumasi, Ghana.
Otsuka, K., Liu, D., and Murakami, N. 1998. Industrial Reform in China: Past
Performance and Future Prospects. Oxford: Clarendon Press.
Paulson, Anna. L., and Townsend, Robert 2004. “Entrepreneurship and Financial
Constraints in Thailand.” Journal of Corporate Finance 10 (2): 229-262.
Ramachandran, V. and Shah, M. K. 1999. “Minority Entrepreneurs and Firm
Performance in Sub-Saharan Africa,” Journal of Development Studies 36 (2): 71-
87.
27
Rosenbaum, P. R. and Rubin, D. B. 1983. “The Central Role of the Propensity Score
in Observational Studies for Causal Effects,” Biometrika 70 (1): 41-55.
Rosholm, Michael; Nielsen, Helena, Skyt; and Dabalen, Andrew 2007. “Evaluation of
Training in African Enterprises.” Journal of Development Economics 84 (1): 310-
329.
Schmitz, H., and Nadvi, K. 1999. “Clustering and Industrialization,” World
Development, 27, 1503-1514.
Schultz, T.W. 1975. “The Value of the Ability to Deal with Disequilibria,” Journal of
Economic Literature 13 (3), 827-46.
Sluwaegen, L. and Goedhuys, M. 2002. “Growth of firms in developing countries:
Evidence form Côte d’Ivoire.” Journal of Development Economics, 68 (1): 117-
135.
Smith, J.A. and Todd, P. E. 2005. “Does Matching Overcome LaLonde’s Critique of
Nonexperimental Estimators? Journal of Econometrics 125 (1-2): 305-353.
Söderbom, M. and Teal, F. 2004. “Size and efficiency in African manufacturing firms:
Evidence from firm-level panel data.” Journal of Development Economics 73 (2):
369-394
Sonobe, T. and Otsuka, K. 2006. Cluster-Based Industrial Development: An East
Asian Model. Basingstoke: Palgrave Macmillan.
Sonobe, T., Akoten, J.E., and Otsuka, K. 2006. “The Development of the Footwear
Industry in Ethiopia: How Different Is It from the East Asian Experience.” FASID
Discussion paper 2006-09-003.
Sonobe, T., Hu, D., and Otsuka, K. 2004. “Process of Cluster Formation in China: A
Case Study of a Garment Town,” Journal of Development Studies, 39 (1), 118-39.
Sonobe, T., Hu, D., and Otsuka, K. 2004. “From Inferior to Superior Products: An
28
Inquiry into the Wenzhou Model of Industrial Development in China,” Journal of
Comparative Economics, 32 (3), 542-63.
Sonobe, T., Hu, D., and Otsuka, K. 2006. “Industrial Development in the Inland
Region of China: A Case Study of the Motorcycle Industry,” Journal of
Comparative Economics, 34 (4), 818-38.
Sonobe, T., Kawakami, M., and Otsuka, K. 2003. “Changing Roles of Innovation and
Imitation in Industrial Development: The Case of the Machine Tool Industry in
Taiwan,” Economic Development and Cultural Change, 52 (1), 103-28.
Van Biesebroeck, J. 2005a. “Exporting Raises Productivity in Sub-Saharan African
Manufacturing Firms.” Journal of International Economics 67(2):373–91.
Van Biesebroeck, J. 2005b. “Firm Size Matters: Growth and Productivity Growth in
African Manufacturing.” Economic Development and Cultural Change 53(3):545–
83.
29
Table 1. Estimates of the Enterprise Population by Sector in the Suame Magazine
Cluster in 2000, 2002, and 2003
a
Garages Metalworking
enterprises
Others
b
Total
2000 4,958 807 2,204 7,969
2002 6,222 990 2,618 9,830
2003 7,847 1139 2,844 11,830
Notes:
a. These estimates are taken from the data base of the Suame branch of the Ghana National
Association of Garages (GNAG). The estimates do not include ancillary trades such as
restaurants and telecommunication shops.
b. “Others” include car body builders, pot makers, sign writers, and some types of welders.
30
Table 2. Characteristics of the Entrepreneurs by Sector
Garages
a
Machinists
b
Manufacturers
c
No. sample enterprises 100 92 45
Age 40.3 40.3 41.7
% from Ashanti 77.0 84.9 80.0
Years of schooling 9.49 11.41 10.53
Apprentice training 99.0 89.1 91.1
Years of apprenticeship 5.61 3.05 4.16
% technical training 24.0 28.3 24.4
% managerial training 4.0 16.3 11.1
% owner abroad 1.0 32.6 15.6
Years of operation 16.4 12.4 14.2
Notes:
a. Garage mechanics include fitters, vehicle body strainghteners, sprayers, and auto
electricians. Fitters diagnose trouble in engine and other auto parts and assign repair work
to specialized mechanics.
b. Machinists are defined here as machining enterprises which do not specialize in engine
block re-boring. They operate lathes, milling machines, screw thread cutters, gear-cutting
machines, and other machine tools, to provide customers with machining services such as
shaft lathing, long shaft turning, gear cutting, and screw thread cutting. They also produce
bolts and nuts, wheel bolts, center bolts, bushings and other fabricated metal parts.
c. Manufacturers produce coffers, carbide welding pots, food processing machinery and
equipment, such as palm oil press, cassava press, rice huller, rice thresher, flour mixing
machine, and other items including cooking stoves for households and large ovens for
industrial use.
31
Table 3. Enterprise Size in 2004 and Growth from 2000 to 2004 by Sector
Garages Machinists Manufacturers
Mean Median Mean Median Mean Median
No of workers 10.2 7.0 5.2 4.5 6.0 6.0
% apprentices 75.9 80.0 61.4 66.7 67.3 75.0
Sales revenue (1,000USD) 14.1 8.0 44.6 30.6 22.6 8.8
Variable cost (1,000USD) 6.0 2.3 20.1 11.7 15.5 8.3
Producer surplus (1,000USD) 8.1 4.7 24.5 16.3 7.1 2.3
Growth in sales revenue (%) -5.0 -5.7 -0.4 -0.9 1.9 0.6
Growth in variable cost (%) 27.1 26.0 0.5 -2.1 5.9 4.8
Growth in producer surplus(%) -13.4 -12.3 -1.0 0.8 -2.6 -1.6
32
Table 4. Material Procurement, Marketing Channels, Capital Stock, 2000, 2004
Machinists Manufacturers
2000 2004 2000 2004
% work time of
entrepreneur allocated
to material procurement
11.3 27.1 12.0 24.1
Material cost-sales
revenue ratio
0.16 0.20 0.32 0.40
% sales to traders and
companies
31.9 35.3 34.6 34.8
Capital stock (USD)
7,364 10,091 3,855 4,026
33
Table 5. OLS Estimates for Machinists, 2004
lnS lnC lnP lnK
Proportion of
sale to
traders and
companies
Workers
Age
-0.018*
(0.010)
-0.030**
(0.013)
-0.011
(0.010)
0.012
(0.008)
-0.006***
(0.002)
-0.023
(0.026)
From Ashanti
0.405
(0.289)
0.192
(0.361)
0.493*
(0.257)
0.359
(0.219)
0.115*
(0.064)
0.500
(0.708)
Years of schooling
0.119***
(0.034)
0.070*
(0.042)
0.150***
(0.030)
0.121***
(0.026)
0.025***
(0.007)
0.198**
(0.082)
Apprentice training
0.399
(0.358)
0.412
(0.447)
0.473
(0.318)
-0.179
(0.272)
0.078
(0.080)
-1.465*
(0.876)
Technical training
-0.074
(0.247)
0.080
(0.308)
-0.157
(0.220)
0.0753
(0.188)
0.077
(0.055)
-0.408
(0.605)
Managerial training
0.566*
(0.301)
0.784**
(0.376)
0.465*
(0.268)
0.268
(0.228)
0.106
(0.067)
2.136***
(0.737)
Abroad-based
owner
0.003
(0.240)
-0.205
(0.299)
0.087
(0.213)
0.521***
(0.182)
0.237***
(0.053)
0.123
(0.586)
Years of operation
-0.022*
(0.011)
-0.016
(0.013)
-0.017*
(0.009)
-0.027**
(0.008)
0.001
(0.002)
0.001
(0.026)
Years of father’s
schooling
0.020
(0.018)
0.002
(0.023)
0.041**
(0.017)
0.020
(0.014)
0.002
(0.004)
-0.0003
(0.045)
Intercept
9.067***
(0.760)
9.339**
(0.948)
7.539***
(0.676)
6.784***
(0.576)
0.002
(0.169)
4.074**
(1.858)
Adj. R-squared 0.183 0.090 0.290 0.399 0.275 0.116
Notes:
Number of observations is 92. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
34
Table 6. OLS Estimates for Manufacturers, 2004
lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Age
-0.001
(0.019)
-0.002
(0.020)
-0.002
(0.024)
-0.033
(0.028)
-0.009*
(0.005)
-0.011
(0.052)
From Ashanti
0.363
(0.440)
0.346
(0.463)
0.520
(0.571)
0.376
(0.647)
0.152
(0.124)
1.559
(1.206)
Years of schooling
0.147*
(0.079)
0.159*
(0.083)
0.160
(0.102)
-0.003
(0.118)
0.036
(0.022)
-0.139
(0.216)
Apprentice training
0.752
(1.007)
1.023
(1.059)
0.980
(1.305)
-1.413
(1.457)
0.393
(0.284)
-2.191
(2.760)
Technical training
0.048
(0.551)
-0.116
(0.580)
0.041
(0.717)
1.697**
(0.814)
-0.240
(0.156)
3.189**
(1.510)
Managerial training
1.276
(0.817)
1.137
(0.859)
2.105**
(1.058)
0.133
(1.218)
0.445*
(0.231)
1.243
(2.239)
Abroad-based
owner
0.364
(0.497)
0.696
(0.523)
0.012
(0.647)
2.447***
(0.733)
0.141
(0.140)
2.612*
(1.363)
Years of operation
0.024
(0.019)
0.023
(0.020)
0.017
(0.025)
-0.004
(0.028)
0.004
(0.022)
0.022
(0.052)
Years of father’s
schooling
-0.008
(0.030)
-0.018
(0.032)
0.011
(0.039)
-0.075
(0.045)
0.002
(0.008)
0.077
(0.083)
Intercept
6.298**
(1.744)
5.819**
(1.834)
4.302*
(2.257)
8.934**
(2.602)
-0.209
(0.492)
6.112
(4.779)
Adj. R-squared 0.219 0.159
0.199
0.354 0.111
0.263
Notes:
Number of observations is 45. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
35
Table 7. OLS Estimates for Garages, 2004
lnS lnC lnP Workers
Age
-0.011
(0.017)
-0.029
(0.020)
-0.006
(0.020)
0.116*
(0.055)
From Ashanti
0.177
(0.311)
-0.023
(0.363)
0.307
(0.351)
-0.662
(0.996)
Years of schooling
-0.022
(0.045)
0.007
(0.052)
-0.045
(0.051)
-0.055
(0.144)
Apprentice training
-0.758
(1.236)
1.090
(1.446)
-1.630
(1.399)
1.872
(3.966)
Technical training
-0.238
(0.297)
0.001
(0.348)
-0.468
(0.336)
1.777*
(0.953)
Managerial training
-0.427
(0.634)
-0.508
(0.742)
-0.126
(0.718)
-3.760*
(2.034)
Years of operation
0.013
(0.015)
0.021
(0.017)
0.010
(0.017)
-0.028
(0.048)
Years of father’s
schooling
0.010
(0.023)
0.012
(0.026)
0.001
(0.026)
0.151**
(0.072)
Intercept
10.13**
(1.34)
7.53**
(1.57)
10.36**
(1.52)
0.405
(4.304)
Adj. R-squared -0.041 -0.033
-0.032
0.043
Notes
Number of observations is 45. Robust standard errors are in parentheses. ***, **, * denote
the statistical significance at the 1%, 5%, and 10% levels, respectively.
36
Table 8. Estimated Probit Models of Training Participation
Technical training
Managerial training
Age
0.032**
(0.013)
0.046**
(0.018)
From Ashanti
0.416
(0.376)
0.504
(0.485)
Apprentice training
-0.118
(0.438)
-0.696
(0.489)
Years of schooling
0.125***
(0.042)
0.111**
(0.052)
Years of operation
0.001
(0.014)
0.018
(0.019)
Abroad-based owner
-0.091
(0.306)
-0.401
(0.443)
Years of father’s schooling
0.009
(0.022)
-0.048
(0.030)
Old site
0.067
(0.287)
-1.187***
(0.391)
Manufacturing sector
-0.095
(0.279)
-0.186
(0.367)
Intercept
-3.695***
(1.033)
-3.380**
(1.307)
LR chi2 (4) 23.73***
32.12***
Log Likelihood -68.05
-40.89
Pseudo R-squared 0.149
0.282
Notes
The number of observations is 137. Marginal effects for dummy variables are calculated as
the effect of a change from 0 to 1. Standard errors are in parentheses. ***, **, and * indicate
the 1, 5, 10 percent levels of statistical significance, respectively.
37
Table 9. PSM estimate of ATT of Technical Training
Matching methods lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Stratification
0.377
0.401
0.398
0.700**
0.048
1.976***
(0.358) (0.337) (0.424) (0.310) (0.059) (0.694)
Nearest Neighborhood
0.206
0.439
0.156
0.626
0.095
2.135**
(0.436) (0.457) (0.519) (0.491) (0.075) (0.996)
Epanechnikov Kernel
0.111
0.285
0.034
0.427
0.050
1.526
(0.361) (0.345) (0.407) (0.322) (0.054) (0.838)
Normal Kernel
0.182
0.322
0.140
0.491*
0.045
1.660*
(0.320) (0.332) (0.408) (0.290) (0.057) (0.879)
Biweight Kernel
0.080
0.274
-0.016
0.447
0.056
1.496*
(0.352) (0.350) (0.407) (0.327) (0.055) (0.839)
Local Linear
-8.81e-05
0.232
-0.060
0.569
0.093
1.902*
(0.451) (0.428) (0.509) (0.441) (0.069) (0.986)
Note. ***, **, * denote the statistical significance at the 1%, 5%, and 10% levels, respectively. In
parentheses are standard errors obtained by bootstraps with 1000 replications and with re-sampling.
The common support condition was imposed.
38
Table 10. PSM estimate of the ATT of Managerial Training
Matching methods lnS lnC lnP lnK
Proportion
of sale to
traders and
companies
Workers
Stratification
0.832**
0.806***
1.143***
0.627
0.099
2.004*
(0.368) (0.350) (0.414) (0.422) (0.076) (1.102)
Nearest Neighborhood
1.469***
1.652***
1.679***
0.751
0.073
2.750**
(0.476) (0.529) (0.560) (0.597) (0.094) (1.299)
Epanechnikov Kernel
0.792**
0.703*
1.142***
0.749
0.089
2.038*
(0.385) (0.425) (0.426) (0.486) (0.076) (1.126)
Normal Kernel
0.778**
1.021***
0.932**
0.501
0.140*
2.047**
(0.327) (0.393) (0.385) (0.395) (0.069) (1.037)
Biweight Kernel
0.818**
0.740
1.161**
0.735
0.081
2.064*
(0.409) (0.434) (0.413) (0.472) (0.080) (1.125)
Local Linear
0.954**
1.211**
1.110*
0.603
0.084
2.225*
(0.404) (0.474) (0.460) (0.484) (0.083) (1.242)
Note. ***, **, * denote the statistical significance at the 1%, 5%, and 10% levels, respectively. In
parentheses are standard errors obtained by bootstraps with 1000 replications and with re-sampling.
The common support condition was imposed.
39
Table A-1. Balancing Test for Impact Evaluation of Technical Training using Local
Linear Matching
Balancing test 1
Mean
t-test
Variable Sample Treated Control % bias
% reduction
|bias|
t p-value
Age
Unmatched
Matched
41.916
45.622
37.783
46.081
37.8
-4.2
88.9
4.62
-0.19
0.000
0.852
From Ashanti
Unmatched
Matched
0.774
0.865
0.788
0.892
-3.3
-6.5
-95.6
-0.41
-0.35
0.681
0.727
Apprentice
training
Unmatched
Matched
0.847
0.811
0.965
0.919
-41.2
-37.8
8.3
-6.24
-1.36
0.000
0.178
Years of
schooling
Unmatched
Matched
11.526
12.892
9.431
12.838
58.4
1.5
97.4
7.20
0.06
0.000
0.949
Years of
operation
Unmatched
Matched
13.709
14.432
13.624
15.730
0.7
-11.4
-1421.5
0.09
-0.45
0.929
0.654
Abroad-based
owner
Unmatched
Matched
0.153
0.297
0.121
0.270
9.2
7.9
15.0
1.17
0.25
0.243
0.800
Years of father’s
schooling
Unmatched
Matched
5.537
6.514
5.233
5.189
5.1
22.1
-336.3
0.63
0.95
0.529
0.343
Old site
Unmatched
Matched
0.674
0.730
0.676
0.649
-0.6
17.3
2897.8
-0.07
0.75
0.944
0.458
Manufacturing
sector
Unmatched
Matched
0.268
0.297
0.229
0.270
9.1
6.2
31.1
1.13
0.25
0.258
0.800
Balancing test 2
Pseudo
R-squared
LR chi2
p-value
Matched 0.041
4.25
0.894
40
Table A-2. Balancing Test for Impact Evaluation of Managerial Training using Local
Linear Matching
Balancing test 1
Mean
t-test
Variable Sample Treated Control % bias
% reduction
|bias|
t p-value
Age
Unmatched
Matched
45.986
48.250
38.006
49.900
75.9
-15.7
79.3
6.00
-0.54
0.000
0.591
From Ashanti
Unmatched
Matched
0.739
0.800
0.789
0.800
-11.5
0.0
100.0
-0.97
0.00
0.331
1.000
Apprentice
training
Unmatched
Matched
0.822
0.750
0.951
0.850
-41.3
-32.0
22.5
-4.52
-0.78
0.000
0.442
Years of
schooling
Unmatched
Matched
12.247
12.850
9.661
12.050
68.4
21.2
69.1
5.89
0.72
0.000
0.476
Years of
operation
Unmatched
Matched
15.644
16.250
13.456
20.000
20.2
-34.6
-71.4
1.55
-0.88
0.121
0.382
Abroad-based
owner
Unmatched
Matched
0.164
0.250
0.124
0.350
11.4
-28.4
-149.2
0.99
-0.68
0.325
0.503
Years of Father’s
schooling
Unmatched
Matched
4.096
4.350
5.401
4.100
-21.6
4.1
80.9
-1.81
0.13
0.070
0.894
Old site
Unmatched
Matched
0.534
0.450
0.688
0.450
-31.8
0.0
100.0
-2.70
0.00
0.007
1.000
Manufacturing
sector
Unmatched
Matched
0.274
0.250
0.234
0.150
9.1
22.9
-151.2
0.77
0.78
0.444
0.442
Balancing test 2
Pseudo
R-squared
LR chi2
p-value
Matched 0.062
3.41
0.946
41
0
1
2
3
4
0 .5 1 0 .5 1
0 1
D
e
n
s
i
t
y
Pr(dtechtrain)
Graphs by dtechtrain
Figure 1. Histogram for estimated propensity score for technical training
0
5
1
0
1
5
0 .2 .4 .6 .8 0 .2 .4 .6 .8
0 1
D
e
n
s
i
t
y
Pr(dnontechtrain)
Graphs by dnontechtrain
Figure 2. Histogram for estimated propensity score for managerial training
doc_873751266.pdf