Description
This study reports the results of a longitudinal ®eld experiment designed to examine the impact of group discussion
when soliciting user requirements of an accounting information system. A total of 346 subjects were randomly assigned
to individual and group treatment conditions. Pre- and post-experiment surveys, administered 12 months apart,
assessed cognitive, motivational, and desired outcome factors. Objective measures of cognition were obtained during a
®ve-week user-input period and actual performance data were collected over 18 months. Results indicate signi®cantly
greater pre-experiment to post-experiment gains in psychological and behavioral variables in the group, as compared to
the individual, condition. Structural equation modeling highlights the key mediating roles of cognition and motivation
on desired user outcomes in a group setting.
Soliciting user-input during the development of an
accounting information system: investigating the ecacy
of group discussion
James E. Hunton
a,
*, Dana Gibson
b
a
School of Accountancy, College of Business Administration, University of South Florida, 4202 East Fowler Avenue, BSN 3403, Tampa,
FL 33620-5500, USA
b
Department of Accounting, College of Business Administration, Texas Woman's University, Denton, TX, USA
Abstract
This study reports the results of a longitudinal ®eld experiment designed to examine the impact of group discussion
when soliciting user requirements of an accounting information system. A total of 346 subjects were randomly assigned
to individual and group treatment conditions. Pre- and post-experiment surveys, administered 12 months apart,
assessed cognitive, motivational, and desired outcome factors. Objective measures of cognition were obtained during a
®ve-week user-input period and actual performance data were collected over 18 months. Results indicate signi®cantly
greater pre-experiment to post-experiment gains in psychological and behavioral variables in the group, as compared to
the individual, condition. Structural equation modeling highlights the key mediating roles of cognition and motivation
on desired user outcomes in a group setting. #1999 Elsevier Science Ltd. All rights reserved.
1. Introduction
Accounting information system design decisions
are not made in isolation; rather, system devel-
opers must consider a host of relevant human
factors, such as the concerns, opinions, and sug-
gestions of a diverse constituency of system users.
One way to include various a?ected users in the
design process is to provide them with opportu-
nities to express, compare, and contrast their
design preferences in group settings during the
requirements de®nition phase of a systems devel-
opment project (Lind, 1995; Lind & Tyler, 1988;
Vessey & Conger, 1993). Many researchers indi-
cate that discussion of this nature is a critical
group participation activity leading to high quality
decisions, primarily due to the collective sharing of
intellectual resources (e.g. Maier, 1963; McGrath,
1991; Osborn, 1957; Stasser & Titus, 1985; Vroom
& Jago, 1988).
The use of group participation is expanding into
all facets of the accounting profession, such as tax
planning, auditing, consulting, systems designing,
and budgeting (Brecht & Martin, 1996; Sutton &
Hayne, 1997). While the trend toward collective
participation is rising across the professional
0361-3682/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PI I : S0361- 3682( 99) 00023- 9
Accounting, Organizations and Society 24 (1999) 597±618
www.elsevier.com/locate/aos
* Corresponding author. Tel.: +1-813-974-6523; fax: +1-
813-974-6528.
E-mail address: [email protected] (J.E. Hunton)
community, accounting research in the area of
participation continues to focus on individuals.
Sutton and Hayne urge accounting researchers to
investigate various forms of group participation in
a wide variety of accounting-related situations.
Specifying design requirements of accounting
information systems is an area where accountants
can add considerable value to group processes
because of their unique, multifaceted role as users,
evaluators, developers, and assurers. Accordingly,
the current study focuses on the e?ectiveness of
facilitating small group discussion in an important
accounting scenario-specifying user requirements
of an accounting information system.
This study is unique along three important
dimensions. First, the relative ecacy of group
versus individual participation strategies is com-
pared. From a theoretical perspective, the current
study enhances our understanding of group pro-
cesses by examining simultaneous relationships
among cognitive, motivational, and outcome con-
structs using a structural equations model (SEM).
We ®nd that the strength of the relationships
among these constructs is signi®cantly greater in
the group, as compared to the individual, condi-
tion. The SEM also indicates that cognition and
motivation are key mediators between user-input
and desired outcomes. To our knowledge, this is
the only longitudinal ®eld experiment in account-
ing and reference disciplines (e.g. psychology and
organizational behavior) with a sucient sample
size to compare individual and group processes
using a SEM. Second, this experiment examines
real actors performing real tasks in a contextually
rich, natural environment. When conducting
longitudinal ®eld experimentation of this nature, it
is dicult to precisely control all internal threats
to validity. However, ®eld experiments are rare in
accounting and reference disciplines, and if inter-
nal threats are controlled with a reasonable degree
of assurance, research results can contribute valu-
able knowledge and understanding to the external
and ecological validity of extant theory (Libby &
Luft, 1993). Finally, most user participation
research relies on self-reports of cognitive percep-
tions and behavioral intentions because it is dicult
to obtain objective data of this nature, particularly
outside of laboratory settings. In the current
study, we obtained and analyzed objective mea-
sures of cognition and user performance. Pairing
objective data with self-reported attitudes and
perceptions provides a solid foundation for asses-
sing convergent validity of latent construct indi-
cators within a research framework.
2. Theory and hypotheses
Extant participation literature in accounting
investigates, for the most part, the impact of indi-
vidual subordinate involvement in in¯uencing and
determining budgets (Shields & Shields, 1998).
The current study expands the scope of accounting
participation research by examining the in¯uence
of user participation, both individual and group,
in determining the speci®cations of an accounting
information system. Many information systems
(IS) researchers believe that user participation,
particularly during the requirements de®nition
phase of a systems development project, is critical
to overall system success (Cushing, 1990; Doll &
Torkzadeh, 1991; Vessey & Conger, 1993). A
variety of IS studies have examined the impact of
user participation on desired attitudes and behavior;
however, study ®ndings are generally inconclusive
(Bailey & Pearson, 1983; Baroudi, Olsen & Ives,
1986; Cavaye, 1995; DeLone & McLean, 1992;
Ives & Olson, 1984). Despite past equivocal
results, researchers continue to believe that the
nature, timing, and extent of user participation are
key factors leading to IS success (Hunton & Price,
1994).
Barki and Hartwick (1994) have examined,
validated and categorized the participation con-
struct for IS users. The following three types of
user participation have been identi®ed: overall
responsibility, user-IS relationship, and hands-on
activity. The former category refers to user activ-
ities and assignments aimed at providing overall
accountability for the systems development pro-
ject, such as leading a project group or selecting
hardware and software. User-IS relationship
re¯ects user activities that are focused on in¯uen-
cing the IS sta?, for example approving IS sta?
work plans or evaluating the work performed by
the IS sta?. Hands-on activity refers to speci®c
598 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
design and implementation tasks performed by
system users. For instance, helping to de®ne
screen layouts and report formats or creating a
user procedure manual are considered to be
hands-on activity. In the current study, accounting
information system users are asked to provide
input to system developers regarding various
aspects of the system during the requirements
de®nition phase of a development project. Hence,
user participation in this study is interpreted as
hands-on activity performed by system users.
Barki and Hartwick (1994) indicate that metho-
dological weaknesses are partially responsible for
equivocal results of past participation studies.
Leana, Locke and Schweiger (1990) and Dipboye
(1990) agree and they provide two important sug-
gestions. First, Leana et al. claim that participa-
tion studies using a single questionnaire at a single
point in time can arti®cially distort the magnitude
of research ®ndings. They recommend that parti-
cipative decision-making researchers should
engage in longitudinal studies in order to improve
measurement reliability. Second, Dipboye notes
that laboratory experiments are used in the
majority of participation studies. He cautions that
methodological artifacts inherent in laboratory
settings contribute to mixed ®ndings across stu-
dies. In order to avoid this problem, he urges
researchers to design and execute studies in ®eld
environments, particularly ®eld experiments. In an
e?ort to address these suggested areas of
improvement, the current study examines the e-
cacy of group versus individual user participation
by conducting longitudinal research in a ®eld set-
ting.
2.1. Group participation
Judgment and decision-making research in
accounting focuses primarily on individuals (Sut-
ton & Hayne, 1997). However, changing business
practices have emphasized the critical role of
groups in a variety of accounting contexts (Ham-
mer & Champy, 1993; Young & Lewis, 1995).
Given the rising prominence and importance of
groups in the accounting profession, it is impor-
tant to investigate the e?ectiveness of various
group participation strategies.
Individual decision-making takes place in three
general phases: (1) gathering relevant information,
(2) generating and evaluating alternatives, and (3)
selecting a course of action (Simon, 1957, 1960).
Analogously, the decision-making phases of indi-
viduals can be applied to groups (Baron, Kerr &
Miller, 1992; Blumberg, 1994). Stasser and Titus
(1985) indicate that group discussion during the
®rst phase of decision-making is a critical partici-
pative activity leading to high quality decisions.
The current study focuses on the ecacy of group
discussion during the requirements de®nition (i.e.
information gathering) phase of a systems devel-
opment project.
When individuals form into a group, each
member typically holds a di?erential set of
knowledge relevant to the problem at hand. A
given member's knowledge set is composed of
multiple items of information. Typically, informa-
tion asymmetry exists among group members
prior to group discussion. The number of group
members who a priori possess a given information
item is referred to as the commonality of the item.
Information items that are known by all group
members are referred to as commonly-held infor-
mation. Items that are known by more than one
but less than all group members are known as
partially-held information. Uniquely-held infor-
mation re¯ects items known by only one group
member.
The ultimate goal of group discussion is for
group members to share their partially-held and
uniquely-held information with other members
(Moreland, Argote & Krishnan, 1996; Stasser &
Titus, 1985, 1987). However, Stasser and Titus
posit that commonly-held information has a
higher probability of being discussed than par-
tially-held information, which has a higher prob-
ability of being discussed than uniquely-held
information. Two important factors that mitigate
this phenomenon, called the common information
sampling bias, are familiarity among group mem-
bers and group member status.
Group members who have gained familiarity
through social interaction, training, and work
experience expend less cognitive e?ort attending to
social conformity cues and place more emphasis
on group performance (Goodman & Leyden,
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 599
1991; Kim, 1996; Levine & Moreland, 1990;
Liang, Moreland & Argote, 1995). As a result,
familiar group members are more likely to
exchange partially-held and uniquely-held infor-
mation because they are less concerned with
acceptance and more oriented toward achievement
(Katzenbach & Smith, 1993; Kerr, MacCoun &
Kramer, 1996). Group members who are peers are
also more likely to share partially-held and
uniquely-held information (Gruenfeld, Mannix,
Williams & Neale, 1996; Hackman & Walton, 1986;
Sanna & Shotland, 1990). Peer group members
freely exchange information with one another
because they are not threatened by unequal power
distribution and evaluation apprehension (Arrow
& McGrath, 1993; Hackman, 1990, 1992; Hol-
lingshead, 1993; Stasser, Stewart & Whittenbaum,
1995; Whittenbaum & Stasser, 1996).
In the current research, the aforementioned
group composition concerns are addressed in the
experimental design in an e?ort to maximize the
ecacy of group discussion. Group members in
this study have a history of working with one
another (familiarity) and they hold equivalent
position levels (peers). The research framework is
presented in the next section.
2.2. Proposed research model
The research model tested in this study is based
on a theoretical framework of accounting infor-
mation systems user participation developed by
Hunton and Price (1994) and supporting research
o?ered by Erez and Arad (1986) and Latham,
Winters and Locke (1994). These studies posit that
participative decision-making strategies (e.g. indi-
vidual versus group) a?ect cognitive and motiva-
tional factors, cognition positively impacts
motivation, and both cognitive and motivational
factors mediate the relationship between partici-
pation and desired user outcomes. These relation-
ships, discussed next, are depicted in Fig. 1.
1
2.2.1. Cognitive factors
In a group discussion environment, one indi-
cator of cognition is the number of discrete ideas
generated by participants (e.g. more ideas suggest
greater cognition) (Haire, Blumberg, Davies &
Kent, 1994). In this study, all subjects record their
ideas regarding ways to improve the existing
accounting information system. In the individual
condition, subjects record their ideas in isolation.
In the group condition, subjects ®rst collectively
discuss relevant issues and then members sepa-
rately record their ideas using the same format as
the individual participants. Because group mem-
bers are likely to share partially-held and
uniquely-held information with one another, we
present the following hypothesis (alternate form):
H1a: The mean number of discrete ideas pro-
vided by group members will be greater
than the mean number of discrete ideas
o?ered by individual participants.
Another important cognitive indicator variable
is the perceived fairness of the decision process
(Hunton & Price, 1994). Process fairness (i.e. pro-
cedural justice) refers to the extent to which deci-
sion making procedures are believed to be in
accordance with accepted and expected norms of
behavior (Thibaut & Walker, 1975). Engaging
a?ected parties in a participation strategy can
enhance their process fairness beliefs (Lind, Kanfer
& Earley, 1990).
1
Cognitive factors, motivational factors, and desired out-
comes are modeled as latent constructs (see Fig. 1). While
latent constructs cannot be measured directly, their presence
can be inferred by observing salient indicator variables. We
recognize that the latent constructs shown in Fig. 1 could be
assessed by observing other variables. For example, one might
represent cognition by assessing knowledge organization,
memory processing capacity, information search strategy, and
other indicator variables outside the scope of the current study.
In structural equation modeling (SEM), path arrows pointing
from the latent construct to indicator variables represent that
the variables are observed manifestations of the underlying
construct. Hence, con®rmatory factor analysis is used in SEM
to determine the extent to which each observed variable loads
on its underlying construct. If arrows point from the measured
variables to the latent construct, SEM will calculate a regres-
sion model, rather than a factor analysis model. The regression
approach to SEM is used when the researcher desires to predict
the value of the latent construct based on the state of measured
variables. In the current study, we are not predicting the value
of the latent construct; rather, we desire to obtain indicators of
the latent construct to facilitate the calculation of path coe-
cients among constructs (see hypotheses H1c, H2c, H3, H4,
and H5 on Fig. 1). In this manner, we can obtain an overall
assessment of the relationships among the latent constructs.
600 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
One way to meaningfully engage users in the
decision process is to provide them with an oppor-
tunity to voice their opinions, preferences, and con-
cerns to decision makers. This form of participation
can provide participants with a sense of control
during the systems development process, particu-
larly if the expression of voice is manifest in the
decision outcome (Leventhal, 1980). Based on the
control-oriented perspective of procedural justice
theory, voice-based participation strategies are
expected to enhance perceptions of process fairness,
as compared to no participation (Lind & Tyler,
1988). In the current study, all subjects are allowed
to participate in information systems development
activities by expressing voice to system designers.
The relational perspective of procedural justice
theory further asserts that when voice is expressed
in a socially interactive environment, as compared
to individuals providing voice in isolation, judg-
ments of process fairness are increased because
voice takes-on symbolic meaning above and
beyond objective control perceptions, such as
Fig. 1. Proposed research model.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 601
enhanced self-esteem, self-con®dence, trust in the
decision maker, and illusory control (Lind &
Tyler, 1988; Lind et al., 1990). As well, expressing
voice in a group setting increases social cognition
of inclusion in the decision process and attach-
ment to the group (Haire et al., 1994). Finally, the
relational perspective posits that voice increases
perceptions of fairness because the collective self-
interest of the group, rather than personal self-
interests, takes precedence (Lind & Tyler). Based
on the previous discussion, the following hypoth-
esis is presented (alternate form):
H1b: When compared to the individual parti-
cipation condition, pre-experiment to
post-experiment gains in process fair-
ness perceptions will be signi®cantly
higher in the group condition.
Based on previously hypothesized gains in the
group condition (H1a and H1b) and the assertion
of Latham et al. (1994) that cognition is enhanced
in group settings, as compared to individual par-
ticipation contexts, the next hypothesis is pre-
sented in alternate form.
2
H1c: The positive relationship between partici-
pation strategy and the latent construct
representing `cognitive factors' will be
signi®cantly greater in the group, as
compared to the individual, condition.
2.2.2. Motivational factors
Researchers in participative decision-making
often frame participation as a means of motivating
a?ected parties to perform target behaviors. While
the relationship between motivation and perfor-
mance has not been clearly demonstrated in prior
research, the link between participation and moti-
vation has been tested and supported in a variety
of studies (see Wagner & Gooding, 1987 for a
review). Two key motivational factors are (a) self-
ecacy beliefs (Locke, 1991) and (b) congruence
between desired and actual participation (Doll &
Torkzadeh, 1989).
Bandura (1986) and Locke (1991) indicate that
self-ecacy is an integral component of motivation.
Self-ecacy perceptions can be aimed at the partici-
pative process (e.g. the extent to which users believe
they can e?ectively contribute to the development of
an information system) and the decision outcome
(e.g. the extent to which participants believe they can
e?ectively use the information system). In this study,
we examine outcome-oriented self-ecacy. In an
information systems setting, Compeau and Higgins
(1995a,b) report a signi®cant, positive relationship
between participation and outcome-oriented self-
ecacy. Lathamet al. (1994) also report increases in
outcome-oriented self-ecacy brought about by
voice-based participation.
The relational perspective of procedural justice
indicates that group participation enhances self-
ecacy levels of group members over their indivi-
dual participants. Providing collective voice to
users in a social setting emphasizes symbolic psy-
chological factors embedded within voice, such as
inclusion, self-esteem, attachment, self-con®dence,
and trust (Lind & Tyler, 1988). These relational
bene®ts coupled with the collective sense of duty,
responsibility, and empowerment arising from
group participation (Jackson, 1992) are expected
to heighten self-ecacy for group members as
compared to individuals. Accordingly, the following
hypothesis is o?ered as follows (alternate form):
2
Hypotheses H1c, H2c, H3, H4, and H5 will be tested using
standardized path coecients produced by SEM, as repre-
sented on Fig. 1. Since latent constructs cannot be directly
measured, SEM ®rst calculates the impact of indicator vari-
ables on their underlying latent constructs through con-
®rmatory factor analysis (i.e. the measurement model). Next,
factor loadings are used to quantify each construct and SEM
employs a series of regression analyses to determine standar-
dized path coecients among model constructs (i.e. the struc-
tural model). Finally, SEM determines overall goodness of ®t
indices that indicate how well the data ®ts the proposed model.
It is important to understand why we did not draw model paths
to and from each indicator variable shown on Fig. 1. First, this
alternative approach would yield a series of path analysis
models (i.e. there would be multiple, independent path models
representing each possible combination of paths among mea-
sured variables). Second, this approach would not control for
covariance between indicator variables within each construct.
Third, it would not be possible to gain a comprehensive picture
of the relationships among the model constructs, as each sepa-
rate path analysis would yield di?erent path coecients and
error terms. Fourth, the SEM approach could not be utilized in
this study due to sample size restrictions and over-identi®cation
problems. Finally, due to the restrictions just noted, it would
not be possible to obtain overall goodness of ®t indices for the
research model; hence, it would be dicult to precisely deter-
mine the comprehensive relationships among latent constructs.
602 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
H2a: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in self-ecacy will be
signi®cantly higher in the group condition.
Individuals desire to participate in decisions that
a?ect their well-being, particularly decisions that
further their self-interest. The strength of desired
participation is positively related to the extent to
which a?ected parties believe the decision out-
come is personally relevant and salient (Alluto &
Belasco, 1972; Lind & Tyler, 1988). When desired
participation exceeds actual (a state of relative
deprivation), a?ected parties become frustrated
and the ecacy of participation is diminished.
When actual participation exceeds desired (a state
of relative saturation), the ecacy of participation
is also negatively compromised, although the
deleterious impact of saturation is not as pro-
nounced as deprivation. Doll and Torkzadeh
(1989) indicate that increased congruence between
desired and actual participation motivates infor-
mation system users to maximize their participa-
tive e?orts. Additionally, participation congruence
is positively related to the formation of con-
sequential user attitudes (e.g. outcome satisfaction
and organizational commitment) and user perfor-
mance (Hunton & Price, 1994). The relational
perspective of procedural justice indicates that
group participation is considered to be more ful-
®lling and personally satisfying than individual
participation; hence, for a?ected parties who
desire more participation, group discussion is
expected to narrow the gap between desired and
actual participation to a greater extent than indi-
vidual participation. Accordingly, the following
hypothesis (alternate form) is presented:
H2b: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in participation con-
gruence will be signi®cantly higher in the
group condition.
Based on the propositions previously set forth
(H2a and H2b) and supporting research ®ndings
from Erez and Arad (1986) indicating that motiva-
tion is signi®cantly higher in group, as compared to
individual, participation strategies, the following
hypothesis is indicated (alternate form):
H2c: The positive relationship between partici-
pation strategy and the latent construct
representing `motivational factors' will be
signi®cantly greater in the group, as com-
pared to the individual, condition.
2.2.3. Relationship retween cognitive and
motivational factors
Hunton and Price (1994), Erez and Arad (1986)
and Latham et al. (1994) suggest a positive rela-
tionship between cognition and motivation. They
reason that a?ected parties gain task relevant
knowledge and receive salient informational cues
(such as increased understanding of alternatives,
constraints, and trade-o?s) when expressing voice.
The appraisal (i.e. thinking) process evoked by
voice-based participation can heighten self-ecacy
beliefs (Bandura, 1986) and improve the con-
gruence between desired and actual participation
(Doll & Torkzadeh, 1989). Theoretical and
empirical support for a stronger association
between cognition and motivation in group versus
individual participation conditions is found in
Latham et al. Based on theory and prior evidence,
we present the following hypothesis (alternate
form):
H3: The positive relationship between the
latent constructs representing `cognitive
factors' and `motivational factors' will
be signi®cantly greater in the group, as
compared to the individual, condition.
2.2.4. Relationship between cognitive factors and
desired outcomes
Cognitive factors can be linked to one's choice
of action and the extent to which that action is
deemed successful (Locke & Latham, 1990). Cog-
nitions are translated into action when individuals
make a volitional choice to consciously appraise
their a priori held knowledge and beliefs. The
appraisal process integrates an individual's values,
needs, desires, and aspirations into an action plan.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 603
As a result, cognitive factors can be directly linked
to attitudinal and behavioral outcome variables. In
an experimental setting, Latham et al. (1994) found
that cognition (i.e. task strategy quality) was sig-
ni®cantly, positively related to performance. Their
study also indicated a stronger relationship
between cognition and performance in group, as
compared to individual, conditions. Accordingly,
we anticipate the following (alternate form):
H4: The positive relationship between the
latent constructs representing `cognitive
factors' and `desired outcomes' will be
signi®cantly greater in the group, as
compared to the individual, condition.
2.2.5. Relationship between motivational factors
and desired outcomes
Meta-analytic reviews of participation studies
have not consistently supported the link between
motivation and consequential outcomes. Locke
(1991) suggests that equivocal results from prior
research are likely due to poor surrogate indica-
tors of motivation. He posits that self-ecacy and
participation congruence (based on expectancy
theory) are valid motivational factors. Supporting
Locke's assertions, Latham et al. (1994) and
Compeau and Higgins (1995a,b) report a sig-
ni®cant, positive relationship between outcome-
oriented self-ecacy and consequential attitudes
and behavior. Doll and Torkzadeh (1989) indicate
a signi®cant link between participation con-
gruence, user satisfaction, and intention to use an
information system. Erez and Arad (1986) and the
relational perspective of procedural justice suggest
a greater relationship between motivation and
performance in group, as compared to individual,
participation conditions. Based on the theory and
evidence just presented, the ®fth hypothesis is
o?ered as follows (alternate form):
H5: The positive relationship between the
latent constructs representing `motiva-
tional factors' and `desired outcomes' will
be signi®cantly greater in the group, as
compared to the individual, condition.
2.2.6. Desired outcomes
The indicator variables examined in this study
representing the latent construct `desired out-
comes' are outcome satisfaction, organizational
commitment, and user performance. These parti-
cular indicator variables were selected primarily
because the management of the state agency who
sponsored this ®eld experiment desired to measure
changes in these factors brought about by the
group and individual user participation strategies.
Outcome satisfaction concerns the users' a?ective
reaction toward the target information system. Doll
and Torkzadeh (1988) examined the issue of user
satisfaction. The ®ve dimensions of user satisfac-
tion they developed and tested were system accu-
racy, ease of use, system timeliness, information
content, and information format. In this study, we
assess the subjects' pre- and post-experiment levels
of user satisfaction along these ®ve dimensions.
Recently, organizational commitment (OC) has
been the focus of research in organizational behavior
(e.g. Aranya, Pollock &Amernic, 1981; Cohen, 1996;
Dunham, Grube & Castaneda, 1994; Jaros, 1995).
OC is de®ned as the strength of an individual's iden-
ti®cation with and involvement in a particular
organization. OC is usually characterized by the
following factors: ``a strong belief in and accep-
tance of the organization's goals and values; a
willingness to exert considerable e?ort on behalf
of the organization; and a de®nite desire to main-
tain organizational membership'' (Porter, Steers,
Mowday & Boulian, 1974, p. 604). Allen and
Meyer (1990) identi®ed three components of com-
mitment: a?ective, normative and continuance.
The a?ective component of OC refers to the
employee's emotional attachment to, identi®cation
with, and involvement in the organization. In this
study, we assess the a?ective component of OC.
3
Locke and Latham (1990) and Locke (1991)
indicate that prior research in participation
demonstrates a weak association between partici-
pative activities and human behavior. Locke and
Latham and Latham and Lee (1986) suggest that
3
Management of the sponsoring State agency and the
researchers were concerned about the length of the pre- and
post-experiment survey instruments. Hence, management chose
to measure only the a?ective dimension of OC.
604 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
research conducted in ®eld settings might help to
provide a more clear link between participative
decision making and consequential behaviors,
especially where the decision outcome and perfor-
mance measures are closely aligned. In this ®eld
experiment, we obtained pre- and post-experiment
objective measures of sales order and customer
billing data input quality. These performance
measures are directly related to the decision out-
come (i.e.the revamped accounting information
system). Based on the theoretical development of
hypotheses H1 through H5, we suggest the ®nal
set of hypotheses (alternate form):
H6a: When compared to the individual partici-
pation condition, pre-experiment to post-
expe-riment gains in outcome satisfaction
will be signi®cantly higher in the group
condition.
H6b: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in organizational com-
mitment will be signi®cantly higher in the
group condition.
H6c: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in sales order data
input quality will be signi®cantly higher
in the group condition.
H6d: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in customer billing data
input quality will be signi®cantly higher
in the group condition.
3. Research method
4
3.1. Scenario
A large state agency permitted the researchers to
conduct a ®eld experiment investigating participa-
tion strategies for developing computerized
accounting information system applications.
5
The
agency is comprised of 46 ®eld oces located
throughout the state. Field oces have little or no
communication with each other, as they are
autonomous in all operational and administrative
respects. Field oce managers report directly to
the agency's central oce located in the capital
city. Computerized accounting information sys-
tems have evolved at ®eld oces with little gui-
dance from the central oce. Accounting
applications at ®eld locations, on the whole, were
poorly designed and sparsely documented. Custo-
mers throughout the state complained about high
levels of order entry and billing errors. Users at
®eld oces realized their accounting systems were
inadequate, as they had expressed their dis-
satisfaction to agency managers and state audi-
tors. Based on user feedback and ®ndings of
recent annual audits, state auditors mandated the
agency to redesign its accounting information sys-
tems in an e?ort to reduce the unacceptably high
data input error rates and correct a multitude of
additional problems. Accordingly, the agency
instituted a ®ve-year plan to revamp all of its
computerized accounting and management infor-
mation systems. The ®rst project was to develop
standardized revenue cycle applications (i.e. sales
order, cash collection, accounts receivable, and
customer billing) at the central information sys-
tems department.
The agency director believed that user input
throughout the development project was critical to
the ultimate success of the system. However, he
was unsure whether users should provide indivi-
dual input to system designers or if it would be
more e?ective for users to discuss their concerns
and preferences in small group environments
before submitting their suggestions to system
designers. Since the director had many more pro-
jects planned in the future, he believed that this
®rst initiative presented an opportunity to explore
the most e?ective participation strategy. Conse-
quently, the agency director asked the researchers
to design a ®eld experiment to assess which strat-
egy, individual or group, yields higher levels of
desired outcomes. In particular, he insisted that
the most critical outcome of interest was the
quality of data input regarding sales order and
customer billing transactions.
4
The research scenario and procedure are the same as
described in Hunton and Gibson (1998) as the data for that
study were collected simultaneously with the current study.
5
The researchers were paid for out-of-pocket expenses only.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 605
3.2. Procedure
At the beginning of the ®rst month of the study
(m
1
), a pre-experiment survey was distributed to
all accounting and sales order clerks throughout
the agency via the agency's intranet. The
researchers initiated the surveys and ensured par-
ticipants that individual responses would be held
in strict con®dence and that only summary mea-
sures would be shown to management. A custo-
mized program presented survey items randomly
to each respondent (to preclude a possible order
e?ect), restricted participants from viewing prior
questions (to preclude respondents from looking
back to see how they answered a similar survey
item and responding in the same way), and pro-
hibited subjects from forwarding results to anyone
other than the originating source, i.e. the
researchers (to ensure con®dentiality).
During the ®rst and second experimental
months (m
1
and m
2
), subjects participated in
developing the revenue cycle applications. An
initial memo was sent from the director to each
participant explaining that the agency's informa-
tion system design group (ISDG) was developing
standardized revenue cycle applications to replace
existing ®eld oce applications. The director
expressed that he and agency management felt it
was critically important for application users to be
involved in the systems design process. The memo
explained that for the next ®ve consecutive Friday
mornings, ®eld oces would not open for business
at their regular time of 8:00 a.m., instead they
would open at 10:00 a.m. Employees would still
report for work at 8:00 a.m., however, they would
spend these two hours expressing their opinions,
concerns, preferences, and suggestions to the
ISDG regarding the applications under develop-
ment. The director made it clear that while
employees could not reasonably expect all of their
individual preferences and suggestions to be
incorporated into the new applications, every
e?ort would be made to include as many of their
recommendations as possible in the new systems.
The ISDG and participants communicated via
the agency's intranet system. Each Friday morn-
ing, participants received a standardized ques-
tionnaire from the ISDG soliciting input regarding
certain aspects of the revenue cycle applications
under development (i.e. sales order, cash collec-
tions, accounts receivable, customer billing). The
questionnaires contained open-ended questions
pertaining to the topic of the week. The ®ve weekly
topics focused on system accuracy, ease of use, sys-
tem timeliness, information content, and report for-
mats. Two participation conditions (individual and
group) were randomized to ®eld oces, resulting in
23 oces assigned to each treatment.
In the individual condition, subjects were asked
to remain at their workstations and answer the e-
mail questionnaire from the ISDG. Field oce
supervisors were present to ensure that interaction
between subjects did not occur during the answer
period. The questionnaire was designed so it could
be completed in about 1 hour. After respondents
completed the instrument, they were instructed to
remain at their desk and work on agency matters
until the oce opened at 10:00 a.m.
In the group condition, the accounting and sales
order clerks met in a conference room, where they
received a copy of the weekly questionnaire and dis-
cussed relevant issues for 1 h. There were no super-
visors present during the meetings inorder tofacilitate
the free exchange of information among clerks; how-
ever, supervisors ensured that all clerks attended the
meetings. The groups were not facilitated by outside
observers. After discussion, group members returned
to their respective workstations and answered the e-
mail questionnaire fromthe ISDG.
After the ®ve weekly meetings concluded, the
software was developed and tested during the
third (m
3
), fourth (m
4
) and ®fth (m
5
) experimental
months. Members of the ISDG conducted eight
hours of standardized training during the sixth
month (m
6
) at each ®eld oce location. At the end
of the sixth month, the revenue cycle applications
were implemented. The researchers administered a
post-experiment survey at the end of the 12th
month (m
12
). Finally post-experiment measures of
performance were tracked for 12 months beyond
installation, concluding at the end of the 18th
experimental month (m
18
).
While it is dicult to attain a high degree of
internal validity in ®eld experimentation, several
controls were incorporated into the current study.
Field oces were randomized to treatment condi-
606 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
tions. Neither ISDG members nor ®eld oce
supervisors were aware of di?erential treatments
across ®eld oces. The following ®ve factors were
held constant between treatment conditions: pre-
and post-experiment survey items; weekly topic
and related questionnaire; time allocated for par-
ticipative activities; training sessions; and decision
outcome (i.e. the ®nal revenue cycle applications).
The procedures by which individuals and group
members engaged in participative activities were
held constant within each treatment condition.
Finally, there were no agency-sponsored meetings
or discussions during the experimental time frame
where representatives from di?erent ®eld oces
could interact with one another.
3.3. Measured variables
Survey items were measured using seven-point,
Likert-scales with semantic anchors on both ends.
Some items were reversed scored, however, all
were converted such that scores of 1=low and
7=high. All variables were assessed using two or
more items. Reliability estimates for two-item
measures were obtained using Pearson's product
moment correlation (r), and reliability estimates
for three or more item measures were obtained
using Cronbach's alpha () (Cronbach, Gleser,
Nanda & Rajaratnam, 1972). Related item mea-
sures were collapsed (i.e. averaged) into single
variable indices.
6
3.4.1. Cognitive factors
The ®rst cognitive factor, number of ideas gener-
ated by clerks throughout the ®ve-week participa-
tionperiod, was ascertained by examining comments
and suggestions provided by all participants. To
accomplish this, we hired three graduate doctoral
students, hereafter referred to as graders, to analyze
the respondents' weekly questionnaires. The graders
®rst independently examined the ®ve questionnaires
submitted by each respondent to identify and de®ne
a common set of 102 discrete ideas. The graders'
initial estimated inter-rater reliability
7
was 0.87.
After reviewing all items, the graders eventually
agreed on a common set of 97 discrete ideas. Next,
the graders independently counted the number of
discrete ideas provided by each subject. Average
inter-rater reliability across all subjects was 0.93.
After reviewing and reassessing disputed items, the
graders reached full agreement on all subjects.
Three survey items assessed the extent to which
subjects believed the process of participating in
developing the revenue cycle applications was fair
(pre-experiment ? 0:95; post-experiment ?0:91).
The fairness questions were adapted from Lind et
al. (1990). For example, one of the three items asked
``What do you think about the way in which your
agency allows you to participate in developing,
modifying, and/or acquiring accounting software
applications?'' (1= Very Unfair, 7=Very Fair).
The three items were averaged to form two process
fairness indicesÐone from the pre-experiment sur-
vey and one from the post-experiment survey.
3.4.2. Motivational factors
The extent of congruence between desired and
actual participation was measured using survey
6
Measured variables were collapsed into indices represent-
ing `average scores' to facilitate the SEM analytic approach.
The number of survey items used to assess measured variables
ranged from 2 to 16 (i.e., for the participation congruence
measure, there were eight `desired' and eight `actual' items).
Further, there were pre-experiment and post-experiment mea-
sures of each variable (except for `discrete ideas'), e?ectively
doubling the number of survey items for each variable. Because
inter-item reliability estimates were relatively high for all mea-
sured variables, collapsing the items into average score indices
does not result in a signi®cant loss of information. Addition-
ally, the sample size in this study is minimal for SEM modeling
and including all pre- and post-experiment measured items in
the model gives rise to power concerns, degree of freedom
issues, and over-identi®cation problems. The researchers ran
alternative tests using summed indices (rather than averaged),
and research ®ndings and interpretations were not signi®cantly
changed for any of the hypotheses.
7
Estimated inter-rater reliability was calculated using
Cronbach's alpha, as described in Kerlinger (1986). Essentially,
the complete set of discrete ideas initially identi®ed by the
raters was ®rst determined (n=102). Next, the 102 ideas were
randomized, to preclude an implied importance rating, and
numbered from 1 to 102. Third, each idea in the set that was
identi®ed by a particular grader was coded as one (1) and ideas
that were not in the initial set were coded as zero (0). After all
three raters' ideas were coded accordingly, a Cronbach's alpha
estimate of reliability was conducted on the assigned values for
all three graders simultaneously. All other inter-rater reliability
estimates reported in this paper were calculated using the same
procedure.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 607
items adapted from Doll and Torkzadeh (1989).
Participants responded to eight items asking the
extent to which they desire to participate in var-
ious activities regarding the development, mod-
i®cation, and/or acquisition of accounting
software applications (pre-experiment ? 0:89;
post-experiment ? 0:93). Eight identical survey
items were used to assess the extent to which par-
ticipants actually do participate in these same
activities (pre-experiment ? 0:94; post-experi-
ment ? 0:88). Participation congruence indices
were determined as follows: variable indices were
calculated by averaging the eight survey items
representing each of the actual and desired parti-
cipation variables; the actual index was subtracted
from the desired index; and the di?erence between
actual and desired indices formed pre- and post-
experiment participation congruence indices.
Three survey items, adapted from Bandura
(1986), examined participants' self-ecacy beliefs
that they could e?ectively use the revenue cycle
applications. For example, one item stated ``I
believe that I can e?ectively use the revenue cycle
applications (sales order, cash collection, accounts
receivable, and customer billing) to process custo-
mer sales order, billing, and payment information.''
(1=Strongly Disagree, 7=Strongly Agree). The
self-ecacy items were averaged to form two
indices, one from the pre-experiment survey
( ? 0:92) and one from the post-experiment sur-
vey ( ? 0:89).
3.4.3. Desired outcomes
Outcome satisfaction was measured using Doll
and Torkzadeh's (1988) end-user computing satis-
faction scale. The items adapted for this study
assessed how respondents felt about ®ve areas of
the revenue cycle applications: system accuracy (2
items); ease of use (2 items); system timeliness (2
items); information content (4 items); and infor-
mation format (2 items). Inter-item reliability esti-
mates for all 12 items were high (pre-experiment
? 0:86; post-experiment ? 0:92) and they were
averaged to form pre- and post-experiment indices.
Organizational commitment was assessed using
the a?ective dimension of the Meyer, Allen and
Smith (1993) scale. For example, one of the six
survey items stated, ``I would be very happy to
spend the rest of my career with this agency''
(1=strongly disagree, 7=strongly agree). Inter-
item reliability was high (pre-experiment ? 0:96;
post-experiment ? 0:87) and the six items were
averaged into pre- and post-experiment organiza-
tional commitment indices.
Human performance was measured by calculat-
ing the quality of sales order and customer billing
data entry transactions of each clerk who partici-
pated in the study. Each transaction type required
a di?erent number of ®elds in which the clerk must
enter data. For example, the pre-development
sales order system required users to enter data into
32 ®elds when adding a new customer and 29 ®elds
when entering a sales order transaction. Assume
that a clerk made errors in 44 ®elds while adding
20 new customers and made errors in 73 ®elds
while entering 50 sales order transactions during a
given month. The clerk's monthly data input
quality rate in this case would be 0.94 [(2090-117)/
2090]. The researchers accessed all monthly trans-
action ®les and computed the data input quality
rates. However, due to con®dentiality concerns,
the state agency would not allow the researchers
to report the total number of input ®elds, errors or
transaction types entered each month.
3.4.4. Gain scores
Gain scores were computed at the individual
level for all variables. Psychological gain scores
represent the di?erence between post-experiment
and pre-experiment variable indices. To compute
gain scores for sales order and customer billing
transactions, we subtracted the average data input
quality rate for the ®nal 6 months of the experi-
ment (m
12
±m
18
) from the average data input
quality rate for the 6 month period prior to
implementing the new revenue cycle applications
(m
1
±m
6
).
8
Data input quality rates for the ®rst 6
months after implementation (m
7
±m
11
) were not
included in the analysis due to observed learning
e?ects.
8
Using the 6 month pre-implementation period as the base-
line measure, the gain scores yield a relative change in data
input quality from pre- to post-implementation.
608 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
4. Results
4.1. Sample characteristics
There were a total of 172 participants in the
individual condition and 174 in the group condi-
tion. Chi-square analyses were conducted to assess
whether proportional di?erences existed between
the individual and group modes based on gender,
education level, ethnic origin, job category (i.e. sales
and accounting), and supervisory performance eva-
luation. No signi®cant di?erences were noted, as
all p-values exceeded 0.24. Additionally, t-tests
were performed to determine the extent to which
mean di?erences existed between the individual
and group modes based on age, months with the
agency, months with the local oce, months in
current position, and months in current grade. As
shown in Table 1, no signi®cant di?erences were
observed.
There were 46 ®eld oces involved in the
experiment. The individual and group manip-
ulations were randomly assigned to 23 ®eld
oces each. At the beginning of the experiment,
394 subjects completed the pre-experiment sur-
vey. When the post-experiment survey was
administered 12 months later, 48 (12%) of the
initial 394 subjects were no longer employed by
the state agency. The human resource manager
of the agency reported a 13.6% turnover for
the same 12 months of the prior year; therefore,
the 12% attrition rate was not considered
abnormal. Of the 48 subjects who left the
agency, 22 were in the individual and 26 were
in the group conditions. We conducted statis-
tical testing to determine the extent to which
demographic factors of the 48 employees who
left the agency were signi®cantly di?erent from
the remaining sample. No di?erences were
found, as the p-values for all tests exceeded 0.5.
The psychological and performance gains
reported in this study re¯ect only the remaining
346 participants.
A total of 60 credit/collection (29 individual, 31
group), 49 cash receipts (24 individual, 25 group),
73 customer billing (37 individual, 36 group), and
164 sales order (82 individual, 82 group) clerks
participated in the study. A chi-square analysis
was conducted to determine whether there were
proportional di?erences in the types clerks
between the two treatment conditions and none
were found (X
2
=0.109, p-value=0.9916). In the
individual (group) conditions, the mean number of
subjects at each ®eld site were 7.47 (7.56).
It should be noted that we did not perform self-
reported manipulation checks of each participation
condition due to the tight controls surrounding the
individual and group treatments. Speci®cally, on-
site supervisors made sure that individuals
received the ISDG survey each Friday morning at
8:00 a.m. and answered all questions in isolation.
In the group condition, supervisors ensured that
group members received the ISDG survey at 8:00
a.m., discussed the survey items in a group setting
for one hour, and then responded to the questions
in isolation. Tests of the research model and
hypotheses are next presented.
4.2. Research model testing
Pre- to post-experiment gains were analyzed
using EQS, a SEM program developed by Bentler
Table 1
Sample characteristics
All subjects Individual mode Group mode t Statistic
a
p Value
Age (years) 36.32 37.01 35.63 0.90 0.37
Months with agency 108.84 109.85 107.83 0.12 0.90
Months with local oce 105.75 106.14 105.37 0.05 0.95
Months in current position 103.09 103.55 102.63 0.06 0.95
Months in current grade 67.58 68.59 66.56 0.22 0.83
a
Based on t-tests of individual versus group means.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 609
(1989).
9
In this study, the SEM model was
designed to test for signi®cant di?erences in incre-
mental changes in model paths from individual to
group conditions following the procedure estab-
lished by Sorbom (1978). Participation was coded
as 0 (individual) and 1 (group) without measure-
ment error. Model paths from participation strat-
egy to cognition and motivation were constrained
at zero (0) in the individual condition. Paths from
cognition to motivation, cognition to outcomes,
and motivation to outcomes were constrained to
be equal to the individual condition. Accordingly,
standardized coecients along model paths represent
incremental di?erences in the group, as compared to
the individual, condition. The statistical sig-
ni®cance of incremental path parameters was tes-
ted using the multivariate LaGrange multiplier
test (Aitchison & Silvey, 1958). Since there was no
theoretical reason to expect di?erential factor
loadings between treatment conditions from mea-
sured variables to latent constructs, all factor
loadings were unconstrained.
10
Incremental stan-
dardized path coecients, results of multivariate
LaGrange multiplier tests on path coecients, and
factor loadings are shown in Fig. 2.
Since there is no generally accepted measure of
overall model goodness of ®t, leading researchers
recommend that multiple criteria be evaluated to
determine how well the data ®ts the model
(Breckler, 1990; Wheaton, 1987). In this study, we
use ®ve goodness of ®t criteria suggested by Ben-
tler (1990) and Schumacker and Lomax (1996).
The ®rst ®t index is the chi-square statistic. This
index compares the proposed model to the general
alternative where all variables are correlated. Sta-
tistically signi®cant values re¯ect a poor model ®t.
However, in large samples the chi-square statistic
is almost always signi®cant because it is a direct
function of sample size. The second and third
goodness of ®t indicators are the normed ®t index
(NFI) and non-normed ®t index (NNFI) as pro-
posed by Bentler and Bonett (1980). The fourth
criterion is the Comparative Fit Index (CFI).
Bentler (1990) indicates that the CFI works well at
all sample sizes and is more stable than the NFI
and NNFI because it is less sensitive to sampling
error. According to Bentler (1990), NFI, NNFI,
and CFI values of 0.90 or greater indicate a good
®tting model. Finally, we report the root mean square
error approximation (RMSEA) (Schumacker &
Lomax, 1996). RMSEA provides an approximation
of overall variance unaccounted for in the model and
values of 0.05 or less are desirable.
Bentler (1990) suggests that, rather than focus-
ing on one ®t index, researchers should look at the
collective evidence generated from multiple ®t
indices to determine the adequacy of model ®t.
Based on the relative size of the goodness of ®t
indicators obtained in this study (see Fig. 2), the
data appears to satisfactorily ®t the research
model.
4.3. Means testing
While the SEM model shown includes all 346
participants, there is valid concern that subjects in
the group condition were not independent from
one another because they discussed relevant issues
before responding to the survey instruments.
Accordingly, one could argue that, conservatively,
the sample size should be 46 (23 ®eld sites per treat-
ment condition) rather than 346. Unfortunately, a
sample size of 46 is too small for SEM calculations.
9
We also ran our analyses using CALIS, LISREL, LIS-
COMP, and SEPATH. While the factor loadings and path
coecients change slightly due to varying estimation processes,
overall ®t indices are satisfactory in all analyses. Hence, our
®ndings are not sensitive to the particular SEM application
used. In preliminary tests, we discovered that our data was not
multivariate normal, as is the case with most data of this nat-
ure. We tested normality using the Kolmogorov±Smirnov test.
Results indicated that these measures failed to approximate a
normal distribution, therefore we used the generally weighted
least squares (GWLS) method to estimate model structural
equation parameters and goodness of ®t indices. The GWLS is
asymptotically distribution free, whereas maximum likelihood
(ML) and generalized least squares (GLS) estimators assume
normally distributed data. Andersen and Gerbing (1988) sug-
gest that EQS is the best modeling approach when data is not
multivariate normal due to its elliptical estimation assumption.
10
Factor loadings determined by SEM are shown on Fig. 2.
For example, the coecients from `cognitive factors' to discrete
ideas (0.82) and process fairness (0.90) suggest that these two
indicator variables load high on and are representative of the
underlying factor, or latent construct. Factor loadings greater
than 0.30 represent a strong association between an indicator
variable and its underlying construct (Schumacker & Lomas,
1996).
610 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
As a way to deal with this concern, the research-
ers compared the results of psychological assess-
ments (i.e. process fairness, self-ecacy,
congruence, satisfaction, and commitment) and
objective performance measures (i.e. sales order
quality and customer billing quality) for both
treatment conditions at the ®eld oce level
(N=23 per treatment condition). Within the
individual and group conditions, post-experiment
means were signi®cantly greater than pre-experi-
ment means (p-value
This study reports the results of a longitudinal ®eld experiment designed to examine the impact of group discussion
when soliciting user requirements of an accounting information system. A total of 346 subjects were randomly assigned
to individual and group treatment conditions. Pre- and post-experiment surveys, administered 12 months apart,
assessed cognitive, motivational, and desired outcome factors. Objective measures of cognition were obtained during a
®ve-week user-input period and actual performance data were collected over 18 months. Results indicate signi®cantly
greater pre-experiment to post-experiment gains in psychological and behavioral variables in the group, as compared to
the individual, condition. Structural equation modeling highlights the key mediating roles of cognition and motivation
on desired user outcomes in a group setting.
Soliciting user-input during the development of an
accounting information system: investigating the ecacy
of group discussion
James E. Hunton
a,
*, Dana Gibson
b
a
School of Accountancy, College of Business Administration, University of South Florida, 4202 East Fowler Avenue, BSN 3403, Tampa,
FL 33620-5500, USA
b
Department of Accounting, College of Business Administration, Texas Woman's University, Denton, TX, USA
Abstract
This study reports the results of a longitudinal ®eld experiment designed to examine the impact of group discussion
when soliciting user requirements of an accounting information system. A total of 346 subjects were randomly assigned
to individual and group treatment conditions. Pre- and post-experiment surveys, administered 12 months apart,
assessed cognitive, motivational, and desired outcome factors. Objective measures of cognition were obtained during a
®ve-week user-input period and actual performance data were collected over 18 months. Results indicate signi®cantly
greater pre-experiment to post-experiment gains in psychological and behavioral variables in the group, as compared to
the individual, condition. Structural equation modeling highlights the key mediating roles of cognition and motivation
on desired user outcomes in a group setting. #1999 Elsevier Science Ltd. All rights reserved.
1. Introduction
Accounting information system design decisions
are not made in isolation; rather, system devel-
opers must consider a host of relevant human
factors, such as the concerns, opinions, and sug-
gestions of a diverse constituency of system users.
One way to include various a?ected users in the
design process is to provide them with opportu-
nities to express, compare, and contrast their
design preferences in group settings during the
requirements de®nition phase of a systems devel-
opment project (Lind, 1995; Lind & Tyler, 1988;
Vessey & Conger, 1993). Many researchers indi-
cate that discussion of this nature is a critical
group participation activity leading to high quality
decisions, primarily due to the collective sharing of
intellectual resources (e.g. Maier, 1963; McGrath,
1991; Osborn, 1957; Stasser & Titus, 1985; Vroom
& Jago, 1988).
The use of group participation is expanding into
all facets of the accounting profession, such as tax
planning, auditing, consulting, systems designing,
and budgeting (Brecht & Martin, 1996; Sutton &
Hayne, 1997). While the trend toward collective
participation is rising across the professional
0361-3682/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PI I : S0361- 3682( 99) 00023- 9
Accounting, Organizations and Society 24 (1999) 597±618
www.elsevier.com/locate/aos
* Corresponding author. Tel.: +1-813-974-6523; fax: +1-
813-974-6528.
E-mail address: [email protected] (J.E. Hunton)
community, accounting research in the area of
participation continues to focus on individuals.
Sutton and Hayne urge accounting researchers to
investigate various forms of group participation in
a wide variety of accounting-related situations.
Specifying design requirements of accounting
information systems is an area where accountants
can add considerable value to group processes
because of their unique, multifaceted role as users,
evaluators, developers, and assurers. Accordingly,
the current study focuses on the e?ectiveness of
facilitating small group discussion in an important
accounting scenario-specifying user requirements
of an accounting information system.
This study is unique along three important
dimensions. First, the relative ecacy of group
versus individual participation strategies is com-
pared. From a theoretical perspective, the current
study enhances our understanding of group pro-
cesses by examining simultaneous relationships
among cognitive, motivational, and outcome con-
structs using a structural equations model (SEM).
We ®nd that the strength of the relationships
among these constructs is signi®cantly greater in
the group, as compared to the individual, condi-
tion. The SEM also indicates that cognition and
motivation are key mediators between user-input
and desired outcomes. To our knowledge, this is
the only longitudinal ®eld experiment in account-
ing and reference disciplines (e.g. psychology and
organizational behavior) with a sucient sample
size to compare individual and group processes
using a SEM. Second, this experiment examines
real actors performing real tasks in a contextually
rich, natural environment. When conducting
longitudinal ®eld experimentation of this nature, it
is dicult to precisely control all internal threats
to validity. However, ®eld experiments are rare in
accounting and reference disciplines, and if inter-
nal threats are controlled with a reasonable degree
of assurance, research results can contribute valu-
able knowledge and understanding to the external
and ecological validity of extant theory (Libby &
Luft, 1993). Finally, most user participation
research relies on self-reports of cognitive percep-
tions and behavioral intentions because it is dicult
to obtain objective data of this nature, particularly
outside of laboratory settings. In the current
study, we obtained and analyzed objective mea-
sures of cognition and user performance. Pairing
objective data with self-reported attitudes and
perceptions provides a solid foundation for asses-
sing convergent validity of latent construct indi-
cators within a research framework.
2. Theory and hypotheses
Extant participation literature in accounting
investigates, for the most part, the impact of indi-
vidual subordinate involvement in in¯uencing and
determining budgets (Shields & Shields, 1998).
The current study expands the scope of accounting
participation research by examining the in¯uence
of user participation, both individual and group,
in determining the speci®cations of an accounting
information system. Many information systems
(IS) researchers believe that user participation,
particularly during the requirements de®nition
phase of a systems development project, is critical
to overall system success (Cushing, 1990; Doll &
Torkzadeh, 1991; Vessey & Conger, 1993). A
variety of IS studies have examined the impact of
user participation on desired attitudes and behavior;
however, study ®ndings are generally inconclusive
(Bailey & Pearson, 1983; Baroudi, Olsen & Ives,
1986; Cavaye, 1995; DeLone & McLean, 1992;
Ives & Olson, 1984). Despite past equivocal
results, researchers continue to believe that the
nature, timing, and extent of user participation are
key factors leading to IS success (Hunton & Price,
1994).
Barki and Hartwick (1994) have examined,
validated and categorized the participation con-
struct for IS users. The following three types of
user participation have been identi®ed: overall
responsibility, user-IS relationship, and hands-on
activity. The former category refers to user activ-
ities and assignments aimed at providing overall
accountability for the systems development pro-
ject, such as leading a project group or selecting
hardware and software. User-IS relationship
re¯ects user activities that are focused on in¯uen-
cing the IS sta?, for example approving IS sta?
work plans or evaluating the work performed by
the IS sta?. Hands-on activity refers to speci®c
598 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
design and implementation tasks performed by
system users. For instance, helping to de®ne
screen layouts and report formats or creating a
user procedure manual are considered to be
hands-on activity. In the current study, accounting
information system users are asked to provide
input to system developers regarding various
aspects of the system during the requirements
de®nition phase of a development project. Hence,
user participation in this study is interpreted as
hands-on activity performed by system users.
Barki and Hartwick (1994) indicate that metho-
dological weaknesses are partially responsible for
equivocal results of past participation studies.
Leana, Locke and Schweiger (1990) and Dipboye
(1990) agree and they provide two important sug-
gestions. First, Leana et al. claim that participa-
tion studies using a single questionnaire at a single
point in time can arti®cially distort the magnitude
of research ®ndings. They recommend that parti-
cipative decision-making researchers should
engage in longitudinal studies in order to improve
measurement reliability. Second, Dipboye notes
that laboratory experiments are used in the
majority of participation studies. He cautions that
methodological artifacts inherent in laboratory
settings contribute to mixed ®ndings across stu-
dies. In order to avoid this problem, he urges
researchers to design and execute studies in ®eld
environments, particularly ®eld experiments. In an
e?ort to address these suggested areas of
improvement, the current study examines the e-
cacy of group versus individual user participation
by conducting longitudinal research in a ®eld set-
ting.
2.1. Group participation
Judgment and decision-making research in
accounting focuses primarily on individuals (Sut-
ton & Hayne, 1997). However, changing business
practices have emphasized the critical role of
groups in a variety of accounting contexts (Ham-
mer & Champy, 1993; Young & Lewis, 1995).
Given the rising prominence and importance of
groups in the accounting profession, it is impor-
tant to investigate the e?ectiveness of various
group participation strategies.
Individual decision-making takes place in three
general phases: (1) gathering relevant information,
(2) generating and evaluating alternatives, and (3)
selecting a course of action (Simon, 1957, 1960).
Analogously, the decision-making phases of indi-
viduals can be applied to groups (Baron, Kerr &
Miller, 1992; Blumberg, 1994). Stasser and Titus
(1985) indicate that group discussion during the
®rst phase of decision-making is a critical partici-
pative activity leading to high quality decisions.
The current study focuses on the ecacy of group
discussion during the requirements de®nition (i.e.
information gathering) phase of a systems devel-
opment project.
When individuals form into a group, each
member typically holds a di?erential set of
knowledge relevant to the problem at hand. A
given member's knowledge set is composed of
multiple items of information. Typically, informa-
tion asymmetry exists among group members
prior to group discussion. The number of group
members who a priori possess a given information
item is referred to as the commonality of the item.
Information items that are known by all group
members are referred to as commonly-held infor-
mation. Items that are known by more than one
but less than all group members are known as
partially-held information. Uniquely-held infor-
mation re¯ects items known by only one group
member.
The ultimate goal of group discussion is for
group members to share their partially-held and
uniquely-held information with other members
(Moreland, Argote & Krishnan, 1996; Stasser &
Titus, 1985, 1987). However, Stasser and Titus
posit that commonly-held information has a
higher probability of being discussed than par-
tially-held information, which has a higher prob-
ability of being discussed than uniquely-held
information. Two important factors that mitigate
this phenomenon, called the common information
sampling bias, are familiarity among group mem-
bers and group member status.
Group members who have gained familiarity
through social interaction, training, and work
experience expend less cognitive e?ort attending to
social conformity cues and place more emphasis
on group performance (Goodman & Leyden,
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 599
1991; Kim, 1996; Levine & Moreland, 1990;
Liang, Moreland & Argote, 1995). As a result,
familiar group members are more likely to
exchange partially-held and uniquely-held infor-
mation because they are less concerned with
acceptance and more oriented toward achievement
(Katzenbach & Smith, 1993; Kerr, MacCoun &
Kramer, 1996). Group members who are peers are
also more likely to share partially-held and
uniquely-held information (Gruenfeld, Mannix,
Williams & Neale, 1996; Hackman & Walton, 1986;
Sanna & Shotland, 1990). Peer group members
freely exchange information with one another
because they are not threatened by unequal power
distribution and evaluation apprehension (Arrow
& McGrath, 1993; Hackman, 1990, 1992; Hol-
lingshead, 1993; Stasser, Stewart & Whittenbaum,
1995; Whittenbaum & Stasser, 1996).
In the current research, the aforementioned
group composition concerns are addressed in the
experimental design in an e?ort to maximize the
ecacy of group discussion. Group members in
this study have a history of working with one
another (familiarity) and they hold equivalent
position levels (peers). The research framework is
presented in the next section.
2.2. Proposed research model
The research model tested in this study is based
on a theoretical framework of accounting infor-
mation systems user participation developed by
Hunton and Price (1994) and supporting research
o?ered by Erez and Arad (1986) and Latham,
Winters and Locke (1994). These studies posit that
participative decision-making strategies (e.g. indi-
vidual versus group) a?ect cognitive and motiva-
tional factors, cognition positively impacts
motivation, and both cognitive and motivational
factors mediate the relationship between partici-
pation and desired user outcomes. These relation-
ships, discussed next, are depicted in Fig. 1.
1
2.2.1. Cognitive factors
In a group discussion environment, one indi-
cator of cognition is the number of discrete ideas
generated by participants (e.g. more ideas suggest
greater cognition) (Haire, Blumberg, Davies &
Kent, 1994). In this study, all subjects record their
ideas regarding ways to improve the existing
accounting information system. In the individual
condition, subjects record their ideas in isolation.
In the group condition, subjects ®rst collectively
discuss relevant issues and then members sepa-
rately record their ideas using the same format as
the individual participants. Because group mem-
bers are likely to share partially-held and
uniquely-held information with one another, we
present the following hypothesis (alternate form):
H1a: The mean number of discrete ideas pro-
vided by group members will be greater
than the mean number of discrete ideas
o?ered by individual participants.
Another important cognitive indicator variable
is the perceived fairness of the decision process
(Hunton & Price, 1994). Process fairness (i.e. pro-
cedural justice) refers to the extent to which deci-
sion making procedures are believed to be in
accordance with accepted and expected norms of
behavior (Thibaut & Walker, 1975). Engaging
a?ected parties in a participation strategy can
enhance their process fairness beliefs (Lind, Kanfer
& Earley, 1990).
1
Cognitive factors, motivational factors, and desired out-
comes are modeled as latent constructs (see Fig. 1). While
latent constructs cannot be measured directly, their presence
can be inferred by observing salient indicator variables. We
recognize that the latent constructs shown in Fig. 1 could be
assessed by observing other variables. For example, one might
represent cognition by assessing knowledge organization,
memory processing capacity, information search strategy, and
other indicator variables outside the scope of the current study.
In structural equation modeling (SEM), path arrows pointing
from the latent construct to indicator variables represent that
the variables are observed manifestations of the underlying
construct. Hence, con®rmatory factor analysis is used in SEM
to determine the extent to which each observed variable loads
on its underlying construct. If arrows point from the measured
variables to the latent construct, SEM will calculate a regres-
sion model, rather than a factor analysis model. The regression
approach to SEM is used when the researcher desires to predict
the value of the latent construct based on the state of measured
variables. In the current study, we are not predicting the value
of the latent construct; rather, we desire to obtain indicators of
the latent construct to facilitate the calculation of path coe-
cients among constructs (see hypotheses H1c, H2c, H3, H4,
and H5 on Fig. 1). In this manner, we can obtain an overall
assessment of the relationships among the latent constructs.
600 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
One way to meaningfully engage users in the
decision process is to provide them with an oppor-
tunity to voice their opinions, preferences, and con-
cerns to decision makers. This form of participation
can provide participants with a sense of control
during the systems development process, particu-
larly if the expression of voice is manifest in the
decision outcome (Leventhal, 1980). Based on the
control-oriented perspective of procedural justice
theory, voice-based participation strategies are
expected to enhance perceptions of process fairness,
as compared to no participation (Lind & Tyler,
1988). In the current study, all subjects are allowed
to participate in information systems development
activities by expressing voice to system designers.
The relational perspective of procedural justice
theory further asserts that when voice is expressed
in a socially interactive environment, as compared
to individuals providing voice in isolation, judg-
ments of process fairness are increased because
voice takes-on symbolic meaning above and
beyond objective control perceptions, such as
Fig. 1. Proposed research model.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 601
enhanced self-esteem, self-con®dence, trust in the
decision maker, and illusory control (Lind &
Tyler, 1988; Lind et al., 1990). As well, expressing
voice in a group setting increases social cognition
of inclusion in the decision process and attach-
ment to the group (Haire et al., 1994). Finally, the
relational perspective posits that voice increases
perceptions of fairness because the collective self-
interest of the group, rather than personal self-
interests, takes precedence (Lind & Tyler). Based
on the previous discussion, the following hypoth-
esis is presented (alternate form):
H1b: When compared to the individual parti-
cipation condition, pre-experiment to
post-experiment gains in process fair-
ness perceptions will be signi®cantly
higher in the group condition.
Based on previously hypothesized gains in the
group condition (H1a and H1b) and the assertion
of Latham et al. (1994) that cognition is enhanced
in group settings, as compared to individual par-
ticipation contexts, the next hypothesis is pre-
sented in alternate form.
2
H1c: The positive relationship between partici-
pation strategy and the latent construct
representing `cognitive factors' will be
signi®cantly greater in the group, as
compared to the individual, condition.
2.2.2. Motivational factors
Researchers in participative decision-making
often frame participation as a means of motivating
a?ected parties to perform target behaviors. While
the relationship between motivation and perfor-
mance has not been clearly demonstrated in prior
research, the link between participation and moti-
vation has been tested and supported in a variety
of studies (see Wagner & Gooding, 1987 for a
review). Two key motivational factors are (a) self-
ecacy beliefs (Locke, 1991) and (b) congruence
between desired and actual participation (Doll &
Torkzadeh, 1989).
Bandura (1986) and Locke (1991) indicate that
self-ecacy is an integral component of motivation.
Self-ecacy perceptions can be aimed at the partici-
pative process (e.g. the extent to which users believe
they can e?ectively contribute to the development of
an information system) and the decision outcome
(e.g. the extent to which participants believe they can
e?ectively use the information system). In this study,
we examine outcome-oriented self-ecacy. In an
information systems setting, Compeau and Higgins
(1995a,b) report a signi®cant, positive relationship
between participation and outcome-oriented self-
ecacy. Lathamet al. (1994) also report increases in
outcome-oriented self-ecacy brought about by
voice-based participation.
The relational perspective of procedural justice
indicates that group participation enhances self-
ecacy levels of group members over their indivi-
dual participants. Providing collective voice to
users in a social setting emphasizes symbolic psy-
chological factors embedded within voice, such as
inclusion, self-esteem, attachment, self-con®dence,
and trust (Lind & Tyler, 1988). These relational
bene®ts coupled with the collective sense of duty,
responsibility, and empowerment arising from
group participation (Jackson, 1992) are expected
to heighten self-ecacy for group members as
compared to individuals. Accordingly, the following
hypothesis is o?ered as follows (alternate form):
2
Hypotheses H1c, H2c, H3, H4, and H5 will be tested using
standardized path coecients produced by SEM, as repre-
sented on Fig. 1. Since latent constructs cannot be directly
measured, SEM ®rst calculates the impact of indicator vari-
ables on their underlying latent constructs through con-
®rmatory factor analysis (i.e. the measurement model). Next,
factor loadings are used to quantify each construct and SEM
employs a series of regression analyses to determine standar-
dized path coecients among model constructs (i.e. the struc-
tural model). Finally, SEM determines overall goodness of ®t
indices that indicate how well the data ®ts the proposed model.
It is important to understand why we did not draw model paths
to and from each indicator variable shown on Fig. 1. First, this
alternative approach would yield a series of path analysis
models (i.e. there would be multiple, independent path models
representing each possible combination of paths among mea-
sured variables). Second, this approach would not control for
covariance between indicator variables within each construct.
Third, it would not be possible to gain a comprehensive picture
of the relationships among the model constructs, as each sepa-
rate path analysis would yield di?erent path coecients and
error terms. Fourth, the SEM approach could not be utilized in
this study due to sample size restrictions and over-identi®cation
problems. Finally, due to the restrictions just noted, it would
not be possible to obtain overall goodness of ®t indices for the
research model; hence, it would be dicult to precisely deter-
mine the comprehensive relationships among latent constructs.
602 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
H2a: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in self-ecacy will be
signi®cantly higher in the group condition.
Individuals desire to participate in decisions that
a?ect their well-being, particularly decisions that
further their self-interest. The strength of desired
participation is positively related to the extent to
which a?ected parties believe the decision out-
come is personally relevant and salient (Alluto &
Belasco, 1972; Lind & Tyler, 1988). When desired
participation exceeds actual (a state of relative
deprivation), a?ected parties become frustrated
and the ecacy of participation is diminished.
When actual participation exceeds desired (a state
of relative saturation), the ecacy of participation
is also negatively compromised, although the
deleterious impact of saturation is not as pro-
nounced as deprivation. Doll and Torkzadeh
(1989) indicate that increased congruence between
desired and actual participation motivates infor-
mation system users to maximize their participa-
tive e?orts. Additionally, participation congruence
is positively related to the formation of con-
sequential user attitudes (e.g. outcome satisfaction
and organizational commitment) and user perfor-
mance (Hunton & Price, 1994). The relational
perspective of procedural justice indicates that
group participation is considered to be more ful-
®lling and personally satisfying than individual
participation; hence, for a?ected parties who
desire more participation, group discussion is
expected to narrow the gap between desired and
actual participation to a greater extent than indi-
vidual participation. Accordingly, the following
hypothesis (alternate form) is presented:
H2b: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in participation con-
gruence will be signi®cantly higher in the
group condition.
Based on the propositions previously set forth
(H2a and H2b) and supporting research ®ndings
from Erez and Arad (1986) indicating that motiva-
tion is signi®cantly higher in group, as compared to
individual, participation strategies, the following
hypothesis is indicated (alternate form):
H2c: The positive relationship between partici-
pation strategy and the latent construct
representing `motivational factors' will be
signi®cantly greater in the group, as com-
pared to the individual, condition.
2.2.3. Relationship retween cognitive and
motivational factors
Hunton and Price (1994), Erez and Arad (1986)
and Latham et al. (1994) suggest a positive rela-
tionship between cognition and motivation. They
reason that a?ected parties gain task relevant
knowledge and receive salient informational cues
(such as increased understanding of alternatives,
constraints, and trade-o?s) when expressing voice.
The appraisal (i.e. thinking) process evoked by
voice-based participation can heighten self-ecacy
beliefs (Bandura, 1986) and improve the con-
gruence between desired and actual participation
(Doll & Torkzadeh, 1989). Theoretical and
empirical support for a stronger association
between cognition and motivation in group versus
individual participation conditions is found in
Latham et al. Based on theory and prior evidence,
we present the following hypothesis (alternate
form):
H3: The positive relationship between the
latent constructs representing `cognitive
factors' and `motivational factors' will
be signi®cantly greater in the group, as
compared to the individual, condition.
2.2.4. Relationship between cognitive factors and
desired outcomes
Cognitive factors can be linked to one's choice
of action and the extent to which that action is
deemed successful (Locke & Latham, 1990). Cog-
nitions are translated into action when individuals
make a volitional choice to consciously appraise
their a priori held knowledge and beliefs. The
appraisal process integrates an individual's values,
needs, desires, and aspirations into an action plan.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 603
As a result, cognitive factors can be directly linked
to attitudinal and behavioral outcome variables. In
an experimental setting, Latham et al. (1994) found
that cognition (i.e. task strategy quality) was sig-
ni®cantly, positively related to performance. Their
study also indicated a stronger relationship
between cognition and performance in group, as
compared to individual, conditions. Accordingly,
we anticipate the following (alternate form):
H4: The positive relationship between the
latent constructs representing `cognitive
factors' and `desired outcomes' will be
signi®cantly greater in the group, as
compared to the individual, condition.
2.2.5. Relationship between motivational factors
and desired outcomes
Meta-analytic reviews of participation studies
have not consistently supported the link between
motivation and consequential outcomes. Locke
(1991) suggests that equivocal results from prior
research are likely due to poor surrogate indica-
tors of motivation. He posits that self-ecacy and
participation congruence (based on expectancy
theory) are valid motivational factors. Supporting
Locke's assertions, Latham et al. (1994) and
Compeau and Higgins (1995a,b) report a sig-
ni®cant, positive relationship between outcome-
oriented self-ecacy and consequential attitudes
and behavior. Doll and Torkzadeh (1989) indicate
a signi®cant link between participation con-
gruence, user satisfaction, and intention to use an
information system. Erez and Arad (1986) and the
relational perspective of procedural justice suggest
a greater relationship between motivation and
performance in group, as compared to individual,
participation conditions. Based on the theory and
evidence just presented, the ®fth hypothesis is
o?ered as follows (alternate form):
H5: The positive relationship between the
latent constructs representing `motiva-
tional factors' and `desired outcomes' will
be signi®cantly greater in the group, as
compared to the individual, condition.
2.2.6. Desired outcomes
The indicator variables examined in this study
representing the latent construct `desired out-
comes' are outcome satisfaction, organizational
commitment, and user performance. These parti-
cular indicator variables were selected primarily
because the management of the state agency who
sponsored this ®eld experiment desired to measure
changes in these factors brought about by the
group and individual user participation strategies.
Outcome satisfaction concerns the users' a?ective
reaction toward the target information system. Doll
and Torkzadeh (1988) examined the issue of user
satisfaction. The ®ve dimensions of user satisfac-
tion they developed and tested were system accu-
racy, ease of use, system timeliness, information
content, and information format. In this study, we
assess the subjects' pre- and post-experiment levels
of user satisfaction along these ®ve dimensions.
Recently, organizational commitment (OC) has
been the focus of research in organizational behavior
(e.g. Aranya, Pollock &Amernic, 1981; Cohen, 1996;
Dunham, Grube & Castaneda, 1994; Jaros, 1995).
OC is de®ned as the strength of an individual's iden-
ti®cation with and involvement in a particular
organization. OC is usually characterized by the
following factors: ``a strong belief in and accep-
tance of the organization's goals and values; a
willingness to exert considerable e?ort on behalf
of the organization; and a de®nite desire to main-
tain organizational membership'' (Porter, Steers,
Mowday & Boulian, 1974, p. 604). Allen and
Meyer (1990) identi®ed three components of com-
mitment: a?ective, normative and continuance.
The a?ective component of OC refers to the
employee's emotional attachment to, identi®cation
with, and involvement in the organization. In this
study, we assess the a?ective component of OC.
3
Locke and Latham (1990) and Locke (1991)
indicate that prior research in participation
demonstrates a weak association between partici-
pative activities and human behavior. Locke and
Latham and Latham and Lee (1986) suggest that
3
Management of the sponsoring State agency and the
researchers were concerned about the length of the pre- and
post-experiment survey instruments. Hence, management chose
to measure only the a?ective dimension of OC.
604 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
research conducted in ®eld settings might help to
provide a more clear link between participative
decision making and consequential behaviors,
especially where the decision outcome and perfor-
mance measures are closely aligned. In this ®eld
experiment, we obtained pre- and post-experiment
objective measures of sales order and customer
billing data input quality. These performance
measures are directly related to the decision out-
come (i.e.the revamped accounting information
system). Based on the theoretical development of
hypotheses H1 through H5, we suggest the ®nal
set of hypotheses (alternate form):
H6a: When compared to the individual partici-
pation condition, pre-experiment to post-
expe-riment gains in outcome satisfaction
will be signi®cantly higher in the group
condition.
H6b: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in organizational com-
mitment will be signi®cantly higher in the
group condition.
H6c: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in sales order data
input quality will be signi®cantly higher
in the group condition.
H6d: When compared to the individual partici-
pation condition, pre-experiment to post-
experiment gains in customer billing data
input quality will be signi®cantly higher
in the group condition.
3. Research method
4
3.1. Scenario
A large state agency permitted the researchers to
conduct a ®eld experiment investigating participa-
tion strategies for developing computerized
accounting information system applications.
5
The
agency is comprised of 46 ®eld oces located
throughout the state. Field oces have little or no
communication with each other, as they are
autonomous in all operational and administrative
respects. Field oce managers report directly to
the agency's central oce located in the capital
city. Computerized accounting information sys-
tems have evolved at ®eld oces with little gui-
dance from the central oce. Accounting
applications at ®eld locations, on the whole, were
poorly designed and sparsely documented. Custo-
mers throughout the state complained about high
levels of order entry and billing errors. Users at
®eld oces realized their accounting systems were
inadequate, as they had expressed their dis-
satisfaction to agency managers and state audi-
tors. Based on user feedback and ®ndings of
recent annual audits, state auditors mandated the
agency to redesign its accounting information sys-
tems in an e?ort to reduce the unacceptably high
data input error rates and correct a multitude of
additional problems. Accordingly, the agency
instituted a ®ve-year plan to revamp all of its
computerized accounting and management infor-
mation systems. The ®rst project was to develop
standardized revenue cycle applications (i.e. sales
order, cash collection, accounts receivable, and
customer billing) at the central information sys-
tems department.
The agency director believed that user input
throughout the development project was critical to
the ultimate success of the system. However, he
was unsure whether users should provide indivi-
dual input to system designers or if it would be
more e?ective for users to discuss their concerns
and preferences in small group environments
before submitting their suggestions to system
designers. Since the director had many more pro-
jects planned in the future, he believed that this
®rst initiative presented an opportunity to explore
the most e?ective participation strategy. Conse-
quently, the agency director asked the researchers
to design a ®eld experiment to assess which strat-
egy, individual or group, yields higher levels of
desired outcomes. In particular, he insisted that
the most critical outcome of interest was the
quality of data input regarding sales order and
customer billing transactions.
4
The research scenario and procedure are the same as
described in Hunton and Gibson (1998) as the data for that
study were collected simultaneously with the current study.
5
The researchers were paid for out-of-pocket expenses only.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 605
3.2. Procedure
At the beginning of the ®rst month of the study
(m
1
), a pre-experiment survey was distributed to
all accounting and sales order clerks throughout
the agency via the agency's intranet. The
researchers initiated the surveys and ensured par-
ticipants that individual responses would be held
in strict con®dence and that only summary mea-
sures would be shown to management. A custo-
mized program presented survey items randomly
to each respondent (to preclude a possible order
e?ect), restricted participants from viewing prior
questions (to preclude respondents from looking
back to see how they answered a similar survey
item and responding in the same way), and pro-
hibited subjects from forwarding results to anyone
other than the originating source, i.e. the
researchers (to ensure con®dentiality).
During the ®rst and second experimental
months (m
1
and m
2
), subjects participated in
developing the revenue cycle applications. An
initial memo was sent from the director to each
participant explaining that the agency's informa-
tion system design group (ISDG) was developing
standardized revenue cycle applications to replace
existing ®eld oce applications. The director
expressed that he and agency management felt it
was critically important for application users to be
involved in the systems design process. The memo
explained that for the next ®ve consecutive Friday
mornings, ®eld oces would not open for business
at their regular time of 8:00 a.m., instead they
would open at 10:00 a.m. Employees would still
report for work at 8:00 a.m., however, they would
spend these two hours expressing their opinions,
concerns, preferences, and suggestions to the
ISDG regarding the applications under develop-
ment. The director made it clear that while
employees could not reasonably expect all of their
individual preferences and suggestions to be
incorporated into the new applications, every
e?ort would be made to include as many of their
recommendations as possible in the new systems.
The ISDG and participants communicated via
the agency's intranet system. Each Friday morn-
ing, participants received a standardized ques-
tionnaire from the ISDG soliciting input regarding
certain aspects of the revenue cycle applications
under development (i.e. sales order, cash collec-
tions, accounts receivable, customer billing). The
questionnaires contained open-ended questions
pertaining to the topic of the week. The ®ve weekly
topics focused on system accuracy, ease of use, sys-
tem timeliness, information content, and report for-
mats. Two participation conditions (individual and
group) were randomized to ®eld oces, resulting in
23 oces assigned to each treatment.
In the individual condition, subjects were asked
to remain at their workstations and answer the e-
mail questionnaire from the ISDG. Field oce
supervisors were present to ensure that interaction
between subjects did not occur during the answer
period. The questionnaire was designed so it could
be completed in about 1 hour. After respondents
completed the instrument, they were instructed to
remain at their desk and work on agency matters
until the oce opened at 10:00 a.m.
In the group condition, the accounting and sales
order clerks met in a conference room, where they
received a copy of the weekly questionnaire and dis-
cussed relevant issues for 1 h. There were no super-
visors present during the meetings inorder tofacilitate
the free exchange of information among clerks; how-
ever, supervisors ensured that all clerks attended the
meetings. The groups were not facilitated by outside
observers. After discussion, group members returned
to their respective workstations and answered the e-
mail questionnaire fromthe ISDG.
After the ®ve weekly meetings concluded, the
software was developed and tested during the
third (m
3
), fourth (m
4
) and ®fth (m
5
) experimental
months. Members of the ISDG conducted eight
hours of standardized training during the sixth
month (m
6
) at each ®eld oce location. At the end
of the sixth month, the revenue cycle applications
were implemented. The researchers administered a
post-experiment survey at the end of the 12th
month (m
12
). Finally post-experiment measures of
performance were tracked for 12 months beyond
installation, concluding at the end of the 18th
experimental month (m
18
).
While it is dicult to attain a high degree of
internal validity in ®eld experimentation, several
controls were incorporated into the current study.
Field oces were randomized to treatment condi-
606 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
tions. Neither ISDG members nor ®eld oce
supervisors were aware of di?erential treatments
across ®eld oces. The following ®ve factors were
held constant between treatment conditions: pre-
and post-experiment survey items; weekly topic
and related questionnaire; time allocated for par-
ticipative activities; training sessions; and decision
outcome (i.e. the ®nal revenue cycle applications).
The procedures by which individuals and group
members engaged in participative activities were
held constant within each treatment condition.
Finally, there were no agency-sponsored meetings
or discussions during the experimental time frame
where representatives from di?erent ®eld oces
could interact with one another.
3.3. Measured variables
Survey items were measured using seven-point,
Likert-scales with semantic anchors on both ends.
Some items were reversed scored, however, all
were converted such that scores of 1=low and
7=high. All variables were assessed using two or
more items. Reliability estimates for two-item
measures were obtained using Pearson's product
moment correlation (r), and reliability estimates
for three or more item measures were obtained
using Cronbach's alpha () (Cronbach, Gleser,
Nanda & Rajaratnam, 1972). Related item mea-
sures were collapsed (i.e. averaged) into single
variable indices.
6
3.4.1. Cognitive factors
The ®rst cognitive factor, number of ideas gener-
ated by clerks throughout the ®ve-week participa-
tionperiod, was ascertained by examining comments
and suggestions provided by all participants. To
accomplish this, we hired three graduate doctoral
students, hereafter referred to as graders, to analyze
the respondents' weekly questionnaires. The graders
®rst independently examined the ®ve questionnaires
submitted by each respondent to identify and de®ne
a common set of 102 discrete ideas. The graders'
initial estimated inter-rater reliability
7
was 0.87.
After reviewing all items, the graders eventually
agreed on a common set of 97 discrete ideas. Next,
the graders independently counted the number of
discrete ideas provided by each subject. Average
inter-rater reliability across all subjects was 0.93.
After reviewing and reassessing disputed items, the
graders reached full agreement on all subjects.
Three survey items assessed the extent to which
subjects believed the process of participating in
developing the revenue cycle applications was fair
(pre-experiment ? 0:95; post-experiment ?0:91).
The fairness questions were adapted from Lind et
al. (1990). For example, one of the three items asked
``What do you think about the way in which your
agency allows you to participate in developing,
modifying, and/or acquiring accounting software
applications?'' (1= Very Unfair, 7=Very Fair).
The three items were averaged to form two process
fairness indicesÐone from the pre-experiment sur-
vey and one from the post-experiment survey.
3.4.2. Motivational factors
The extent of congruence between desired and
actual participation was measured using survey
6
Measured variables were collapsed into indices represent-
ing `average scores' to facilitate the SEM analytic approach.
The number of survey items used to assess measured variables
ranged from 2 to 16 (i.e., for the participation congruence
measure, there were eight `desired' and eight `actual' items).
Further, there were pre-experiment and post-experiment mea-
sures of each variable (except for `discrete ideas'), e?ectively
doubling the number of survey items for each variable. Because
inter-item reliability estimates were relatively high for all mea-
sured variables, collapsing the items into average score indices
does not result in a signi®cant loss of information. Addition-
ally, the sample size in this study is minimal for SEM modeling
and including all pre- and post-experiment measured items in
the model gives rise to power concerns, degree of freedom
issues, and over-identi®cation problems. The researchers ran
alternative tests using summed indices (rather than averaged),
and research ®ndings and interpretations were not signi®cantly
changed for any of the hypotheses.
7
Estimated inter-rater reliability was calculated using
Cronbach's alpha, as described in Kerlinger (1986). Essentially,
the complete set of discrete ideas initially identi®ed by the
raters was ®rst determined (n=102). Next, the 102 ideas were
randomized, to preclude an implied importance rating, and
numbered from 1 to 102. Third, each idea in the set that was
identi®ed by a particular grader was coded as one (1) and ideas
that were not in the initial set were coded as zero (0). After all
three raters' ideas were coded accordingly, a Cronbach's alpha
estimate of reliability was conducted on the assigned values for
all three graders simultaneously. All other inter-rater reliability
estimates reported in this paper were calculated using the same
procedure.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 607
items adapted from Doll and Torkzadeh (1989).
Participants responded to eight items asking the
extent to which they desire to participate in var-
ious activities regarding the development, mod-
i®cation, and/or acquisition of accounting
software applications (pre-experiment ? 0:89;
post-experiment ? 0:93). Eight identical survey
items were used to assess the extent to which par-
ticipants actually do participate in these same
activities (pre-experiment ? 0:94; post-experi-
ment ? 0:88). Participation congruence indices
were determined as follows: variable indices were
calculated by averaging the eight survey items
representing each of the actual and desired parti-
cipation variables; the actual index was subtracted
from the desired index; and the di?erence between
actual and desired indices formed pre- and post-
experiment participation congruence indices.
Three survey items, adapted from Bandura
(1986), examined participants' self-ecacy beliefs
that they could e?ectively use the revenue cycle
applications. For example, one item stated ``I
believe that I can e?ectively use the revenue cycle
applications (sales order, cash collection, accounts
receivable, and customer billing) to process custo-
mer sales order, billing, and payment information.''
(1=Strongly Disagree, 7=Strongly Agree). The
self-ecacy items were averaged to form two
indices, one from the pre-experiment survey
( ? 0:92) and one from the post-experiment sur-
vey ( ? 0:89).
3.4.3. Desired outcomes
Outcome satisfaction was measured using Doll
and Torkzadeh's (1988) end-user computing satis-
faction scale. The items adapted for this study
assessed how respondents felt about ®ve areas of
the revenue cycle applications: system accuracy (2
items); ease of use (2 items); system timeliness (2
items); information content (4 items); and infor-
mation format (2 items). Inter-item reliability esti-
mates for all 12 items were high (pre-experiment
? 0:86; post-experiment ? 0:92) and they were
averaged to form pre- and post-experiment indices.
Organizational commitment was assessed using
the a?ective dimension of the Meyer, Allen and
Smith (1993) scale. For example, one of the six
survey items stated, ``I would be very happy to
spend the rest of my career with this agency''
(1=strongly disagree, 7=strongly agree). Inter-
item reliability was high (pre-experiment ? 0:96;
post-experiment ? 0:87) and the six items were
averaged into pre- and post-experiment organiza-
tional commitment indices.
Human performance was measured by calculat-
ing the quality of sales order and customer billing
data entry transactions of each clerk who partici-
pated in the study. Each transaction type required
a di?erent number of ®elds in which the clerk must
enter data. For example, the pre-development
sales order system required users to enter data into
32 ®elds when adding a new customer and 29 ®elds
when entering a sales order transaction. Assume
that a clerk made errors in 44 ®elds while adding
20 new customers and made errors in 73 ®elds
while entering 50 sales order transactions during a
given month. The clerk's monthly data input
quality rate in this case would be 0.94 [(2090-117)/
2090]. The researchers accessed all monthly trans-
action ®les and computed the data input quality
rates. However, due to con®dentiality concerns,
the state agency would not allow the researchers
to report the total number of input ®elds, errors or
transaction types entered each month.
3.4.4. Gain scores
Gain scores were computed at the individual
level for all variables. Psychological gain scores
represent the di?erence between post-experiment
and pre-experiment variable indices. To compute
gain scores for sales order and customer billing
transactions, we subtracted the average data input
quality rate for the ®nal 6 months of the experi-
ment (m
12
±m
18
) from the average data input
quality rate for the 6 month period prior to
implementing the new revenue cycle applications
(m
1
±m
6
).
8
Data input quality rates for the ®rst 6
months after implementation (m
7
±m
11
) were not
included in the analysis due to observed learning
e?ects.
8
Using the 6 month pre-implementation period as the base-
line measure, the gain scores yield a relative change in data
input quality from pre- to post-implementation.
608 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
4. Results
4.1. Sample characteristics
There were a total of 172 participants in the
individual condition and 174 in the group condi-
tion. Chi-square analyses were conducted to assess
whether proportional di?erences existed between
the individual and group modes based on gender,
education level, ethnic origin, job category (i.e. sales
and accounting), and supervisory performance eva-
luation. No signi®cant di?erences were noted, as
all p-values exceeded 0.24. Additionally, t-tests
were performed to determine the extent to which
mean di?erences existed between the individual
and group modes based on age, months with the
agency, months with the local oce, months in
current position, and months in current grade. As
shown in Table 1, no signi®cant di?erences were
observed.
There were 46 ®eld oces involved in the
experiment. The individual and group manip-
ulations were randomly assigned to 23 ®eld
oces each. At the beginning of the experiment,
394 subjects completed the pre-experiment sur-
vey. When the post-experiment survey was
administered 12 months later, 48 (12%) of the
initial 394 subjects were no longer employed by
the state agency. The human resource manager
of the agency reported a 13.6% turnover for
the same 12 months of the prior year; therefore,
the 12% attrition rate was not considered
abnormal. Of the 48 subjects who left the
agency, 22 were in the individual and 26 were
in the group conditions. We conducted statis-
tical testing to determine the extent to which
demographic factors of the 48 employees who
left the agency were signi®cantly di?erent from
the remaining sample. No di?erences were
found, as the p-values for all tests exceeded 0.5.
The psychological and performance gains
reported in this study re¯ect only the remaining
346 participants.
A total of 60 credit/collection (29 individual, 31
group), 49 cash receipts (24 individual, 25 group),
73 customer billing (37 individual, 36 group), and
164 sales order (82 individual, 82 group) clerks
participated in the study. A chi-square analysis
was conducted to determine whether there were
proportional di?erences in the types clerks
between the two treatment conditions and none
were found (X
2
=0.109, p-value=0.9916). In the
individual (group) conditions, the mean number of
subjects at each ®eld site were 7.47 (7.56).
It should be noted that we did not perform self-
reported manipulation checks of each participation
condition due to the tight controls surrounding the
individual and group treatments. Speci®cally, on-
site supervisors made sure that individuals
received the ISDG survey each Friday morning at
8:00 a.m. and answered all questions in isolation.
In the group condition, supervisors ensured that
group members received the ISDG survey at 8:00
a.m., discussed the survey items in a group setting
for one hour, and then responded to the questions
in isolation. Tests of the research model and
hypotheses are next presented.
4.2. Research model testing
Pre- to post-experiment gains were analyzed
using EQS, a SEM program developed by Bentler
Table 1
Sample characteristics
All subjects Individual mode Group mode t Statistic
a
p Value
Age (years) 36.32 37.01 35.63 0.90 0.37
Months with agency 108.84 109.85 107.83 0.12 0.90
Months with local oce 105.75 106.14 105.37 0.05 0.95
Months in current position 103.09 103.55 102.63 0.06 0.95
Months in current grade 67.58 68.59 66.56 0.22 0.83
a
Based on t-tests of individual versus group means.
J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618 609
(1989).
9
In this study, the SEM model was
designed to test for signi®cant di?erences in incre-
mental changes in model paths from individual to
group conditions following the procedure estab-
lished by Sorbom (1978). Participation was coded
as 0 (individual) and 1 (group) without measure-
ment error. Model paths from participation strat-
egy to cognition and motivation were constrained
at zero (0) in the individual condition. Paths from
cognition to motivation, cognition to outcomes,
and motivation to outcomes were constrained to
be equal to the individual condition. Accordingly,
standardized coecients along model paths represent
incremental di?erences in the group, as compared to
the individual, condition. The statistical sig-
ni®cance of incremental path parameters was tes-
ted using the multivariate LaGrange multiplier
test (Aitchison & Silvey, 1958). Since there was no
theoretical reason to expect di?erential factor
loadings between treatment conditions from mea-
sured variables to latent constructs, all factor
loadings were unconstrained.
10
Incremental stan-
dardized path coecients, results of multivariate
LaGrange multiplier tests on path coecients, and
factor loadings are shown in Fig. 2.
Since there is no generally accepted measure of
overall model goodness of ®t, leading researchers
recommend that multiple criteria be evaluated to
determine how well the data ®ts the model
(Breckler, 1990; Wheaton, 1987). In this study, we
use ®ve goodness of ®t criteria suggested by Ben-
tler (1990) and Schumacker and Lomax (1996).
The ®rst ®t index is the chi-square statistic. This
index compares the proposed model to the general
alternative where all variables are correlated. Sta-
tistically signi®cant values re¯ect a poor model ®t.
However, in large samples the chi-square statistic
is almost always signi®cant because it is a direct
function of sample size. The second and third
goodness of ®t indicators are the normed ®t index
(NFI) and non-normed ®t index (NNFI) as pro-
posed by Bentler and Bonett (1980). The fourth
criterion is the Comparative Fit Index (CFI).
Bentler (1990) indicates that the CFI works well at
all sample sizes and is more stable than the NFI
and NNFI because it is less sensitive to sampling
error. According to Bentler (1990), NFI, NNFI,
and CFI values of 0.90 or greater indicate a good
®tting model. Finally, we report the root mean square
error approximation (RMSEA) (Schumacker &
Lomax, 1996). RMSEA provides an approximation
of overall variance unaccounted for in the model and
values of 0.05 or less are desirable.
Bentler (1990) suggests that, rather than focus-
ing on one ®t index, researchers should look at the
collective evidence generated from multiple ®t
indices to determine the adequacy of model ®t.
Based on the relative size of the goodness of ®t
indicators obtained in this study (see Fig. 2), the
data appears to satisfactorily ®t the research
model.
4.3. Means testing
While the SEM model shown includes all 346
participants, there is valid concern that subjects in
the group condition were not independent from
one another because they discussed relevant issues
before responding to the survey instruments.
Accordingly, one could argue that, conservatively,
the sample size should be 46 (23 ®eld sites per treat-
ment condition) rather than 346. Unfortunately, a
sample size of 46 is too small for SEM calculations.
9
We also ran our analyses using CALIS, LISREL, LIS-
COMP, and SEPATH. While the factor loadings and path
coecients change slightly due to varying estimation processes,
overall ®t indices are satisfactory in all analyses. Hence, our
®ndings are not sensitive to the particular SEM application
used. In preliminary tests, we discovered that our data was not
multivariate normal, as is the case with most data of this nat-
ure. We tested normality using the Kolmogorov±Smirnov test.
Results indicated that these measures failed to approximate a
normal distribution, therefore we used the generally weighted
least squares (GWLS) method to estimate model structural
equation parameters and goodness of ®t indices. The GWLS is
asymptotically distribution free, whereas maximum likelihood
(ML) and generalized least squares (GLS) estimators assume
normally distributed data. Andersen and Gerbing (1988) sug-
gest that EQS is the best modeling approach when data is not
multivariate normal due to its elliptical estimation assumption.
10
Factor loadings determined by SEM are shown on Fig. 2.
For example, the coecients from `cognitive factors' to discrete
ideas (0.82) and process fairness (0.90) suggest that these two
indicator variables load high on and are representative of the
underlying factor, or latent construct. Factor loadings greater
than 0.30 represent a strong association between an indicator
variable and its underlying construct (Schumacker & Lomas,
1996).
610 J.E. Hunton, D. Gibson / Accounting, Organizations and Society 24 (1999) 597±618
As a way to deal with this concern, the research-
ers compared the results of psychological assess-
ments (i.e. process fairness, self-ecacy,
congruence, satisfaction, and commitment) and
objective performance measures (i.e. sales order
quality and customer billing quality) for both
treatment conditions at the ®eld oce level
(N=23 per treatment condition). Within the
individual and group conditions, post-experiment
means were signi®cantly greater than pre-experi-
ment means (p-value