^Academy of Management /oumai

1986, Vol, 29, No, 3, 536-561.

MATRIX ORGANIZATION: A SOCIAL EXPERIMENT

WILLIAM F. JOYCE
Dartmouth College

A S€>cial experiment was used to assess the effects of matrix structure on

organizational processes, role perceptions, and work attitudes. Tests of
hypotheses employed a nonequivalent coDtrol group design with statisti-
cal procedures simulating a complementary, quasi-experimental design
termed treatment-effect correlations. Implementing matrix structure
caused predicted increases in the quantity of communications, but
decreased the quality of these communications. Negative effects on rele-
vant role perceptions, work attitudes, and coordination also occurred.
Implications of these findings are discussed in relation to the literature
on matrix organization, and suggestions for further research are
advanced.

This study presents evidence concerning the effects of matrix structure on

organizational processes, role perceptions, and work attitudes. Previous
researchers have frequently hypothesized that matrix organization increases
capacity for information handling and decision making within organizations
by establishing formal, lateral channels of communications that complement
and supplement existing hierarchical channels (Davis & Lawrence, 1977;
Galhraith, 1973; Hrebiniak & Joyce, 1984), Creating these channels violates
classical principles of management concerning unity of command and scalar
chains of authority and also complicates decisions concerning delegation by
making responsibilities unclear or ambiguous. Studies have suggested that
these violations cause increased role conflict and ambiguity and produce
negative effects on work attitudes like job satisfaction and involvement (Butler,
1973; Reeser, 1969),
Although the effects of matrix structure have been consistently described
in this way, alternative effects could easily be hypothesized. Delbecq and
Filley (1974) suggested that matrix structures provide opportunities for
"advanced work and coUegial interaction with a prestigeful set of colleagues"
and that "prestige often accrues to the project and to personnel associated
with it" (1974: 67). Such effects would not be expected to decrease job satis-
faction and involvement. Empirical evidence that could serve to resolve
such issues is sparse, despite frequent references to matrix structures in the
academic literature and their apparently widespread use.

The author wishes to acknowledge the help of a number of individuals who contributed to
this paper in several ways, especially Douglas Garber. Lawrence Hrebiniak, Jeffrey Kerr, Peter
Lorenzi, John Slocum, and the anonymous reviewers.

536
1986 Joyce 537

Evan (1971) classified research strategies into case studies, field surveys,
and laboratory and field experiments. When applied to research concerning
matrix structures, this classification reveals that, although several case
studies and field surveys have yielded interesting and provocative results, no
organizational experiments have been conducted to test propositions con-
cerning matrix structure. Evan failed to find a single instance in which
investigation of an organizational problem proceeded from case study through
field study to organizational experiment. He took this not as an indication
that the process lacked "feasibility or utility, but rather as an indication of a
glaring deficiency in the field" (1971: 2). This article reports the results of a
field experiment conducted to assess the effects of implementing a matrix
structure on organizational processes, role perceptions, and work attitudes.

REVIEW AND PROPOSmONS

Organizational Processes
Calbraith (1973) argued that matrix organization represents one of several
alternatives for dealing with the requirements for information processing
imposed by high levels of task uncertainty. When these requirements exceed
the capability of basic bureaucratic structures, one response is to increase an
organization's capacity for information processing by creating new, formal,
lateral channels of communication complementing existing formal, hierar-
chical ones. This choice allows formal communications to take place not
only through an existing hierarchy and chain of command, but also through
lateral channels.
Occupants of key lateral roles in these structures, such as program, pro-
duct, project, or business managers, have responsibility and authority for
lateral decision making. Although variations in the degree of lateral re-
sponsibility and authority occur in practice, matrix arrangements generally
imply a rough equivalence in power between hierarchical and lateral di-
mensions of the structures (Davis & Lawrence, 1977; Delbecq & Filley, 1974;
Galbraith, 1973; Pfeffer, 1981).
The first intended consequence of implementing a matrix structure is to
increase the quantity of communication by adding formal but nonhierarchical
channels, thus augmenting an organization's basic capacity for information
processing. However, a matrix structure should influence the quality or style
of communication as well. Davis and Lawrence (1977) noted, for example.
that
project or business managers [in matrix organizationsl do not
unilaterally decide. They manage the decision process so that
differences are aired and trade-offs are made in the interests of
the whole. They must use their knowledge, competence, relation-
ships, force of personality, and skills in group management to
get people to do what is still necessary for project success (1978:
87).
Consequently, most researchers have argued that a matrix structure should
increase the participative nature of formal communications. Delbecq and
538 Academy of Management Journal September

Filley (1974) also argued that "because of the developmental nature of the
work of the project group, a participative colleague relationship is signifi-
cant for problem-solving success" [1974: 59). Similar propositions are com-
mon in the literature (Galbraith, 1973; Lawrence & Lorsch, 1967).
Matrix managers are likely to employ a more participative style than do
other managers and may also employ directive or forcing behavior some-
what more. Delbecq and Filley noted that the "function of the project man-
ager is to keep the eye of the group firmly focused on the project task and
make sure that problems and difficulties are confronted rather than smoothed
over. This may even involve some forcing behavior" (1974: 59). Similar
implications may also be found in Lawrence and Lorsch (1967). In summariz-
ing their findings regarding factors important to obtaining high levels of
integration, they wrote: "All of these data taken together suggest that while
heavy reliance on confrontation is essential [for obtaining coordination], it
may also be important... to have a back up mode that relies on some forcing
. . . . Forcing means that the managers will at least reach some decision"
(1967: 78). Commenting on matrix structures in particular, Davis and Law-
rence (1977) wrote: "In an effective matrix organization, confrontation is the
primary mode of conflict management, and forcing or compromise the sec-
ond most frequently used mode. Forcing, although reducing trust, does at
least assure that some action is taken" (1977: 105). Burns and Stalker (1961)
also presented an argument with a similar conclusion (1961: chap. 7).
Thus, the literature suggests that implementing a matrix structure should
improve coordination by increasing the amount and frequency of formal
communication through creating and legitimating new, formal, lateral chan-
nels of communication. As this happens, previously informal communica-
tion concerning task-related matters is legitimized and carried in these new
channels. A decrease in the frequency of informal task-related communica-
tion should accompany the increase in formal communication as informal
channels become redundant. In addition, the participative and directive
quality of formal communications within an organization should increase as
workers engage in confrontation and problem solving to balance and respond
to multiple supervisors and conflicting objectives.

Role Perceptions and Work Attitudes ,'

The effects on organizational processes discussed above should affect
individuals' role perceptions and work attitudes. Increases in role conflict
are frequently hypothesized to result from matrix arrangements (Galbraith,
1973; Hrebiniak & Joyce, 1984; Pfeffer, 1981). The increased demands of dual
supervision in matrix structures contribute to role overload by causing con-
flicts between multiple role expectations and individuals' abilities to satisfy
such expectations. In addition, the demands of lateral and hierarchical influ-
ence structures may be inconsistent, generating a type of role strain called
intersender role conflict. These conditions combine to produce a generalized
state of role conflict (Rizzo, House, & Lirtzman, 1970),
1986 Joyce 539

TABLE 1
Propositions Concerning the Effects of Implementing a Matrix Structure
on Organizational Processes, Role Perceptions, and Work Attitudes
Predicted Effects
Variables of Matrix Structure
Organizational processes
Frequency of formal communication Increases
Frequency of informal communication Decreases
Amount of formal communication Increases
Participative quality of communication Increases
Directive quality of communication "• Increases
Coordination Increases
Role perceptions - •
Role conflict • * Increases
Role ambiguity • - Increases
Work attitudes
Satisfaction with work Decreases
Satisfaction with supervision Decreases
Satisfaction with co-workers Decreases
Satisfaction with pay Decreases
Satisfaction with promotions Decreases
Job involvement Decreases

Butler (1973) suggested that such conflict is especially likely in matrix

structures owing to (1) requirements for joint decision making in situations
of high task interdependence. (2) competition and bargaining for scarce human
and material resources, [3) obscurely defined power and authority relation-
ships, and (4) differences in managerial values and objectives, Pfeffer (1981)
argued that a certain amount of conflict is designed into matrix structures,
presumably to create confrontation; if managed productively, this process
results in higher levels of coordination and an ability to make complex
decisions more effectively than do simple bureaucratic procedures (Lawrence
& Lorsch, 1967).
Burns and Stalker [1961) argued that conflict results from change. More
conflict arises in matrix structures because, as Delbecq and Filley proposed,
the "essential feature of the project participant position is its temporary
character" (1974: 63). Because projects end and new ones begin, a recurring
process of change is inherent in matrix structures. Adopting a matrix struc-
ture also increases role ambiguity. In such arrangements, responsibility for
task execution rests with both lateral and hierarchical managers, making
explicit definition of their respective responsibilities difficult. Reeser (1969)
found that engineers with varying degrees of exposure to matrix manage-
ment showed greater role ambiguity, greater dissatisfaction with personnel
practices and supervision, and less loyalty to their organizations than engi-
neers working in functional structures. Reeser's results suggest that matrix
structures affect work attitudes to the extent that such structures create
negative role perceptions in the forms of role conflict and ambiguity, Rizzo
and colleagues (1970) reported significant negative correlations between these
540 Academy of Management Journal , September

types of role strain and both job satisfaction and involvement, a finding
consistent with Reeser's results. Table 1 summarizes the propositions concern-
ing the effects of matrix structure on organizational processes, role percep-
tions, and work attitudes,

STRATEGY OF INVESTIGATION
As the discussion in the preceding section illustrates, although some
propositions concerning the effects of matrix structure tend to predominate
in the literature, others could logically be advanced. The lack of relevant
experimental evidence perpetuates this ambiguity.
A related question concerns whether such experimental research should
be conducted in laboratories, to maximize precision and control, or whether
it should be conducted in the field, where some precision and control may
be sacrificed to realism, generalizability, and potential sharing of results.
These conflicting objectives can, perhaps, best be served through field experi-
mentation using social experiments. A social experiment "is a type of manipu-
lative experiment where the changes are produced, not by the scientist's
intervention, but by that of the policy maker or practitioner" (Kaplan, 1964:
164), In a social experiment, a policy maker produces the manipulation and
a researcher studies its effects. Social experiments offer a number of signifi-
cant advantages for improving understanding of matrix structures. Their
results are likely to be generalizable because they minimize scale and thresh-
old effects (Kaplan, 1964: 164) and because the phenomena of interest are
studied where they ordinarily occur. Campbell (1971) noted that, for experi-
ments on administrative principles, social experiments provide "a basis of
experimentation that is closer to home, where the extrapolation is smaller,
and where the likelihood of valid extrapolation is greater" (1971: 170).
Social experiments also minimize artificiality and the possibility that the
process of experimentation itself is producing the observed results. A research
situation and an experimenter's behavior can lead participants to respond
and behave differently than they would otherwise (King, 1974; Milgram,
1965; Orne, 1962; Ome & Evans, 1965), Participants in social experiments
"may not care that they are subjects because far stronger motivations have
been tapped" (Kaplan, 1964: 168),
For these reasons, social experiments provide a useful and desirable
method for studying the effects of matrix structures on organizational
processes, role perceptions, and work attitudes. Every method has a drawback,
however; for social experiments, it is frequently the lack of a comparable
control group. The next section describes how this potentially serious prob-
lem was overcome with a design using a nonequivalent control group to
bolster the internal validity of this study's conclusions.
Experimental Design
In social experiments, researchers do not control the administration of the
experimental stimuli. Lacking this control, they require an alternative, which
1986 Joyce 541

may take the form of a quasi-experimental design whose purpose is "to

introduce the logic of experimentation when control over the scheduling of
experimental stimuli is lacking" (Roos, 1973). For social experiments, one
attractive logic involves bracketing an event by taking measures of variables
of interest before and after administration of stimuli for both the individuals
exposed to the manipulation (the experimental group) and individuals not
so exposed (the control group). Bracketing identifies experimental and con-
trol groups that are generally nonequivalent but are as similar as possible in
important respects. This design is often referred to as the nonequivalent
control group design (Campbell & Stanley, 1966), or as the untreated control
group with pretest and posttest (Cook & Campbell, 1976), Although not true
randomized experiments, such designs control for most of the threats to
internal validity identified in the two sources just cited, with the exceptions
noted in the next paragraph. Coupled with the strengths of social experiments
with respect to external validity, a design employing a nonequivalent con-
trol group provides a needed complementary focus on internal validity.
Despite the many strengths of this design, it has weaknesses that require
researchers to employ a laborious method of investigation in which threats
to internal validity are examined one by one. It is important to note that
nonequivalent control group designs control for all but three of the threats
to internal validity Cook and Campbell (1976) identified. These three,
which analysis and interpretation of results must address, are selection-
maturation interaction, scaling effects, and local history. In this experiment,
the effects of local history, described as "events other than the treatment
which affect the experimental group, but not the control group, or vice
versa" (Cook & Campbell, 1976: 251), are negligible. Diaries maintained over
the course of this experiment^ revealed no such threats to internal validity.
Scaling was not a serious threat either; social experiments minimize scale
effects, as discussed in the previous section. However, selection-maturation
interaction was a significant potential threat to internal validity. Use of
statistical procedures that simulate a complementary quasi-experimental
design, termed treatment-effect correlations, addressed this threat. This type
of design controls for selection-maturation interaction. Using it in combina-
tion with a nonequivalent control group effectively compensates for the
limitations that would be present if either design were used alone. The
details of the statistical procedures appear in the Methods section.
Description of the Change ^ ,
The social experiment conducted here was made possible by the reorgani-
zation of the Engineering Division of an aircraft manufacturing firm. The
firm undertook this reorganization to improve control of projects, the devel-
opment times of engineering programs, and the motivation of the division's
employees. The management's rationale was that assigning personnel to

' During this time, the author was employed full-time by the host organization to observe
and measure the effects of the reorganization.
542 Academy of Management /ournal September

projects that overlaid their basic functional workgroups would allow them
greater autonomy and facilitate identification with projects. This would in
turn increase motivation, shorten development times, and improve control
of projects through the new role of project manager. Prior to the reorganization,
the Engineering Division was organized functionally, with separate groups
for engineering design, drafting, and engineering tests and substantiation.
Nature of the work. The Engineering Division was responsible for the
design and testing of new and modified aircraft. After design prototypes had
flown and passed certification tests specified by the Federal Aviation
Administration (FAA), the firm's Production Division took responsibility for
quantity production of the new models. Prior to this transfer of responsibility,
three major groups within the Engineering Division played complementary
roles. Engineering Design developed new models. Because aircraft are very
complex and require the application of considerable technical expertise,
departments within this group had been formed to concentrate on different
components of aircraft. Some specialized in wing, fuselage, or tail assembly
design, and others had responsibility for design of power plants or electrical
systems. Thus, the overall task of designing a complete airplane was frag-
mented and the several departments constituted for this purpose were
interdependent. Ultimately, the wing had to mate with the fuselage, the
empennage had to provide sufficient control power to maneuver the aircraft,
and the power plant had to deliver sufficient thrust to propel it at the desired
cruise speed and payload.
In order to ensure that products met these design criteria, the Tests and
Substantiations Group conducted evaluations to assess each aircraft's perfor-
mance and structural integrity, as marketing specifications and governmen-
tal regulations required. The group had three departments to accomplish
these tasks. The Aerodynamics Department conducted performance analy-
ses and wind tunnel tests. The Flight Test Department actually flew proto-
type versions of new models to determine experimentally whether or not
they met their design specifications and the extent to which they satisfied
FAA requirements regarding controllability and aerodynamic safety. The
Structures Department determined the loads that an aircraft must sustain in
flight and conducted tests to ensure that structures proposed by the Design
Group could actually bear these loads in practice. When the results of aerody-
namic or structural analyses or tests suggested a problem with a design,
appropriate departments in the Design Group made modifications, and new
tests were performed before the design became final. Although the greatest
interdependence between departments occurred within the Design Group
and within the Tests and Substantiations Group, there was significant inter-
dependence between departments in different groups as well.
The Drafting Group became involved in the process as a final design for
an aircraft neared completion. Draftsmen transformed design drawings that
had been used for prototype construction into production drawings incorpo-
rating all the design changes necessary to produce an FAA-certifiable aircraft
that met desired performance specifications. In contrast to the work per-
1986 Joyce 543

formed in the engineering groups, this work, although intricate, was quite
routine. When changes to design drawings were minor, as they often were, a
draftsman would simply trace the designer's work, paying a bit more atten-
tion to scale and the quality of the finished drawing. Interdependence among
draftsmen was relatively low even when individuals were working on highly
interdependent parts of an airplane, because the design drawings from which
they worked already fit together as a result of previous design and analysis.
Reorganization. The reorganization created a matrix structure by estab-
lishing a formal project management system overlaying these three major
groups of the Engineering Division, The firm appointed a project manager for
each of the aircraft development projects in progress. These people were
responsible for overall project integration. Their task was to ensure that
personnel effectively attended to the interdependencies that every project
involved by providing a management perspective oriented toward whole
aircraft rather than toward their subsystems and components. Project manag-
ers were assigned greater influence and formal power than the department
managers. Department managers were to provide technical direction to their
employees who were also assigned to project managers, but project managers
actually directed these employees' activities on the projects. This arrange-
ment is the classical division of labor between lateral and hierarchical man-
agers in matrix structures (Davis & Lawrence, 1977). Project managers had
the authority to overrule any of the department managers, even on technical
matters, although it was expected that they would exercise this perogative
judiciously. The role of project manager was intended to resemble that of the
owner of a small engineering or manufacturing firm, and incumbents were
viewed as totally responsible for the success or failure of their projects.
Intervention. The firm took several actions to support this reorganization,
the new role of project manager, and the intended balance of power between
project and functional management, All three groups of the Engineering
Division shared some aspects of this intervention; other aspects of the treat-
ment differed for the three groups.
The common aspects of the intervention included preparation of new
organizational charts and memoranda summarizing the purposes of the
change, broad role descriptions for the new project management function,
and brief descriptions of how the new system was intended to function. The
firm used several methods to communicate this information, including distri-
bution of the memoranda; addresses by the Vice President of Engineering,
who encouraged questions and comments from division personnel; and small
meetings between managers of functional groups and their subordinates,
beginning at the top of the organization and flowing down the hierarchy.
These procedures were followed uniformly across both engineering groups
and the Drafting Group.
However, the treatment received by the Drafting Group differed from the
treatment of the two engineering groups. In addition to undergoing the admin-
istrative changes, drafting personnel were physically relocated to project
centers where they sat together with other draftsmen working on the same
544 Academy of Management /ournal September

project. Personnel from the Engineering Design and Testing and Substantia-
tion Groups were not relocated. Thus, although draftsmen now worked in
close proximity to the newly appointed project managers and away from
their functional managers, the engineers continued to work in the same areas
as they had in the earlier structure, close to their accustomed supervisors
and distant from the new project managers, in what were essentially intact
functional groupings. Physical relocation involved three significant dimen-
sions of change for the draftsmen: (1) separation from their previous physical
work settings with a consequent need to adjust to new work settings and
locations, (2) distancing from accustomed supervision and close exposure to
new and unfamiliar sources of leadership, and (3) mixing of personnel as
functionally based groups were disbanded and reformed around projects.
Engineering personnel did not experience any of these changes.
Differences between the drafting and engineering groups. In addition to
the differences in the treatments received, the drafting and engineering groups
also varied along a number of other dimensions. As would be expected,
members of the Drafting Group were not as highly educated as memhers of
the engineering groups and had heen recruited primarily from the local
community. They also received hourly pay rather than salaries, and they
performed more routine tasks, as can be seen in the descriptions of their
work. In order to confirm this observation, three judges rated the degrees of
task variability within the engineering and drafting groups using scales drawn
from the Organization Assessment Instrument (Van de Ven & Ferry, 1980).
Results indicated that participants in the engineering groups performed tasks
that were intermediate in variahility, primarily through the use of discretion-
ary programming (Van de Ven & Delhecq, 1974). In contrast, the Drafting
Group performed tasks low in variability, with few exceptional cases.
Another important difference was that the members of the engineering
groups had some previous exposure to the use of lateral relations. Liaison
roles (Galbraith, 1973) were used in these groups to coordinate information
relevant to aircraft development projects. Individuals filling these roles func-
tioned primarily as expediters who ensured that necessary information was
available to others to aid in their decisions. Although the liaison role had
formalized responsihility, role incumbents had relatively little legitimate,
reward, or coercive power; consequently, they had to rely on knowledge or
personal sources of influence (Hrebiniak & Joyce, 1984).
Finally, the engineering and drafting groups also varied in the extent to
which they displayed intragroup differences. Generally, the work performed
by the engineering groups was more heterogeneous than that of the Drafting
Group. Few substantial differences existed in drafting work on different
components of an aircraft—an individual working on electrical systems could
be redeployed to wing or fuselage drafting with little loss of efficiency. This
work was homogeneous in that it was primarily drafting, regardless of the
particular aspect of the aircraft heing worked on. However, it was generally
not possihle for engineers to suhstitute for one another, even within each
engineering group, A structural engineer could not easily perform the work
 Joyce 545

of an aerodynamiclst, nor could a wing designer easily design electrical

systems.
Assessing the relative "size" of the change. The differences between the
drafting and engineering groups, in conjunction with the differences in treat-
ment that they received, suggest that the "size" of the change they experi-
enced varied. This raises the issue of how to characterize size. Beyer and
Trice (1978) provided a useful framework for assessing the degree of change
involved in a particular intervention, proposing that it can be assessed along
four dimensions: pervasiveness, magnitude, innovativeness, and duration.
The number of people expected to change their behavior and the amount of
time that they will be behaving in new ways indicates pervasiveness (Beyer &
Trice, 1978: 18). In this study, pervasiveness was high for both the engineer-
ing and drafting groups, because all their members were expected to change
their behavior as a result of the reorganization. In addition, because thty
would be functioning witbin the constraints of a new organizational structure,
this change would affect them during most of their activities. Duration refers
to the time span over which a change is intended to persist. The duration of
this change was also similar for the engineering and drafting groups, because
it was proposed that the new structure would last indefinitely.
Although the pervasiveness and duration of the change as experienced
by engineering and drafting personnel were similar, the magnitude and
innovativeness of the change varied for the two. Magnitude refers to the
displacement of existing organizational states, and innovativeness refers to
the degree to which past behaviors provide routines useful in a process of
change (Beyer & Trice, 1978: 19). Because lateral relations already existed
within the engineering groups, workers there sensed only a change in the
degree of lateral managers' influence. However, within the Drafting Group,
where no lateral structures existed previously, workers would perceive a
totally new—and strong—source of influence. These differences in the treat-
ment's interaction with existing structural arrangements implied a higher
magnitude of change for individuals in the Drafting Group,
They would also see the change as more innovative than the engineers
would. Draftsmen generally had less formal education and less exposure to
the work practices and procedures of other aerospace firms than engineers.
Gonsequently, they had little previous experience, either academic or
practical, to draw upon in adapting to the change. In contrast, engineering
personnel with experience in other firms had often worked hefore under
this type of management system, which is widely used in this industry.
Although not all members of the engineering groups shared such experience,
it was present to a degree lacking in the Drafting Group.
In summary, the change as experienced by both the drafting and engi-
neering groups could be characterized as high in hoth pervasiveness and
duration. Magnitude and innovativeness differed, with the Drafting Group
experiencing higher levels of both of these dimensions during the interven-
tion.
546 Academy of Management Journal September

METHODS
Data Collection
Questionnaires designed to measure process as well as behavioral out-
comes of the reorganization were distributed to all members of the Engineer-
ing Division whom the reorganization affected directly. The questionnaires
were distributed approximately four weeks prior to the beginning of the
reorganization, in order to avoid a possible interaction between testing and
the intervention, and again six months after the change began. Six months
represented the longest time span during which management could main-
tain the conditions of the experiment without further changes that would
have confounded the design,
A control group of industrial engineers and production specialists who
worked for the company and who were located in another building was also
surveyed before and after the change. These individuals were subject to all
the same administrative policies and procedures as the experimental groups,
with the exception of the matrix reorganization. Their physical separation
from the reorganization addressed threats to internal validity from compensa-
tory rivalry and diffusion of treatment (Gook & Gampbell, 1976),
The questionnaires, which took about 25 minutes to complete, were
administered at special sessions at which the researcher was present. Respon-
dents completed them anonymously, placed them in sealed envelopes, and
deposited them in collection boxes before leaving the room in which the test
was administered. Numbers of respondents within the groups at the pretest
were: engineering, 33; drafting, 21; and control, 25, At the posttest, engineer-
ing had 31 respondents; drafting, 16; and control, 26,
Appendix A presents the details of measurement for variables concern-
ing organizational processes, role perceptions, and work attitudes that were
investigated to test the propositions summarized in Table 1.
Statistical Procedures
The differences in the treatments received by the drafting and engineer-
ing experimental groups suggested that their data be analyzed separately;
presentation of the experiment's results reflects this separation.
The analysis of results from quasi experiments is complex. Results
obtained from designs with nonequivalent control groups have been interpre-
ted using analyses conducted between waves of an experiment, within waves,
or by some combination of the two methods. Between-wave analysis, which
involves tracking individuals longitudinally, is not appropriate for experi-
ments in which assignment to the experimental and control groups is not
under an experimenter's control, because nonrandom pretest differences
between groups are likely. Instead, Gook and Gampbell (1976) recommended
within-wave analyses to assess treatment effects in nonequivalent control
group designs. Within-wave analyses can also be shown to be mathemati-
cally equivalent to an alternative quasi-experimental design, treatment effect
correlations (see Appendix B). Because such a design is robust with respect
1986 Joyce 547

to threats from selection-maturation interactions, using it is tantamount to

including an additional complementary quasi experiment that compensates
for the primary weakness of designs using a nonequivalent control group.
The specific statistical procedure employed was to test for significant
differences hetween the mean scores of experimental and control groups
within waves of data collected both before and after tbe reorganization, or
treatment. If comparison of the pretest and posttest analyses of variance
indicated that overall treatment effects existed, Scheff6 tests of post hoc
differences between the means of the experimental and control groups were
conducted within waves and examined across waves to isolate specific treat-
ment effects. Three patterns of significance between pretest and posttest
(within wave) ANOVAs would indicate a possible treatment effect: (1) pre-
test is nonsignificant, posttest is significant; (2) pretest is significant, posttest
is nonsignificant; (3) pretest is significant, posttest is significant. The first
two patterns would provide clear evidence that the treatment had an effect
hy causing or diminishing differences between the experimental and control
groups. The third is more equivocal; an effect could be present if the pattern
of significant differences between the experimental and control groups dif-
fered in posttest and pretest. Such variation cannot be detected by ANOVA
and requires post hoc inspection of means, as is described in the Results
section.
These analytic methods result in more conservative tests for the effects
of the treatment on the dependent variables than do tests that require the
matching of pretest and posttest scores for each individual. Therefore, statis-
tically significant results are also more likely to be practically significant. An
additional strength is that there is no compelling reason to specifically iden-
tify individuals in the analyses. Respondents can complete questionnaires
anonymously, which minimizes perceived risk and encourages candid and
valid responses.

RESULTS
Table 2 presents the intercorrelations among all variables employed in
this study. Significant relationships that are generally consistent with previ-
ous research and theory exist among several of the measures. Role conflict
and ambiguity were negatively related to both job involvement and satisfac-
tion and positively related to the frequency of informal lateral communication.
This result is consistent with the argument advanced in the first main section,
that deviations from the use of formal, hierarchical communication paths
introduce conflicting sources of information, thus heightening conflict and
ambiguity. Furthermore, the positive correlation between coordination and
role ambiguity (r = ,23, p< ,05) suggests that such an increase may introduce
valuable new perspectives and bring consequent improvement in coordina-
tion.
The interesting relationships between coordination and the otber vari-
ables also merit comment. The results show negative relationships between
548 Academy of Management Journal September

q
r

,34
.17

,01
.08
m o o ID
o o
o a

I I

I "i I ^
.^
D* ~ —
CC
•=

«
o a r
e - e " E
C n = 1=^ O
£• = S ' y "
o 2 o -.p ^
u I u -g o u
1986 Joyce 549

coordination and both frequency and amount of formal communication (r ==

-,22. p < .05; r = -,32, p < .01, respectively), as well as between coordina-
tion and the directive quality of communication (r = -.29, p < .01). This
finding is also consistent with information processing tbeory (Galbraitb,
1973; Hrebiniak & Joyce, 1984), whicb suggests tbat as levels of information
processing increase, a point wbere reorganization is required is reached. At
this point, existing hierarchical channels cannot carry tbe information
required for effective decision making and attempts at coordination through
increased use of an overloaded structure are nominal.
Process Variables
The results shown in Table 3 and Figure 1 indicate that implementing a
matrix structure bad significant effects on all process variables. Tbe effects
on tbe quantity of formal and informal task-related communications were as
predicted, but were confined to the Drafting Group, where tbe degree of
cbange was more extensive. However, coordination in this group actually
declined, perbaps owing to accompanying unfavorable decreases in the par-
ticipative and directive quality of communication.
It was proposed tbat the amount and frequency of formal communica-
tion would increase through tbe legitimation of lateral channels for task-
related communication, resulting in a decrease in the frequency of infor-
mal task-related communication. Tbe frequency of formal communication
increased in the Drafting Group, but remained relatively stable in tbe engineer-
ing and control groups. Differences among the three groupings were signifi-
cant at both the pretest and posttest (Fpj-e = 7.33, p < .01; Fpost = 13-42, p <
,01) because of higher levels of communication within botb of tbe experimen-
tal groups. However, tbese differences were even stronger at tbe posttest
because the frequency of formal communication increased witbin tbe Draft-
ing Group.
Effects of the treatment were also noted for tbe amount of formal com-
munication (Fpre = 1-18, n,s,; Fpost = 5-87, p < .01), again because of an
increase in tbe amount of communication witbin tbe Drafting Group at tbe
posttest (p < .05). These results indicate tbat both tbe amount and frequency
of formal communication sbowed predicted increases in quantity as a result
of implementation of matrix structure.
The frequency of informal task-related communications also declined as
predicted within the Drafting Group, Whereas this group reported a higher
frequency of informal communication than the control groups at the time of
the pretest (p < ,05), these differences were not significant at tbe posttest
(F pre = 3.69, p < .05; Fpost = 1-90. n.s.).
The cbange in structure altered tbe quality as well as the quantity of
formal communication, but in an unpredicted and unfavorable direction.
Effects for participative communication (Fpre = 1.87, n.s.; Fpost = 5,66,
p < .01) were noted because of a decrease in the level of this variable witbin
tbe Drafting Group compared to tbe engineering (p < .05) and control
(p < ,01) groups, Tbe use of directive communication also decreased in the
550 Academy of Management Joumal September

o o o q 9 9
VV V V V V

V V

o o CO in O O

0) CO in in

B«
a II
fj o O
m CO (O
cb CD i n I"
c S3
9 o.
ed
u
« 9
• XI
13
OS CO CO i n
1
objO

£ S £ S
(1. 0. 0. a.

I I
^ .2 Sao.
1986 Joyce 551

FIGURE 1
Means of Process Variables
Before and After Matrix Reorganization

26 4.5
24
22 4.0
20
3.5
18
16 Con 3.0
14 Con

Pre Post Pre Post

a. Frequency of Formal Communication b. A
Amount of Formal Communication

18 r 10
—
16 • Drf 9
Eng
14 8
Cnr,
12 - 7

1 1
Pre Post Pre Post
c. Frequency of Informal Communication d. Participative Quality of Communication
8
20 Con
7 19 —— ""
18
6 17
16 -
5 15 - ^ ^ Drf
14 _
I 1 1
Pre Post Pre Post
e. Directive Quality of Communication f. Coordination

"Drf = drafting, Eng = engineering, and Con = control; pre = pretest and post = posttest.
552 Academy of Management /ournaJ September

Drafting Group, resulting in the elimination of pretest differences with the

control group (p < .05) and no significant differences among the groups at the
posttest (Fpre = 5-76. p < .01; Fpost = 0.186, n.s.).
Coordination decreased in the Drafting Group (Fpre = 0.192, n.s.; Fpost =
7.63, p < .01), despite increases in the quantity of communication in that
group. Coordination was significantly lower when contrasted with the engi-
neering (p < .01) and control (p < .01) groups at the posttest; no significant
differences had existed at the pretest.
These results have interesting implications regarding implementation of
matrix structures. Although the reorganization hrought predicted improve-
ments in the quantity of communications in one experimental group, the
change unfavorahly affected the quality of communications, and correspond-
ing decreases in coordination occurred. This suggests that implementing a
matrix structure must favorably affect both the quantity and quality of com-
munications for it to result in such desirable outcomes as improved
coordination.
Role Perceptions and Work Attitudes
Table 4 and Figure 2 present the results of analyses of the effects of
the reorganization on role perceptions and work attitudes. Significant effects
were noted for role ambiguity, job involvement, satisfaction with work, satis-
faction with co-workers, and satisfaction with promotions. Changes within
the Drafting Group were responsible for these effects, most of which were
negative in direction and all of which were unfavorable.
Levels of role ambiguity among the experimental and control groups
differed at both the pretest and posttest (F pre = 10.48, p < .01; Fpost = 8.46,
p < .01). However, the changing pattern of significant post hoc comparisons
between groups indicates a significant treatment effect on the level of role
ambiguity within the Drafting Group. These respondents were higher than
the control group in role ambiguity after (p < .01), but not before the change.
Moreover, although at the pretest ambiguity was lower in Drafting Group
than in the engineering groups (p < .01), increased ambiguity in the Drafting
Groups eliminated this difference at the posttest.
These results are consistent with the observed decreases in both the
participative and directive quality of formal communications in this group.
Both direction from supervisors and participation in decision making rele-
vant to tasks clarify roles. When use of both of these methods for obtaining
role clarity declines, as in the present case, role ambiguity should increase.
The reorganization also affected job involvement (Fpre - 0.23, n.s.;
f^post = 6.38, p < .01). Involvement decreased within the Drafting Group, but
scores for both the engineering and control groups remained stable, resulting
in significant differences between the Drafting Groups and the others (p < .01
for both).
The new structure also negatively affected job satisfaction. Significant
treatment effects were noted for satisfaction with work (Fpre = 1.04, n.s.;
= 3-30, p < .05), co-workers (Fpre = 1-75, n.s.; Fpost = 3.48, p < .05),
1986 Joyce 553

u
VV VV

CO (O O O CM in CO M o „- u
m in
O CO Or- O CO CM d

"I
II
(M f O O -
CO BO
d d .s
v 'C
u

I a c
iii
c
5
o o a otl O O O
in CO CO q ai i n CO
tv CM tn CO CM
in CM n
<
O D.
" E
•B O
C "
SO U

CO (M in

na to
O CO CO So
CL a. a. a. a. a. O.CX. 040, OiCU CLOu

a
e 6 •3
5
-1—1

G
o
"f V V
S a a
554 Academy of Management Journal September

FIGURE 2
Means of Role Perception and Work Attitude
Variables Before and After Matrix Reorganization

5.0 Drf 5,0

Eng
4.0 4.0
Con
3.0 3.0

2.0 2.0

Post Pre Post

b. Role Conflict

45

40
Con
35
Drf
30 Eng
45 -
Drf J
Pre Post Pre Post
c. Job Involvement d. Satisfaction with Work
45 -
45
40 -
Con 40
^ ^ Drf
ting 35
30 - 30
1 1 _L
Pre Post Pre Post
e. Satisfaction with Supervision f. Satisfaction with Co-workers

lOi-
lor

Post Pre Post

g. Satisfaction with Pay h. Satisfaction with Promotions

Drf = drafting. Eng = engineering, and Con = control; pre = pretest and post = posttest.
1986 Joyce 555

and promotions (Fpre = 6.91, p < .01; Fpost = 15.3. p < .01). Satisfaction with
work decreased within the Drafting Group, but levels within the engineering
and control groups remained stable. Work satisfaction was significantly lower
in the Drafting Group than in the control group at the posttest (p < .05); such
a difference did not exist prior to the change. Satisfaction with co-workers
similarly declined for the Drafting Group. Following the reorganization, sig-
nificant differences existed between the drafting and engineering groups
(p < .05); such differences did not exist prior to the change.
Satisfaction with promotions also declined within the Drafting Group.
This result is particularly interesting given the low pretest levels of this
variahle. Because of the design of this experiment, scaling was not a serious
threat to the validity of the conclusions in general. However, the very low
levels of satisfaction with promotions within both experimental groups sug-
gests the plausibility of such a threat in this case. Specifically, a basement
effect is possible (Cook & Campbell, 1976) because there is less possihility of
negative movement on this scale (range 0-27) when scores are very near its
lower extreme. The strength of the treatment's effect is evident here; satisfac-
tion with promotions declined even though pretest scores were in the very
lowest ranges of the instrument. Following the reorganization, the Drafting
Group scored significantly lower than the control group (p < .01) on this
criterion, whereas no significant differences had existed at the pretest.
The results for the drafting group are consistent with hypotheses.
However, the engineering groups again showed few effects from the change,
and there was an unexpected lack of evidence for effects of implementing a
matrix structure on perceived role conflict in any group.

DISCUSSION
The results of this study indicate that matrix structure has strong effects
on relevant organizational process variables, role perceptions, and work
attitudes. These results differ in some respects from the popular view of the
effects and costs of matrix organization as reflected in the propositions of
this study.
Two unexpected results of this research merit discussion. The first is its
failure to find any effect of the matrix reorganization on role conflict in
either of the experimental groups. This is notable given the dominance of
this hypothesis in the literature.
One possible explanation of this finding is the complexity of the con-
struct of role conflict. As defined and measured by Rizzo and colleagues
(1970). generalized role conflict is produced by inconsistency in self- and
other behavioral expectations (intersender conflict), behavioral standards for
self and roie (person-roie conflict), multiple role requirements (intrarole
conflict), and performance capability and role demands (role overload). Imple-
menting a matrix structure may not affect all of these. It would not be likely
to affect person-role conflict. Role overload would be affected if time pres-
sures resulting from structural change and project activity were high.
556 Academy of Management Journal September

However, in this case such pressures were low, hecause the firm instituted
the change when project activity was slack to allow available resources to be
concentrated on the transition in structure.
What should have changed were levels of intersender and intrarole
conflict. However, the extent to which the demands of members of a role set
are recognized and perceived as legitimate can affect these conflicts. The
high level of role ambiguity in the Drafting Group, perhaps the result of
decreases in participative and directive communication, may have blocked
changes in intersender and intrarole conflict. The lack of conflict effects
among engineers is symptomatic of and consistent with the overall lack of
treatment effects in this group. This is the second unusual result of this
study.
Differences in the treatment received and characteristics of the engineer-
ing and drafting groups may account for the differences in results for the
groups. For the Drafting Group, the reorganization involved physical
relocation, proximity to new superiors, and disbanding and reformation of
functional groups around projects. Higher levels of task variability within
the engineering groups had previously necessitated more complex coordinat-
ing mechanisms in the form of liaison roles. Consequently, engineers, unlike
draftsmen, had some previous exposure to lateral relations. One possible
interpretation of the results reported here is that when the magnitude and
innovativeness of a change (Beyer & Trice, 1978) are small, reorganization
has fewer behavioral consequences than when the change is larger along these
dimensions. In this case, the ohserved differences in the effects of the treat-
ment on the engineering and drafting groups suggest that important dimen-
sions of the size of a change may include changes in physical locations of
organizational members and in their locations relative to authority figures
and co-workers, as well as changes in administrative procedures. The Draft-
ing Group scored higher on all of these dimensions, suggesting some relation-
ship between the size of an organizational change and the degree of behav-
ioral and attitudinal adjustment of persons engaged in the change.
The lack of process effects within the engineering groups supports this
view. No effects emerged among engineers for role perceptions, work attitudes,
or process outcomes. Differences in the treatment received by the engineer-
ing and drafting groups suggest a view of organizational design and change
that is more complex than a simple structural view. Realization of the intended
process outcomes of the change in the engineering groups may have required
more than a simple adjustment in administrative structure. The primary
intervention in this case was structural; supporting secondary interventions—
such as physical relocation or another means—may be necessary to imple-
ment a primary change (Hrebiniak & Joyce, 1984).
This observation is particularly interesting in light of the inconsistent
results obtained for the quantity and quality of communication within the
Drafting Group. Apparently, the combination of matrix support systems—
physical relocation and proximity to supervision—and the use of a new
1986 Joyce 557

administrative structure was sufficient to produce the intended increase in

communication in this group. No changes in the quantity of communication
were noted in the engineering groups, where only administrative structure
was altered. This suggests that both matrix structure and matrix systems are
required to produce changes in the quantity of communication. However,
systems and structure were insufficient to change the quality of communica-
tion. Davis and Lawrence (1977) suggested that matrix management requires
not only matrix structure and support systems, but a supportive culture as
well. The lack of favorable change in the quality of communication in the
Drafting Group may reflect the absence of such a culture. This research
provides evidence supporting the view that implementing complex changes
requires multiple interventions. This requisite complexity may help explain
the growing disenchantment with matrix structures in the popular literature
(e.g., Peters & Waterman, 1982).
This study also raises an interesting theoretical issue for future research
concerning the degree to which lateral channels of communication can and
should substitute for hierarchical channels. Current thinking is that lateral
channels should supplement rather than replace existing hierarchical
channels (Galbraith. 1973; Hrebiniak & Joyce. 1984). Lateral channels are
overlaid on existing hierarchical channels, and new lateral structures depend
on what Hrebiniak and Joyce termed the "basic operating structure" (1984:
149). It may be that effective matrix arrangements substitute new lateral
channels of communication for hierarchical ones, altering basic operating
structures as well, and creating a situation in which both lateral and hierar-
chical dimensions of matrices depend on one another and neither can stand
alone. Although this image has intuitive appeal, it runs counter to current
thinking concerning the elaboration of complex organizational forms like
matrix structures and deserves further research.
The complexity in the treatment administered in any reorganization,
and in this study in particular, raises interesting issues concerning the
generalizability of social experiments. Although external validity should be
higher than with many other types of research design, the peculiarities of
particular interventions raise an issue. From what and to what can research-
ers employing such designs generalize? As Gaibraith noted, the problem
concerns providing useful information regarding "a specific organization
providing specific services at a specific point in time to specific clients in a
specific location while employing a specific group of people" (1977: 7).
Gharacterizing the size of a change along dimensions such as those Beyer
and Trice (1978) proposed may facilitate informed generalization by avoid-
ing specification of the change to such a degree that similarities are obscured,
inconsequential differences are aggrandized, and generalization is rendered
impossible. The question of just what is inconsequential, however, is
empirical, and further research explicitly measuring Beyer and Trice's
dimensions of change would be valuable in exploring the usefulness of such
schemes for facilitating generalizability.
S58 Academy of Management Journai September

A related point concerns the extent to which the time elapsed between
pretest and posttest can explain the results of this study. Lewin (1947) argued
that as organizations go through periods of change, an initial disruption
occurs that diminishes over time, providing the change is implemented
properly. However, current theory does not allow specifying the time required
for the negative effects of change per se to diminish. It is possible that to
some extent such effects confound the results of this experiment. Such con-
founding is less likely the longer that experimental conditions can be
maintained. Since other events may also impinge on a study like this one,
potential threats from these other sources must be weighed against the desir-
ability of minimizing change-related effects through longer time spans of
measurement and of administering posttests when interpretability is highest.
Such a procedure was followed in this research.
Finally, the relative desirability of social experiments as a method for
learning about organizations should be considered. Although the complexity
of treatments and confounding with change-related effects can make general-
ization difficult, researchers widely view such experiments as desirable.
Nevertheless, they are still rare, which is problematic for several reasons.
Social experiments afford a degree of internal and external validity that is
not possible within the more frequently used concurrent study. Although
social experiments do not automatically control for all threats to internal
validity, they do require investigators to think systematically about what
conclusions they may reasonably draw and thus to productively consider
issues that more conventional inquiries rarely raise.
Moreover, social experiments are more likely to produce results that
have usable implications. The strength of manipulations and their conse-
quent realism facilitate external validity, and conclusions regarding the effects
of treatments on relevant criteria are causal. Thus, practical uses of the
results of such experiments are more likely than in simpler concurrent stud-
ies that do not have these features.
Finally, social experiments fill an important theoretical gap identified
by Evan (1971). An individual researcher may not and need not pursue an
issue from case study, through survey and laboratory experiment, and on to
field experiment, but such a process should emerge for each issue as results
accumulate across studies. For many issues, the most serious gap in this
process is a lack of field experiments, particularly social experiments. Future
research should address this need.

REFERENCES
Beyer, J.. & Trice, H, 1978. Implementing change: Alcoholism policies in work organizations.
New York: Free Press.
Burns, T., & Stalker, G.M. 1961. The management of innovation. London: Tavistock,
Butler, A, 1973. Project management: A study of organizational conflict. Academy of Manage-
ment Joumai, 16: 84-102.
1986 Joyce 559

Campbell, D. T. 1971. Administrative experimentation, institutional records, and nonreactive

measures. In W, G, Evan (Ed.), Organizational experiments: Laboratory and field research:
169-179. New York: Harper & Row Publishers,
Campbell, D, T,, & Stanley, J, 1966. Experimental designs and quasi-experimental designs for
research. Chicago: Rand McNally & Co.
Cook, T, D,, & Campbell. D. T, 1976, The design and control of quasi-experiments in field
settings. In M. D. Dunnette (Ed,), Handhook of industrial and organizational psychology:
223-326. Chicago: Rand McNally & Co,
Davis, S. M., & Lawrence, P, R. 1977, Matrix. Reading, Mass,; Addison-Wesley.
Delbecq, A,, & Filley, A. C. 1974, Program and project management in a matrix organization:
A case study. Madison: Graduate School of Business, University of Wisconsin.
Draper, N. R., & Smith, H, 1966, Applied regression analysis. New York: John Wiley & Sons.
Evan, W, G, 1971, Introduction: The organizational experiment. In W. G. Evan (Ed,), Organiza-
tional experiments: Laboratory and field research: 1-5, New York: Harper & Row
Publishers,
Galbraith, J. R, 1973. Designing complex organizations. Reading. Mass.: Addison-Wesley.
Galbraith. J. R. 1977. Organization design. Reading, Mass.: Addison-Wesley.
Georgopoulos, B, S., & Mann. F. C. 1962. The community general hospital. New York: Macmil-
lan Book Publishing Go,
Hrebiniak, L. G,, St Joyce, W. F. 1984, Implementing strategy. New York: Macmillan Book
Publishing Go.
Kaplan, A, 1964. The conduct of inquiry. New York: Chandler,
King, A, S. 1974. Expectation effects in organizational change. Administrative Science QutiT'
terly, 19: 221-230.
Lawrence, P. R., & Lorsch, J. W. 1967. Organization and environment. Boston: Graduate School
of Business Administration, Harvard University,
Lewin, K. 1947, Frontiers of group dynamics. Human Relations, 1(1): 5-41. '
Lodahl, T, M, & Kejner, M. 1965, The definition and measurement of job involvement. Journal
of Applied Psychology, 49: 24-33.
Milgram, S. 1965. Some conditions of obedience and disobedience to authority. Human Relations,
18(1): 50-76,
Ome. M, T. 1962. On the social psychology of the psychological experiment. American Psycholo-
gist, 19: 760-783.
Ome, M. T., & Evans, F. J. 1965, Social control in the psychological experiment: Anti-social
behavior and diagnosis, Joumai of Personality and Social Psychology, 1(3): 169-200,
Peters, T,, & Waterman, R, 1982. In search of excellence. New York: Harper & Row Publishers.
Pfeffer, J. 1981, Power in organizations. Marshfieid. Mass.: Pitman Publishing.
Reeser, C. 1969, Some potential human problems of the project organization. Academy of
Management Joumai, 12: 459-467.
Rizzo, J. R,, House R, )., & Lirtzman. S, I. 1970. Role conflict and ambiguity in complex
organizations. Administrative Science Quarterly. 15: 150-163.
Roos, L. L. 1973, Panels, rotation, and events: Some uses of quasi-experimentation. In I. A.
Caporaso & L. L. Roos, Jr. (Eds,), Quasi-experiments: Testing theory and evaluating pol-
icy: 82-104. Chicago: Northwestern University Press.
Simon. H. A. 1947. Administrative behavior. New York: Free Press.
 560 Academy of Management Journal September

Smith. P. C, Kendall. L., & Hulin. C. 1969, The measurement of satisfaction in work and
retirement: A strategy for the study study of attitudes. Chicago: Rand McNally & Go.
Van de Ven, A, H., & Delbecq, A, L, 1974. A task-contingent model of work unit structure.
Administrative Science Quarterly. 19(2): 183-197.
Van de Ven, A, H., & Ferry, D, L. 1980, Measuring and assessing organizations. New York:
Wiiey-Interscience.

APPENDIX A
Measurement
Organizational process. Goordination was measured using scales developed hy Georgo-
poulos and Mann (1962). Scores range from 7 to 35, with high scores indicating higher levels of
coordination. Items tap relevant aspects of coordination, including the extent to which various
work activities fit together, are not in conflict, are well-timed, and collectively support organiza-
tional objectives. Coefficient a = ,81,
Communication was also measured using scales developed by Georgopoulos and Mann.
The frequency of both formal and informal, job-related, verbal communication was measured
(a = .86 and a = .84,range7-35);thetotalamountoffonnal communication was also measured.
This instrument also contains a 7-item scale measuring the extent of the directive or partici-
pative quality of communication. To confirm the factor structure for quality, these items under-
went a principal components factor analysis with varimax rotation, which yielded three factors.
Factor 1, with items relating to providing direction and giving orders, was designated the
directive quality dimension (a = .74, range 2-10), Factor 2, with items measuring solicitation of
suggestions, requests for information, and support, represented the participative quality dimen-
sion (a = .64, range 3-15). Factor 3, containing two items concerning criticism and excess or
unnecessary information (a = ,55), was not included in further analysis because it was irrelevant
to the propositions tested.
Role perceptions and work attitudes. Role conflict and ambiguity were measured using an
instrument developed by Rizzo, House, and Lirtzman (1970). Role conflict refers to the compati-
bility or incompatibility between individuals' values and role hehavior, time and resources and
role behavior, and self- or other behavioral expectations; role ambiguity refers to the extent to
which individuals are clear regarding their goals, authority, responsibility, and expectations for
role performance (a = ,87 and a = ,64;range 1-7, with 7 indicatinghigh conflict orambiguity).
lob satisfaction was measured using the Joh Descriptive Index (Smith. Kendall, & Hulin.
1969). For the five facets of satisfaction included, a = .87 for satisfaction with work, a = .72 for
supervision, a = .82 for co-workars, a = .75 for pay, and a = .80 for promotions. Scores for
satisfaction with work, supervision, and co-workers could range from 0-54; those for pay and
promotions, from 0-27, with high scores indicating higher satisfaction.
Job involvement was measured using an instrument developed by Lodahl and Kejner (1965)
having 20 items with response formats ranging from 1. low involvement, to 4, high involvement.
Scale scores could therefore range from 20-80 [a = .86).

APPENDIX B
Treatment Effect Correlation
This term refers to a quasi-experimental design in which measures of a treatment are
correlated with measures of the criterion taken before and after administration of the treatment.
The resulting statistic is termed a treatment-effect correlation. If no systematic differences
among treatment and control groups exist, the correlation between the treatment, coded as a
dummy variable, and the criterion will not differ significantly from 0, If exposure to the treat-
ment has affected the criterion, the treatment-effect correlation between this same dummy
variable and the posttest measure should be significant.
1986 Joyce 561

The procedure is believed to control for selection-maturation interaction because increased

variation in the criterion scores would depress the correlation. Such would be the case in
selection-maturation, where the variation among scores at a posttest would be higher than that
at a pretest because one group is maturing more quickly. Thus, a significant treatment-effect
correlation would provide evidence for the effects of a manipulation despite any aggravating
effects due to selection-maturation interaction working to depress the obtained correlation.
For a single-treatment, single-control-group case in which the levels of the treatment can be
coded using dummy variables (0,1) and then correlated with the criterion, the significance of
the resulting point-biserial correlation is mathematically equivalent to that obtained by conduct-
ing a t-test for the differences between the pretest means of the experimental and control groups.
Where multiple levels of a treatment exist, the multivariate generalization of the treatment-
effect correlation would be the regression of the criterion on a set of dummy variables represent-
ing membership in the multiple-treatment and control groups. This regression is mathemati-
cally equivalent to one-way analysis of variance conducted among treatment and control groups
withinwavesof a study (Draper & Smith. 1966), the method of analysis employed here. Compar-
ing the results of these analyses across pretest and posttest therefore yields the same informa-
tion as treatment-effect correlations computed in the conventional way, thereby simulating this
design and controlling for selection-maturation interaction.

William F. Joyce is an associate professor of strategy and organization theory at the

Amos Tuck School of Dartmouth College. He obtained his Ph.D. degree in organiza-
tional hehavior from the Pennsylvania State University in 1977. His research interests
focus on interdisciplinary prohlBms in strategy implementation, organizational design,
and organizational climates.