Coordination and control in a government agency: Contingency and institutional theory perspectives on GAO audits

Abstract

Drawing on both contingency and institutional theories, a study examines how professionals in an institutionalized environment are coordinated and controlled and what forces shape the structures organizations adopt for this coordination and control. Hypostheses are developed and tested with data from a study of 96 audit teams in the US General Accounting Office (GAO). Results show that the more institutionalized the environment: 1. the more the organization relies on a bureaucratic mode of control, 2. the greater the task difficulty and team interdependence and the more the organization relies on personal and group modes of control, and 3. the more the organization relies on personal and group modes of control to improve audit-team performance.

Drawing on both contingency and institutional theories, this paper examines how professionals in an institutionalized environment are coordinated and controlled and what forces shape the structures organizations adopt for this coordination and control. According to contingency theory, the coordination and control of organizational members is shaped by the task environment and the technical nature of the work they perform, while institutional theory proposes that an organization's need to demonstrate conformity to institutionalized expectations of rational practice also influences its choice of control and coordination mechanisms. We use both theoretical perspectives to develop hypotheses that are tested with data from a study of 96 audit teams in the United States General Accounting Office (GAO). Results show that the more institutionalized the environment, (a) the more the organization relies on a bureaucratic mode of control, (b) the greater the task difficulty and team interdependence and the more the organization relies on personal and group modes of control, and (c) the more the organization relies on personal and group modes of control to improve audit-team performance.*

Several theories have been used to explain differences in the rationalized, formal structures of organizations that are used to coordinate and control their members. Two of the most prominent theories, contingency theory and institutional theory, take almost opposite approaches to understanding the factors that lead to the development of different formal structures. Contingency theory suggests that the demands imposed by technical tasks in the organization encourage the development of strategies to coordinate and control internal activities, and institutional theory suggests that the expectations regarding appropriate organizational forms and behavior that are expressed in the wider social environment promote the development of an organization's formal structure. According to the contingency perspective, an organization focuses on its technical activities and shapes its work processes to promote those activities while protecting them from disturbances in the external task environment. More specifically, according to work-unit contingency theory, when the coordination and control practices applied to work-unit members coincide with the nature of the tasks to be performed, work-unit performance improves (Drazin and Van de Ven, 1985; Gresov, 1989).

In contrast, according to the institutional perspective, an organization, particularly a government or professional organization, gains legitimacy by conforming to external expectations of acceptable practice while separating its internal technical activities from externally directed symbolic displays. An organization thus may ceremonially adopt elements of formal organizational structure, such as standardized coordination and control practices, to demonstrate the rationality of its operations to external constituents, rather than to control organizational members (Meyer and Rowan, 1977; DiMaggio and Powell, 1983). Despite the apparent inconsistency between these two perspectives, theorists working within each recognize its interrelation with the other. Drazin and Van de Ven (1985: 516-517), who used contingency theory, recognized that many internal coordination and control practices may become institutionalized over time and thus may be unresponsive to the task technology the organization uses. Similarly, Gresov (1989: 439) observed that certain coordination and control practices may serve a symbolic role while others may serve an instrumental role in achieving control and improving performance.

The impact of external and internal contingencies has also been recognized in institutional theory. Meyer and Rowan (1977) theorized that organizations may use relatively formalized control practices for symbolic purposes while actually exercising control over their members via more idiosyncratic and social means. Scott (1987: 507-509) has observed that contingency and institutional theory explanations, when applied separately, offer only an incomplete understanding of the different roles played by various coordination and control practices that are used in contemporary organizations but that both theories together could be used to understand better the instrumental and symbolic roles fulfilled by coordination and control practices.

The purpose of this paper is to examine empirically the connection between the external conditions implied by institutional theory and the internal structure implied by work-unit contingency theory. Our focus is on professionals in an institutionalized environment and what forces shape the coordination and control mechanisms in the organization they work in. This paper responds to calls by Drazin and Van de Ven (1985) and Gresov (1989) for research in a professional setting and research that examines individuals in positions at a higher level in the hierarchy than those in prior research. Also, by examining audit teams in the United States General Accounting Office (GAO)--an organization whose audit oversight responsibility, expressed in terms of the expenditures of the government agencies audited, is perhaps the largest in the world (Sexton, 1986; GAO, 1991)--we have an opportunity to verify the truth of observations by Meyer and Rowan (1977), DiMaggio and Powell (1983), and Scott (1992: 139) that institutional forces may be most powerful in government and professional settings.

CONTINGENCY AND INSTITUTIONAL THEORY INTERPRETATIONS OF CONTROL AND COORDINATION

Structural contingency theory has long dominated the research directed at understanding the relationships between task environment, technology, and work-unit structure. A central theme in contingency theory is that work-unit information-processing needs must match information-processing capabilities in terms of the coordination and control practices used, and this theory thus focuses on the type of practice that works best in a particular situation (Galbraith, 1973; Scott, 1992). In a series of studies, Van de Ven and his associates (e.g., Van de Ven and Delbecq, 1974; Van de Ven. 1976; Drazin and Van de Ven, 1985) found that work units facing high task uncertainty tend to use loose or collegial coordination and control practices, while work units facing low or moderate task uncertainty use more structured or bureaucratic coordination and control practices.

Van de Ven, Delbecq, and Koenig (1976) suggested three modes of coordination and control that are used to manage work units: the bureaucratic mode, the personal mode, and the group mode. The bureaucratic mode relies on standardized tasks to constrain and channel the activity of each work-unit member. The integrating mechanisms include preestablished plans, schedules, and forecasts; formalized rules, policies, and procedures; and standardized information and communication systems. Given a well-known task technology, stable and repetitive tasks, and clear goals, this mode requires little verbal communication among members. The personal mode of control relies on supervisors providing feedback to subordinates, only partially codified standard operating procedures, and both horizontal and vertical channels of communication to permit an exchange of ideas and achieve coordination and control. The personal mode gives team members the discretion to modify tasks in response to situational demands and is suitable for managing recurring tasks that present enough variations and exceptions to require adjustments in work methods. The group mode is much more social and is characterized by heightened mutual adjustment based on feedback, high professional discretion, and the exercise of autonomy. Work-unit members are more interdependent and communicate laterally more frequently. It is appropriate for managing tasks, problems, or issues sufficiently unique to require work processes tailored to the situations (Drazin and Van de Ven, 1985: 525).

Both internal and external contingency factors have been used in explaining the mix of coordination and control practices work units use (Dewar and Hage, 1978; Altman, Valenzi, and Hodgetts, 1985). Although theorists view internal contingency factors in a variety of ways (Thompson, 1967; Scott, 1992), basic to all are three underlying dimensions: task variability, task difficulty, and task interdependence. Task variability has been defined as the degree to which one perceives stimuli as familiar or unfamiliar and is expressed in such terms as the number of exceptions or nonroutine cases a work unit encounters (Thompson, 1967; Van de Ven, Delbecq, and Koenig, 1976). Task difficulty denotes the degree of complexity encountered in a task, the amount of thinking required to complete a task, and the availability of a knowledge base that provides guidance in completing the task (Van de Ven, Delbecq, and Koenig, 1976). Task interdependence denotes the extent to which the work requires collaboration among supervisors and various work-unit members. Interdependence has been further categorized by levels as pooled (when interdependence among work-unit members is minimal and subject to direct supervision), sequential (when the output of one member is the input for the next), reciprocal (when each member contributes input to the others), and team (when the work-unit members work collaboratively to complete a project) (Thompson, 1967: 54-55). High degrees of task variability, task difficulty, and task interdependence have been found to be associated with the use of personal and group modes of coordination and control (Van de Ven, Delbecq, and Koenig, 1976).

Among the most frequently researched external contingency variables is size, typically defined as the number of work-unit members (Fry and Slocum, 1984). Research results, although mixed, suggest that large size is associated with the bureaucratic mode of control (Dewar and Hage, 1978; Scott, 1981). Going beyond this structural-property interpretation, Scott (1992: 258-267) described two other facets of size. First, size may be a contextual variable, related more to an organization's interface with its environment; large size thus may signal a complex relationship with the environment. Second, size may be an institutional phenomenon, in that larger organizations are more visible to a variety of external constituents and thus are subject to institutional pressures. Scott (1992) concluded that the expected effect of size is paradoxical. Under the structural-property definition, increasing size would call for increasing routinization and the application of the bureaucratic mode of control. Under contextual and institutional definitions, increasing size would be interpreted by organizational members as increasing the complexity or visibility of the work they perform, thereby calling for the use of personal and group modes of control. Although we do not include hypotheses below on the effect of size, because of conflicting prior interpretations, we included size in our empirical analysis in order to examine these conflicting interpretations and provide a more robust model.

Environment, the second external contingency factor, has typically been discussed in contingency theory in terms of a "task" or "technical" environment. This factor is often thought of as the context immediately surrounding the organization within which the work unit functions and is often referred to through such polarities as "stable" versus "dynamic" and "simple" versus "complex" (Thompson, 1967; Altman, Valenzi, and Hodgetts, 1985). Generally, the more dynamic and complex the environment, the more reliance the work unit places on the personal and group modes of control. Empirical work using contingency theory has not considered the effects of an institutional environment (Meyer and Rowan, 1977; Scott, 1987), however, as this paper does.

Three specific refinements of work-unit contingency theory are useful for developing our theoretical model. First, earlier theorists proposed that the three modes of coordination and control need not be substitutes for one another but, rather, may be complements to each other. Following a theoretical position established by Thompson (1967), Van de Ven, Delbecq, and Koenig (1976: 323) reasoned that the bureaucratic, personal, and group modes of control may be used in context-driven combinations and concluded that a critical research issue is to isolate the situational factors that give rise to certain combinations. Their empirical results suggested, for example, that increased interdependence among work-unit personnel was associated with the additive use of all three coordination and control modes.

Second, more recent work-unit contingency theory research has progressed from merely examining the association between work-unit coordination and control practices and internal and external contingency factors to studying the effect of these relationships on performance. Arguing that work-unit performance improves when a "fit" exists between context and design, researchers have tended to focus on examining one internal contingency factor at a time, such as task uncertainty (Drazin and Van de Ven, 1985) and work-unit interdependence (Ito and Peterson, 1986). Advancing this work and using Van de Ven's original data, Gresov (1989) found greater efficiency in work units in state-level employment-security offices in which a fit existed between two contingency factors (task uncertainty and horizontal dependence and the modes of coordination and control), than in those without such a fit.

Third, the latest work-unit contingency theory research includes what has come to be known as the "systems" approach (Drazin and Van de Ven, 1985: 519-522; Gresov, 1989: 437-440). In contrast to the reductionism of other variants of the theory, the systems approach considers holistically the effect of multiple and possibly conflicting contingencies on work-unit performance in order to promote a more complete understanding of work-unit design. Gresov (1989: 433, 439), for example, recognized that organizational coordination and control practices are social constructions that may become institutionalized over time. He reasoned that seemingly conflicting contingencies like those calling for the bureaucratic mode of control versus those calling for the group mode of control may be resolved without having to redesign work-unit tasks. By recognizing, for example, that a set of coordination and control practices represents a "menu." Gresov (1989) theorized that management can select certain practices to serve a symbolic role in institutionalized contexts and other practices to serve a more instrumental role for actually achieving control and improving performance, even though the two selections may seem to be in conflict. Such theorizing about the symbolic and the instrumental roles of coordination and control practices has been further developed within institutional theory.

Institutional theorists generally reason that because an organization depends for survival on the support of external constituents, it must conform to accepted social norms; this reasoning applies especially in the case of government organizations (Meyer and Rowan, 1977; Meyer et al., 1978; DiMaggio and Powell, 1991; Scott, 1987). These institutionalized expectations are expressed in a broad class of elements that includes rules, blueprints for action, standard operating procedures, impersonal prescriptions, rationalizing techniques, formalization, and documentation (Meyer and Rowan, 1977: 341-343; Scott. 1987: 497-499). Meyer and Rowan (1977: 348) proposed that the formalized coordination and control systems of organizational structure serve as "rationalized and impersonal prescriptions that identify various social purposes [demonstrating institutional conformity] as technical ones [affecting control] and specify in a rule-like way the appropriate means to pursue these technical purposes rationally." Thus, irrespective of the task characteristics of internal work-unit operating processes, institutional forces contribute to a heightened emphasis on the relatively documentable, bureaucratic mode of control as a form of procedural conformity.

Institutional theorists also recognize a conflict between conformity to institutionalized standard operating procedures and efficiency criteria, especially in government and professional sectors, where the taken-for-granted form of organization is bureaucracy and survival is mainly a matter of legitimacy and only secondarily of actual performance (Meyer and Rowan, 1977: 352; Scott and Meyer, 1983). Thus, one expects the bureaucratic mode of control to be unrelated to the work unit's actual performance (DiMaggio and Powell, 1983: 147; Scott, 1987: 504; Powell, 1988: 115). Two hypotheses proceed from our reasoning:

Hypothesis 1a (H1a): Highly institutionalized contexts promote reliance on the bureaucratic mode for controlling work-unit processes.

Hypothesis 1b (H1b): Within highly institutionalized contexts, use of the bureaucratic mode of control does not promote work-unit efficiency.

The adoption of the bureaucratic mode of control projects an image of tidiness and clarity, although it tends to underrepresent the indeterminacy, complexity, and messiness of internal operations. Procedures applied behind the scenes to control work-unit members must still address the efficiency of the organization and its internal work units (Meyer and Rowan, 1977: 353; Scott, 1987: 504; 1992: 267-269). But the structures that demonstrate institutional conformity can conflict with those that foster efficiency and are sensitive to specific, nonstandardized problems unique to professional organizations. This is particularly true for professional auditing organizations. The idiosyncratic, internal operating processes inherent in audit tasks, which themselves differ in terms of task variability, task difficulty, and task interdependence, tend to be accommodated by means of less formalized control practices (Abbott, 1988). The personal and group modes of control, being more organic and better suited to informally aligning task determinacy with more fluid control mechanisms (Gresov, 1989), thus would tend to improve the performance of the work unit. From a work-unit contingency theory perspective, Drazin and Van de Ven (1985: 517) noted that "structure and process variables not prescribed [institutionalized] at the macro level are left to the particularistic control of the subunit. Only these variables should interact with context to explain variations in performance." We propose two additional hypotheses:

Hypothesis 2a (H2a): Work-unit processes in organizations that operate in highly institutionalized contexts will have higher levels of (a) task variability, (b) task difficulty, and (c) member and supervisor interdependence and greater reliance on the personal and group modes of control than those not in such contexts.

Hypothesis 2b (H2b): The more consistent the alignment between institutionalized expectations, task characteristics, and work-unit design, the higher is the work-unit efficiency.

Our two sets of hypotheses thus result from a blend of contingency and institutional theories. Contingency theory predicts that dynamism and complexity arising in the task environment will give rise to interdependence and complexity in work-unit tasks. It also predicts that a fit between these task conditions, complexity, and interdependence and control and coordination practices will promote work-unit performance. In contrast, the hypotheses suggest that contingency and institutional theories together predict that multiple and complex demands in the institutionalized environment give rise to bureaucratization (H1a), but bureaucratization does not promote work-unit performance (H1b). The complementary use of contingency and institutional theories also predicts that institutionalized environments are associated with work-unit task complexity (H2a), and when the complexity is handled with more social control practices, work-unit performance improves (H2b). Thus contingency and institutional theories together suggest that both bureaucratic and social means of control will exist in an organization and, further, that their use will serve the differential functions of demonstrating ceremonial conformity and actually controlling internal work processes, respectively. Figure 1 presents a model depicting the relationships between the variables expressed in the hypotheses that respond to the call by contingency theorists to consider multiple contingency factors (Schoonhoven, 1981; Drazin and Van de Ven, 1985; Gresov, 1989). (Figure 1 omitted) The model includes a greater number of external and internal factors than the most ambitious empirical treatment to date in the literature (Gresov, 1989). This study itself counters the observation that institutional theory is imprecise as to which components of an organization's institutional forces may or may not exert an influence (Scott, 1987: 501; Tolbert, 1988: 101, 109). By specifying a number of these forces in our model and testing them empirically, we add to the institutional literature, much of which has been only theoretical. Thus, examining contingency and institutional explanations of work-unit processes empirically promises to extend both perspectives .

METHOD

RESEARCH CONTEXT

We carried out our research in the United States General Accounting Office (GAO). The Budget and Accounting Act of 1921 established the GAO as an independent agency within the legislative branch of the federal government, granting it broad oversight powers to audit a national budget that today exceeds one trillion dollars. The GAO's principal resource is its professional personnel. Over 75 percent of the GAO's own $400 million budget underwrites the salaries and related expenses for its 5,000 members (Sexton, 1986; GAO, 1991). As an independent auditor of the U.S. government, the GAO is accountable only to the U. S. Congress. The GAO also audits any organization outside the federal government that receives funding of over $25,000 a year, and there are GAO employees all over the world--anywhere that U.S. government money is spent. Since the 1920s the GAO has continuously expanded the scope of its activities and has helped immensely to broaden the scope of the traditional audit function. At present, the GAO is involved in conducting three kinds of audits: financial and compliance audits, efficiency and effectiveness audits, and program evaluation audits. In the accounting and auditing literature, the latter two types of audits are often collectively referred to as operational audits. Our concern in this study is with audit teams conducting such operational audits. While financial and compliance audits are conducted mostly along the lines of audits conducted by CPA firms, efficiency and effectiveness audits and program evaluation audits take the concept of governmental accountability a step further by examining not only whether a governmental agency has used and managed its financial and nonfinancial resources economically and efficiently but also whether the program objectives as established by the legislature have been achieved or not.

One can describe the GAO as a heteronomous professional organization in that its professional employees are subject to administrative controls that circumscribe individual discretion and autonomy (Scott, 1992: 254). The work of GAO members is thus subject to routine supervision as required in its own codified auditing standards (GAO, 1988: 37). The GAO's core activities--auditing of the federal government's myriad agencies--are performed by temporary audit teams, which are work units whose personnel is reconfigured for each audit. Typically, a GAO audit team comprises auditors with accounting and auditing background and includes three hierarchical levels: group directors, evaluators-in-charge, and staff accountants. The auditor in charge of an audit team is called the group director, and the immediate audit-team supervisor is known as the audit manager or the evaluator-in-charge.

DATA COLLECTION

Interviews with high-level GAO administrators in Washington, D.C. helped us identify audits that varied in size, task complexity, and the degree to which the agency audited was important to external constituents. We decided to include in our sample only the audits that were completed during the last six months. We restricted the time frame during which the audits were completed to enable the respondents to recall the details on audit-specific questions with relative ease and accuracy. The final selection of the sample of audits, which used a theoretically based sampling plan (Lincoln and Guba, 1985: 102), was based on these interviews, a review of GAO archival material, and review of the extensive press coverage of GAO audit reports. For our final sample of 96 audits, the average audit team size was 4.2, with a range of 2 to 13 members. The number of sites at which an audit was conducted ranged from 1 to 14. We also used the interviews to ascertain the appropriateness to the GAO context of existing measures of such things as coordination and control practices and to develop new measures needed for our study.

MEASURES

The control and coordination variables, bureaucratic, personal, and group as well as task variability, task difficulty, task interdependence were measured using an instrument developed and validated in two studies, by Van de Ven and Delbecq (1974) and Van de Ven, Delbecq, and Koenig (1976). Validity and reliability diagnostics appear in Appendix A. We used two separate scales of four items each, as in Van de Ven, Delbecq, and Koenig (1976: 333-334), to measure task variability and task difficulty, rather than using the expanded, 40-item scale developed by Van de Ven and Ferry (1980: 159-168), because most of the additional items in the expanded scale were not relevant to the current study. Thus care should be taken when comparing our results with results of studies using the 40-item scale.

Unlike prior studies (Gresov, 1989; Gresov, Drazin, and Van de Ven. 1989), we did not aggregate task variability and task difficulty. Because segregating the two measures yielded superior measurement diagnostics, we analyzed them separately (see Appendix A). We measured audit-task interdependence with a six-item scale used by Van de Ven, Delbecq, and Koenig (1976: 334-336) that assessed job dependence on audit-team supervisor and on other audit-team members. We measured the three dimensions of coordination and control--bureaucratic, personal, and group--for each audit by asking the audit-team members to respond to an eight-item Likert scale.

Following Van de Ven (1976), Kimberly (1976), Dewar and Hage (1978), and especially Scott (1992: 258-259), we measured size along two dimensions, using information obtained from the GAO master assignment sheets, which lists all audits and teams assigned to them, along with the deadlines by which various phases of the audit need to be completed and the deadline for the final audit report. Consistent with Scott's structural-property interpretation, we defined size as the total number of auditors assigned to each audit team. Consistent with his environmental interface and institutional interpretations, we defined size also as the number of locations in which the audit was conducted. This second measure also conforms with various size measures, as identified by Van de Ven (1976: 68-69), with respect to an organization's number of programs, sections, and units. More importantly, this second measure is a proxy for auditee size, in terms of the number of locations at which its operations are conducted. If a large number of congressional districts are served by the agency being audited, the audit is visible to more external constituents, such as members of Congress and newspapers.

Institutional theorists disagree about how an institutional environment should be conceptualized and which of its features are most salient (Scott, 1992). Because an institutional environment "cannot be defined a priori, but must be defined on the basis of empirical investigation" (DiMaggio and Powell, 1983: 148), we held semistructured interviews with GAO administrators to discuss the elements of the GAO environment to which they were most sensitive. The public is one such element. Because the GAO serves as an agency of Congress, reports its findings directly to Congress, and receives virtually all of its funding from Congress, Congress itself is another important element. Given the extensive newspaper coverage that can attend a GAO audit, the press is yet another important element, and given specific requirements that the GAO solicit comments of the agency audited before issuing a final report (GAO, 1988), an audited agency is the fourth important element. A review of GAO archival material corroborated the interview evidence and confirmed the importance of these four elements. Based on these interviews and archival material, we developed an 11-item scale. Nine of the items used a 5-point Likert scale, and the other two offered a yes/no response choice. The administrators we interviewed judged this nine-item scale (reproduced in Appendix B, panel A) to be highly relevant to understanding the GAO's institutional environment. and no participants completing the questionnaire questioned its relevance.

We developed two open-ended questions (Appendix B, panel B) to measure audit efficiency related to the timely completion of two GAO activities: the audit itself and the distribution of the resulting report. Prior to the start of the audit a detailed budget for each milestone is developed, in terms of the staff days needed to complete an audit. At the same time, either the requestor of the audit or the GAO itself establishes a deadline for delivery of the final audit report. The respective questions to measure audit efficiency were also developed from interviews with GAO officials, and an extensive review of internal memoranda and GAO policies corroborated their critical importance.

SAMPLE AND QUESTIONNAIRE

Three hundred fifty-five questionnaires were distributed to each member of 105 audit teams that included group directors, evaluators-in-charge, and staff accountants. A total of 331 questionnaires (93 percent) were returned. After eliminating partially completed questionnaires or those from teams for which we lacked a response from each rank of the team, the final sample consisted of 299 respondents from 96 audit teams representing all but one regional office out of the 15 U.S. regional offices. Although the response rate was very high, we still checked for response bias by examining the temporal distribution of returned questionnaires and comparing completed and partially completed questionnaires and found no evidence of it (Oppenheim, 1966). On average, the respondents were 42 years old, with 15 years of experience at the GAO. The average government service (GS) rank exceeded level thirteen. Forty-three percent of the participants had education beyond the baccalaureate degree, and degrees were across a number of disciplines.

Each respondent received the name of a specific audit on which he or she had worked in the past six months and was asked to use that audit (referred to as "the most recently completed audit") as a point of reference. Following the approach of Van de Ven, Delbecq, and Koenig (1976), we averaged the scores for each audit team. Given audit-team members' professional orientation and length of GAO service, we weighted the scores equally across team members in computing team scores.

DATA ANALYSIS

We used LISREL to test hypotheses, for two reasons. First, it allowed us to examine multiple contingencies to better understand coordination and control in a social context, and, consistent with the systems approach of contingency theory, it allowed us to focus on the overall pattern of relationships among internal and external factors (Drazin and Van de Ven, 1985; Gresov, 1989). LISREL explicitly differentiates between observed and latent variables, and the latter are not assumed to be measured perfectly by the former. It requires both explicitly stating the correspondence between observed and latent variables and specifying relationships among theoretical constructs precisely, and all the assumptions necessary to the connections or patterned relationships among constructs are made clear. LISREL thus avoids the distortions introduced when one adopts a more piecemeal strategy (see Joreskog and Sorbom, 1989; Hayduk. 1987; Hughes, Price, and Marrs, 1986).

The second reason for using LISREL was that it allowed us to estimate measurement and structural models simultaneously through confirmatory factor analysis and two-stage, least-squares estimation (Goldberger, 1972). Measurement error is estimated through a measurement model component that allows one to conduct confirmatory factor analysis. Because the researcher must specify expectations a priori, this capability represents a powerful alternative to exploratory factor analysis. We conducted confirmatory factor analysis on the latent variables depicted in Figure 1. Relative fit statistics showed that the latent variables depicted there could not be expressed in alternative ways. Diagnostic results suggest that task variability and task difficulty, for example, should not be combined into a single "task complexity" measure, as in prior research (Gresov, 1989; Gresov, Drazin, and Van de Ven, 1989). In most cases, a single factor solution proved superior to multiple factor structures. For task interdependence, however, the model that separated work-unit member interdependence from supervisor interdependence yielded better psychometric properties than the model that combined them (see Appendix A).

RESULTS

We used a "theory trimming" approach to the LISREL analysis. This approach involves the iterative deletion of structural equation relationships found not to be significant individually until the best fitting model is produced (Hayduk, 1987). For simplicity, the model depicted in Figure 2 includes only the paths found to be significant at the .05 level. (Figure 2 omitted) The structural equation and measurement model results associated with this analysis are presented in Table 1. (Table 1 omitted) We also performed additional LISREL analyses that are not reported here. In the first of these, we eliminated the institutional environment construct to examine the impact of size more closely. Then, to examine the possible influence of individual auditor experience on the reported relationships, we divided the auditor sample into more-experienced and less-experienced subsamples, using as measures of experience the participant's length of service at the GAO and GS (government service) rank. We then ran concurrent LISREL analyses of the less-experienced and more experienced subsamples. A discussion of the results of these additional analyses is in the Discussion section below.(1)

Results of the LISREL analysis are presented in Figure 2. The paths depicted in Figure 2 support H1a (p < .01), suggesting that audits conducted on agencies operating in institutionalized settings are associated with greater use of the bureaucratic mode of control than agencies not in those settings. Results also support H1b, indicating that this greater use of the bureaucratic mode of control is not associated with variation in the efficiency of the audit.

The results also partly support H2a: Audits conducted in institutionalized contexts are perceived to exhibit higher task variability (p < .01), task difficulty (p < .01), and supervisor interdependence (p < .01). Figure 2 suggests, however, that only task difficulty, supervisor interdependence, and member interdependence are associated with the personal or group modes of control; task variability is not associated with increased reliance on either of these modes. This last finding would have been lost had we combined task variability and task difficulty into task complexity, as in prior contingency theory research. Hypothesis 2b was fully supported. In highly institutionalized environments, paths between both the personal and group modes of control and efficiency are significant at p < .01.

There are three notable nonhypothesized results in Figure 2. First, more institutionalized audit contexts are directly associated with an increased use of the personal mode of control, indicating that this mode plays a critical role in such settings. Second, the inclusion of the institutional environment factor has not completely swamped the effect of size; larger audits are associated with higher perceived task variability and difficulty. In turn, task difficulty is positively associated with the group mode of control, which is positively associated with performance and in harmony with the effects of the institutional environment. Third, Figure 2 shows that member interdependence is associated both with increased use of the group mode of control, as hypothesized, and also directly with the bureaucratic mode of control.

DISCUSSION AND CONCLUSIONS

This is the first empirical study that combines contingency and institutional theory perspectives to examine the effect of institutional forces in a work-unit setting. The LISREL results demonstrate that the two perspectives can be combined to study and understand the relationships in a government professional organization between such elements as institutionalized setting, size, task characteristics, work-unit interdependence, coordination and control, and work-unit performance. The results also support the institutional theory position that government and professional organizations apply bureaucratic control that is unrelated to work-unit performance, while backstage they coordinate and control and improve work-unit performance through the more social and idiosyncratic personal and group modes. Consequently, this study contributes to an understanding of the connection in heteronomous, professional organizational settings between external conditions, expressed using institutional theory, and internal structure, expressed using work-unit contingency theory.

Some nuances and limitations in our empirical work point to a number of directions for future research. As evident by the absence of legitimacy from Figures 1 and 2, the first limitation is that we did not specifically examine legitimacy, an additional potential outcome of audit-team processes. We did examine conformity to the general expectation of bureaucratic practice in public sector organizations, as well as the effect of team processes on performance, but we did not hypothesize or test the effect of this conformity on achieving legitimacy, although the final reports from all of the 96 audit teams were ultimately accepted by the GAO's central administration and were formally published. We originally reasoned that achieving legitimacy in the eyes of external constituents is a more macro, organization-level phenomenon that is not achieved solely at a micro, work-unit level by adopting a bureaucratic mode of control. Our results indicate, instead, that legitimacy is achieved by adopting a range of rationalizing practices, of which bureaucratic control is but one exemplar. In addition, legitimacy may be achieved by having appropriately credentialed work-unit members, with degrees in appropriate disciplines (e.g., accounting and engineering), certifications (e.g., C.P.A.), and an appropriate number having graduate degrees (Scott, 1992: 198-212). We did not examine the effect of conformity on achieving legitimacy but found, nevertheless, a number of indicators that the GAO has legitimacy. These indicators include the length of time the GAO has been in existence (established in 1921); its annual budgetary appropriations from Congress (currently $400 million annually); the extensive and positive newspaper coverage its audits enjoy (in 1991 and 1992, 433 citations appeared in the New York Times, Wall Street Journal, Los Angeles Times, and Washington Post indices); the extensive use of its audit reports in congressional hearings (the GAO was called to offer 277 pieces of congressional testimony in 1991; GAO, 1991); and the intensive use of GAO reports by powerful members of Congress, such as Representative John Dingell, chairman of both the House Energy and Commerce Committee and its Subcommittee on Oversight and Investigations (Rogers, 1990). While we believe that bureaucratic control helps the GAO maintain its legitimacy, which is quite high, we nevertheless recommend that future research be directed at studying the effect of organizational-level and work-unit level actions on gaining legitimacy.

A second limitation concerns our restricted operational definitions and measures of coordination and control, particularly as they relate to the personal and group modes. Here, we sought to examine the idiosyncratic and social means by which organizations actually achieve control and improve performance, as well as those forces influencing the application of these means. While this theoretical construct was relevant, and Van de Ven and his associates' approach offered a useful first approximation and yielded promising empirical results, we believe it may have underrepresented what institutional theorists call the idiosyncratic and social (Meyer and Rowan, 1977: 356-359). On this point, Drazin and Van de Ven (1985) noted that for an organization to survive, its structure and processes should match its culture. environment, technology, and size. Our analysis of the personal and group modes of control failed to embrace more theoretically ambitious concepts such as organizational culture, which could usefully be included in future research, along with more extended definitions than ours of the idiosyncratic and social processes of achieving control.

Third, we focused on the societal expectations of formalized, rationalized structure as embodied in the bureaucratic mode of control. While this focus conforms to a prominent line of reasoning, it represents but one of the many forms of societal expectation (Scott and Meyer, 1983; Scott, 1987: 500-502). For example, our sample included a significant number of auditors with graduate work up to the Ph.D. across a number of disciplines. These achievements may indicate not only instrumental competence but also a ceremonial conformity to societal expectations that call for such credentials Scott, 1992: 211-212). Similarly, idiosyncratic and social controls may be applied not merely for instrumental purposes but also to comply with the general expectations of society as to how professional work should be orchestrated (Abbott, 1988). The multiple expectations that flow through institutionalized work processes should be the subject of future research.

Fourth, we defined size as the number of work-unit members, with the inference that bureaucratic control and size should vary directly (Fry and Slocum, 1984). We also interpreted size as involving environmental and institutional interfaces and defined it as the number of locations across which the audit had been conducted, with the inference that size varies directly with the personal and group modes of control (Scott, 1992: 258). As Figure 2 shows, the empirical results suggest that the environmental interface dominates. This supports the institutional interpretation, in that size varied directly with task variability and task difficulty but was not directly related to bureaucratic control. The unreported LISREL analyses that excluded the institutional measure, which overwhelmed size as reported in Figure 2, suggested a much more pervasive size influence, again, with larger-sized audits being directly associated with task variability, task difficulty, and task interdependence and consequently with the personal and group modes of control. Future research should examine a more extended concept of size and its interrelationship with institutional forces. This might be pursued in governmental sectors by including such additional measures of size as agency funding, a variable unavailable to us during our fieldwork.

Fifth, consistent with cognitive psychology research (Chiesi. Spilich, and Voss, 1979; Chi, Glaser, and Rees, 1982), which posits that more experienced people learn to penetrate surface features and develop a more robust understanding of underlying relationships, it may be that the experience level of individual auditors exerts an influence on the relationships depicted in Figure 2. The findings of the unreported LISREL analyses, described above, that probed the effect of experience generally corroborate the results reported in Figure 2, but more experienced people perceived that there were more pervasive, direct and indirect influences of the institutionalized environment construct than did their less-experienced counterparts. In support of this finding, Meyer and Rowan (1977) reasoned that organization leaders tend to buffer lower-level members from the effects of institutional forces. Future research might usefully examine the effects of social actors' experience within institutionalized settings.

Finally, we conducted our empirical work in a heteronomous professional organization within the government, a context particularly subject to institutional pressures. Thus, while researchers should be wary of generalizing our findings to less institutionalized environments, previous research on work-unit contingency theory, such as that done with Van de Ven's data from state-level employment security offices, may have underrepresented the effect of institutional forces as an omitted variable. Research, both theoretical (DiMaggio and Powell, 1983: 155-166) and empirical (Tolbert and Zucker, 1983; Covaleski, Dirsmith, and Jablonsky, 1985), from an institutional theory perspective, points to an institutional effect within state and local governments and related organizations. If we are to understand how these organizations really function, we have to expand our perspectives on them to include the institutional environment in which they operate.

(1) Detailed descriptions of the trimming process, including further analysis of size and the effect on the experience level of each audit team member, are available from authors on request.

REFERENCES

Abbott, Andrew 1988 The System of Professions. Chicago: University of Chicago Press.

Altman, Steven, Enzo Valenzi, and Richard M, Hodgetts 1985 Organizational Behavior: Theory and Practice. New York: Academic Press.

Chi, M. T., R, Glaser, and E, Rees 1982 "Expertise in problem solving." In R. Sternberg (ed.), Advances in the Psychology of Human Intelligence: 213-231. Hillside, NJ: Erlbaum.

Chiesi, H. L., G, J. Spilich, and J. F. Voss 1979 "Acquisition of domain-related information in relation to high and low knowledge." Journal of Verbal Learning and Learning Behavior, 18: 257-273.

Covaleski, Mark, Mark W. Dirsmith, and Stephen Jablonsky 1985 "Traditional and emergent theories of budgeting: An empirical analysis." Journal of Accounting and Public Policy, 4 (Winter): 277-300.

Dewar, Robert, and Jerald Hage 1978 "Size, technology, complexity, and structural differentiation: Toward a theoretical synthesis." Administrative Science Quarterly, 23: 111-136.

DiMaggio, Paul J., and Walter W. Powell 1983 "The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields." American Sociological Review, 48: 147-160.

1991 "Introduction." In W. W. Powell and P. J. DiMaggio (eds.), The New Institutionalism in Organizational Analysis: 1-38. Chicago: University of Chicago Press.

Drazin, Robert, and Andrew H, Van de Ven 1985 "Alternative forms of fit in contingency theory." Administrative Science Quarterly, 30: 514-539.

Feldt, Leonard S., David J, Woodruff, and Fathi A. Salih 1987 "Statistical inference for coefficient alpha." Applied Psychological Measurement, 11(March): 93-103.

Fry, Louis W., and John W, Slocum 1984 "Technology, structure, and work group effectiveness: A test of a contingency model." Academy of Management Journal, 27: 221-246.

Galbraith, Jay 1973 Designing Complex Organizations. Reading, MA: Addison-Wesley.

General Accounting Office 1988 Standards of Audit of Governmental Organizations, Programs, Activities and Functions. Washington, DC: U.S. Government Printing Office.

1991 Comptroller General's 1991 Annual Report. Washington, DC: U.S. Government Printing Office.

Goldberger, Arthur S. 1972 "Econometrics and psychometrics: A survey of commonalities." Psychometrika, 36: 83-157.

Gresov, Christopher 1989 "Exploring fit and misfit with multiple contingencies." Administrative Science Quarterly, 34: 431-452.

Gresov, Christopher, Robert Drazin, and Andrew H. Van de Ven 1989 "Work-unit task uncertainty, design and morale." Organization Studies, 10: 45-62.

Hayduk, Leslie 1987 Structural Equation Modeling with LISREL: Essentials and Advances. Baltimore: Johns Hopkins University Press.

Hughes, Marie A., R. Leon Price, and Daniel W. Marrs 1986 "Linking theory construction and theory testing: Models with multiple indicators of latent variables." Academy of Management Review, 11: 128-144.

Ito, Jack K., and Richard B. Peterson 1986 "Effects of task difficulty and inter-unit interdependence on information processing systems." Academy of Management Journal, 29: 139-149.

James, Lawrence R., Robert G. Damaree, Stanley A. Mulaik, and Robert T. Lodd 1992 "Validity generalization in the context of situational models." Journal of Applied Psychology, 77: 3-14.

Joreskog, Karl G., and Dag Sorbom 1989 LISREL 7. Chicago: SPSS Inc.

Kimberly, John R. 1976 "Organizational size and the structuralist perspective: A review, critique, and proposal." Administrative Science Quarterly, 21: 571-597.

Lincoln, Yvonna S., and Egon G. Guba 1985 Naturalistic Inquiry. Newbury Park, CA: Sage.

Meyer, John W., and Brian Rowan 1977 "Institutionalized organizations: Formal structure as myth and ceremony." American Journal of Sociology, 83: 310363. Meyer, Marshall, and Associates 1978 Environments and Organizations. San Francisco: Jossey-Bass.

Oppenheim, Abraham N. 1966 Questionnaire Design and Attitude Measurement. New York: Basic Books.

Powell, Walter W. 1988 "Institutional effects on organizational structure and performance." In L. G. Zucker (ed.), Institutional Patterns and Organizations: 115-136. Cambridge, MA: Ballinger.

Rogers, David 1990 "Man of Capitol: Representative Dingell wields wide power to probe much of industry." Wall Street Journal, March 5: A1, A7.

Schoonhoven, Claudia B. 1981 "Problems with contingency theory: Testing assumptions hidden within the language of contingency theory." Administrative Science Quarterly, 26: 49-77.

Scott, W. Richard 1981 Organizations: Rational, Natural, and Open Systems. Englewood Cliffs, NJ: Prentice-Hall.

1987 "The adolescence of institutional theory." Administrative Science Quarterly, 32: 493-511.

1992 Organizations: Rational, Natural, and Open Systems, 3rd ed. Englewood Cliffs, NJ: Prentice-Hall.

Scott, W. Richard, and John W, Meyer 1983 Organizational Environments: Ritual and Rationality: 129-153. Beverly Hills, CA: Sage.

Sexton, R. 1986 "GAO: Twenty years of change." GAO Review, Winter 17-21

Thompson. James D, 1967 Organizations in Action. New York: McGraw-Hill.

Tolbert, Pamela S. 1988 "Institutional sources of organizational culture in major law firms." In L. Zucker (ed.), Institutional Patterns and Organizations: 101-113. Cambridge, MA: Ballinger.

Tolbert, Pamela S., and Lynne G. Zucker 1983 "Institutional sources of change in the formal structure of organizations: The diffusion of civil service reform, 1880-1935." Administrative Science Quarterly, 28: 22-39.

Van de Ven, Andrew H. 1976 "A framework for organization assessment." Academy of Management Review, 1: 64-78.

Van de Ven, Andrew H., and Andre L. Delbecq 1974 "A task contingent model of work-unit structure." Administrative Science Quarterly, 19: 183-197.

Van de Ven, Andrew H., Andre L. Delbecq, and Richard Koenig, Jr. 1976 "Determinants of coordination modes within organizations." American Sociological Review, 41: 322-338.

Van de Ven, Andrew H., and Diane L. Ferry 1980 Measuring and Assessing Organizations. New York: Wiley.

APPENDIX A

Coordination-and-control questionnaire diagnostics. Van de Ven and Ferry (1980: 180) reported a coefficient alpha of .70 for the combined construct; for our GAO sample, this coefficient was .53. Despite this somewhat low reliability, we decided to retain these measures, without deleting any, to combine the contingency and institutional theory perspectives more faithfully. The somewhat low reliability in an institutional context may mitigate against an institutional effect (i.e., increase alpha error): LISREL confirmatory factor analysis demonstrated that a three-factor model that separated the latent constructs of bureaucratic, group, and personal modes of coordination and control maximized fit statistics. This model had the following results: GFI = .945, AGFI = .883, chi-square = 81.80, d.f. = 17, root mean square residual = .072, coefficient of determination of X variables = .819. This model was a significantly better fit (p < .01) than any nested alternative. All lambda loadings on the latent constructs were significant at p < .01.

Task characteristics. Van de Ven and Ferly (1980: 159-168) reported alpha coefficients of .81, .72, and .59 for task variability, task difficulty. and interdependence, respectively; for our GAO sample, these coefficients were .47, .80, and .62. Unlike prior researchers (e.g., Gresov, Drazin, and Van de Ven, 1989), we maintained the distinction between task difficulty and task variability. The theoretical distinction appeared to have merit; test diagnostics turned out to he superior for the segregated approach. Confirmatory factor analysis results for task variability indicated that a one-factor model was the best expression of the data (GFI = .971, AGFI = .857, chi-square = 19.44, d.f. = 2, root mean square residual = .046, coefficient of determination of X variables = .871. The same was true for task difficulty (GFI = .970, AGFI = .849, chi-square = 21.14,d.f. = 2, root mean square residual = .074, coefficient of determination of X variables = 0.747). While interdependence was not very well represented by a single factor, a two-factor model differentiating interdependence with the supervisor and team-member interdependence produced stronger results (GFI = .976, AGFI = .938, chi-square = 24.47, d.f. = 8, root mean square residual = .040, coefficient of determination of X variables = .947. In all cases, all loadings were significantly related to the latent construct at p < .01.

Institutional context. The eleven-item institutional environment scale we developed was specific to the GAO's environment (Appendix B. panel A). We conducted confirmatory and exploratory factor analysis (Cronbach alpha = .81), imposing a one-factor solution that resulted in an acceptable level of consistency between each measurement item and the latent factor. A single factor is also more consistent with the construct as theorized and is parsimonious. Confirmatory factor analysis revealed the superiority of a one-factor model (GFI = .877, AGFI = .795, chi-square = 193.87, d.f. = 27, root mean square residual = .076, coefficient of determination of X variables = .864), with all items loading on the latent construct at p < .01.

In addition, this scale was consistent with both the archival and interview data. This consistency may be attributable to the organization's recognition of the uncertainty posed by the institutional environment, evident in the archival material and interviews, and its concomitant development of appropriate information and planning capabilities, its reliance on rich media sources (e.g., direct contact with Congress and use of relevant legislation and the press), and extensive boundary-spanning activities as the auditor of other governmental agencies. This scale should nevertheless be considered exploratory and only suggestive of how future researchers may seek to address the institutional measurement problem.

Performance. Because the efficiency metric included two items (Appendix B, panel B), confirmatory factor analysis and Cronbach alpha were not relevant. The Pearson chi-square statistic (appropriate because of the yes/no component of these items) was 54.23 (p < .00) suggesting that this combination is appropriate.

In view of the institutional theory argument, it may be that audits performed in a highly charged environment may alter the reliability of operational measures of dependent and independent variables, suggesting thereby that the issue of reliability is potentially more complex than is recognized in the contingency theory literature (for a related, general discussion, see James et al., 1992). In order to assess this possibility, audits examined were partitioned into "highly charged" and "low charged" categories. The reliabilities were recomputed and (statistics computed (Feldt, Woodruff, and Salih, 1987) for all measures discussed here. Although reliabilities were not significantly different (p < .05), researchers examining the effect of institutional theory constructs using existing scales should address this issue.

Although LISREL allows one to conduct confirmatory factor analysis as an integral part of the main analysis and therefore represents a powerful alternative to exploratory factor analysis, a separate confirmatory factor analysis was nevertheless conducted to permit comparisons with prior research. Table A.1 presents this analysis, while Table A.2 presents the correlation matrix of the variables studied. The results of these analyses corroborate those obtained from the LISREL analyses reported in the paper. (Table A.1 and A.2 omitted)

APPENDIX B: Questionnaire

Panel A: Institutional Environment Scale

[Participants were asked to respond to the following items, with reference to their most recently completed audit, on a 5-point scale, on which 1 = little or no extent; 2 = some extent; 3 = moderate extent; 4 = great extent; and 5 = very great extent.]

1. To what extent was the public interested in the findings of the audit?

2. To what extent was the Congress (e.g., its members or its committees) interested in the findings of the audit?

3. To what extent were the press or the media interested in the findings of the audit?

4. To what extent were the officials of the audited agency interested in reviewing and discussing the findings of the audit report prior to its submission to the requestor? (If there was no requestor, did they still show any interest?) 5. To what extent were you required to devote greater attention and time to ensure that detailed working papers, documents, and other records were properly kept?

6. To what extent was emphasis placed on the documentation of methodology (i.e., as to how you conducted the audit and arrived at findings)?

7. To what extent was the audit controversial and visible to important parties external to the GAO?

8. To what extent, in your opinion, did the audit deal with politically sensitive issues?

9. How often, in your opinion, did you, other audit team members, or any other representative of the GAO have contact with the news media concerning the audit?

[Participants were asked to check "Yes" or "No" in response to the following two items, with respect to the most recently completed audit.]

10. Did the requestor of the audit direct you not to share the audit findings with the audited agency before he or she could review them?

11. Did the audited agency make more requests to review the findings of the audit than it normally does for its other audits?

Panel B: Audit Team Efficiency Scale

[Participants were asked to check "Yes" or "No" in response to the following two items, with respect to the most recently completed audit.]

1. Did you complete the audit within the budgeted staff days! If not, how many more staff days did you take?

2. Were you able to deliver the audit report to the requestor (if any) on time?

* The authors thank Richard Bord. Anthony Hopwood, James Thies. N. S. Umanath. Marshall Meyer, and the anonymous reviewers for their useful suggestions. They also thank the United States General Accounting Office and its personnel for their cooperation and the Institute of Internal Auditors Research Foundation and the Price Waterhouse Foundation for their financial support.