WK3-410-CH9

9 A Preliminary Theory of Interorganizational Network Effectiveness: A Comparative Study of Four Community Mental Health Systems Keith G. Provan H. Brinton Milward This chapter presents the results of a comparative study of interorganizational networks, or systems, of mental health delivery in four U.S. cities, leading to a preliminary theory of network effectiveness. Extensive data were collected from surveys, interviews, documents, and observations. Network effectiveness was assessed by collecting and aggregating data on outcomes from samples of clients, their families, and their case managers at each site. Results of analyses of both quantitative and qualitative data collected at the individual, organizational, and network levels of analysis showed that network effectiveness could be explained by various structural and contextual factors, specifically, network integration, external control, system stability, and environmental resource munificence. Based on the findings, we develop testable propositions to guide theory development and future research on network effectiveness. The study of relations between organizations has been a major concern of organization theorists for at least the past 25 years. While most of the work in this area has focused on the determinants or predictors of interorganizational relations (see Oliver, 1990, for a review), as an understanding of the phenomenon has grown, the unit of analysis has gradually shifted from the dyad to the organization set, to the network. Especially in recent years, the study of organizational networks has proliferated. Much of this interest has been generated by an emerging recognition by academics that businesses, as well as organizations in the not-for-profit and public sectors, are increasingly turning to various forms of cooperative alliances as a way of enhancing competitiveness and effectiveness that would not be possible through the traditional governance mechanisms of market or hierarchy (Powell, 1990). While a good deal of what has been written about networks has been atheoretical, discussing the advantages of networks or examining issues of measurement and analysis, considerable theory-based research has also emerged (e.g., Cook, 1977; Burt, 1980; Granovetter, 1985; Jarillo, 1988; Williamson, 1991; Cook and Whitmeyer, 1992; Larson, 1992; Provan, 1993). In the organization theory literature, work on networks has been guided primarily by two theoretical perspectives: resource dependence, and related exchange perspectives, and transaction cost economics, with most recent work focusing on the latter approach. Each of these perspectives offers both complementary and contrasting views about the network form. For the most part, however, each perspective focuses essentially on the organizational antecedents and outcomes of network involvement, with little attention paid to the network as a whole, except for its governance and structure. This organizational view is understandable, since it is organizations that make up a network and organizations that stand either to lose or benefit by network involvement. In both the transaction cost and resource dependence literatures, for instance, the motivation and rationale for cooperative, interorganizational integration of activities and services is at the organizational level, either for reasons of efficiency related to reduced transaction costs (Williamson, 1985) or to gain resources and power (Pfeffer and Salancik, 1978). Individual organizations make strategic choices to form or become part of a cooperative network of other organizations when it appears that the advantages to such an arrangement, especially enhanced survival capacity (Uzzi, 1994), outweigh the costs of maintaining the relationship, including any potential loss of operating and decision autonomy. Thus, much of what is known about the rationale for network involvement is based on an extension of the literature on interorganizational relations. Absent from these views, however, is a focus on nonstructural outcomes of the network as a whole. Even in the general network literature, which places a heavy emphasis on network-level properties and structures (Aldrich and Whetten, 1981; Knoke and Kuklinski, 1982; Marsden, 1990; Scott, 1991), issues of network outcomes and effectiveness are mostly ignored. While focus on organizational effectiveness is clearly appropriate when outcomes can readily be attributed to the activities of individual organizations, not all problems can be solved by the actions of individual organizations. Particularly in the area of community-based health care and social services for such groups as the homeless, people with severe mental illness, drug and alcohol abusers, and the elderly, a focus on organizational outcomes is insufficient, because such outcomes reflect only how well individual providers are performing their particular component of the many services needed by their clients. If the overall well-being of clients is a goal, then effectiveness must be assessed at the network level, since client well-being depends on the integrated and coordinated actions of many different agencies separately providing shelter, transportation, food and health, mental health, legal, vocational, recreational, family, and income support services. Although individual agencies obviously have an important role to play in service delivery, and some agencies will clearly be more involved and provide higher-quality services than others, if overall client well-being depends on receiving different services provided by multiple agencies, client outcomes should be explainable by focusing on network-level activities and structure. The critical issue, both for clients and system-level planners and funders, is the effectiveness of the entire network of service providers, not whether some agencies that are part of the network do a better job than others in providing a particular component of service. Obviously, a network may be well integrated and still be ineffective if individual provider agencies do a poor job. At the opposite extreme, even though the agencies in a system may provide excellent services, overall outcomes may be quite low if clients can only gain access to some of these services. The prevailing view among many service professionals, policymakers, and researchers is that by integrating services through a network of provider agencies linked through referrals, case management, and joint programs, clients will gain the benefits of reduced fragmentation and greater coordination of services, leading to a more effective system (Warren, Rose, and Bergunder, 1974; Rogers and Whetten, 1982; Department of Health and Human Services, 1991; Goldman et al., 1992; Alter and Hage, 1993). In the public and not-for-profit sectors, where a public interest motive is involved, network outcomes are especially salient, and the rationale for organizations cooperating to accomplish system goals rather than organizational ends is often stronger than in the private sector, even when specific incentives to integrate and cooperate are weak. For key external groups like policymakers and funders, as well as for service professionals and clients themselves, emphasis is often on achieving outcomes that enhance the overall well-being of clients, without regard to whether the goals of individual provider organizations are met. Such outcomes may or may not have any direct impact on the effectiveness of many of the organizations that make up the network and may even result in a reduction in the number of organizations that provide services to a particular clientele if services are provided more efficiently. Arguments for the effectiveness of integrated networks of organizations have been particularly prevalent in discussions of the community-based care of people with severe mental illness (Windle and Scully, 1976; Turner and TenHoor, 1978; Tessler and Goldman, 1982; Grusky et al., 1985; Morrissey, Tausig, and Lindsey, 1985; Goldman et al., 1992; Alter and Hage, 1993). Much of the rationale for the need for integrated, coordinated services for these individuals stems from the failure of efforts to serve the many patients released from state mental hospitals since the 1960s. Deinstitutionalized patients were to draw on the services offered by community health and human service organizations, but little effort or money was expended to ensure that needed services were available or that clients would be able to move among community agencies to get these services. Since long-term hospitalization was no longer socially acceptable or politically feasible, the alternative treatment ideology that emerged was based on the integration of services into a system or network of providers (Bassuk and Gerson, 1978; Weiss, 1990). Integrated systems of community agencies can provide the full range of health and human services needed by severely mentally ill adults in a way that ensures continuity of care (Dill and Rochefort, 1989). Such continuity is particularly critical for this illness, which is estimated to afflict between 1.7 and 4 million Americans (National Institute of Mental Health, 1991), not only because of clients' multiple needs but because of the disorienting nature of the illness itself. This latter problem makes it extremely difficult for clients to be responsible for ensuring that their own treatment is effectively coordinated across a variety of autonomous agencies that are often scattered geographically. In addition, through coordination, an integrated system supposedly minimizes duplication of services by multiple provider agencies while increasing the probability that all essential services are provided somewhere in the system and that clients will have access to these needed services. Thus, system structure, particularly integration among service providers, is presumed to have a strong impact on outcomes for the mentally ill. Many of the original ideas behind services integration for the mentally ill were incorporated under the National Institute of Mental Health's (NIMH) Community Support Program, begun in 1977, although the specific form that integration was to take was purposely left vague (Grusky et al., 1985). Because funding for implementing this program was limited and no mandates were imposed, mental health professional groups and NIMH frequently relied on normative pressures (supported by some demonstration grants) to influence local integration behavior through establishment of a so-called community service ethic (Weiss, 1990). NIMH continues to call for research on the topic (Steinwachs et al., 1992), firmly believing that integration of the broad range of services needed by the severely mentally ill is critical for favorable client outcomes. Unfortunately, although the rationale for network involvement may be strong, there is almost no empirical evidence to support the presumed relationship between integration and network effectiveness or to indicate what network characteristics are associated with effective outcomes. Consistent with the ideas discussed above, our study was organized around a single fundamental research question: What, if any, is the relationship between the structure and context of mental health networks and their effectiveness? We operationalized effectiveness using a multitrait, multiperspective methodology designed to assess the overall well-being of severely mentally ill clients collectively served by the agencies that make up the health and human service delivery system in each of four mid-sized U.S. cities. From the study, we draw conclusions for theory and research on the effectiveness of organizational networks and for issues of health care policy that apply to service delivery through integrated and coordinated networks of organizations. Research Methods This study uses what Yin (1984) has described as a case survey approach, in which multiple levels of analysis (individual, agency, and network levels) are used to develop an in-depth picture of a single case. We used this approach in each of four systems, so that our study has a comparative case research design. We draw on extensive qualitative and questionnaire data collected from hundreds of individuals and organizations that we first aggregate by system to reflect the properties of that system and then compare. Research Sites and Criteria for Selection The community mental health systems in four U.S. cities were studied: Tucson, Arizona; Albuquerque, New Mexico; Providence, Rhode Island; and Akron, Ohio. Our choice of city was guided by three selection criteria. First, we wanted to choose cities of roughly comparable size that would be large enough to contain the full range of organizations, clients, and services existing in nearly all medium to large cities while not being so large that data collection and analysis would be unmanageable. Our cities ranged in size from 369,000 (Providence) to 667,000 (Tucson), using 1990 census data from the 1991 Statistical Abstract of the United States. Second, we wanted to choose cities having at least one core mental health agency whose role it was to provide services and to coordinate the services of other community agencies, if only through case management. These dual roles are typical of the core agency in most medium- to large-sized communities, although significant variance exists in funding, structure, actual services provided, and coordination emphasis. Finally, we used a most similar/most different research design (Przeworski and Teune, 1970), selecting two cities from states with relatively high levels of per capita mental health spending and two from states with low mental health spending. This would allow comparisons based on the relative resource munificence of the broader state system in which our community systems were embedded. We chose this approach because the majority of funding for community mental health flows to localities through the state (Mechanic and Surles, 1992). Using data reported by Torrey et al. (1990), resource munificence was operationalized as per capita mental health spending by each state's mental health authority. Spending was high in Rhode Island ($52.34) and Ohio ($45.33) and low in New Mexico ($23.79) and Arizona ($19.76).   Author's Note: This chapter is a reprint from Administrative Science Quarterly, 40(1), 1–33, 1995. This research was funded by a grant from the National Institute of Mental Health (R01-MH43783). The authors would like to thank Mike Berren, Dick Bootzin, A. J. Figueredo, Julie Sebastian, Lee Sechrest, and Ken Sublett for their contributions to this project and Terry Anfiburgey, Rob Burns, Christine Oliver, Chuck Windle, and three anonymous ASO reviewers for their many valuable comments on various drafts of the paper. Network Data Collection Procedures In each of the four cites, we used identical data and data collection procedures, based on a modified version of methods used in a pilot study (Provan and Milward, 1991). Data were collected in 1991 and early 1992. After each site was identified and we had confirmed initial cooperation by the core mental health agency, we traveled to each site and met with system leaders, including heads of key mental health providers, funders, and support groups. At this meeting, we explained procedures, sought further cooperation, and presented, discussed, and refined a list of organizations in the system. It was our intent to develop a comprehensive list of all agencies in the community that provided services to adults with severe mental illness. Our primary method for determining inclusion was what Laumann, Marsden, and Prensky (1983) referred to as positional, where an agency had to provide actual services or funding to our target population. Besides agencies having mental health as a prime goal, the mental health system at each site included agencies providing permanent and temporary housing and shelter, counseling, legal, food, transportation, income assistance, employment, rehabilitation, medical, drug and alcohol abuse, and recreation services. For many of these agencies, severely mentally ill (SMI) adults made up only a fraction of the total clientele served, but each agency and its services were an important piece of the entire community mental health system in that city. Consistent with the positional selection criterion, agencies were selected because of their involvement with SMI clients and not because they were already linked to other service providers for the SMI. Thus, we made every possible attempt to avoid including only those agencies that were already well integrated into the mental health network. From the list of mental health provider agencies that we generated and refined at each site, we developed a questionnaire for use at all four sites, with identical questions related to agency characteristics and network involvement. The main network question listed all agencies initially identified as being part of the system, at each site (43–55 agencies, depending on the site). The questionnaire was then mailed to each agency, addressed to the agency head or a key staff member, depending on who would be most knowledgeable about SMI adult services. The survey was accompanied by a cover letter from us as well as a letter of support signed by the head of the city's core agency, primary mental health funder, or both. About 10 days after mailing, we telephoned each agency to resolve any initial problems and set up an interview. At this point, a number of agencies at each site were eliminated when it became apparent that their level of involvement with the target population was so minimal that inclusion in the study made little sense. We then visited each site, jointly interviewing members of every agency in the system. One purpose of these in-depth meetings was to review questionnaire items and responses to ensure that respondents were interpreting them as we intended. The interviews also allowed us to eliminate several more agencies that should not have been included in the first place and to add additional ones, if mentioned by three or more respondents. This modified snowball procedure (Knoke and Kuklinski, 1982: 23) ensured capturing of any network agencies we might otherwise have missed. Finally, and most important, the interviews enabled us to collect much qualitative information on respondents' views and impressions of the city's mental health system, its history and functioning, and the agency's role in that system. We took extensive notes at each interview, which we discussed and compared and compiled into a final set of field notes. We made no attempt to structure these interviews, except that we made every effort to pursue issues that respondents frequently mentioned as important at each site. Questionnaires were either collected after the interview or returned by mail. We followed up with telephoning and additional interviewing to collect missing data and check data that appeared to be inaccurate when we compared questionnaire responses with field notes. In addition, we collected extensive mental health funding data from documents provided by state agencies and through interviews with state agency officials responsible for community mental health funding. Overall, the comprehensive data collection procedure took one year to complete but resulted in final response rates of 92 percent in Albuquerque (N = 35) and Providence (N = 35), 97 percent in Tucson (N = 32), and 100 percent in Akron (N = 36), with few missing data. Nearly two years after initial data collection began, as a final step in the process, we visited each of the sites once again, after all data were collected and initially analyzed, to present and discuss findings with system members. We did this as a reality check to ensure that our major conclusions were not inconsistent with their understanding of system operation. Client Data Collection Procedures: Network Effectiveness Network effectiveness was assessed using data collected from a 5 percent random sample of adult SMI clients in each community: 64 in Tucson, 59 in Albuquerque, 62 in Providence, and 80 in Akron. SMI clients were defined as adults (over 18) whose emotional or behavioral functioning is severely and persistently impaired in a way that interferes with their capacity to remain in the community without supportive treatment. Decisions about who qualified for this designation were made by mental health professionals at each site using standard evaluation criteria published in the Diagnostic and Statistical Manual of the American Psychiatric Association (3rd ed.), commonly referred to as the DSM-III-R. By using a random sample and by focusing only on individuals whose illness was severe and chronic, yet whose current problems were not sufficiently acute to warrant institutionalization, we were confident that our client populations would be comparable, although not identical in illness severity across sites. The limited data on severity we were able to collect in Tucson and Akron showed no significant differences in clients at these two sites. Clients were selected directly from the coded rosters of the core agency in each city and were only identified by name once they were actually selected for the sample. Clients were then contacted, through their case manager, and asked to participate in the study. Those who agreed met individually with our trained personnel who, through structured interviews, recorded clients' responses to questions concerning quality of life, adjustment, psychopathology, and satisfaction with services. The same information was obtained about the clients from a family member, if one was available, and from the clients' case manager or therapist. A clinical psychologist who was part of the research team, and had conducted such work previously, trained the interviewers and coordinated the client data collection effort. Measurement Network effectiveness. Comparative research on the effectiveness of organizational networks is virtually nonexistent, except for recent work by Morrissey et al. (1994) that relied on global perceptions of network performance and work by Lehman et al. (1994) that was inconclusive. As noted earlier, part of the problem is that most researchers are more concerned with organizational outcomes, even when the organizations studied are part of a network. In addition, actual assessment of network effectiveness, like organizational effectiveness, is extremely problematic (see Goodman and Pennings, 1977, and Cameron and Whetten, 1983, for reviews of organization-level problems). The early effectiveness literature often focused on the accomplishment of predetermined organizational goals that were ill-defined and often in conflict or on the receipt of scarce and valued resources (Yuchtman and Seashore, 1967) with minimal concern for process, outcomes, or a diversity of views. A more recent alternative has been to consider the views of an organization's multiple constituencies, or key stakeholders (Zammuto, 1984). This approach has been a considerable advance in thinking about effectiveness, although actual measurement has still proved troublesome, owing primarily to problems in determining how to evaluate the needs of key constituent groups and how they should be weighted, relative to those of other groups. Despite problems, it is this approach that we have adopted here. In community mental health networks, as in any publicly funded system, there are many constituencies whose views must be considered in policy decisions. These include clients, families, service professionals, state-level policymakers, funders, agency staff and administrators, and taxpayers, just to name the most prominent ones. For this research, we felt it was critical to tie effectiveness measures to enhanced client well-being. This outcome reflects what is most certainly a top priority of these constituent groups. While it is reasonable to argue that these groups will have additional goals that are perhaps more self-serving (e.g., maximizing career opportunities, enhanced influence, increased resources), and each may have different beliefs as to the means of treatment and service, disagreement over the fundamental goal of improved client outcomes is unlikely. In addition, and of substantial importance for this study, we had to choose a measure of effectiveness that reflected the outcomes of network-level activities and structures. To assess client outcomes, we selected a series of items from several standard assessment instruments, most notably, the Colorado Client Assessment Record and the New York Functioning Scale, and modified them to fit our study. Pretesting of methods and measures was done as part of an earlier project (Bootzin et al., 1989). Items focused on quality of life, client satisfaction, psychopathology, and level of functioning. Consistent with a multiple-constituency approach, we collected outcome data from three different groups: the clients themselves, their families (if available), and the clients' case managers or therapists. We believed these three groups would have the most complete understanding of client outcomes, although each might have a somewhat different perspective (Bootzin and Sadish, 1986). Hence, we used a multimeasure, multiperspective approach. Once data collection was complete, we pooled results from all subjects from all four sites and did factor analyses to test for the existence of similarities in perspectives both within and across the three groups. Initial analyses showed that each group maintained its own perspective, having only modest overlap with the other two groups. When each group was analyzed separately, four distinct perspectives ultimately emerged. The clients themselves assessed their outcomes in two different ways: One factor comprised items related to their quality of life, including various economic, social, and daily living services, while a second factor was made up of items related to their psychopathology and physiological status. These factors, labeled QofL/satisfaction and psychiatric/medical status, respectively, explained 48 percent of the variance in all the client-reported items. In contrast to client self-assessment, only one factor emerged for the family perspective (42 percent of variance) and one factor for that of the case manager (53 percent of variance). These were labeled overall quality of life. As a final step in the analysis of outcome data, we computed factor scores for each of the four factors. We then regrouped respondents by city and computed mean factor scores for each site so that each system could be compared according to its factor score for each of the two clients, one family, and one case manager perspectives. These final factor scores, which we used as our measures of network effectiveness, are reported in Table 9.1. In general, both families and the clients themselves considered the network to be most effective in meeting client needs in Providence and least effective in Tucson and Akron. Scores for Albuquerque were more equivocal but generally fell in the mid-range. In contrast, case managers/therapists felt best about client outcomes in Albuquerque and Tucson and worst in Akron, with Providence being slightly negative. As shown in Table 9.1, most of these differences were statistically significant. It is important to keep in mind, however, that these scores only reflect relative differences in effectiveness across the sites, not absolute evaluations of client well-being. Outcome scores were statistically unrelated to differences in client age, gender, race, and ethnicity. Demographic data for clients across all four sites were as follows: mean age = 42.3 years (range = 19 to 78); gender = 50.9 percent female; race = 64.9 percent white, 18.5 percent Afro-American, 14 percent Hispanic, and 2.6 percent other. We originally hoped to develop a single measure of network effectiveness, combining the perspectives of clients, families, and case managers, but the multiple-constituency approach resulted in multiple views of effectiveness. Nonetheless, the pattern of factor scores across sites made it possible to group findings so as to make comparisons and draw conclusions. The patterns of results for the two client factors and the single family factor were sufficiently similar to warrant combining them into a single dimension of network effectiveness. We averaged the three factor scores for clients and families for each site to enable us to place that site on a continuum of relative effectiveness. This method showed Tucson to be the least effective system (mean = −.19), followed by Akron (−.12), then Albuquerque (−.02), and then Providence (+.42), which was rated as significantly more effective than the other sites, especially Tucson and Akron. While all predictors of network effectiveness are presented and discussed below in terms of their relationship to both the client/family and case manager views of effectiveness, our primary discussion and interpretation of results focuses on the combined client and family perspective. Client and family views were generally consistent across sites, offering convergent validity for the measures used. In addition, it is the experiences of families and, particularly, the clients themselves that ultimately matter the most, since they are the ones who actually receive the system's services, are most knowledgeable about the client's overall well-being, and must live with the results. It is this combined client and family view that agency and program heads were most concerned with in the follow-up site visits we made to report our findings. Table 9.1 Network Effectiveness Comparisons: Factor Scores of Client Outcome Scales (N = 265)* *Sample size indicates the total number of clients from and about whom outcome data were collected at each site. †When a mean factor score from a particular constituent group is significantly different (p < .05) from the score from that group at another site, a letter (or letters) follows the score, indicating the significantly different comparison site (i.e., P = Providence). A letter in parentheses indicates p < .10. Network integration. As discussed earlier, the basic question underlying our research is whether network structure, particularly integration and related issues of coordination, is related to network effectiveness. Unfortunately, the concept of integration is ill-defined, making operationalization difficult and interpretation of outcomes confusing (Aldrich, 1978; Bolland and Wilson, 1994). Because the research was exploratory, we chose to consider integration rather broadly, in an attempt to determine which aspects, if any, were related to network effectiveness. Integration was generally considered to focus on issues of both interconnectedness among provider agencies and the extent to which provider agencies are integrated and coordinated through a central authority (Morrissey et al., 1994). Our operationalization is consistent with the general network structure concepts of density and overall centralization. According to Scott (1991: 92), these concepts “refer to differing aspects of the overall ‘compactness’ of a graph (i.e., a network). Density describes the general level of cohesion in a graph; centralization describes the extent to which this cohesion is organized around particular focal points. Centralization and density, therefore, are important complementary measures.” In more general terms, density is simply a measure of the extent to which all network organizations are interconnected, or linked to one another, and reflects network cohesiveness. Centralization refers to the power and control structure of the network, or whether network links and activities are organized around any particular or a small group of organizations. Overall centralization of the network is an extension of Freeman's (1979) concept of point centrality, which refers to the centrality of individual network members. Our density and centralization measures of network integration and coordination are reported below, as are our findings on the relationship between these measures and network effectiveness. The basic building block of any network study is the linkages among the organizations that make up the network. To develop measures of agency integration through both density and centralization, every agency in each system was first listed alphabetically on a questionnaire insert that was unique to that system. Key informants from each agency surveyed were asked to indicate whether their agency was involved over the previous year with every other agency listed, through each of five different types of service links: referrals sent, referrals received, case coordination, joint programs, and service contracts. Respondents were not asked about frequency of involvement except for referrals sent and received, where they were asked to consider only referral activity occurring “with some regularity.” While other types of links, such as friendship ties or information links, were possible, our concern was solely with links specifically related to service delivery. In addition, the questionnaire and the face-to-face interviews continually emphasized reporting only those links involving SMI adults. Thus, our results exclude non-SMI-service linkages among agencies that may provide SMI services. For instance, a hospital and a community center may refer clients to one another and provide important services to the mentally ill, but their referral activity might not involve the mentally ill clients they serve. The distinction we make between construction of a general organizational linkage network and a more focused service implementation network represents what we believe to be an important advance in network methodology, which in the past has often focused on organizational links without fully considering whether or not the type of linkage measured is relevant for the particular question being researched (see also Hjern and Porter, 1981). We validated the reported linkages in two ways. First, whenever possible, at least two agency respondents were asked to attend the data collection interview. Discrepancies in opinions about the link were discussed and resolved, usually in favor of the service professional at the meeting rather than the head administrator, who might have only limited knowledge concerning operating-level links such as referrals. Second, after data were collected, we confirmed each link (Marsden, 1990). Thus, for instance, a joint program for SMI adults listed by Agency A with Agency B would only be counted as a viable link if Agency B also indicated that the link existed. Results of Network Analysis Network Density The overall density of each network was computed in two ways. First, we computed a mean agency service link score by adding confirmed scores for each of the five types of linkages for all agencies in each system and dividing by the total number of agencies in that system. While this service link density score reflects the depth, strength, or durability of network integration (what network analysts refer to as multiplicity), moderately high scores may be achieved by relatively few agencies integrating all or most of their SMI services. Thus, we computed a second score, organizational link density, that indicates the mean number of agency links of any of the five types that occurred in each system. This second score is a measure of organizational integration through SMI service links rather than a measure of the strength of this integration across the network. Dividing the total number of linkages in each network by the number of agencies in that network, the maximum service link score for any network is 5(n − 1), while the maximum organizational link score is n − 1. Service and organizational link density scores for the entire network at each site and their relationship to each of the two perspectives for network effectiveness are reported in Table 9.2. Table 9.2 Relationships Between Network Effectiveness and Services Integration Measured by Overall Network Density* *Numbers in parentheses after each city's name refer to the client-outcome-based effectiveness score for that site and for the particular perspective listed. The mean density scores for each network were first statistically compared with the scores at each of the other sites, using t-tests. Despite apparent differences between Providence and the other three sites, the only significant finding was that organizational link scores for Albuquerque were marginally higher than for Providence (p < .10 for a two-tailed test). Most revealing, however, was that there was little relationship between density-based integration and network effectiveness, regardless of the measures of density or effectiveness used. Network density scores for Tucson, Albuquerque, and Akron were all extremely close, despite differences in effectiveness at each site. The one strong relationship found was that the most effective network based on client and family views, Providence, was also by far the least integrated system, a finding that is completely contrary to the prevailing assumptions and beliefs of mental health professionals about the value of integrating services. While these findings were unexpected in view of prevailing assumptions about integration, we resisted attempts to draw conclusions without first considering findings from our centrality-based measures of integration described below. Network Centralization We next shifted our attention to issues of integration through centralized control of the network. A real problem with integration through decentralized linkages among providers is that system coordination can be exceptionally complex, unless the network is quite small (Provan, 1983). Because there are limits to how many links an organization can handle effectively, the number of service linkages per organization in moderate to large networks may well be no greater than the number per organization in small networks. Large networks with many decentralized links will have difficulty organizing and effectively operating as a complete system, depriving clients of the potential benefits of integration across the full range of agencies and services offered in a large community. As an alternative, centralized control of integration within larger networks allows the coordination of integration across many agencies as well as closer monitoring of services. This is the logic behind the formation of a federation structure (Provan, 1983; D'Aunno and Zuckerman, 1987). With this in mind, we examined two aspects of overall network centralization: core agency centrality and concentration of influence. Core agency centrality. As the primary organizer and integrator of services in most communities, a core mental health agency, often a community mental health center, is typically at the center of all activities relating to services for individuals with severe mental illness. When the core agency is central in the flow of services, the network can operate more efficiently, as the other providers need not devote much time and effort to the task of coordinating the services they provide with the many other agencies in the network. Whether such a centralized system of integration is particularly effective, however, and whether highly centralized networks are more effective than relatively decentralized ones is not known. To measure core agency centrality, we computed service and organizational link density scores for only the core agency (see Table 9.3). Network density scores, excluding the core agency, were also recomputed for all agencies in each system. These calculations revealed some important differences across sites, as shown in Table 9.3. Statistical comparisons of means using two-tailed t-tests revealed significant differences between the core agency centralization scores (via service links) for Tucson and Providence (p < .001), Akron (p < .001), and Albuquerque (p < .05) and between Albuquerque and both Providence (p < .05) and Akron (p < .10). In contrast to the findings reported in Table 9.2, when the core agency was excluded from this analysis, we also found significant differences in network density when we compared across sites. Significant differences were found for service link density scores between Providence and Tucson (p < .10), Albuquerque (p < .10), and Akron (p < .10) and for organizational link density scores between Providence and Tucson (p < .10), Albuquerque (p < .05), and Akron (p < .05). Table 9.3 Comparisons of the Structure of Integration: Network Centralization Through the Core Agency and Core-Excluded Density Scores* *Sample size indicates the number of agencies from which data were collected at each site. Figures in parentheses indicate links of each type expressed as a percentage of the maximum possible number of links of that type within each system. Most important, the pattern of scores across sites for network centralization through the core agency were substantially different than density scores. Providence had the highest core agency linkage scores (97 percent of all agencies in the system were linked to the core through an average of 2.41 types of service links) but the lowest level of linkages, or density, among all agencies other than the core, indicating a highly centralized control system. Akron's core agency was also highly central in its network, though somewhat less so than Providence (94 percent of all agencies were linked to the core through an average of 2.31 links). Akron's network density scores were far closer to those of Tucson and Albuquerque than to Providence, however, reflecting a system integrated in two ways; Albuquerque's core agency centrality score was lower than the two eastern sites (88 percent of agencies linked to the core through an average of only 1.82 links), while density-based integration among the other agencies was quite high. Finally, Tucson exhibited characteristics of a highly decentralized but integrated system, with the lowest core agency linkage scores of any site (only 65 percent of all agencies linked to the core through an average of 1.12 links), coupled with the highest core-excluded network density score through service links. Concentration of influence. We also assessed integration through centralization by examining the network influence structure. We speculated that integration and coordination of services across the network would be enhanced when influence over decisions related to SMI clients was concentrated in a single organization. Under this condition, whether agencies were actually linked to the highly influential organization, agency actions would likely reflect the influential organization's preferences and policies about SMI services. When agencies in a system act in ways consistent with the views and expectations of a single organization, centrally controlled, coordinated actions are attainable. Although past studies of networks have relied on linkage-based measures of centrality to assess power within the network (Cook, 1977; Cook and Yamagishi, 1992), this approach assumes that structural position is equivalent to actual influence, which may not be true. To determine influence in each system, respondents at each agency were first asked to list up to six other organizations in the system “whose needs, goals, decisions, and/or expectations… are generally taken into consideration by your agency” when major decisions are made concerning services to SMI adults. Responses from all agencies in each system were then totaled, and each agency received an influence score indicating the number of times it was mentioned by others. Scores were converted to reflect the number of mentions as a percentage of total possible mentions in each system, thereby allowing comparisons across systems. To assess concentration of influence, we examined the percentage scores of the five most influential agencies in each system. We then considered two separate aspects of influence, consistent with the approach used by Eisenhardt and Bourgeois (1988: 743) in their case study of firms in high-velocity environments. First, we determined the most influential agency in each system. As shown in Table 9.4, in Tucson, Akron, and Providence this agency received mentions by between 71 and 76 percent of respondents. In contrast, the most influential agency in Albuquerque received only 56 percent of possible mentions. Second, we examined the distribution of influence among the top players, focusing on the distance between the scores of the most influential agency and the one (ones) in second place. If the second-place agency did not have more than half the influence score of the top agency, then we considered that system to have concentrated influence, as was the case in Providence and Albuquerque. Because Albuquerque's top score was so much lower than the most influential agency at the other three sites, however, we labeled that system concentrated/weak and Providence as concentrated/strong. Akron was labeled as moderately dispersed, since its two second-place (tied) agencies received relatively high influence scores (mentioned by 54 percent of respondents). Tucson was labeled as dispersed since four other agencies in the system were two thirds as influential as the most influential player. Table 9.4 Concentration of Influence Over Mental Health Decisions (Top Five Most Influential Agencies at Each Site)* Notes: MH, mental health; HS, human service; AMI, Alliance for the Mentally Ill. *Influence scores reflect the number of times each agency was reported as being influential in mental health decisions as a percentage of the total number of agencies in that network. Combining findings on the two centrality-based measures of network integration with the effectiveness results allows us to make some conclusions that go beyond what could be said from a consideration of density-based measures only, at least for client and family assessments of effectiveness. As discussed earlier, density measures were unrelated to effectiveness, except that Providence, the least cohesive system, had the highest effectiveness score. In contrast, when focusing on integration through network centralization, a distinct pattern emerged. When influence over mental health decisions was highly concentrated in a single core agency, as in Providence, client outcomes were highest. At the other extreme, when influence was widely dispersed among a number of agencies, as in Tucson, effectiveness was lowest. Akron's influence structure, which was more dispersed than any system other than Tucson's, had outcomes that were also lower than all but Tucson's. Finally, influence in Albuquerque, while far weaker than in Providence, was more concentrated than either Tucson or Akron, and consistent with this finding, its network effectiveness score on client and family assessment of outcomes was higher than Tucson and Akron but lower than Providence. Conclusions from the linkage-based measure of centralization were consistent with this pattern, with the exception of Akron: We found the highest effectiveness in Providence, which was highly centralized through the core agency, and the lowest effectiveness in Tucson, which was decentralized, while Albuquerque, a moderately decentralized system, had a mid-range effectiveness score. Akron, the anomaly, was a centralized system with low effectiveness. Discussion Network Effectiveness: Client and Family Perspective The clear linear relationship between our influence-based measure of centralization and client and family assessments of network effectiveness, coupled with the findings for the linkage-based measure of centralization, which seemed to explain effectiveness for all systems except Akron, might lead to the conclusion that a positive tie between network integration and effectiveness is most likely when integration and coordination occur from the top down but not when agencies take it upon themselves to integrate their services. This conclusion is consistent with those of Goldman, Morrissey, and colleagues from their research on mental health systems funded by the Robert Wood Johnson Foundation (Goldman, Morrissey, and Ridgely, 1994; Morrissey et al., 1994). While such a conclusion would not be inaccurate, it is based only on partial information about the systems under study. Using our interviews and observations, we were able to supplement the quantitative network findings with qualitative insights, allowing us a deeper understanding of why some mental health networks are more effective than others. Our qualitative data from Providence were entirely consistent with the network structure results. The city's core mental health agency was easily the most powerful player in the system. It was the focus of discussion in nearly all our interviews with other agencies in the system, although many viewed its powerful role in an uncomplimentary way, variously describing the center as “bureaucratic,” “insensitive,” and “arrogant.” As described by its director, the primary mission of the core agency was to provide psychiatric and case management services to all individuals with severe mental illness. In fact, unlike any of the other systems, the core agency in Providence provided its own psychiatrists to staff the psychiatric beds of several area hospitals, thereby controlling patient treatment even during institutionalization. Through control of both case management and outpatient psychiatric services, the core agency could also control which services offered by the other community agencies the SMI clients received and when they received them. Perhaps of greatest importance, Providence's core agency had had for many years a unique and direct relationship with the state's mental health authority, in which all funding for the needs of the severely mentally ill would be paid directly to the core agency. In addition, if services were needed that the core agency did not provide, such as housing or vocational rehabilitation, the state could contract with other agencies for such services only after core agency approval. Agencies not having such approved contracts might provide certain services to the SMI, but these agencies would not receive state funding expressly for that purpose. This “monopoly model,” as we referred to the Providence system, made integration among the various providers largely unnecessary. Integration and coordination of the system was still critical for attaining effective client outcomes, since the core agency did not provide many of the services needed by clients. Where integration occurred, it was through the core agency, which controlled access to mental health funding and clients and was the obvious focal point for mental health services throughout the community. An explanation of why network centralization through the core agency had a positive impact on client effectiveness was perhaps best stated by the head of the state's department of vocational rehabilitation, who said that because “[the core agency's] clients already have their housing, medication, and most other needs met when they come to us, it is easier to effectively rehabilitate them.” Providence also had the advantage of relatively high state funding, enabling the system to provide a level of services that might not be possible in Tucson or Albuquerque. This high funding was allocated to a local system that, because of its strong central control, could function relatively efficiently with little duplication and little questioning of decisions. In addition, direct control over funding to a single core provider allowed the state to monitor closely the outcomes of the core agency, while the core could closely monitor the activities of the other agencies in the network through its role as network gatekeeper. Finally, the Providence system was highly stable. The core agency had been in place as a mental health center for over 20 years, and its director had been in place since its founding. Most of the directors of the other key agencies had known each other for years and had longstanding working relationships with the core agency. As a result, the system and its operation were well-known to agency professionals, resulting in little uncertainty for them or their clients. The situation in Tucson, the least effective system, was substantially different. Unlike the stable system in Providence, the system in Tucson was undergoing rapid and substantial change at the time we collected data. The changes were brought about in large part by what was referred to by one respondent as “a revolution from below,” in which a key system leader joined forces with the local chapter of the Alliance for the Mentally Ill (AMI), a client advocacy group formed by families of the SMI, to put pressure on the state to change Tucson into a case-managed, capitated system headed by a single agency responsible for both funding and service delivery. The revolution was only partially successful, resulting in formation of a new but scaled-back core agency, primarily responsible for case management. Although it contracted with local doctors and the psychiatric unit of a local hospital to provide psychiatric treatment for its clients, the core agency provided few substantive mental health services on its own. Thus, while it served as a key coordinator of SMI services in the community, it had no monopoly on services. In fact, even the core agency's role in Tucson was spread among several other agencies, most notably a traditional community mental health center and another multiservice mental health provider that was set up primarily to serve the city's large Hispanic population. The most powerful local agency in Tucson was a private not-for-profit agency that was the local funding entity. Unlike Providence, where funding was direct from the state to the core provider agency, all treatment dollars flowed from the state's department of health services to Tucson's funding entity, which made all local funding decisions. Instead of working closely with local providers, however, especially the new core and the two other mental health agencies, the funding entity used its control over resources to build its own power, both in Tucson and at the state level. Treatment dollars were provided, but the system was poorly funded, and little if any monitoring of services took place. One respondent described the fund allocation and monitoring system as being akin to “leaving money on a tree stump and seeing who claimed it.” That the funding entity and its president were not strongly committed to goals of client service was amply demonstrated by a scandal that erupted shortly after our data collection. It was discovered that the entity's president was diverting precious treatment dollars to a fund to purchase a new building, something that outraged the mental health community but was perfectly legal under the state's poorly specified system of contracting.1 Because of these problems, the mental health system in Tucson operated largely on the goodwill and professionalism of those who operated the key provider agencies. The director of one of these providers pointed this out to us, noting that the heads of the three main providers described above were all good friends and were willing to work together. The other main integrating force in Tucson was the local chapter of the Alliance for the Mentally Ill. Because of the interpersonal skills and visibility of its president and the presence of several powerful people on its board, AMI was able to organize local provider agencies and client families into an effective force for lobbying the state. AMI also published a regular newsletter that further helped to knit together both providers and families and to point out problems with the new system. All this decentralized integration of the network was critical to maintain minimal levels of acceptable services in a system that was underfunded and had no strong central authority to pull things together. Our observations and interviews in Tucson led us to the conclusion that decentralized integration was critical for helping a weak system limp along but was not sufficient to result in the high levels of client outcomes that were observed in Providence. Based on the quantitative findings, Albuquerque fell between the highly centralized and effective Providence system and the decentralized and ineffective Tucson system. Albuquerque's network effectiveness was particularly surprising in light of the state's weak funding of and commitment to care for the severely mentally ill. The simplistic, structural explanation for this is that effectiveness is enhanced with greater levels of network integration through centralization but not through decentralized links among providers. Since Albuquerque's centralization was between Tucson's and Providence's, its effectiveness should also be intermediate, as we found. Yet this explanation says nothing about why Albuquerque's system was structured as it was, how effectiveness might have been causally related to structure, and what other factors may have contributed to the relative effectiveness of the Albuquerque system. Unlike any of the other three systems we studied, case management, a key service-level integration mechanism, was formally provided to severely mentally ill individuals by four different agencies. Although the core agency was by far the largest of these agencies and served an estimated “90 percent of the city's adult SMI population at one time or another,” centralized integration and coordination of the system through the core agency was not really possible, since the core agency did not manage the treatment plans of many of these clients. Thus, while the other three case management agencies and many other agencies both sent and received clients to and from the core, each also developed its own subnetwork of linkage partners. The agencies in Albuquerque were dependent on and thus linked to the core for many of the services it offered, including most emergency and inpatient care, but the core agency did little to actively integrate the system. It is telling that only one agency maintained four or more of the five types of service links we measured with the core agency, as opposed to six agencies in Akron and eight in Providence (see Table 9.3). As one agency head told us, “The [core agency] isn't the glue that holds the system together. They are unresponsive to the community, aren't protective, and do only what will be funded.” System decentralization was enhanced by the state's mechanism for funding. Unlike any of the other three systems, the four main case management agencies in Albuquerque received state mental health funds from three primary sources, and any agency in the city could apply for these funds. Because funding was direct, with no local intermediary as in Tucson, the state had somewhat greater control over how resources were spent than in Tucson. Because funding was fragmented, however, with each fund source having its own priorities and funding preferences, the state was unable to centralize mental health activities and services around a single agency by concentrating resources on that agency, as in Providence. The largest share of state funding did go to the core agency, giving it resources and subsequent services that other agencies could not provide in a poorly funded system. The core agency thus became a central player in the system, even though its management did not seek out a strong integrative role. The integrative shortcomings of the core agency appeared to be made up for by other agencies in the system, several of which viewed the idea of an integrated and coordinated system of services far more seriously than the core agency. Thus, Albuquerque displayed some of the decentralized elements of Tucson while still having a single, large, multiservice core agency. Albuquerque also had a system that, despite its poor funding and integrative shortcomings, had not undergone any significant changes in many years. Clients, families, and agency professionals thus understood the system and could work around its shortcomings. A good example of this was offered by one respondent at the city's jail, who told us, Because evaluation and intake at [the core agency] can take five hours of a police officer's time on weekend nights, the officer will slap a mentally ill person with a petty offense like disorderly conduct to get them into the jail's psychiatric unit. Because it takes so much less time to “book” the person at the jail, it is often the mental health facility of first resort, rather than the last. After the person's emergency needs are taken care of, they are sent on to [the core agency] or another agency. Agency service professionals in Albuquerque knew how the system operated and were able to work with the core agency when its services were needed or around the core agency when it could be avoided. The findings in Akron are somewhat more difficult to explain. Structurally, the system was closest to Providence, having service linkages that were centralized through a core agency. In addition, both systems were in states that provided relatively munificent resource environments. Yet in terms of client and family assessments of effectiveness, the system was closer to Tucson. Only concentration of influence, which was moderately dispersed in Akron, seems to fit the linear relationship between aspects of network structure and effectiveness found with the other three systems. Using our qualitative data, however, we can offer several explanations, primarily centered around the issue of stability, to help make sense of the discrepancy in findings for Akron. Like Tucson, the community mental health system in Akron was new. Passage of Ohio's Mental Health Act of 1988 significantly altered system funding and structures in Ohio's cities. In Akron, three mental health centers had served the community in each of three separate catchment areas, each competing for clients and state funding. After passage of the Mental Health Act, an Alcohol, Drug Abuse, and Mental Health Board was established as the local contracting, monitoring, and integrative entity for all state- and local-government-funded services for mental health, as well as for drug and alcohol abuse. Thus, like Tucson, but unlike the other two systems, fiscal control by the state was indirect, in this case, through a public rather than a private not-for-profit entity. The ADM board, as it was known, hired as its president an outsider activist who was a self-described “tinkerer.” Within a year, he stopped all government funding to two of the three mental health centers, effectively shutting them down, while revising the role of the third center to focus on clients with acute, rather than severe, mental illness and to perform intake evaluations for all clients. He then created and funded a new core agency as the sole provider of and case manager for those with severe mental illness. While the new core agency was very much at the center of service delivery to the SMI population in Akron, it was apparent that effective integration and coordination of services was still very much evolving and that there was considerable confusion about the new system. As one agency director told us, “The major changes in the mental health system in the past few years means we don't really have a handle on how the system works.” Another said, “The system here is confusing to understand, and I don't see how people with mental illness can comprehend it.” One problem was that the system for handling emergency services was not only confusing, involving different agencies at different times of the day, but was creating animosity among the agencies involved, particularly between the acute-care agency and its supporters and the new core agency. Another problem was that there was still considerable underlying resentment stemming from the recent system restructuring. Not only was there a change in the organizations that provided key services and funding, creating uncertainty and loss of resources for some agencies, but there was also a shift in treatment philosophy. This shift was a conscious effort by the ADM president first to emphasize the community-based care and treatment of all but the most severely mentally ill, as opposed to emphasizing acute-care patients, and, second, to place less emphasis on a traditional psychiatric/medical orientation and more emphasis on a psychosocial, community support philosophy exemplified by the case management approach of the core agency. Agencies not supporting the new approach still established service links to the core agency, not just because of its services but also because the core (through the ADM board) formally controlled and managed the treatment plans of and funding for all officially designated SMI clients. Thus, while the system in Akron was structurally nearly as centralized through the core agency as Providence, uncertainty about the system was still high, and many agencies were not yet committed to a system dominated by the new core agency. This interview-based conclusion is consistent with the quantitative finding that while Akron was centralized through the core agency, it al