Short-Term Versus Long-Term Coalitions in the Policy Process: Automotive Pollution Control,...

Introduction

Political scientists have long been interested in the formation and the maintenance of political coalitions-here defined as a group of actors coordinating their behavior to some extent in order to achieve a common, or complementary, political objective (Hula, 1999). In public policy, these coalitions-often consisting of diverse sets of actors, including legislators, agency officials, interest group leaders, and researchers-are critical because of their importance in passing and implementing major policy initiatives (Kingdon, 1984; Sabatier & Jenkins-Smith, 1993).

Research on coalitions in the policy process has produced varying results regarding their stability over time and resilience in the face of potentially disruptive events. On one hand, there is evidence of relatively long-term coalitions in health care (Marmor, 1970), energy policy (Jenkins-Smith, St. Clair, & Woods, 1991; Jenkins-Smith & St. Clair, 1993; Wildavsky & Tenenbaum, 1981), banking (Worsham, 1997), civil rights (Hula, 1999), and environmental policy (Sabatier & Brasher, 1993). Equally numerous, however, are portraits of relatively fluid coalitions in farm policy (Browne, 1988), health care (Heinz, Laumann, Nelson, & Salisbury, 1993), energy policy (Heinz et al., 1993), air pollution control (Ackerman & Hassler, 1981), and regulation in general (Ripley & Franklin, 1979).

There are all sorts of theoretical reasons to expect variation in coalition stability across policy sector: the number of actors involved, the density of network ties, and so on (Hinckley, 1982; Hula, 1999). In addition, however, there are at least two potential methodological reasons for the mixed evidence. First, most scholars have not been very careful to distinguish situations in which (a) fundamental beliefs/interests are at stake (e.g., the proper distribution of authority between government and markets) versus those involving (b) more secondary and instrumental beliefs (e.g., the proper level of a payment program).1 Second, very few of these studies use systematic methods to measure coalition behavior and stability over time. The systematic work has been primarily cross-sectional in scope (e.g., Heinz et al., 1993), and most of the longitudinal work has utilized primarily qualitative methods of data acquisition and analysis that are subject to quite varying interpretations (e.g., Ackerman & Hassler, 1981; Ripley & Franklin 1979; Wildavsky & Tenenbaum, 1981; Marmor, 1970).

This article seeks to address both points in the analysis of coalition stability over time. We apply a conceptual framework-the Advocacy Coalition Framework (Sabatier & Jenkins-Smith, 1993, 1999)-that distinguishes fundamental from secondary beliefs/interests to examine coalition stability and resilience in the face of potentially disruptive events. This guides our empirical focus to the following questions:

1. Are coalitions in domestic policymaking relatively short-term and fleeting in duration-for example, lasting only long enough to pass a piece of legislationor do they tend to endure for periods of a decade or more?

2. Is agreement within a coalition based on relatively narrow policy preferences or on broad beliefs concerning, for example, the proper role of government or the relative priority of different values within the policy subsystem?

We then employ systematic methods of data acquisition and analysis to evaluate coalition groupings in automotive pollution control policy. Similar to JenkinsSmith et al. (1991), our principal data source is the content of organizations' testimonies at congressional hearings. We evaluate change in coalition membership over time, by examining how organizations alter their publicly expressed policy beliefs relative to others. Assessments are made both in response to two sequences of major events-the 1973-74 Oil Embargo and the combined impact of the 1979 Iranian Revolution and 1980 election outcomes-and in general over a 26-year period.

The next two sections briefly discuss alternative perspectives of coalition formation and maintenance that support either short- or long-term views of political coalitions. We do so first in general and then with particular attention to the Advocacy Coalition Framework (ACF). The bulk of the article applies two hypotheses drawn from the ACF to the analysis of testimony before the U.S. Congress on automotive pollution control between 1963 and 1989. The concluding section summarizes our results and examines their generalizability to the study of coalitions in the policy process.

I. Competing Perspectives of Coalition Formation and Maintenance

A. Short-Term Coalitions

Perhaps the best known theory of political coalitions is that of Riker (1962,1980). Riker's theory starts from the premise that actors choose coalition size to maximize the benefit of a single victory to each coalition member. "Minimum winning coalitions" achieve this objective by splitting the spoils of political success (e.g., ministerial portfolios) among the smallest number of coalition members. The goal is to maximize net benefits at the margin, that is, in the short-term. Seeking to maximize average benefits over the long-term would be irrational, particularly if the future is quite uncertain.

This reasoning can be easily extended to interest group coalitions by noting that additional coalition members increase the degree of compromise necessary to form the coalition while reducing the extent that individual members can claim credit for resulting political victories The logic for "minimum winning coalitions" is particularly attractive in decentralized political systems-such as the United States-in which issues rise and fall on various government agendas with some frequency. When new issues tip the balance of power in a coalition's favor, incentives exist for peripheral members to be excluded. When the issue balance turns against the coalition, new members must be found to increase the coalition's resources. These pressures to realign coalitions are particularly strong when a major exogenous shock to the policy subsystem-such as the election of a new president or a major economic perturbation-substantially alters the political resources of existing coalitions.

An alternative rationale for short-term, fluid coalitions comes from Heclo's (1978) concept of "issue networks." If most actors in an issue domain are either small organizations with uncertain staying power or individuals who flit from organization to organization, the instability of members will make coalition formation difficult and short-term (Hula, 1999).

B. Long-Term Coalitions-Reciprocity and Values

In contrast to the incentives for a short-term coalitions, there are at least two reasons to expect that coalitions will be relatively long-term enterprises.2

The first mechanism promoting long-term coalition stability is derived from arguments incorporating multiple time periods into the rational actor formulation of the coalition problem. The logic of these multi-period models depends on a specific type of contingent contract termed reciprocity (Axelrod, 1970; Putnum, 1993). It states: "If I resist the temptation to defect today, I expect you to resist the temptation next year. In the long run, we will both benefit if we can count on each other to help preserve, and build upon, past victories." The reasoning of these models is particularly relevant to situations in which the nature of the task requires long-term cooperation. These would include logrolling on the allocation of public works projects (Ferejohn, 1974), the implementation of major legislation (Mazmanian & Sabatier, 1989), and the maintenance of institutions to provide common property resources (Ostrom, 1990). Actors pursuing relatively similar-or, at least, compatible-policy objectives in these situations should perceive that their long-term average benefits requires maintaining fairly stable coalitions.

A second factor promoting stable coalition composition is the existence of fixed basic values. One is more likely to choose coalition partners who espouse ideologies relatively similar to one's own because of the increased probability of interacting, developing trust, and finding common ground with those individuals. Several scholars suggest that fundamental ideological preferences resulting from political socialization (e.g., basic value priorities, whose welfare counts, the proper role of the state) are the critical values that shape long-term coalitions (Axelrod, 1970; deSwaan, 1973; Browne & Dreijamis, 1982; Franklin & Mackie, 1984; Merson, 1996). Organizational missions may also play important roles in promoting long-term stability when studying coalitions that include administrative agencies and interest groups because these missions often imply relatively fixed basic values along with policy strategies for achieving those values (see Browne, 1988).

II. The Advocacy Coalition Framework

The Advocacy Coalition Framework (ACF) looks at the role of coalitions in policy change within policy subsystems over periods of a decade or more (Sabatier & Jenkins-Smith, 1993, 1999). It contains two arguments relevant to this article.

First, the ACF starts with the assumption that policy elites have well-integrated policy belief systems that link fundamental substantive and distributional values, perceptions of the severity and causes of policy problems, and perceptions of the proper approaches to be used in addressing those problems. These beliefs result from the individual's socialization, education, and organizational and institutional experiences. They are reinforced-and therefore remain relatively stable-because of the tendency of all actors to screen out information that is dissonant with their preexisting beliefs (Lord, Ross, & Lepper, 1979). Thus people in different coalitions will interpret the same piece of evidence quite differently, leading to suspicion regarding the motives underlying the "perverse" interpretation of evidence by opponents. Hostility is further exacerbated by the tendency of actors to remember losses/defeats more than gains/victories (Quattrone & Tversky, 1988). The end result is that members of the opposing coalition are perceived to be both more suspicious and more powerful than they actually are-the "devil shift" (Sabatier, Hunter, & McLaughlin, 1987). This increases the costs of short-term defections to the opposing coalition and enhances the benefits of long-term solidarity to one's own coalition. Thus, when dealing with coalitions of interest groups, agencies, legislators, and researchers over periods of a decade or more (Sabatier, 1998, Table 2), the ACF predicts:

Hypothesis 1 : On major controversies within a policy subsystem, the lineup of allies and opponents will be rather stable over periods of a decade or so.3

From this perspective, short-term, shifting coalitions are much less likely than Riker perceives-at least in high-conflict situations.

Second, to manage the potential complexity of elite belief systems, the ACF distinguishes between three levels of beliefs that differ in their topical scope (Sabatier & Jenkins-Smith, 1999). Deep core beliefs are the general normative and ontological beliefs that span numerous policy subsystems (e.g., the traditional Left/Right cleavage). Policy core beliefs consist of important normative and perceptual beliefs spanning an entire policy subsystem (e.g., identification of which stakeholders' welfare is of primary importance, orientation on basic value priorities within the subsystem). Secondary aspects are the instrumental decisions and information searches necessary to implement policy core beliefs in a portion of the entire subsystem (e.g., perceived seriousness of a problem in a specific locale, evaluation of a specific program).

Any model that assumes instrumental rationality implies that individual actors hold their ultimate goals to be more important than their contingent means for achieving those goals. Following this reasoning, the ACF predicts that actors make concessions on secondary aspects (instrumental beliefs) before altering their policy core values (ultimate ends). These concessions cause secondary aspects to be less stable over time, thus:

Hypothesis 2: The policy core beliefs that individuals and organizations espouse will be more stable over time than secondary aspects.4

Nevertheless, the presumption that elite belief systems are well-integrated implies that considerable stability over time should exist for both types of beliefs.

Because of the emphasis on stable coalitions with stable policy core beliefs, the ACF predicts that major policy change within a subsystem (defined as change in the policy core elements of government programs) requires an "exogenous perturbation" to significantly alter the resources or beliefs of coalitions within the subsystem. These perturbations can take the form of (1) changes in socioeconomic conditions (e.g., serious inflation or recession), (2) changes in public opinion, (3) a change in the system-wide governing coalition, or (4) outputs from other subsystems. The opportunity provided by this perturbation must, however, be skillfully exploited by the minority coalition if it is to gain power, since the dominant coalition will almost certainly resort to a variety of delaying strategies in an effort to "ride out" the shock (Sabatier & Jenkins-Smith, 1993, Chap. 10; see also Baumgartner & Jones, 1993; True, Jones, & Baumgartner, 1999).

III. A Brief History of Automobile Pollution Control in the United States

We use automotive pollution control policy as a focus for our study of stability in policymaking coalitions. Whereas regulation of air pollution in the United States includes both mobile sources (e.g., cars, trucks) and stationary sources (e.g., factories, refineries), development of public policy governing these two areas has involved two separate (although somewhat related) policy subsystems. Traditionally, mobile and stationary sources involve separate titles of federal and California law and are usually administered by separate sections of federal, state, and local pollution control agencies. Further, the principal target groups in mobile source control are only minor actors in a stationary source pollution control. Thus, the automotive air pollution control subsystem provides a cogent boundary for the study of subsystem dynamics (Sabatier, 1998, pp. 111-112).

We divide our analysis into three time periods prior to the 1990 Amendments to the Clean Air Act. The first period describes the formation and solidification of automotive air pollution control as a policy subsystem during the 1960s, whereas the later two separate the effects of two sets of exogenous events that influenced the direction of subsystem policy, one in 1973-74, the other in 1979-80.

A. Formative years (1955-1970)

In the 1950s and early 1960s, two processes gradually altered people's perception of the air pollution problem (Davies & Davies, 1975; Krier & Ursin, 1977; Sabatier, 1975). First, evidence began to accumulate that air pollution was unhealthy, rather than simply being a minor irritant or an indicator of prosperity. second, work in California indicated a new pollutant, photochemical smog/oxidant/ozone, for which the automobile was the principal source.

Because photochemical smog was the principal pollution problem in California and was perceived to be more serious there than elsewhere, California took the lead on the regulation of automobile pollution (Doyle, 2000; Krier & Ursin, 1977). The California Motor Vehicle Pollution Control Board was created in 1960, and it soon promulgated regulations requiring that all new vehicles sold in the state after 1963 be equipped with a crankcase blowby device and that 1966 vehicles have exhaust emission controls (reducing hydrocarbon emissions about 50% from that in uncontrolled cars). In 1965 the Federal Government passed the Motor Vehicle Pollution Control Act giving the Secretary of Health Education and Welfare (HEW) the authority to issue emission regulations, and he quickly mandated that the California regulations be applied nationwide by the 1968 model year. The Clean Air Act of 1970 passed in the wake of dramatic increase in public concern with pollution and the resulting competition of major elected officials to capitalize on the divergence between public sentiment and public policy (Ingram, 1978; Jones, 1975). In early 1970, President Nixon proposed legislation to reduce automotive tailpipe emissions 90% by 1980. Senator Edmund Muskie, the unquestioned congressional expert on pollution control, took the Nixon initiatives as a challenge to his "turf"particularly since he was widely expected to challenge Nixon in the 1972 presidential election. Over the summer of 1970, his Senate Subcommittee on Pollution substantially strengthened the controls on both stationary and mobile sourcesincluding shortening the deadline for meeting the auto emission standards from 1980 to 1975 (1976 for nitrogen oxides)-and his bill quickly passed the Congress and was signed into law.5

B. Implementation Delays of the 1970s

The 1970 Amendments represented an exercise in "technology forcing": the automobile companies would have to invent new technologies to achieve the mandated 90% tailpipe emission reductions. After a few years, it was clear that some sort of catalytic converter was likely to be the answer, but the long lead times between design and production in the automobile industry meant that the 1975/76 deadlines were infeasible. Thus the 1970s were marked by a repeated dance in which the automobile companies would ask for an extension on the ultimate emissions standards, together with the imposition of weaker interim emissions standards, while environmental groups and their allies sought to maintain the original deadlines and increase sanctions for missing the deadlines.

The conflict was exacerbated by the 1973-74 Arab oil embargo. The embargo put a strain on the U.S. economy in general, but also had a specific and focused impact on the U.S. auto industry. The dramatic increase in oil price hurt the sales of "gas guzzlers" produced by the American auto companies and favored more energy-efficient imports. The down-turn in the domestic auto industry strengthened the industry's antistandards position (Mazmanian & Sabatier, 1989, Chap. 4), especially considering that most tailpipe controls threatened to increase the price of automobiles and further reduce fuel economy. In the wake of the embargo, the auto industry was able to delay implementation of the ultimate tailpipe standards in 1973, 1974, and 1976.

Two relatively new issues gained prominence in the late 1970s, as it became apparent that the emission reductions achieved by the auto industry would be insufficient to reach air quality goals. The first dealt with the topic of "in-use emissions." Beginning in the mid-1970s, testing of vehicle emissions by the Environmental Protection Agency (EPA) revealed that pollution control systems deteriorated rather rapidly after only a few years. In many cases, this was attributed to either poor vehicle maintenance or to deliberate dismantling of pollution control systems by owners (White, 1980). EPA proposed that vehicles should be subjected to inspection and maintenance (I/M) testing programs, and this requirement was included in the 1977 Clean Air Amendments. The second issue surrounded the importance of transportation control plans. It had long been recognized that reducing vehicle emissions would require both (a) reductions in emissions per vehicle mile (via emission standards) and (b) reductions in vehicle miles traveled (VMT). The latter would involve fundamental changes in American lifestyle, particularly reducing commuting trips in single-occupant automobiles. EPA tried repeatedly throughout the 1970s to encourage state and local governments to impose "transportation control plans" to reduce VMT, but these received only a lukewarm reception in Congress (Mazmanian & Sabatier, 1989, Chap. 4).

C. Reagan-Bush Years

In 1979-80, the automotive pollution control subsystem was subjected to two related events: (a) a dramatic rise in oil prices because of the Iranian Revolution and (b) the election of Ronald Reagan as president. As a classic example of an exogenous shock, the 1979 Iranian Revolution had a two-fold effect on the automotive pollution control system in the United States. First, it produced another oil shortage, with the consequent rise in gasoline prices, recession in the automobile industry, and pressure to relax automotive pollution controls. Second, President Carter's response to the Iranian development was perceived unfavorably by the public, contributing to the election of Ronald Reagan as president and a Republican majority in the Senate in November 1980.

In 1981-82, Reagan sought to relax the automobile emissions standards and other provisions in the Clean Air Act but was rebuffed by moderate Republicans in the Senate (particularly John Chafee of the Environment Subcommittee) and Henry Waxman in the House (Tobin, 1984). The latter 1980s were marked by repeated EPA efforts to improve I/M programs, to reinvigorate transportation control measures, and to strengthen evaporative emissions controls, all of which were substantially achieved in the 1990 Clean Air Amendments (Bryner, 1993; Grant, 1995).6

In sum, the automotive pollution control subsystem provides a good case to study coalition stability over time. First, its relatively long history with frequent congressional hearings provides a data source to evaluate actors' espoused policy positions over time. Second, its exposure to at least two major exogenous shocksthe 1973-74 Oil Embargo and the 1979 Iranian Revolution-provide challenges for coalition maintenance.

IV. Database and Methods

We analyze coalition stability in automotive pollution control policy based on the stated positions of organizational representatives testifying before Congress. Assessment of coalition membership is based on congruence in these expressed believes. We label the resulting coalitions "belief coalitions" because we lack a systematic measure of whether organizations espousing similar beliefs also coordinated their behavior. In the concluding section, we shall present some qualitative evidence of the degree of coordination among coalition partners. In this section, we describe our methods for selecting hearings and testimonies to code, evaluating the content of these testimonies, constructing measures of relevant beliefs, and finally, analyzing the similarity in organizations' scores on belief measures over time.

A. Selection of Hearings and Organizations

The database used in this article consists of a total of 273 testimonies by 179 different people presented at 20 sets of congressional hearings covering the mobile source aspects of air pollution control between 1963 and 1989. These hearings represent a subset of almost 60 sets of congressional hearings that occurred over the period from 1954 through 1989 dealing with automotive air pollution control and the Clean Air Act and its amendments.

We selected hearings in eight periods when either major amendments to the Clean Act were considered or major implementation reviews of mobile source provisions were conducted. These periods span the following years: 1963-1965, 1969-1970, 1973, 1975, 1977, 1981-1982, 1987, and 1989. We selected hearings within these years using two main criteria: (1) existence of corresponding House and Senate hearings (as the two chambers have frequently differed on mobile source controls) and (2) presence of representatives from major auto companies (which proved to be unimportant, since they were at virtually all 60 hearings).

Our objective is to examine change in organizations' publicly expressed policy positions over time. Thus, we focus on organizations that appeared at least three times within the twenty hearings selected. These include: U.S. Environmental Protection Agency (US EPA), U.S. Department of Health, Education, and Welfare (US HEW), State and Territorial Association of Pollution Program Administrators (STAPPA), California Air Resources Board, League of Cities, the big three U.S. automobile companies (Ford, General Motors, and Chrysler), American Petroleum Institute, National Clean Air Coalition, Sierra Club, and Natural Resources Defense Council (NRDC). The unit of analysis is the organization's testimony at a hearing (or set of hearings) held by a congressional committee during the period of time defined by the individual Government Document Serial Number. Individuals not affiliated with major organizations or agencies were not coded.

Table 1 lists the eight periods used in our analysis, as well as the frequency of testimonies from representatives of important organizations. Although representatives of the auto companies were the most frequent testifiers (54 appearances), other categories-including US EPA/HEW, environmental and health groups, California air pollution agencies, oil company representatives, county and city organizations, and catalyst manufacturers-also testified at least 12 times each. Representatives of the oil companies (usually American Petroleum Institute), automobile companies, and EPA/HEW testified during all seven periods. In addition, note that, although regulation of automotive emissions was an exclusively federal matter during the 1970s and 1980s (except for California), there were 71 testimonies from officials of subnational governments. These data certainly demonstrate that the automotive pollution control subsystem is intergovernmental in scope.

B. Coding Frame

The coding frame used in this project is very similar to those used in studies of land use and water quality in the Tahoe Basin (Sabatier & Brasher, 1993) and of Outer Continental Shelf Leasing Policy (Jenkins-Smith & St. Clair, 1991 JenkinsSmith et al., 1993). It consists of 139 variables with responses on five-point scales that attempt to encompass the range of possible topics raised over the 26-year period of our analysis. Items not addressed in a particular hearing testimony were coded as "missing/not mentioned."

The coding frame was divided into two parts. The first section contained 17 policy core items relating to fundamental normative orientations and perceptions regarding the entire breadth/scope of automotive air pollution control policy. The items dealt with the proper scope and level for governmental responsibility, seriousness of the air pollution problem and the threat that air pollution control presented to the economy, perceptions about causes of the air pollution problem, and the importance of different values for public health, economic welfare, and environmental quality. The second section included 122 items dealing with the secondary aspects of elite belief systems pertaining to only a subset of automotive pollution control topics. These included: specific air quality and emission standards (and deadlines), the costs and benefits of automotive emission controls, inspection and maintenance programs, the need and effectiveness of transportation control measures (TCMs), potential economic incentives, and an evaluation of various actors.7

Given the length and complexity of the coding frame, inter-coder reliability tests between the three coders were conducted at regular intervals. Inter-coder reliability was measured as the percentage agreement both on decisions to code an item as mentioned (versus missing) and on the specific numeric code given. On the decision when to code or not to code, all three coders agreed over 93% of the time. On the total set of items coded affirmatively by at least two coders, the coders were within one coded value over 94% of the time and were identical over 80% of the time. These results provide reasonable confidence that the coding results are intersubjectively reliable.

C. Belief Measures

Many of the issues discussed over the 26-year period were either fleeting or only of interest to a small number of organizations.8 We focus our analysis on issues that were mentioned consistently over time and beliefs that received significant attention in at least one of the sets of hearings.

We created additive belief scales to analyze related items that received considerable attention across all time periods (i.e., mentioned in more than 20% of the 273 testimonies). This permits us to provide a scale score for a testimony even if the person testifying mentions only one component item in the scale. To reduce the impact of any missing items on the scale values, we normalized all potential scale items (zero mean and unit variance).9

We generate two secondary aspect scales that address support or opposition to tailpipe standards. We included items in these scales based upon their conceptual consistency and high inter-item correlations (r ranging from 0.59 to 0.91 when component items mentioned in the same testimony). We label the scales "Absolute Tailpipe Standards" (Cronbach's a = 0.88) and "Interim Tailpipe Standards" (Cronbach's a = 0.97). High values on both scales identify support for stricter tailpipe standards.

We generate two policy core scales that address separate dimensions of support/opposition to automotive air pollution control in general. Because the policy core items in our coding frame were not as tightly related conceptually as the tailpipe standard items, we used a principal components factor analysis to help identify the structure underlying the six most commonly mentioned policy core items. This identified two underlying factors that separately explained 52% and 22% of the variance. The loadings on the first factor suggested creation of a four-item additive "Government Control of Automotive Air Pollution" scale (Cronbach's a = 0.87). High values on this scale denote general support for government action to reduce pollution emitted form the automobile. The loadings on the second factor identified a two item additive scale for "Federal Government Regulation" (r = 0.46 between the two items). High values on this scale denote support for federal government regulation as the specific means for dealing with problems associated with automotive air pollution.

Because the policy debate surrounding secondary aspects on automotive policy control was broader than simply tailpipe emissions standards, we also included analysis of four specific issues that were either mentioned consistently over time or mentioned frequently in one of our historical periods. The four individual items used in this analysis are Crankcase BIowby Standards, Auto Inspection/ Maintenance Standards, Transportation Control Measures, and Evaporative Emissions Standards.

D. Analytic Method

We use cluster analysis to examine the consistency with which different organizations espouse similar beliefs over time. Cluster analysis is a set of tools that partitions a set of objects (in our case testimonies) into two or more clusters such that the objects within clusters are similar and objects in different clusters are dissimilar (Hintze, 1996). We perform separate cluster analyses for policy core beliefs and for secondary aspects in each of the eight time periods, resulting in sixteen separate analyses.

We utilize a partitioning method of cluster analysis to classify testimonies. This subclass of methods distributes objects between k clusters. We choose the appropriate number of clusters for each analysis by comparing the "goodness-of-fit" of the results for different values of k. The use of a "goodness-of-fit" criterion allows us to apply a consistent method for determining the numbers of clusters across all sixteen analyses. The assessments of fit were made based on the average silhouette values of all clustered objects, resulting in either two or three clusters in all analyses.10 Our application of a partitioning approach provides an advantage over hierarchical methods of cluster analysis (either agglomerative or divisive) because these techniques would require subjective visual examination of output (usually dendrograms or banner plots) in order to assess the numbers of clusters for all sixteen analyses on a "goodness-of-fit" basis.11 Further, partitioning methods do not suffer from the problem of strict sequential classification, whereby initial (and erroneous) decisions to join objects into clusters cannot be undone in further iterations (Kaufman & Rousseeuw, 1990).

The specific algorithm we use is called "Fuzzy Clustering" (Kaufman & Rousseeuw, 1990). This technique has the additional advantage of not requiring individual objects (testimonies) to be strictly classified within a single cluster; rather, objects are allowed to be partially classified into (or be members in) multiple clusters. This provides information regarding the extent to which a given object is either central or peripheral in any given cluster. We implement this algorithm based on Euclidean distances between objects (testimonies). To ensure that the distances can be calculated between each dyad of testimonies, we select the subset of testimonies for each separate analysis that all address either the policy core scale or secondary aspect scale/item that is most discussed.12

We present our results for given hearings in tables that categorize testimonies of organizations into clusters-using either policy core or secondary aspect belief measures. The tables present the proportion of each testimony's cluster membership that resides in this dominant cluster. Testimonies with higher membership values are more central to the cluster; those with lower membership values are more peripheral. The average silhouette value provides a measure of the goodness of fit for the cluster analysis as a whole.

V. Results

We present cluster results by first examining the periods for which our theoretical expectations predict the greatest instability in detail. These include the two sets of hearings in the formative years of the automotive pollution control subsystem, and two sets of hearings after major exogenous events-the 1973-74 Embargo and the combined impact of the 1979 Iranian Revolution and 1980 electionsaffected the balance of power within the policy subsystem. Coalition stability during these difficult periods would be strong support for our first hypothesis. We then summarize our results across all eight periods. Here we uncover whether the dominant pattern over 26 years is one of organizations consistently falling on the same side of various issues over time. Further, we can examine whether there is more stability across issues that reflect policy core beliefs than secondary aspect beliefs.

A. Coalition Membership in Periods of Expected Instability

The four time periods in our database in which we would expect the most instability are 1963-65, 1970, 1975, and 1981-82. The first two periods-1963-65 and 1970-fall within the "Formative Years" of the automotive pollution control subsystems and provide a baseline for evaluating coalition structure in later time periods. They are prior to the technology forcing standards of the 1970 Clean Air Act and the heated conflict of the 1970s. Because they are relatively early in the conflict over tailpipe standards, we would expect relatively amorphous coalitions in the formative years (Sabatier, 1998). The second two periods-1975 and 1981-82-are the most directly subsequent to the 1973-74 oil Embargo and the sequence of events culminating in the 1980 election outcomes, which both threatened the balance of power within the policy subsystem.13

1963-Spring 1970: The Formative Years. Table 2 presents the data for the 1963-65 period, that is, the hearings leading up to the 1965 Motor Vehicle Control Act. The left columns provide the data for coalitions with respect to policy core beliefs, and the right hand column presents the coalitions with respect to secondary aspects. We can see there are three coalitions with respect to the policy core beliefs: (a) a Federal Pollution Control cluster defined by relatively high positive scores on both the Pro-Governmental Control of Air Pollution Scale (mean = 0.23) and the Pro-Federal Regulation Scale (mean 1.19), (b) a Local Pollution Control cluster defined by a high score on the Pollution Control Scale (mean = 0.52), but a negative mean on the Federal Regulation Scale (-0.51), and (c) an Anti-Pollution Control Cluster with low scores on both scales (means of -1.03 and -0.60, respectively). In parentheses are found the number of testimonies in that coalition during a particular period with scores on that scale. For example, in 1963-65, 7 people in the Federal Pollution Control cluster had scores on the Governmental Control of Pollution Scale, and 7 had scores on the Federal Regulation Scale.

The membership of the first two coalitions based on policy core beliefs is not surprising. The Federal Pollution Control Cluster consisted of US HEW, Senator Muskie, the American Public Health Association, the American Municipal Association, and the California Motor Vehicle Pollution Board. The Local Pollution Control Cluster consisted of a variety of state and local pollution control officials who, at the time, were hostile to an active federal presence. The Anti-Pollution Control Cluster contained a rather bizarre set of actors ranging from the auto companies to public health groups and several local pollution control agencies. Note, moreover, that the membership scores were generally in the 0.4-0.6 range, indicating that 40-60% of a group's testimony put it in that coalition-which obviously left 60-40% in the other two coalitions. There were also several organizations with testimonies that put them in multiple coalitions. And the average silhouette value was only 0.43 (values between 0.25 and 0.5 indicate only marginal underlying structure, see Kauf man & Rousseeuw, 1990). All of this suggests a relatively fluid set of coalitions at the policy core level during this period.

With respect to secondary aspects, the coalitions are certainly more cohesive (as represented by membership scores between 0.7 and 1.0 and an average silhouette value of 0.558), but the number of organizations testifying on these topics are too few to warrant much attention.

Table 3 deals with the testimonies presented on the Nixon. Administration bill in the winter-spring of 1970.14 At the policy core level, the belief coalitions are becoming more cohesive. Almost all the membership scores exceed 0.7, indicating that the actor's testimony is falling largely within a single coalition. There are now two coalitions, not three, although within the Pollution Control Cluster the support for a strong governmental role (mean = 0.92) is stronger than for a strong federal role (mean = 0.21). The Pollution Control Cluster consists of mostly expected members-US HEW, Senator Muskie, Representative Rogers, the AFL-CIO, and several public health groups-plus a few rather curious ones, particularly the Chamber of Commerce. The Anti-Pollution Control Cluster is now clearly dominated by the auto companies and American Petroleum Institute, but also contains a couple public health groups. Note that Gould Inc., a catalyst manufacturer, presented testimonies that put it in different coalitions-a finding that occurs during several sets of hearings in our dataset. As in 1963-65, the cluster analysis of secondary beliefs provided relatively high membership scores, but the number of organizations testifying on these topics was again relatively small.

In general, examining the 1963-65 policy core coalitions corroborates a finding from research on Lake Tahoe water policy (Sabatier & Brasher, 1993) that, in the formative years of a subsystem, coalitions will be relatively fluid. These coalitions become more distinct in 1970.

Response to the 1973-74 oil Embargo. Table 4 presents the cluster results for the 1975 hearings. It identifies three policy core clusters. At one extreme is a Pollution Control cluster composed largely of environmental and public health groups, state pollution control officials from California and New York, Senator Muskie and Representative Rogers, and several local government associations. The only anomaly is the American Petroleum Institute. On the other extreme is an AntiPollution Control cluster composed of those with strong opposition to any governmental role and consisting largely of the auto companies, the Chamber of Commerce, and the American Petroleum Institute. The third is a Moderate cluster consisting of those with mixed beliefs about the stringency of pollution control efforts (mean = -0.17 on the Governmental Control Scale).15 This cluster appears to consist of organizations from the two opposing coalitions who adopted more moderate policy positions during the 1975 period. As one might expect, it is a rather diverse group consisting of several US EPA testimonies, several testimonies by the NRDC (a prominent environmental group very active in air pollution), General Motors, the United Auto Workers, the National Academy of Sciences, and the state pollution control administrators. All three clusters are reasonably cohesive (with most membership scores > 0.7). However, several organizations (including American Petroleum Institute, US EPA, General Motors, Gould Inc.) presented testimonies that classify them in both the Moderate cluster and one of the extreme clusters.

On secondary aspects, there were only two clusters. The Strict Tailpipe Standards cluster wanted relatively strict ultimate and interim exhaust standards and was composed primarily of the usual members of the environmental coalition, plus a few new "members" (particularly the National Parking Association and the National Academy of Sciences). The Weak Tailpipe Standard cluster consisted of the auto companies, American Petroleum Institute, and the United Auto Workers. Apparently, the United Auto Workers was willing to espouse moderate views on general policy core beliefs, but was strongly opposed to stricter tailpipe standards because of the negative economic repercussions on its members. US EPA and Gould Inc. presented testimony that classified them in both clusters.

Evidence from Table 4 suggests that the 1973-74 oil Embargo put severe strains on members of the environmental coalition. While both the membership scores and the number of organizations in two separate policy core clusters is similar in 1975 and 1970, the number of policy core clusters changed from two to three, and NRDC appears in the policy core Moderate cluster. This is unlike the results of similar research at Lake Tahoe (Sabatier & Brasher, 1993), where the only important exogenous shock-the election of Jerry Brown as California governor in 1974-led to an increased hardening into two distinct coalitions during the 1970s.

Response to the Iranian Revolution and the 1980 Elections. Table 5 presents the results for the 1981-82 hearings. At the policy core level, we find a Pollution Control cluster that consists of environmental and public health groups, subnational pollution control officials, the AFL-CIO, and, barely, Representative Waxman (membership score = 0.53). The second cluster is an Anti-Pollution Control cluster that includes auto companies, the American Petroleum Institute, the Reagan EPA, and, somewhat unusually, the associations of governors and counties. The coalitions on secondary aspects are similar, with a few exceptions: the Governors Association moves into the Strict Tailpipe Cluster, Representative Dingell (a congressman from Michigan) emerges in the Weak Tailpipe Cluster, and, most curiously, Representative Waxman shows up in the Weak Tailpipe Cluster.16

Focusing solely on the four periods presented here, we see relatively strong support for the contention that coalitions will remain stable over periods of a decade or more (Hypothesis 1). The line-up of allies and opponents in 1981 strongly resembled that in 1970. This is particularly remarkable given the threat posed by the Reagan administration to the policy core of the Pollution Control cluster. Virtually no one defected. In the next section, we examine summary analyses that incorporate cluster results for all eight sets of hearings in our database.

B. Summary of Coalition Membership, 1963-1989

Our results on the composition of belief coalitions in the four time periods above identifies two separate types of changes in coalition structure that have implications for coalition stability over time. The first relates to the divergence between the Federal Pollution Control and the Local Pollution Control policy core clusters in the 1963-65 period. This fracture disappeared in 1970, as the active organizations appeared to find common ground prior to the major policy change in the 1970 Clean Air Act. This divergence does not reappear in later periods, suggesting that Pollution Control cluster members were able to find and maintain common ground on this issue over the period of twenty years. The convergence in coalitions lends support to the expectation of coalition stability over a decade or more (Hypothesis 1). It also corroborates results from Sabatier and Brasher's (1993) study of change in environmental policy at Lake Tahoe in which coalitions were quite amorphous in the formative stages of the subsystembut subsequently coalesced.

The second type of structural change in belief coalitions results from the emergence of a "moderate" cluster in the later periods. Figure 1 begins to summarize our findings on this topic for the 21 organizations whose testimonies dealt with both the critical policy core and secondary aspect items in at least three separate hearing testimonies in the 1970 and later periods.17 It is a ternary plot displaying the organization's average cluster membership scores in the Pollution Control, Moderate, and Anti-Pollution Control policy core clusters across all testimonies after the 1963-65 period. Each cluster is represented by one of the triangle's vertices. The closer an organization falls to a vertex, the greater the group's average membership score in the cluster that vertex represents. The three regions on the plot identify the cluster in which each organization has the largest average membership value.18

On policy core scales, 15 of the 21 actors who testified repeatedly were classified with high membership scores in either of the two extreme clusters. Seven of these actors were in the Pollution Control cluster. Sierra Club (Sierra) and CA Air Resources Board (CARB) both had greater than 80% of their membership in this cluster. National Clean Air Coalition (NCAC), League of Cities (Cities), National Governors Association (Governors), and the South Coast Air Quality Management District (SCAQMD) all had greater than 70% of their cluster membership classified in the Pollution Control cluster, and the American Lung Association (ALA) had nearly 70%. The eight other organizations were clumped near the Anti-Pollution Control vertex. The Motor Vehicle Manufacturers Association (MVMA), American Petroleum Institute (API), Chamber of Commerce (Commerce), and Chrysler all had more than 70% of their average membership score in the Anti-Pollution Control cluster. Ford and General Motors (GM) had greater than 60% of their average membership in this cluster, and Honda and the United Auto Workers (UAW) had nearly 60%.

Of the remaining six organizations, one actor-the National Academy of Sciences (NAS)-had most of its average membership in the Moderate cluster. The other five organization's average membership scores classified them in the Pollution Control cluster overall, but they also had significant membership scores in the other clusters as well. Of these, the State and Territorial Air Pollution Program Administrators (STAPPA) was the closest to the more central members of the Pollution Control Cluster, with almost 57% of its membership in the Pollution Control cluster and the remainder split evenly between the other two clusters. NRDC had a lower average membership score in the Anti-Pollution Control cluster (15%) than half of the central members of the Pollution Control cluster, but the remainder of its membership scores were split almost evenly between the Pollution Control and the Moderate clusters. The National Association of Counties (Counties) had low membership scores in the Moderate cluster, but was almost evenly split between the two extreme clusters. Finally, US EPA/HEW and Gould Inc. seemed to distribute their testimonies the most evenly between the three clusters.

Figure 2 is a ternary plot of displaying the organizations' average cluster membership scores in the Stricter Standards, Brokering, and Weaker Standards secondary aspect clusters across all testimonies after the 1963-65 period. The results here are similar to those in Figure 1 in two main respects. First, the average membership scores of most organizations are skewed in favor of one of the two extreme clusters (i.e., Stricter Standards or Weaker Standards). Eleven organizations have greater than 70% of their average cluster membership in either of these two clusters. This includes Sierra Club, NRDC, National Clean Air Coalition, South Coast Air Quality Management District, and League of Cities for the Stricter Standards cluster, and Chamber of Commerce, Ford, Chrysler, General Motors, United Auto Workers, and Motor Vehicle Manufacturers Association for the Weaker Standards cluster. Four additional organizations have greater than 60% of their average cluster membership in one of the extreme clusters. This includes CA Air Resources Board, National Academy of Sciences, and STAPPA for the Stricter Standards cluster, and American Petroleum Institute for the Weaker Standards Cluster. American Lung Association almost meets the 60% cutoff with 59.9% of its average cluster membership in the Stricter Standards Cluster. Second, 17 of the 21 organizations are in the same regions in Figure 1 as they are in Figure 2, suggesting that there is consistency across the policy core and secondary aspect beliefs expressed by the organization in public.

While the data appear to support the long-term coalition perspective in aggregate, we also find several exceptions. The first class of exceptions occurs when organizations change their policy positions to take advantage of possibilities of short-term gains, similar to Riker (1962,1980). Honda, for example, entered the pollution control debate in our dataset in 1973 when it testified in favor of more stringent standards than those supported by the American auto manufacturers (Honda's three testimonies in Figure 2). At this time, the Honda CVCC engine promised to meet hydrocarbon and carbon monoxide standards with only a moderate relaxation of the 1976 nitrogen oxides requirements. The standards offered the potential for gaining market share by selling engines and/or technology to U.S. manufacturers. However, the CVCC was not chosen as the dominant control technology, and when Honda next appears in our database in 1989, they testify twice in a manner consistent with the policy core Anti-Pollution Control cluster (see Figure 1).

Gould Inc. provides another example of a Rikerian coalition change. Gould Inc., a manufacturer of a catalyst-based emissions control device, first enters our dataset in the 1970 hearings. Although there is considerable variation in how this company is classified between the three different clusters (as evidenced by its relatively central position in Figures 1 and 2), there is an underlying trend in its distribution between clusters over time. In 1970, the content of its testimonies were skewed toward the antipollution control and moderate clusters (67% for the policy core and 100% for the secondary aspects). This declines to 50% in the Anti-Pollution Control and Moderate Clusters for both the policy core and secondary aspects in 1973. In 1975 (the last period Gould Inc. testifies in our database), evidence mounted against the practicality of the CVCC engine for the U.S. auto industry and the catalyst technology emerged as the control method of choice. Not surprisingly, Gould Inc.'s membership in the more control-oriented clusters increased so that it was classified in the policy core Pollution Control cluster in 67% of its testimonies (2 of 3) and in the secondary aspect Strict Standards cluster in 80% of its testimonies (4 of 5). As with Honda, the most parsimonious explanation for this shift is the benefits of stricter standards to the company's bottom line.19

The second type of exception to the pattern of long-term coalition stability results from changing presidential administrations. US EPA (and its precursors in US HEW) is the only federal agency active in our dataset in the 1970 and later periods. Overall, US EPA/HEW is classified in all three clusters, resulting in its central positions in Figures 1 and 2. The expressed policy views of US EPA/HEW, however, vary over time with the ideology of leaders of the executive branch. The relatively environmental Nixon administration and the Democratic Carter administration both mark the 1970s. During hearings in these periods, US EPA/HEW is classified in either the moderate- or the control-oriented clusters in 100% and 82% of their testimonies on policy core (n = 14) and secondary aspect (n = 11) issues, respectively. After the Reagan-Bush administration, US EPA's expressed beliefs are always classified in the policy core Anti-Pollution Control (n = 4) and secondary aspect Weak Standards (n = 4) clusters. These results are consistent with JenkinsSmith and St Clair (1991) and Jenkins-Smith et al. (1993).

The third class of exceptions to the long-term stability perspective was peak associations with diffuse interests. This includes both the National Governors Conference/Association and the National Association of Counties. The National Governors Conference is clearly in the Pollution Control policy core cluster in Figure 1 but is evenly split between the Moderate and Weak Standards secondary aspect clusters. The National Association of Counties is split almost evenly between the Pollution Control and Anti-Pollution Control policy core clusters in Figure 1 and does have a majority of its average membership in any of the three secondary aspect clusters in Figure 2.

In short, the majority of actors testify consistently over long periods of time on the basis of belief systems combining policy core and secondary aspects. These data provide rather strong support for Hypothesis 1, that coalitions will remain relatively stable over a period of a decade or more. The exceptions-that is, those operating on the basis of shifting conditions and/or rather supple belief systems-include two corporations (Gould Inc. and Honda), the US EPA/HEW, and two associations of elected officials (the Governors and Counties). This stability of coalitions is particularly remarkable given that the subsystem went through two major exogenous shocks: the 1973-74 Oil Embargo and the 1979-80 combination of the Iranian Revolution and the Reagan election. Although several actors did soften their positions to form more moderate clusters after major exogenous events, few organizations switched positions to support the opposing coalition.

The second hypothesis-that organizations would demonstrate greater stability over time on policy core beliefs than on secondary aspects-was not confirmed by the data. Although belief coalitions based on both types of beliefs appear relatively consistent over time, the data do not suggest that one category is clearly more stable. Perhaps this is because the most frequently discussed secondary aspects in our case-tailpipe standards-affect virtually the entire subsystem and represent the clearest example of the technology forcing strategy of the 1970 Clean Air Act. Thus, they are very similar in scope and salience to the policy core items. In fact, the evidence presented here, in combination with that in Zafonte and Sabatier (1998), convinced Sabatier and Jenkins-Smith (1999) to place policy preferences that are subsystem wide in scope and salient over a long period into the policy core as "policy core policy preferences."

Apart from reconceptualizing the tailpipe standards, we suspect that Hypothesis 2 is correct, but we simply cannot adequately test it using our data base. Our data base tends to screen out issues that were raised in less than three hearings. This would include issues such as the viability of transportation controls in various locales, as well as inspection and maintenance programs in different states. Even on broader issues, such as the general desirability of transportation control programs, inspection and maintenance programs, and evaporative emissions controls, those issues were raised much less frequently than tailpipe controls in all three periods examined in this study (1970,1975, and 1981-82). Although the general issues were mentioned over time, they were much less salient than tailpipe standards to most members of the auto pollution control subsystem. This would suggest that, although stable, they were of secondary importance to most members of the policy subsystem. On the other hand, a determined skeptic would say, "You still haven't proven Hypothesis 2 is correct." And that person would be right.

C. How Much Coordination Was There?

Our analysis thus far has produced evidence of two groups of actors, each espousing relatively similar beliefs from the 1960 through the 1980s. The first "belief coalition" was composed of environmental and public health groups; the California Air Resources Board (CARB); state pollution control agencies (STAPPA); the South Coast (Southern California) Air Quality Management District (SCAQMD); and a few members of Congress, particularly Senator Muskie and Congressman Waxman. It believed that air pollution was a serious public health issue, favored technology forcing as a basic strategy, supported a strong national control program, and consistently favored stricter tailpipe standards, transportation control measures, inspective and maintenance programs, and evaporative emissions controls as control strategies. The second belief coalition consisted of the American auto companies, the American Petroleum Institute, perhaps Honda, the United Auto Workers (UAW), and Congressman John Dingell from Michigan. It was hostile to governmental regulation in general and particularly to technology forcing as a basic strategy. It was preoccupied with the economic costs of regulation and continually opposed any new controls.

Unfortunately, the data presented thus far only deal with espoused beliefs. There has been no evidence that people with similar beliefs actually coordinated their behavior. And without some evidence of coordinated behavior, no "coalitions" can be said to have existed (Hula, 1999). This is reminiscent of Edella Schlager's (1995) criticism of the ACF's assumption that similar beliefs automatically produce coordination.

Although we have no systematic data on this topic, we have interviewed 10 long-term actors in the automotive pollution control subsystem to ascertain how much coordination of testimony and other lobbying activity occured during the 1970s and 1980s.20 Here is what we found:

1. Environmental Belief Coalition

* Environmental and public health groups constantly exchanged information, developed strategies together, and have had a formal organization-the National Clean Air Coalition (which also includes the League of Women Voters and the Steelworkers Union)-that has met biweekly since 1973.

* STAPPA sometimes met with NCAC and had positions very similar to environmental groups. Particularly in the 1970s, it was heavily reliant on CARB for information.

* CARB was definitely viewed by environmental groups as a key ally. It often had better information than the US EPA and were more willing to share it. As an independent regulatory commission, it had to be careful about appearing fair and neutral. But some members periodically met with enviros. CARB preferred to be seen as a little more "moderate" than environmental groups.

* Senator Muskie's office did not try to coordinate testimony, but it was viewed by environmental groups as a major ally. During 1960 and 1970s, it relied heavily on CARB for technical and political advice.

2. Anti-Pollution Control Coalition

* Big 3 automakers coordinated their development of pollution control devices in 1960s; this led to a 1969 consent decree that prohibited coordination under anti-trust law (Doyle, 2000). During 1970s the Big 3 were very nervous about doing anything together. But they exchanged a lot of information because they felt very threatened by all the regulations. Senator Muskie's office prohibited Auto Manufacturers Association from testifying; each company did it separately. After the consent decree was vacated by the Reagan administration, the Big 3 developed close ties for the rest of the 1980s.

* Japanese automobile firms were minor actors in 1970s, except for Honda. They became more important in the 1980s, especially Toyota. Congressman Dingell was viewed as the major congressional protector of the auto industry (Doyle, 2000). He helped coordinate the Big 3 testimony and lobbying strategy in 1981-82.

* API and oil industry shared the antiregulation ideology of auto industry and supported them in tailpipe emission hearings. But they were major opponents on several major issues in 1970s and 80s: removing lead from gasoline in the early 1970s, evaporative emission controls in the late 1980s, and reformulated gasoline in the 1980s and 1990s. Several auto executives did NOT view them as allies.

* United Auto Workers definitely were viewed as an opponent by environmental groups and as an (untrustworthy?) ally by the auto companies in the late 1970s and 1980s.

In sum, the "core" members of each coalition (environmental groups and Big 3 automakers) clearly coordinated their strategies when not prohibited by law. STAPPA was a secondary member of the environmental coalition with some coordination. CARB was a core member in terms of sharing technical information but had to be somewhat circumspect in lobbying campaigns before EPA and Congress. The two legislators (Muskie and Dingell) were probably secondary members of their respective coalitions. The most problematic case was the oil firms. Although they shared the auto firms' antiregulatory ideology, they nevertheless were opponents on enough big policy issues that they weren't viewed as allies by the auto companies.21

VI. Conclusion

This article has used testimony presented at congressional hearings to examine the structure and stability of belief coalitions regarding U.S. automobile pollution control policy in the period from 1963 to 1989. The evidence presented here generally supports the basic expectation of stable coalitions in the policy process rather than shifting, short-term coalitions of convenience. Our results on the types of beliefs that are more stable over time is more mixed.

Our research builds upon the previous work of Jenkins-Smith and St. Clair (1991) and Jenkin-Smith et al. (1993)-arguably the best existing study on coalition stability in the policy process-in at least one important way. We broaden the range of actors examined to include not only national interest groups and federal agencies that intervene regularly but also (a) state and local agencies and (b) actors who testify periodically but much less frequently.22 Both sets of actors should produce strains on coalition stability, the first because gubernatorial (and other) elections might produce changes in testimony before Congress, the second because actors who intervene relatively infrequently do not have the same incentives as "regulars" to be sensitive to the costs of defecting from the coalition. Nevertheless, although several of these groups-notably, the national associations of governors and countries-switched coalitions with some frequency, the addition of these two new groups of actors did not materially alter the overall impression of general coalition stability from 1970 to 1989.

In addition to the substantive findings of coalition stability in both policy core and secondary aspect beliefs, this article demonstrates the utility of systematic measurement of organizational behavior when examining coalition composition over time. Qualitative case studies by their very nature run the risk of analyzing "interesting" behavior in depth, at the expense of understanding general patterns. The interesting behavior in our case appears when organizations shift from what appears to be their "traditional" or expected coalitions (e.g. US EPA/HEW, Honda, Gould Inc.). However, drawing the conclusion that coalition structure is fluid based on these findings would, in our view, ignore the stability in coalition membership exhibited by most organizations. Conversely, quantitative studies that assert coalition membership based on attendance at hearings without systematic examination of the content of the testimony presented (e.g., Worsham, 1997) assume away the interesting variation found by applying either more qualitative methods or the quantitative content analysis used in this article. This assumption is especially dangerous when the ultimate goal of the research project is to assess the stability of coalition structure. The methodology applied in this article is able to identify both the overall pattern of coalition membership over time along with the interesting exceptions to those patterns (e.g., US EPA/HEW, Honda, Gould Inc.).

We believe that our findings concerning coalition stability should be generalizable to a relatively broad range of policy areas, at least within the United States. First, we have seen that the automotive pollution control subsystem contains the full range of subsystem actors from national agencies and interest groups to members of Congress, national associations of government, specific state and local agencies, individual corporations, and even the National Academy of Sciences. This diversity of actors is characteristic of most domestic policy arenas. Second, our observation of stable coalitions occurs in the face of two relatively large exogenous perturbations-the 1973-74 Oil Embargo and the 1980 elections-that substantially affected the policy debate. In our case, coalition members responded to these significant events while maintaining a relatively stable set of coalition partners. However, a few actors did moderate their espoused policy stances for a few years.

When examining the results from our case, it is important to note that automotive pollution control policy is characterized by substantial divergence between coalitions on broad, underlying beliefs, i.e., those that we term policy core. Coalition composition may be more fluid in cases in which there is broad consensus on policy core issues (e.g., farming should/should not be subsidized), but disagreement over more specific issues e.g., which industries should reap the benefits of subsidized markets (Browne, 1988). In these cases, actors are probably better able to empathize with policy positions taken by their "opponents" and, therefore, less likely to attribute nefarious motives to their actions. This should reduce belief hardening and resulting stability reinforced by the "devil shift" process identified by Sabatier et al. (1987).

VII. Methodological Appendix-Scale Construction

Policy Core Beliefs. We performed a factor analysis with varimax rotation on the correlation matrix of six policy core items. This identified two major dimensions underlying these variables. The first reflects level of support for Government Control of Air Pollution. This scale contains the following four standardized items (with their numeric coding values and factor loadings):

"Technology forcing is an appropriate strategy to help control air pollution." Codes range from "Strongly Oppose" (1) to "Uncertain/Mixed" (3) to "Strongly Favor as General Strategy" (5). Rotated Factor Loading = 0.892.

"Overall threat of air pollution control on the economy." Codes range from "Very Beneficial" (1) to "No Net Impact" (3) to "Extremely Serious Threat" (5). Rotated Factor Loading = -0.847.

"Overall seriousness of air pollution as a national health/environmental quality problem." Codes range from "Not Serious at All" (1) to "Moderately Serious" (3) to "Extremely Serious" (5). Rotated Factor Loading = 0.780.

"Scope of governmental versus private (market) activity." Codes range from "Much More Market" (1) to "Status Quo" (3) to "Much More Government" (5). Rotated Factor Loading = 0.761.

Recall that someone who testified on at least one of these items would get ranked on the scale (in point of fact, the mean number of component items was 1.86 when this scale was mentioned).

The second policy core scale reflects level of support for Federal Government Regulation. This scale contains the remaining two standardized items (with their numeric coding values and factor loadings):

"Proper level of government responsibility" Codes range from "Primarily Local" (1) to "Primarily State" (3) to "Primarily Federal" (5). Rotated Factor Loading = 0.915.

"Proper level of government responsibility" Codes range from "Much Less Federal" (1) to "Status Quo" (3) to "Much More Federal" (5). Rotated Factor Loading = 0.738.

The mean number of component items mentioned for this scale was 1.49.

Secondary Aspects. We focused our analysis on 10 secondary aspects, creating two four item scales and leaving three items by themselves. The inter-scale item correlations were sufficiently high (r > 0.59 for all six bivariate comparisons in the first scale, and r > 0.70 for all six bivariate comparisons in the second scale) that we made scale construction decisions based upon both the correlation matrix and the logical consistency of the individual items.

The first scale deals with evaluation of the Strictest (Ultimate) Tailpipe Standards in Law. Each variable in this scale addresses the following item for Tailpipe Standards (Overall), Nitrogen Oxide Standards, Carbon Monoxide Standards, and Hydrocarbon Standards:

"Evaluation of the strictest/ultimate tailpipe standards in current law." Codes range from "Much Too Strict" (1) to "OK" (3) to "Need to be Much Stricter" (5).

The mean number of component items mentioned for this scale was 2.29.

The second scale focuses on evaluation of the Interim Tailpipe Standards. Each variable in this scale address the following item for Tailpipe Standards (Overall), Nitrogen Oxide Standards, Carbon Monoxide Standards, and Hydrocarbon Standards:

"Evaluation of the continuation of current year standards for the next two years." Codes range from "Much Too Strict" (1) to "OK" (3) to "Need to be Much Stricter" (5).

The mean number of component items for this scale was 2.21 when it was mentioned.

The four individual items that enter our analysis are secondary aspects that evaluate policy issues that were either mentioned consistently over time, or mentioned frequently in one of our historical periods. They are:

Evaluation of Current Auto Inspection/Maintenance Programs from "Much Too Strict" (1) to "OK" (3) to "Need to be Much Stricter" (5).

Evaluation of Current Crankcase (Blowby) Standards from "Much Too Strict" (1) to "OK" (3) to "Need to be Much Stricter" (5).

Position on TCMs in general to reduce VMT from "Strongly Oppose" (1) to "Neutral/Need More Information" (3) to "Strongly Support" (5).

Evaluation of Current Evaporative Standards (e.g., on-board fuel recovery systems) from "Much Too Strict" (1) to "OK" (3) to "Need to be Much Stricter" (5).