Examining Patterns of Political, Social Service, and Collaborative Involvement of Religious Congregations: A Latent Class and Transition Analysis

Transcription

1 DOI /s Examining Patterns of Political, Social Service, and Collaborative Involvement of Religious Congregations: A Latent Class and Transition Analysis Nathan R. Todd Jaclyn D. Houston Ó Society for Community Research and Action 2012 Abstract This investigation examines typologies of congregations based on patterns of congregational political and social service activities and collaborative partners. Based on a latent class analysis of a national random sample of 2,153 congregations, results indicated four distinct types of congregations with unique patterns of political, social service, and collaborative partnerships labeled: (a) Active, (b) Not Active, (c) Social Service Not Political, and (d) Political Not Social Service. Moreover, congregational characteristics such as religious tradition and clergy characteristics predicted membership in certain types. A latent transition analysis using an additional 262 congregations revealed distinct patterns of how congregations changed types across a nine year period. Results showed both congregational continuity (e.g., Not Active congregations remained Not Active) and change (e.g., Active congregations were likely to change type membership). This study advances congregational research by examining congregational types, what predicts certain types, and how congregations change types across time. Implications for future research and partnership with religious congregations also are discussed. Keywords Religious congregations Social services Politics Collaboration Latent class analysis Latent transition analysis N. R. Todd (&) J. D. Houston Department of Psychology, DePaul University, 2219 N. Kenmore Avenue, Chicago, IL 60614, USA ntodd@depaul.edu Introduction Religious congregations are local social settings where individuals gather to practice and pursue spiritual ends (Chaves 2004; Pargament 2008; Pargament and Maton 2000) and are woven into the social fabric of society with approximately 150 million people attending over 330,000 congregations across the US (Linder 2010). A growing body of research moves beyond personal religious beliefs and participation to focus on the congregation as the unit of analysis. For example, studies examine how congregations serve as empowering community settings (Maton 2008; Maton and Salem 1995), organize for social change (Speer et al. 2003), and the characteristics of congregations that promote civic engagement (Schwadel 2005) and social justice (Todd and Allen 2011). Of interest to community psychologists, previous research has examined the core activities of congregations to understand how they (a) provide social services (Chaves and Tsitsos 2001), (b) are politically involved (Beyerlein and Chaves 2003; Putnam and Campbell 2010), and (c) collaborate to achieve shared goals (Ammerman 2005; Maton et al., in press). Understanding these dimensions of congregational activity is important to comprehend the diverse ways they interface with society and the community. However, few studies jointly consider multiple types of congregational activities (e.g., political, social service, collaborations) to understand the combinations of activities offered by congregations or how activities change across time. This study advances research by using latent class and latent transition analysis to examine combinations of activities within congregations, what predicts certain combinations, and how activities change across time. Beyond differences in activities, religious congregations in the US are diverse social settings along dimensions of

2 (a) size, (b) composition of members by age, race/ethnicity, and (c) education (Dougherty 2003; Putnam and Campbell 2010), (d) composition of leaders and leadership structure (Chaves 2004), (e) location in the US and proximity to cities, and (f) if the congregation belongs to a denomination (i.e., a formal national organization such as Methodist or Southern Baptist; Chaves 2004). Pargament and Maton (2000) note congregations differ in their mission, organizational structure, and particular theological beliefs which results in different community and spiritual goals, activities, and priorities. One major difference among congregations is in faith tradition (e.g., Muslim, Christian, Jewish, Baha i); however, there also is diversity within each faith tradition. For example, within Christianity, sociologists have identified major categories of denominations based on historical developments and particular theological beliefs such as biblical literalism (Ammerman 2005; Putnam and Campbell 2010; Steensland et al. 2000; Wuthnow 1996). According to Steensland et al. (2000), denominations are classified as Black Protestant, Evangelical Protestant, Mainline Protestant, and Catholic. Steensland et al. (2000) demonstrate social and political differences between these groups which lends further support to the importance of ideological differences among congregations (Wuthnow 1996). Overall, it is crucial to examine this diversity to better understand how differences among congregations predict patterns in the types of activities coordinated by congregations in their community. We now situate the current study in the literature on congregational political and social service activities and collaborations while highlighting the importance of congregational diversity and the need to examine congregations across time. Congregations and Political Activity Research has examined distinct types of congregational political activity such as (a) congregational lobbying, advocacy, and marching (e.g., Beyerlein and Chaves 2003), (b) conducting voter drives and distributing voter guides (e.g., Chaves 2004), and (c) providing a context for political discussion, education, and training in skills for civic engagement (e.g., Chaves 2004; Putnam and Campbell 2010; Schwadel 2002; Verba et al. 1995). However, based on a national random sample of over 1,200 congregations, overt political activity is peripheral to most congregations, a finding which counters the stereotypic image of congregations as deeply involved in political activity (Chaves 2004). Even if not directly involved, Putnam and Campbell (2010) assert that congregations serve as settings where political socialization takes place through informal friendships forged in the congregation and indeed members of the same congregation tend to be politically homogeneous (Campbell 2004; Wald et al. 1990). Among politically involved congregations, many differences in the type (lobbying, voting, discussions) and content (for or against certain issues such as abortion) of political activity emerge based on the race and religious tradition of the congregation (Chaves 2004). The current study focuses on the type rather than content of political activity, though literature is plentiful that examines personal religious belief and political attitudes (Layman 1997; Putnam and Campbell 2010; Wald and Calhoun-Brown 2011), religious movements and politics (Manza and Wright 2003; Wald et al. 1989), and the interface of religion and politics in the US (e.g., Green et al. 1996; Wald and Calhoun-Brown 2011). Although congregational political activity is the exception rather than the norm (Chaves 2004), there are differences among politically involved congregations that follow race and religious tradition. Chaves (2004) and colleagues (Beyerlein and Chaves 2003) found that Black Protestant and Catholic congregations exhibit more political activity than other traditions and engage in distinct ways. Black Protestant congregations are more likely to engage in voter registration drives and to distribute voter guides (Beyerlein and Chaves 2003; Brown 2006; Chaves 2004), likely connected to the link between the civil rights movement and Black congregations (Calhoun-Brown 2000; Liu et al. 2009; Moore 1992). Catholic congregations are more likely to engage by lobbying elected officials or participating in marches or demonstrations (Chaves 2004), which may connect to a few key issues such as abortion (e.g., O Connor and Berkman 1995). White Evangelical congregations are more likely to distribute voter guides, and specifically voter guides from the Christian Right which may reflect one way the Christian Right works through congregations to influence electoral politics (Chaves 2004; Manza and Wright 2003; Woodberry and Smith 1998). Mainline Protestant congregations engage by housing political discussions and linking members to political opportunities, but are less likely to engage in ways that influence voting (e.g., Wuthnow and Evans 2002). In general, larger congregations engage in more political activity (Beyerlein and Chaves 2003). Overall, this research shows the importance of demographics in how congregations engage politically. Congregations and Social Services Pargament and Maton (2000) and Maton et al. (in press) describe how congregations have different missions and strategies based on their theological beliefs and religious traditions, which shape outreach efforts to their local community. Maton et al. (in press) describe that some congregations serve as the site for community intervention by providing physical or mental health programs, while others, about 58 % of congregations according to Chaves

3 and Tsitsos (2001), may serve as a community resource by providing social services that meet direct needs such as food, clothing, or shelter. Other research reports higher levels (up to 90 %) of congregational involvement in social services which may be due to oversampling of larger congregations that are more likely to provide services (Chaves and Tsitsos 2001) or that even if congregations do not directly provide social services they may help supply volunteers for such efforts (Putnam 2000). Differences among congregations in social service provision correspond with differences in congregational demographics and religious tradition. Chaves (2004) and Chaves and Tsitsos (2001) note that congregations in poor neighborhoods, those with more college graduates, and those with lower incomes do more social services. The gender of congregational leadership also is important as Stewart-Thomas (2009) reports congregational participation in social services is four times greater when the primary religious leader is a woman rather than a man. Findings are mixed as to whether racial composition predicts congregational social service provision (Brown 2008; Cavendish 2000; Chaves 2004); however, Tsitsos (2003) finds differences in types of programs where African American congregations are more likely to focus on mentoring and educational programs. Tradition may account for how Mainline Protestant and Catholic congregations exhibit similar levels of social service activity but significantly more activity than Evangelical Protestant congregations (Chaves and Tsitsos 2001). Furthermore, congregations that self-identify as theologically liberal report an average of two more social service programs than theologically conservative congregations (Chaves 2004). Overall, congregational demographics and religious tradition predict the amount of social service provision. Congregations and Collaborative Partners Ammerman (2005) notes that congregations often collaborate with community partners to achieve common goals, such as social services provision, political advocacy, or to host other neighborhood and civic groups. Indeed, Chaves (2004) notes that 84 % of congregations engaged in social service activities did so in collaboration with an organization outside of their congregation and qualitative interviews with religious leaders note collaboration as key for community engagement (Todd and Rufa 2012). Congregations may partner with other congregations or religious organizations, government, or other secular non-profit organizations (Ammerman 2005; Chaves 2004; Chaves and Tsitsos 2001; Pargament and Maton 2000). Congregations also may play a role in faith-based community organizing (Wood 2003; Wood and Warren 2002) as exemplified by congregations networking together to promote social and political change (Robinson and Hanna 1994; Speer et al. 2003; Todd 2012). However, certain congregations are more likely to collaborate or may collaborate with different partners. For example, congregations that are large, liberal, and educated are more likely to collaborate (Chaves 2004; Chaves and Tsitsos 2001). More specifically, Mainline Protestant and African American congregations are more likely to partner with secular organizations (Ammerman 2005) or government (Chaves 1999, 2004) whereas Conservative Protestant and Catholic congregations are more likely to partner with other congregations or religiouslybased organizations (Ammerman 2005). Understanding if and with whom congregations collaborate for social service provision provides a window into how and with whom congregations provide social services and sheds further light onto how congregations are woven into the fabric of local communities. Congregational Types: Integrating Political, Social Service, and Collaborative Partners Extant research on congregational political, social service, and collaborative partners examines these dimensions separately. Yet conceptual scholarship discusses congregations as whole entities that engage in combinations of these activities. For example, Pargament and Maton (2000) note that congregations have different theologies, missions, and organizational structures that result in distinct cultures and pathways to community outreach such as through social action, service, conversion, or avoidance. Becker (1999), in a qualitative study of conflict in 23 congregations, proposed 4 major religious cultures that structured congregational activity such as (a) house of worship, (b) family, (c) community, and (d) leader where each culture resulted in different congregational activities. These observations resonate with research that notes congregations have distinct missions such as serving the needy, spreading the faith, changing the world, or defending morality (Ammerman 2005; Roozen et al. 1984). This suggests that there are certain types of congregations; thus, examining congregational activities that include political, social service, and collaborative partners allows for unique types to emerge based on the combination of all three types of activities. For example, congregations that provide social service activities alone or with collaborators may also have distinct patterns of political activities; thus, examining all three together gives a more holistic sense of congregations and the unique ways they interface with their communities. Empirical research on congregational types is limited. In one notable exception, Green (2007) used a stratified random sample of US congregations to group congregations into five types based on the number and kind of their social

4 service programs: (a) comprehensive activity, (b) extensive activity, (c) moderate activity, (d) specialized activity, and (e) limited activity. These types of congregations varied by geographic region (e.g., extensive type in Northeast and Midwest) and demographic characteristics (e.g., greater activity in congregations with more revenue and diversity in terms of member age, race, and income). Moreover, congregations that were more active reported more collaborations. However, these types were only based on social services, and to our knowledge no empirical study to date has jointly examined the patterns of political, social service, and collaborative activities. Such a contribution provides a more holistic and nuanced understanding of congregations and their contribution to their local community. Change in Congregational Activities Across Time Like any organization, congregations change over time. Ammerman (1997) documents a process of change for 23 congregations across the US and notes how the religious ecology within and around congregations prompted congregational change such as (a) changing neighborhood ethnic and racial demographics, (b) the presence of new congregations, (c) changing economic conditions at local and national levels, and (d) denominational changes or shifts in leadership. The emergence of religious movements such as the Religious Right also may influence local congregational belief and practice (Manza and Wright 2003). Within the congregation beliefs may shift over time (e.g., becoming more open to LGBT individuals and issues) or congregational conflict may prompt changes in mission, identity, and activities (Ammerman 1997, 2005). Although there may be continuity over a short period of time, congregations likely evolve to meet new congregational and community needs. Chaves and Anderson (2008) examine changes from 1998 to 2007 using the National Congregations Study and find similar levels of congregational political, social service, and collaborative activity across these 9 years; however, they did observe a net increase in voter guide distribution and they did not examine individual congregational change. Research is needed to examine how congregations change over time such as by examining how congregations shift types across time. Present Investigation The current study uses latent class analysis (Collins and Lanza 2010) to see if distinct types of congregations emerge based on patterns of congregational political and social service activities and collaborative partners. Latent class analysis is conceptually similar to cluster analysis where distinct clusters (i.e., types) are formed based on responses to multiple questions (Steinley and Brusco 2011). Congregational characteristics (e.g., size, clergy, religious tradition, member composition) are then used to predict membership in certain types. Latent transition analysis (Collins and Lanza 2010) is used to see how congregations transition from one type to another across two measurement occasions spaced 9 years apart (1998, 2007). To date, little empirical work examines combinations of congregational activities and partners and little is known about congregational change across time. Thus, this study contributes to the extant research on the role of congregations in providing social services in the community, congregational political involvement, and who congregations work with to achieve their goals. Findings have potential to further elucidate the role of congregations in local communities and how they offer resources for social services and political engagement. Method Participants and Procedures The data used in this research are from the first (1998) and second (2007) waves of the National Congregations Study (NCS; Chaves and Anderson 2008; Chaves et al. 1999) and are available online at The NCS surveyed a national random sample of congregations by having General Social Survey (GSS) participants in 1998 and 2007 who attended a religious service in the last year provide the name and location of their congregation. These congregations were then recruited for the NCS with the assumption that congregations identified by a random sample of individuals from the GSS would result in a random sample of congregations. The congregational response rates in 1998 and 2007 were 80 and 78 % respectively. Congregational data were collected via an interview of a key informant, usually the religious leader of the congregation. Data collected in 1998 consist of 1,234 congregations and the data from 2007 consist of 1,506 congregations. However, out of these congregations, 262 participated both in 1998 and This is because a random sample of congregations (stratified by race, urban/rural location, size, and religious tradition to increase representativeness) was drawn from the participating congregations in 1998 in order to create panel data with the same congregations participating at two timepoints. Additional sampling details are available at or from study investigators (Chaves and Anderson 2008; Chaves et al. 1999). In this study we combined congregations that reported data for congregational activities and were sampled in either 1998 or 2007 into one dataset which we call the Full sample (N = 2,153). We also used the panel dataset consisting of the 262

5 congregations providing data both in 1998 and 2007, which we call the Panel sample. Thus, the Full sample only contains independent congregations that provided data in 1998 or 2007 but not both, whereas the Panel sample contains all congregations providing data at two time-points. Congregational demographics for the Full and Panel samples are presented in Table 1. Measures All study measures were assessed through interviews with key informants who reported on their religious congregation (Chaves et al. 1999). Although direct information about the reliability and validity of survey items is not available (e.g., most measures are single items), Chaves Table 1 Frequencies of congregational demographics Full dataset Panel data 98 Panel data 07 Variable Frequency (%) Frequency (%) Frequency (%) Year (42.2) 262 (100.00) ,244 (57.78) 262 (100.00) Region East coast 347 (16.12) 47 (17.94) 47 (17.94) Midwest 410 (19.04) 70 (26.72) 70 (26.72) South 863 (40.08) 100 (38.17) 100 (38.17) West coast 533 (24.76) 45 (17.18) 45 (17.18) Census tract Urban 440 (20.45) 169 (64.50) 176 (67.17) Suburban 315 (14.64) 37 (14.12) 40 (15.27) Rural 1397 (64.91) 56 (21.37) 46 (17.56) Clergy gender Man 1,962 (94.69) 233 (93.95) 226 (86.26) Woman 110 (5.31) 15 (6.05) 21 (8.02) Clergy race White 1,616 (78.56) 203 (81.85) 197 (75.19) Person of color 441 (21.44) 45 (18.15) 65 (24.81) Tradition Catholic 521 (24.20) 65 (24.81) 65 (24.81) Evangelical protestant 809 (37.58) 94 (35.88) 96 (36.64) Black protestant 263 (12.22) 26 (9.92) 26 (9.92) Mainline protestant 480 (22.29) 70 (26.72) 68 (25.95) Non-Christian 80 (3.72) 7 (2.67) 7 (2.67) Political orientation Conservative 1,471 (72.21) 124 (49.80) 133 (52.67) Middle of the road 767 (37.65) 100 (40.16) 89 (33.97) Liberal 159 (7.81) 25 (10.04) 26 (9.92) Theological orientation Conservative 1,175 (56.03) 133 (50.76) 138 (52.67) Middle of the road 709 (33.81) 98 (37.40) 89 (33.97) Liberal 213 (10.16) 24 (9.16) 26 (9.92) et al. (1999) provide an in-depth discussion of steps that were taken to assure reliable and valid data. For example, questions were intentionally designed to ask about directly observable congregational characteristics (e.g., the presence or absence of specific programs or activities) rather than more intangible questions about congregational values, goals, or mission (e.g., questions about if the congregational values helping the poor) to decrease the likelihood of inaccurate reporting. See Chaves et al. (1999) for further information. Congregational Activities Key informants were asked to report if their congregation participated in or had groups related to six political activities. Responses were coded 0 (not present) or 1(present). The six political activities were the following: (a) marches (group to organize or participate in a demonstration or march either in support of or opposition to some public issue or policy); (b) lobbying (group to organize or participate in efforts to lobby elected officials of any sort); (c) political discussion (group to discuss politics); (d) political opportunities (people at worship services have been told of opportunities for political activity, including petition campaigns, lobbying, or demonstrating); (e) voter guides (voter guides [have] been distributed to people through your congregation); and (f) voter registration (groups for an effort to get people registered to vote). For social service activities, the key informant was asked, Has your congregation participated in or supported social service, community development, or neighborhood organizing projects of any sort within the past 12 months? If programs and activities were named, key informants were instructed to describe the programs in an open ended response. As described by Chaves (2004) and Chaves and Tsitsos (2001), researchers recorded these responses and then coded them into various types of congregational programs and activities. In the current study we focus on three social service activities regarding the following: (a) food (feeding the hungry), (b) clothing (clothing or blankets, including rummage sales), and (c) housing (home building, repair, or maintenance). Responses for each congregation were coded as 0 (not present) and 1 (present). For each social service program that was identified, key informants reported if they carried out these programs alone or in collaboration with any other groups or organizations, with the informant identifying the two most important collaborators for each program. Chaves and Tsitsos (2001) describes how each of these responses was coded to note the type of collaborating organization (e.g., non-profit), and an aggregated variable was created for

6 each congregation as to the types of organizations they collaborated with. In this study we focus on three types of collaborators: (a) other congregations (collaborated with other congregations on social service projects); (b) government (collaborated with government on social service projects); and (c) non-profit (collaborated with a non-profit on social service projects). For each collaborator, responses for each congregation were coded 0 (collaboration not present) or 1 (collaboration present). Thus, in this study collaboration is for social service programs and these items indicate whether congregations are partnering with specific types of community collaborators (congregations, secular, or government) in carrying out social services. All variables in the current study were present in both 1998 and 2007 and were collected using similar procedures. Frequencies of activities are displayed for each sample in Table 2. Congregational Demographics The same demographic variables were present in 1998 and 2007 and were collected via self-report from the key informant. Congregational size was assessed by asking participants How many persons would you say are associated in any way with the religious life of this congregation counting both adults and children, counting both regular and irregular participants, counting both official or registered members and also participating nonmembers. What is the total number of persons associated with this congregation to any degree at all? Because of the skewed distribution of responses, we used the natural log transformation of congregational size for analyses. Congregational age was assessed by the year the congregation was founded. Congregational education was assessed by the percent of the congregation with a four-year degree or more. Congregational racial composition was assessed as the percent of the adult participants in the congregation that were White (non-hispanic). The region in which the congregation was located was coded by researchers based on geographic state into four areas of (a) East Coast, (b) Midwest, (c) South, or (d) West Coast. For analysis these responses were dummy coded with South used as the reference group. Furthermore, the census tract of the congregation (at either the 1990 or 2000 census depending on the year of data collection) was used to classify the proximity of a congregation to a city as follows: (a) Urban, (b) Suburban, or (c) Rural. For analyses, these responses were dummy coded with Urban serving as the reference group. Congregational religious tradition was classified using Steensland et al. s (2000) categories of (a) Roman Catholic, (b) Evangelical Protestant, (c) Black Protestant, (d) Mainline Protestant, (e) Non-Christian. Responses were dummy coded with Evangelical Protestant serving as the reference group. Clergy gender was coded 0 (man) and Table 2 Frequencies of congregational activities Full dataset Panel data 98 Panel data 07 Variable Frequency (%) Frequency (%) Frequency (%) Food Present a 1,046 (54.88) 131 (64.53) 122 (46.56) Not present a 860 (45.12) 72 (35.47) 140 (53.44) Housing Present 686 (35.99) 95 (46.80) 104 (39.69) Not present 1,220 (64.01) 108 (53.20) 158 (60.31) Clothing Present 370 (19.41) 51 (25.12) 37 (14.12) Not present 1,536 (80.59) 152 (74.88) 225 (85.88) Marches Present 406 (18.95) 49 (18.77) 47 (17.94) Not present 1,737 (81.05) 212 (81.23) 215 (82.06) Lobbying Present 258 (12.03) 35 (13.41) 47 (17.94) Not present 1,887 (87.97) 226 (86.59) 215 (82.06) Political discussions Present 297 (13.83) 33 (12.60) 49 (18.85) Not present 1,850 (86.17) 229 (87.40) 211 (81.15) Political opportunities Present 683 (31.98) 96 (36.78) 93 (35.63) Not present 1,453 (68.02) 165 (63.22) 168 (64.37) Voter guides Present 538 (25.20) 71 (27.31) 77 (29.50) Not present 1,597 (74.80) 189 (72.69) 184 (70.50) Voter registration Present 440 (20.57) 33 (12.60) 68 (26.25) Not present 1,699 (79.43) 229 (87.40) 191 (73.75) Congregations collaborate Present 681 (33.12) 92 (40.89) 88 (33.59) Not present 1,375 (66.88) 133 (59.11) 174 (66.41) Government collaborate Present 268 (13.04) 31 (13.78) 28 (10.69) Not present 1,788 (86.96) 194 (86.22) 234 (89.31) Non-profit collaborate Present 617 (30.01) 88 (39.11) 80 (30.53) Not present 1,439 (69.99) 137 (60.89) 182 (69.47) a Present refers to the presence or absence of congregational activities, programs, or collaborations 1(woman). Clergy race was coded 0 (White) and 1 (clergy of color: comprised of Black/African American, Hispanic, Asian/Pacific Islander, and other). Theological conservatism was assessed by asking the respondent to identify the congregation as theologically (a) Conservative, (b) Middle of the road, or (c) Liberal. Responses were dummy coded with Conservative as the reference group. Political conservatism was assessed by asking the respondent to identify the congregation as politically (a) Conservative, (b) Middle

7 of the road, or (c) Liberal. Responses were dummy coded with Conservative as the reference group. The year that the congregation participated in the NCS was coded as 0 (1998) and 1 (2007). Frequencies for each sample are given in Table 1. Data Analysis We used latent class analysis (LCA) and latent transition analysis (LTA) to examine study hypotheses (Collins and Lanza 2010). We conducted LCA using the Full dataset to look for distinct classes (i.e., types) of congregations. LCA is similar to cluster analysis but uses categorical rather than continuous response items and is based on a latent model (Steinley and Brusco 2011). In LCA the classes are considered part of a latent (i.e., unobserved) variable that is measured indirectly with two or more categorical indicator variables (Collins and Lanza 2010). In this study, the 12 congregational program variables served as observed, categorical (present or absent) indicator variables. The purpose of the current LCA analysis was to find the unique latent classes (i.e., types) that best capture the response patterns in the data such that congregations responding in a similar manner to the 12 program items are grouped in the same class. In LCA each class is described by probabilities. For example, LCA provides the latent class prevalences which are the probabilities of membership in each latent class. A latent class prevalence of.25 would indicate that 25 % of congregations in the sample would fall into that class. In addition, within each class the item-response probability of answering yes to each observed variable is given. Itemresponse probabilities closer to one indicate a greater likelihood that congregations in the class endorsed the item, and probabilities closer to zero indicate lower likelihood of congregations in the class endorsing the yes response. Examination of these item-response probabilities shows what responses characterize the class. The closer the itemresponse probabilities are to zero or one, the greater the degree of homogeneity or the extent to which members of the latent class provide the same observed response. Comparing item-response probabilities among different classes shows how separated or different the classes are. Homogeneous and separated classes are desirable and facilitate interpretation (Collins and Lanza 2010). Plotting the itemresponse probabilities for each class is a useful way to understand the composition of each class and to make comparisons between classes. After determining the number of latent classes, we used multinomial logistic regression to test how covariates predicted latent class membership (Collins and Lanza 2010; Lanza et al. 2007). We examined odds ratios to understand the size of effect of a covariate on the probability of one class membership versus a reference class (Collins and Lanza 2010; Lanza et al. 2007). We used LTA with the Panel dataset to examine how the prevalence of latent class membership changed across time and the probability that congregations changed from one latent class in 1998 to another class in 2007 (Collins and Lanza 2010; Lanza and Collins 2008). In LTA, latent status prevalences (i.e., probability of membership in each latent class) and item-response probabilities (i.e., likelihood that congregations in a class endorsed an item) are interpreted in the same manner as in LCA and are used to confirm the number and interpretation of classes just like in LCA. However, in LTA we follow Collins and Lanza (2010) and now refer to a latent class as a latent status to indicate that membership in a class may be temporary, and that congregations may move statuses across time. Also, LTA introduces a new set of transition probabilities that show the patterns of change between latent statues (Collins and Lanza 2010). These transition probabilities show the likelihood that congregations starting in one status in 1998 moved to another status in This information is used to understand if and how congregations changed status membership. LCA and LTA analysis were conducted in SAS using PROC LCA (Lanza et al. 2007) and PROC LTA (Lanza and Collins 2008). Both procedures handle missing data using Full Information Maximum Likelihood which assumes data are missing at random (MAR; Schafer and Graham 2002); however, cases missing values on one or more covariates are removed from regression analysis (Collins and Lanza 2010). Additional details regarding the LCA and LTA analyses are provided throughout the results. LCA has been used in other community psychology research (Cammack et al. 2011; Connell et al. 2010; Sipsma et al. 2012) as has LTA (Lambert et al. 2010). Results How Many Classes of Congregations? We followed recommendations of Collins and Lanza (2010) to determine the number of latent classes. We first examined models with one through six latent classes and then compared multiple indices (i.e., G 2, AIC, BIC, ABIC, Loglikelihood) to determine the best-fitting model. Table 3 shows the number of parameters estimated, the G 2 statistic, the degrees of freedom, and fit statistics. P-values are not reported because the degrees of freedom are too large in each model (Collins and Lanza 2010). Although G 2 decreases as more latent classes were added, the BIC begins to increase and the other fit indices begin to level off after the four class solution. The four class solution provided relatively homogeneous and separated latent classes that yielded parsimonious interpretation. The four class solution also evidenced good identification as eight of ten random starts converged to the lowest log-likelihood value (Collins and Lanza 2010).

8 Table 3 Information for selecting number of congregational types (Full sample N = 2,153) Number of latent classes Number of parameters estimated G 2 * df AIC BIC ABIC Log-likelihood ,045 4,083 4,069 4,137 4,099-13, ,738 4,070 2,788 2,930 2,851-12, ,316 4,057 2,392 2,608 2,487-12, ,190 4,044 2,292 2,582 2,420-12, ,102 4,031 2,230 2,593 2,390-12, ,018 4,018 2,172 2,609 2,365-12,132 * p-values not reported because the degrees of freedom are too large Based on these considerations the four class solution was selected. Because data were collected from congregations in 1998 and a separate cohort of congregations in 2007, we followed procedures outlined by Collins and Lanza (2010) to examine measurement invariance to establish that the number and structure of latent classes were identical across years. Analyses (available from the authors) indicated the structure of the classes to be similar across years and thus we combined data from 1998 and 2007 into one sample of 2,153 congregations. Latent class prevalences and item-response probabilities are shown in Table 4 and plotted in Fig. 1 to aid in interpretation of each class. Characteristics of each latent class are now discussed in terms of homogeneity (i.e., the extent to which members of the latent class provide the same observed response) and separation (i.e., how the overall pattern of response probabilities differentiates the classes) as discussed by Collins and Lanza (2010). Class A: Active The Active class was characterized by high probabilities of responding yes to the social, political, and collaboration items. Even though there was a lower probability of responding yes to some items (e.g., clothing, government collaboration), these items had higher probabilities compared to those in other classes. Overall, the Active class was comprised of congregations providing various social services, active in multiple political dimensions (marching, announcing political opportunities, and voting activities) and collaborating with multiple community partners. The prevalence (i.e., probability of latent class membership) of membership in the Active class was.09, meaning that 9 % of congregations fell into this class. Class B: Not Active The Not Active class was characterized by a low probability of responding yes to the social, political, and collaboration items. This class was characterized by Table 4 Item response probabilities for congregational types (Full sample N = 2,153) Latent class Political Not Social Service Not Active congregations not providing social services, not being politically active, and not collaborating. The prevalence of membership in the Not Active class was 32 %. Class C: Social Service Not Political Active Latent class prevalences Item-response probabilities Food Clothing Housing Marches Lobbying Political discussion Political opportunities Voter guides Voter registration Congregations collaborate Government collaborate Non-profit collaborate Social Service Not Political Item-response probabilities[.5 in bold to facilitate interpretation. The probability of an absent response can be calculated by subtracting the item-response probabilities shown above from 1 The Social Service Not Political class was characterized by congregational activities focused on social services and collaboration but not political activity. The prevalence of membership in the Social Service Not Political class was 40 %.

9 Fig. 1 Congregational types based on social service, political, and collaborators using the Full sample N = 2,153 Proportion of Yes Response Social Service Political Congregational Activities Collaborators Active Not Active Social Service Not Political Political Not Social Service Class D: Political Not Social Service The Political Not Social Service class was characterized by activities focused on political activities involving voting (voter registration and voter guides) and announcements of political opportunities. Moderate endorsements were present for marches and providing food. Overall, this class was comprised of congregations focused on voting related political activities with little social service activities or collaboration. The prevalence of membership in the Political Not Social Service class was 19 %. Comparing Classes In general, the Active class was engaged in social, political, and collaborative activities whereas the Not Active class was not engaged in any of these activities. The Social Service Not Political class focused on social services with moderate collaboration but no political activities. The Political Not Social Service class had voting related political activities with little social service activities or collaborations. What Predicts Class Membership? We examined a multinomial logistic regression model with a latent class dependent variable to determine what congregational characteristics predicted specific class membership. Results are shown in Table 5 using odds ratios as an indication of the size of effect. Odds ratios show the change in odds of membership in one latent class compared to the reference class for a one-unit change in the covariate (Collins and Lanza 2010). For example, an odds ratio of 2.05 in predicting the Active versus Non Active class for Catholic versus Evangelical Protestant congregations would be interpreted that for Catholic congregations the odds of being in the Active latent class relative to the Non Active latent class was 2.05 times the corresponding odds for Evangelical Protestant congregations. This indicates Catholic congregations are about twice as likely as Evangelical congregations to be in the Active rather than the Non Active class. Congregations providing data on all covariates from the full sample were used for analysis with a resulting sample of 1,701 congregations. Visual inspection of the four class solution using this reduced sample confirmed a similar class structure as earlier analyses (available upon request from the authors). Congregational Size and Age Congregations with more participants were more likely to be in the Political Not Social Service, Active, and Social Service Not Political classes than the Not Active class. Also, larger congregations were likely to contain a political element as larger congregations were more likely to be in the Active or Political Not Social Service classes compared to the Social Service Not Political class. Older

10 Table 5 Predicting membership in congregational type Political Not Social Service Active Social Service Not Political Political Not Social Service Social Service Not Political Social Service Not Political Not Active Not Active Not Active Active Active Political Not Social Service Variable Odds ratio (95 % CI) Odds ratio (95 % CI) Odds ratio (95 % CI) Odds ratio (95 % CI) Odds ratio (95 % CI) Odds ratio (95 % CI) Intercept 0.02* (0.01, 0.05) 0.00* (0.00, 0.01) 0.09* (0.04, 0.20) 11.7* (3.01, 45.80) 57.2 (16.69, ) 4.87* (1.86, 12.77) Year 0.93 (0.73, 1.19) 0.87 (0.64, 1.18) 0.68* (0.55, 0.84) 1.07 (0.77, 1.50) 0.79 (0.59, 1.06) 0.73* (0.58, 0.93) Size 1.71* (1.52, 1.91) 1.85* (1.60, 2.15) 1.41* (1.28, 1.56) 0.92 (0.79, 1.07) 0.76* (0.66, 0.88) 0.83* (0.74, 0.92) Age 1.00 (0.99, 1.00) 1.00 (0.99 f, 1.01) 1.00* (1.00, 1.01) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) East Coast a 1.47* (1.03, 2.11) 1.15 (0.72, 1.83) 0.84 (0.60, 1.17) 1.28 (0.79, 2.08) 0.73 (0.47, 1.14) 0.57* (0.40, 0.81) Midwest a 1.29 (0.95, 1.76) 1.58* (1.06, 2.36) 1.11 (0.85, 1.44) 0.81 (0.53, 1.24) 0.70 (0.48, 1.02) 0.86 (0.64, 1.15) West Coast a 0.94 (0.66, 1.34) 1.21 (0.78, 1.86) 0.89 (0.66, 1.20) 0.78 (0.48, 1.24) 0.74 (0.49, 1.12) 0.96 (0.68, 1.35) Suburban b 0.74 (0.51, 1.06) 0.79 (0.49, 1.27) 0.74* (0.55, 0.99) 0.93 (0.55, 1.57) 0.93 (0.59, 1.47) 1.00 (0.70, 1.43) Rural b 0.82 (0.59, 1.15) 0.82 (0.50, 1.32) 0.74* (0.56, 0.97) 1.01 (0.60, 1.69) 0.91 (0.56, 1.46) 0.90 (0.64, 1.27) Percent White 1.00 (0.99, 1.00) 0.99 (0.99, 1.00) 1.01* (1.00, 1.01) 1.00 (1.00, 1.01) 1.01* (1.01, 1.02) 1.01* (1.00, 1.01) Education 1.00 (0.99 f, 1.01) 1.01* (1.01, 1.02) 1.01* (1.00, 1.01) 0.99* (0.98, 0.99) 0.99* (0.99, 0.99 f ) 1.01* (1.00, 1.01) Clergy gender g 1.69 (0.98, 2.92) 0.75 (0.37, 1.53) 1.24 (0.78, 1.97) 2.25* (1.06, 4.76) 1.65 (0.84, 3.22) 0.73 (0.44, 1.23) Clergy race h 1.26 (0.79, 2.02) 0.69 (0.39, 1.21) 0.75 (0.47, 1.17) 1.83* (1.03, 3.25) 1.08 (0.62, 1.87) 0.59* (0.37, 0.95) Catholic c 1.49* (1.01, 2.20) 2.05* (1.22, 3.47) 0.73 (0.50, 1.06) 0.72 (0.42, 1.24) 0.36* (0.22, 0.59) 0.49* (0.34, 0.71) Black protestant c 2.08* (1.18, 3.67) 1.48 (0.64, 3.40) 1.14 (0.63, 2.06) 1.40 (0.59, 3.33) 0.77 (0.33, 1.80) 0.55 (0.30, 1.01) Mainline protestant c 0.78 (0.52, 1.19) 1.79* (1.07, 2.99) 1.84* (1.36, 2.50) 0.44* (0.25, 0.78) 1.03 (0.63, 1.68) 2.35* (1.59, 3.47) Non-Christian c 0.43 (0.18, 1.01) 1.43 (0.65, 3.14) 0.63 (0.32, 1.24) 0.30* (0.12, 0.78) 0.44* (0.21, 0.94) 1.46 (0.63, 3.40) Politically MR d 0.96 (0.71, 1.29) 1.93* (1.32, 2.81) 1.48* (1.14, 1.91) 0.50* (0.34, 0.74) 0.77 (0.54, 1.10) 1.54* (1.16, 2.04) Politically Lib d 1.50 (0.82, 2.75) 5.00* (2.69, 9.28) 1.22 (0.71, 2.10) 0.30* (0.16, 0.57) 0.24* (0.14, 0.42) 0.81 (0.46, 1.42) Theologically MR e 0.60* (0.44, 0.81) 1.11 (0.75, 1.64) 0.96 (0.74, 1.26) 0.54* (0.36, 0.81) 0.87 (0.60, 1.26) 1.61* (1.20, 2.16) Theologically Lib e 0.99 (0.55, 1.76) 2.22* (1.23, 4.01) 1.17 (0.72, 1.90) 0.44* (0.24, 0.82) 0.53* (0.31, 0.90) 1.18 (0.70, 1.99) Log-likelihood =-9,700.94, N = 1,701. MR = Middle of the Road, Lib = Liberal Reference group is South. b Reference group Urban. c Reference group Evangelical Protestant. d Reference group Politically Conservative. e Reference group Theologically Conservative. f Number is not rounded to show the confidence does not interval include g 0 = man, 1 = woman. h 0 = White, 1 = person of color a

11 congregations were likely to be in the Social Service Not Political versus Not Active class. Thus, both the size and age of the congregation predicted class membership. Geographic and Neighborhood Location The geographic location in the US of the congregation also predicated class membership. East Coast congregations were more likely than Southern congregations to be Political Not Social Service rather than Not Active or Social Service Not Political. Midwest congregations were more likely than Southern congregations to be in the Active versus Not Active Class. There were no differences between West Coast and Southern congregations. Regarding neighborhood location, congregations located in Urban areas were more likely than Suburban or Rural congregations to be in the Social Service Not Political versus Not Active class. Congregational Demographics Congregations with a higher percentage of White congregants were more likely in the Social Service Not Political class versus the Not Active, Active, or Political Not Social Service classes. Congregations with higher average education were more likely to be in the Active than the other three classes. Higher education also predicated membership in the Social Service Not Political rather than Not Active or Political Not Social Service classes. Congregations with clergy who were women or people of color were more likely to be in the Political Not Social Service versus Active class, and congregations with a clergy who was a woman were more likely to be in the Political Not Social Service versus Social Service Not Political class. Congregational Tradition Compared to Evangelical Protestant congregations, Catholic congregations were more likely to be in the Political Not Social Service and Active rather than Not Active classes and also more likely to be in the Active or Political Not Social Service classes rather than Social Service Not Political. Black Protestant congregations were more likely than Evangelical congregations to be Political Not Social Service rather than Not Active. Mainline Protestant congregations were more likely than Evangelical congregations to be Active rather than Not Active or Political Not Social Service and more likely to be Social Service Not Political rather than Not Active or Political Not Social Service. Non-Christian congregations were more likely than Evangelical congregations to be Active than Political Not Social Service or Social Service Not Political. Congregational Political and Theological Conservatism Political Middle of the Road and Liberal congregations were more likely than politically Conservative congregations to be in the Active versus Not Active or Political Not Social Service classes. Politically Middle of the Road congregations were more likely than Conservative congregations to be in the Social Service Not Political versus Not Active or Political Not Social Service classes. Politically Liberal congregations were more likely than Conservative congregations to be in the Active versus Social Service Not Political class. Theologically Middle of the Road congregations were more likely than theologically Conservative congregations to be in the Not Active or Social Service Not Political versus Political Not Social Service class. Both theologically Middle of the Road and Liberal congregations were more likely than theologically Conservative congregations to be in the Active rather than Political Not Social Service class. Finally, theologically Liberal were more likely than theologically Conservative congregations to be in the Active rather than Not Active or Social Service Not Political classes. How Do Congregations Transition Across Time? To examine if and how congregations transitioned between classes from 1998 to 2007, we used latent transition analysis (Collins and Lanza 2010) with the panel data set (N = 262 congregations). Following recommendations of Collins and Lanza (2010), we first replicated the four class solution by examining models with two through six latent classes and constrained the item-response probabilities to be equal across time (analyses available upon request). Thus, the same four class solution is now used in the latent transition analysis, but we follow the nomenclature of Collins and Lanza and call the classes statuses to indicate that congregations may transition from one status to another across time. As shown in Table 6, latent status prevalences remained similar for the Not Active and Social Service Not Political statuses. However, prevalence of the Political Not Social Service status increased from 1998 to 2007 whereas the prevalence of the Active status decreased. The transition probabilities showed that congregations in the Political Not Social Service status in 1998 were likely to maintain membership in this status in Congregations in the Not Active status in 1998 were likely to stay in that status, or to shift to the Political Not Social Service status. Congregations in the Active status in 1998 were not likely to stay in this status, but instead had about equal likelihood of making a transition to the Political Not Social Service or Social Service Not Political statuses. Congregations in the Social Service Not Political status in 1998 were moderately