Introduction

Participation in community activities is commonly held to be both an individual and social good. For example, we know that psychological benefits, such as empowerment and sense of community, disproportionately accrue to those who actively participate in community activities, including community organizing (Florin and Wandersman 1990; Chavis and Wandersman 1990; Hughey and Speer 2002; Speer and Hughey 1995). Research also points to benefits incurred by individuals from participation and civic engagement, including increases in tolerance and empathy (Mendelberg 2002), trust (Whiteley 1999; Stolle et al. 2005), skill development (Prestby et al. 1990), individual well-being (Wollebæk and Selle 2002) and social learning (Florin and Wandersman 1984).

At an aggregate level, participation and engagement have been linked to functioning civil societies and other collective benefits (Boyte 2003; Putnam et al. 1993). We know from research that community organizing is one of the most effective vehicles for citizens seeking to build social capital and make change at the local level (Warren 2001; Swarts 2008). As an approach that spans community betterment and social change (Maton 2008), community organizing has been particularly successful at meaningfully engaging ordinary people in local decision-making processes and achieving local change (Speer et al. 2003; Brooks 2001; Pilisuk et al. 1996).

We do not understand, however, very much about what causes some people to stay involved in activities such as community organizing over time, while others fall into inactivity. The number of community organizing groups has grown over recent decades (Swarts 2008), and organizing groups now engage a small but significant percentage of the publicFootnote 1 (Wood and Warren 2002). Research on differences in overall levels of community participation has focused primarily on demographics and individual resources (e.g., Verba et al. 1995; Musick et al. 2000). Community, organizational and other contextual factors contributing to differences in voluntary activities have been highlighted as promising but underdeveloped areas of inquiry (Wilson 2000).

Previous research that has gone beyond demographic understandings of community participation can be classified into five branches. The first branch examines differences in beliefs and personality variables as predictors of community participation (Cohen et al. 2001; Bekkers 2005). The second branch of this research is focused on psychosocial variables in relation to community participation, including psychological sense of community (Ohmer 2007; Peterson et al. 2008) and empowerment (Speer and Hughey 1995). The third branch examines community and organizational contexts such as leadership and collective efficacy (Foster-Fishman et al. 2007) and social ties (McAdam and Paulsen 1993) as predictors of organizational involvement and community participation. The fourth branch of this research focuses on characteristics of human and physical geography such as economic heterogeneity (Oliver 1999), residential stability, defensible space (Perkins et al. 1996) and other physical environmental characteristics (Perkins et al. 1990). Finally, the fifth branch examines macro-structural characteristics such as the intensity and scope of the voluntary sectors of different nations (Wollebæk and Strømsnes 2008) and historical social changes (Putnam 1995).

Social change organizations, including grassroots community organizing groups, depend on the sustained involvement of their membership (Chetkovich and Kunreuther 2006; Mondros and Wilson 1994). Much of the effectiveness of community organizing initiatives, in particular, stems from these initiatives’ ability to consistently draw large groups of people together. The long-term impacts of organizing on neighborhoods and cities can be difficult to identify and disentangle from a data-analytic perspective (Speer et al. 2010). Participation in organizing is a more immediate outcome from the process, and can be understood as a proximal outcome for organizational power, which, when effective, then leads to neighborhood and metropolitan change (Speer 2008).

Few studies make use of actual documented participatory activities, over time, in specific organizations. Previous research has frequently used self-reports or organizational reports of membership and participation (e.g., Putnam 1995; Perkins et al. 1996). This study, in contrast, uses longitudinal data on uniquely identified individual participants through the use of sign-in sheets at group meetings. As forms of civic participation, community organizing activities can be broadly distinguished from other nonprofit and voluntary groups by an emphasis on grassroots process (Robinson and Hanna 1994; Staples 2004), the exercise of social power (Speer et al. 2003; Christens et al. 2008) and by their pursuit of social transformation (Maton 2000). More specifically, Wood and Warren (2002) identify six distinguishing characteristics of faith-based community organizing groups: they are faith based, broad based, locally constituted, multi-issue, professionally staffed, and political but nonpartisan. This study reports results from a longitudinal study of congregation-based community organizing in which individual participation patterns were tracked for five years. Each act of participation was recorded through the use of meeting sign-in sheets. Context is taken into account analytically through the use of three-level longitudinal growth modeling (Singer and Willett 2003; Luke 2004).

Specifically, this study addresses whether differences in participatory contexts, such as neighborhoods, networks, and settings, can account for divergent outcomes in attrition or sustained participation in organizing. Beyond assessing the relative influence of different contextual factors on sustained participation, this study examines a further question, with a subsample from this study that includes individual demographics. This allows a comparison of the relative impacts of context and behavior settings and the more commonly investigated individual characteristics.

Methods

Research Design

The data were gathered as part of a multi-site quasi-experimental study of local community organizing conducted over 5 years. When individual participants attended group meetings related to community organizing, they indicated their attendance on a sign-in sheet. The sign-in sheets were copied and mailed to the research team, who then entered the information into a database in which every individual participant was uniquely identified. In addition, the staff of the organizations periodically logged into an online database to record their one-to-one meetings with individual participants. These data on one-to-one meetings were joined with the data on individual participants. Two waves of telephone survey data were also collected for a subset of individual participants (wave one at year 2.5, wave two at year 5), along with a geographically balanced random sample of non-participating residents in each city. For the results reported in this paper, the survey data shed light on the relative impact of individual demographic characteristics on participation.

Setting and Sampling

There were 3,435 group meetings held over 5 years by 115 congregations, which were part of five local organizing initiatives in five cities. All five initiatives were affiliated with the PICO National Network (PICO stands for People Improving Communities through Organizing), and were thus employing a similar organizing model. The local initiatives were selected to represent both regional and urban/rural diversity: one research site was in a major metropolitan area in the eastern US, one was in a mid-sized city in the eastern US, two were in mid-sized cities in the Midwest, and one was in a rural/suburban area in the western US. The organizing in these initiatives was focused on improvements in community quality-of-life issues including violence prevention, education, neighborhood development, affordable housing, access to health care, and the prevention of predatory lending.

In total, 11,528 individuals participated a total of 40,304 times in group meetings. Additionally, there were 15,043 one-to-one meetings reported between staff and participants. Thus, overall involvement over 5 years, including participation in one-to-ones, was 55,347. These numbers are likely to be slightly understated, due to individuals failing to sign into meetings. However, the participation that is recorded in the data represents the entire universe of participants who signed into meetings, even once, in all 115 congregations involved in the local organizing initiatives.

Participants

Although individual-level demographic characteristics were not gathered for the entire sample, the sub-sample that participated in the survey (n = 461; response rate 47%) allows comparisons between participants in organizing and their randomly selected neighbors. When compared to a random sample of residents, the participants in the community organizing initiative were, on average, slightly older, had completed more formal education, and reported lower levels of family income. Among the 11,528 people who participated in organizing meetings, the mean number of attendances was 3.4 (SD = 12.3; max = 514). When those attendances are aggregated into annual counts of attendance, the mean annual attendance count is .68 (SD = 3.2; max = 182).

Measures

The criterion variable for this study is participation in community organizing group meetings. Aspects of the participatory contexts were used as predictors. Specifically, the aspects of context included in this study are the compositional characteristics of the neighborhoods in which participation is occurring (neighborhood compositional characteristics), differences in the involvement of participants’ networks of affiliation (affiliation network engagement), and the types of meetings that participants are attending (meeting types or ‘setting phenotypes’).

Participation in Community Organizing

Data on individual participation in group meetings were aggregated into a count variable for each year—which was then transformed into an binary indicator of participation/nonparticipation in group meetings. Although the use of the binary variable does not take full advantage of the availability of information on the depth of individual participation, it also has both substantive and statistical advantages. A substantive advantage is that the model fit is not disproportionately influenced by the small minority of heavily engaged participants. A statistical advantage is that it resolves the issue of the non-normality in the distribution of participation due to the presence of a large number of zeros. Employment of a binary transformation of the dependent variable requires the use of a nonlinear probability model, which makes no assumptions about the normality of the distribution of the variable.

Neighborhood Compositional Characteristics

Previous research has highlighted the importance of neighborhood characteristics for community participation (Sampson et al. 1997; Perkins et al. 1990). At the neighborhood level, this study investigates the effects of median income (Duncan et al. 2003; Perkins et al. 1996), economic heterogeneity (Oliver 1999; Swaroop and Morenoff 2006), and residential stability (Perkins and Long 2002; Irwin et al. 1999) using census data (STF3, Tables P24, P52, H6 and H7, US Census Bureau 2000). Economic heterogeneity is measured using an index of qualitative variation, equivalent to the likelihood that any two households within the selected geography will fall into different income categories, as opposed to the same category. The geographic unit of analysis for the neighborhood is the census tract. All neighborhood measures were converted to discrete scores (quintiles) for parsimony and to avoid over-specification of the level three models.

Affiliation Network Engagement

As people participate in meetings, they interact with others, forming networks of affiliation, or ‘two-mode networks’ (Hanneman and Riddle 2005). Scrutinizing these networks allows an understanding of interdependence (Kelly 1966). From a data analysis standpoint, individual participants are connected to events, and then connected through the event to other individuals. These individuals, in turn, are tied to still other events. Rather than an inquiry into the structure of the networks of affiliation formed as individuals attend events, the present study uses information on the levels of attendance of affiliated individuals over time. Accordingly, affiliation network engagement represents the level of involvement of the other attendees at meetings that an individual attends during the year in which the attendance occurs, i.e., the organizing engagement of an individual’s network of affiliation.

Group Meeting Types (or ‘Setting Phenotypes’)

As parts of the community organizing process, different group meeting types (e.g., action meetings or research action meetings) have shared characteristics and differences (Speer et al. 1995). Each type of meeting creates different social climates (Trickett and Moos 1973) or characteristics of a setting. The characteristics shared across all organizing activities can be understood as the setting’s genotype, whereas various types of meetings within the organizing process have characteristics that vary according to the setting phenotype (Luke et al. 1991). Different phenotypes—action, research, planning, etc.—provide differing sets of expectations and availability of differing sets of roles. They also create different understandings of the organizing process, and, hypothetically, differences in future participation patterns.

The most publicly visible, and largest type of group meeting held by organizing groups are action meetings. Multiple research meetings—known as research actions—are held leading up to an action. These meetings involve leaders and key members of institutions with knowledge on the topic that the group is pursuing, and involve gathering information pertinent to the community issues around which the organizing is taking place, and, in particular, a search for tensions or contradictions between stated goals and applied practices. For example, directors of local government agencies will often be invited to a research action, where each organizing participant will pose a question that has been developed by the group in advance. Planning meetings are also held leading up to many actions and research meetings, and after such meetings for evaluation/reflection purposes. Other types of internal meetings may be held, and participants from the organizing network may attend meetings held by other entities (i.e., city council meetings). Frequencies of group meeting types and average attendance levels across all organizing initiatives in the study were: 240 local organizing committee meetings per year with a mean of 10.3 participants (SD = 6.8); 132 planning meetings per year with a mean of 6.6 participants (SD = 5.6), 82 research action meetings per year with a mean of 8.1 participants (SD = 8), 12 action meetings per year with a mean of 115 participants (SD = 105), 88 federated meetings per year with a mean of 9 participants (SD = 10.3), and 131 other meeting types per year with a mean of 14.46 (SD = 34).

One-to-One Meetings

One-to-ones are face-to-face meetings between participants in the organizing process that last approximately 30 min. Based on understandings from qualitative research, the number of one-to-one meetings that an individual has in the current year is also hypothesized to be positively associated with greater likelihood of future participation in organizing (Christens 2010). One-to-one meetings are emphasized by professional staff as an important tool for leadership development within the organizing process since the PICO model of community organizing stresses that grassroots leadership is developed through interpersonal relationships (Keddy 2001). Among participants in group meetings, the mean number of one-to-ones was .88 (SD = 5.94) over 5 years.

Individual Demographics

The full growth model for the participants in the survey subsample (Model F; N = 461) incorporates demographic variables at the individual level (level-2), which are often associated with civic engagement and participation. Income is measured in an ordinal scale composed of seven categories of annual family income. The mean response for participants in the survey falls into the category of $35,000 to $45,000 (SD = 2.2 categories). Education was measured on an ordinal scale composed of five categories. The mean response falls between some college and a completed college degree (SD = 1.2 categories). Gender is represented by a binary variable, and two-thirds of respondents identified as women.

Analytic Approach

Multilevel modeling is an analytic approach suited to the task of capturing the effects of context (Luke 2005). The current study uses a longitudinal analysis so that level-1 captures the effect of time, and permits the use of time-varying covariates. For instance, the number of one-to-one meetings held with a participant in a given year and the control variable that accounts for the prior year’s participation fluctuate each year for each participant. The time-varying level-1 models are nested within level-2 models, in which each participant is a unit of analysis. These models are nested within level-3 models, in which each group is a unit of analysis. Contextual variables such as neighborhood compositional characteristics are modeled at level-3. Without a multilevel approach, it would be difficult to account for shared contextual effects, or the effects of time.

Results

Preliminary Analyses

Among the 11,528 people who were involved in the organizing initiatives, the mean number of times individuals participated in group meetings was 3.4 (SD = 12.3; max = 514) over 5 years. When those attendances are aggregated annually, the mean annual attendance is .68 (SD = 3.2; max = 182). These figures demonstrate the skewed distribution of participation in group meetings. A few people participate very often; a large number of people participate very infrequently. This is understandable when considering the population of participants, which includes staff and leaders, as well as people only attending a single group meeting during the five-year timeframe. Of course, it is not possible, using only the current data, to determine whether these people attending only once have prior records of participation, or whether they have permanently ceased to participate in organizing.

7,833 out of 11,528 (nearly 68%) of participants who attended one meeting attended only once—meaning that they did not return to a group meeting within the timeframe of this study. There are at least two caveats with the use of this number as a generalization for the community organizing process. First, many of these individuals attended a group meeting for the first time during the fifth year of the study, and there may not have been enough follow-up time to detect their continued participation. Second, many of those who did return to a group meeting during the study’s timeframe may have been active for long periods of time preceding the start of the study, which would inflate the appearance of sustained participation within that group of individuals.

To better understand the patterns of individual participation in group meetings, it is helpful to examine a subset of the data that is less affected by these two caveats. In an attempt to remove the effect of highly involved individual participants, the subset is composed only of the people who did not participate in the first or second years, but participated during the third year (removing the attendees from the first 2 years drops 5,312 individuals or 46%—and removing those who participated for the first time after year three drops 4,160 individuals, or 36%). Looking only at the remaining 18% of people who participated in group meetings during the third year for the first time during the study’s timeframe (n = 2,058), the mean number of attendances is 2.4 (SD = 5.3). From this subset of first time attendees in year three of the study, 1,383 people participated in group meetings only once during the remaining 3 years—67.2%. In combination with the figures for the entire data set, this provides support for the conclusion that around two-thirds of first time attendees in community organizing group meetings never return after participating once (Table 1).

Table 1 Participation frequencies for first time attendees in year three (N = 2,058)
Full size table

Longitudinal Growth Models

Models were first fit to the data for the entire population of individuals (n = 11,528) (Models A, B, and C). Then, a similar set of models were fit to the data for the survey subsample (n = 461), (Models D, E, and F) so that individual demographic information could be understood relative to other independent variables. Before fitting the models with substantive predictors, unconditional means models and unconditional growth models were fit to the data. The unconditional means models (A and D) do not take time into account. The unconditional growth models (B and E) take time into account, but include no substantive predictors. Both sets of unconditional models establish baselines, which then allow growth models with substantive predictors to be compared to these baselines (C and F).

All the growth models are probability models that use a log transformation of the binary dependent variable. This transformation of the level-one outcome variable avoids the assumption of normality in the distribution. The logit link function for the level-1 dependent variable is given by:

$$ \eta_{ijk} = {\text{Log}}[\varphi_{ijk} /1 - \varphi_{ijk} ]; $$
$$ \varphi_{ijk} = {\text{Prob}}(Y_{ijk} = 1|\pi_{jk} ) $$

The models are built using an extension of the multilevel model (Goldstein 1995)—the generalized hierarchical linear model (GHLM; Hox 1995; Luke 2004) for longitudinal data. Estimation of all models uses full Penalized Quasi-Likelihood (PQL) (Breslow and Clayton 1993) in the HGLM function of the software program HLM3 version 6 (Raudenbush et al. 2004). The composite specification for the full conditional model (Model F) is given by:

$$ \begin{aligned} \eta_{ijk} = & {\text{logit(}}Y_{ijk} )= \gamma_{000} + \gamma_{001} (resid.\;stability_{k} ) + \gamma_{002} (median\;income_{k} ) \\ & + \gamma_{003} (heterogeneity_{k} )+ \gamma_{010} (income_{jk} )+ \gamma_{020} (gender_{jk} )\\ & + \gamma_{030} (education_{jk} )+ \gamma_{100} (year_{ijk} )+ \gamma_{200} (attendance_{ijk} )\\ & + \gamma_{300} (one{ - }to{ - }ones_{ijk} )+ \gamma_{400} (network_{ijk} )+ \gamma_{500} (research_{ijk} )\\ & + \gamma_{600} (action_{ijk} )+ r_{0jk} + u_{00k} \\ \end{aligned} $$

A comparison of model fit between Model C, and the baseline unconditional model, Model A, supports the rejection of the null hypothesis, indicating that Model C provides a superior fit to the data (χ 2 = 1,788; df = 9; p < .001) (Table 2).

Table 2 Parameter estimates for full sample (models A, B & C)
Full size table

Of the substantive predictors, prior participation in group meetings has the greatest effect, though its theoretical importance is limited—it is included in the model as a control. The number of one-to-one meetings in which an individual takes part in the current time-period, and their attendance at one or more research actions in the present time-period are significantly positively predictive of future participation in group meetings. Conversely, attendance at one or more actions in the current time-period is significantly negatively predictive of future participation in group meetings. Controlling for all other predictors in the model, the variable for affiliation network engagement—which measures average level of outside attendance of co-attendees—is not statistically significant. Neither are the neighborhood-level (level-3) independent variables measuring neighborhood stability, median income, and income heterogeneity at the census tract level.

To test the effects of individual demographic characteristics, a second set of models (D, E, and F) were specified for the subset of participants who responded to the telephone survey, providing demographic information. Participants in the survey sub-sample participated in group meetings an average of 14 times over 5 years (SD = 26.1; max = 230 attendances). The mean of the indicator for participation/nonparticipation in each year is .48—meaning that participation is slightly higher for the survey subsample than for the equivalent figure for the entire sample of participants. The mean number of one-to-ones held by participants in the survey sub-sample was 3.95 (SD = 10.06; max = 77 one-to-ones) over 5 years, with a mean of .98 one-to-ones per individual per year.

As in the case of the models for the full sample, the comparison of model fit between Model F, and the baseline unconditional model, Model D, supports the rejection of the null hypothesis, indicating that Model F provides a superior fit to the data (χ 2 = 474; df = 13; p < .001) (Table 3). Notably, the demographic characteristics of individual participants are non-significant as predictors of future participation in group meetings, controlling for other independent variables in the model. Attendance at action meetings is non-significant for the survey sub-sample, as opposed to the negative effect of attending action meetings found for the entire sample. Other differences in the models for the survey sub-sample include that a participants’ affiliation network engagement is a significant predictor of future participation in group meetings. Also, neighborhood median income is found to be negatively predictive of participation at level-3.

Table 3 Parameter estimates for survey sub-sample (Models D, E & F)
Full size table

Figure 1 displays the model-based (Model F) predicted probabilities of participation in group meetings during the next year by those who do not attend a research action in the current year, and those who attended a research action in the current year. Since the majority of participants do not return to organizing activities, the model predicts lower likelihood of participation over time for all participants. However, participants in research actions in year one have a higher predicted probability of future participation in group meetings, with the remaining variables in the three-level model held constant. The model-based display reinforces the potential utility of the findings for practice in grassroots community organizing. A 28% difference in predicted probabilities of future attendance is relatively large for variation in a single independent variable, and the effect endures over time. Similar model-based displays have been produced for the other independent variables with significant effects on future participation. These displays have been shared and discussed with community organizers and grassroots leaders as part of an ongoing cycle of action research.

Fig. 1
figure 1

Predicted probabilities of participation over time by attendance of research action in year one

Full size image

Conclusions and Discussion

The findings from this study point to the importance of context—particularly participatory settings and the development of interpersonal relationships—for sustaining participation in community organizing. An alternative interpretation of these findings would be that individuals with certain proclivities or dispositions are selected or self-select into certain types of activities. While the design of this study does not permit us to rule out such an interpretation, other longitudinal studies have found that behavioral change precedes psychological change rather than vice versa (Christens et al. in press), thus supporting the idea that socialization through participation in different settings is a plausible causal mechanism for the effects observed in this study. Broadly, these findings have implications for: 1. Practice in grassroots community organizing efforts; 2. Practice in voluntary action, more broadly and; 3. Theory on participation and engagement in civil society.

Implications for Practice in Grassroots Community Organizing

A descriptive finding of particular interest in this study concerns the attrition rates in community organizing. First-time attendees in year three of the study were used to select only new attendees who had not attended in the first 2 years. Amongst this subset of the sample, only a third of first-time attendees ever returned to a second meeting during the next 3 years. The multilevel longitudinal modeling of participation in group meetings provides insights into the ways that grassroots community organizing efforts might be more intentional in involving participants. For example, findings highlight the importance of face-to-face meetings, which are heavily emphasized in the training of leaders in community organizing, particularly in relational models of community organizing (Warren 1998). Findings from this study provide the first empirical support for the premise that these meetings designed to build interpersonal relationships are significant predictors of future attendance at group meetings, while controlling for other variables, such as overall prior attendance rate in the previous year. Face-to-face meetings in relational organizing models are thus reinforced as tools for maintaining and deepening participation among members, while controlling for other variables at the event, individual, and affiliate levels.

Attendance at research actions are also found to be of particular influence among the meeting types in predicting probability of future participation in group meetings. Unlike attendance at action meetings, which came out as a significantly negative predictor (in Model C) or as a non-significant predictor (in Model F), attendance at a research action increases the predicted probability of future attendance. A hypothesized mechanism for this effect is drawn from behavior setting theory (Barker and Gump 1964), which posits that characteristics of settings impact the behaviors of individuals in those settings (Altman and Rogoff 1987). Research action meetings involve participants directly in action-oriented analyses of community issues and conversations with key community leaders. They are also relatively small meeting types in which many participants play a role. Thus, the hypothesized mechanism for the effect of participating in research actions on future participation in group meetings is the opportunity role structure (Maton and Salem 1995; Peterson and Hughey 2002) of the setting. These findings extend behavior setting theory by pointing to the lasting impact of setting differences, which significantly impact individual attendance at group meetings in subsequent years.

Implications for Practice in Voluntary Action

Studies of mobilization and social movements have often retrospectively examined the historical circumstances that have allowed movements to flourish and succeed (Meyer and Corrigall-Brown 2005). An emerging focus is on nonprofit social change organizations and the practices they can employ to build not only toward short-term instrumental gains, but toward long-term sustainability and potential for participation in social movements (McCarthy and Walker 2004; Chetkovich and Kunreuther 2006). Sustaining volunteer participation in efforts for locally focused social change is a key element in this process, and implications from this study may have relevance to other mobilization and voluntary action efforts.

Specifically, a finding that is particularly relevant (and perhaps surprising) is that large action meetings are negatively predictive of future participation in community organizing. If this finding holds true across different types of voluntary action organizations, it means that leaders should evaluate their processes to ensure that volunteers, particularly those new to the organization, are attending smaller and more participatory meetings alongside larger meeting types (which may include large fundraisers, rallies, or demonstrations). While these types of meetings are important in most models of voluntary action, current findings suggest that they should be balanced with other smaller meeting types that involve volunteers in first-hand research on issues and interpersonal relationship development.

The relational model of organizing employed by the organizations in this study is explicit in its emphasis on human development and leadership development (Keddy 2001). This study employed longitudinal research methods and was therefore able to reveal the role that different parts of the process play in the change and continuity of activity within the organization (Lorion 1990). The results suggest that the components of the community organizing process that are focused on relationship development and the development of the skills and perspectives of members play an important role in sustaining member participation over time. This provides support for developmental models employed by social change organizations (Chetkovich and Kunreuther 2006; Polletta 2002; Foster-Fishman et al. 2006), and for developmental approaches to research on community processes (Kieffer 1984).

Implications for Theory on Participation and Engagement in Civil Society

The findings from this study challenge the prevailing wisdom on the importance of demographic characteristics at the individual and neighborhood levels for civic engagement and the development of social capital (Verba et al. 1995; Musick et al. 2000). More important, findings show, are the contexts of individual participation, relationships between individuals, and group-level processes. These findings have implications not just for the study of community organizing, but for applied work across the voluntary sector, and for theory on civil society and participatory democracy. Moreover, these findings are directly contrary to recent findings that attribute social capital formation to institutionalization in the voluntary sector rather than face-to-face contact (Wollebæk and Strømsnes 2008).Footnote 2

The portions of the community organizing processes that were found to be positively predictive of future participation in group meetings were those in which (a) participants were spending time face-to-face and developing interpersonal relationships with other participants, and (b) participants were interactively conducting inquiries on local community systems. These findings suggest the importance of attention to relationships and collaborative inquiry for sustaining and deepening participation in community action processes. The development of relationships across generations or socioeconomic and other demographic difference facilitates the development of social trust (Flanagan 2003). Collaborative inquiry processes like those studied here are supportive of reflective learning processes, or what some have called reflection-in-action (Schön 1984) or simply education (Dewey 1916). It seems, then, that the portions of organizing processes that we would expect to be most supportive of human development are the same ones that are most advantageous from an organizational or community development perspective, adding to the body of evidence supporting the claim that human and community development processes are linked inextricably (Christens et al. 2007). With these linkages in mind, grassroots organizations and systems change efforts should carefully consider their processes and become more intentional about enhancing development among their participants.

Limitations and Future Research Directions

This study found that some of the most frequent predictors of variance in participation levels, individual-level demographic characteristics, were non-significant as predictors when contextual variables were added to the model. One interpretation from this finding is that more studies should explore the relationship between context and participation. However, the larger context for this study, grassroots community organizing, is noted as a particularly effective strategy for engaging people across lines of race/ethnicity, education, social class, and religion. Findings from this study concerning demographics should therefore be explored in other contexts. A second limitation of this study is that findings differed between the full sample and the survey subsample. It is quite possible that the survey respondents’ characteristics differed from those of the full sample. Future research should therefore seek to obtain demographic data for all participants.

A third limitation is that although it involved a large sample of individual participants, the sample size of organizing initiatives was not large enough for meaningful statistical analysis. The multilevel models demonstrate that there is significant shared group variance at level-3, but the current study does not explain this variance, other than to show that neighborhood compositional characteristics do not have a meaningful impact. It is likely that characteristics of the organizations and their processes (e.g., leadership, issue selection) would help to explain shared variance in participation at the congregational and initiative levels, and future research should investigate these issues using a larger sample of organizations.

Finally, a fourth limitation concerns alternative plausible causal interpretations of the findings. As noted, a recent study found that participation preceded psychological empowerment (Christens et al. in press), supporting the notion that participants are socialized into beliefs and patterns of behavior through participation in one-to-ones and certain group meeting types rather than selected through a combination of predispositions, preferences and requests from others to participate. Nevertheless, both mechanisms remain plausible and future examination of these mechanisms or, alternatively, a process of reciprocal causality, are warranted. Future studies should test not only unidirectional and bi-directional temporal relationships to better understand causality; they should explore such relationships in voluntary processes beyond congregation-based community organizing. Additionally, future studies should compare different ways of measuring community participation (e.g., comparing self-report to organizational records of participation). This study both answers and echoes the calls for increased attention to context in community-based research, and continued theoretical and methodological work to enhance our ability to understand context (e.g., Shinn 1990; Luke 2005; Christens and Perkins 2008).