Community Meetings As a Predictor of Collective Action: Applying Propensity Score Matching to Strengthen Causal Inference

Background:  Community-based organizations often consider resident engagement in community meetings and events to be stepping stones to increased levels of participation that culminates in collective action by neighbors to facilitate a positive community change.  One of the challenges of testing this hypothesis is that individuals who attend community meetings may not be similar to individuals who do not attend these meetings on a wide range of socio-demographic variables.  These differences make it difficult to draw causal inferences about the effect of attending community meetings on future collective action activities.  This study utilizes propensity score matching to address this issue allowing for the ability to make a stronger causal statement about the role of community meetings in facilitating collective action. 

Method: This study draws on the United States Census Bureau’s Current Population Study for September 2015.  Propensity score matching with the MatchIt package in R was used to match individuals who stated that they had attended a community meeting within the previous twelve months with those who had not on the socio-demographic variables of volunteering, family income, number of hours per week the respondent was engaged in paid employment, whether or not the respondent owned a business, sex of the respondent, marital status of the respondent, number of children living in the household, educational attainment of the respondent, self-reported race, and whether or not the respondent identified as Hispanic.  Nearest neighbor matching with a caliper of 0.10 resulted in a final matched sample of 9,534.  Next, a logistic regression was used to predict the effect of attending a community meeting on working with neighbors to fix or improve something (i.e., collective action), while controlling for all covariates included in the propensity score calculation.  Post hoc leave-one-out sensitivity analysis was completed to assess the degree to which individual variables confounded the relationship between attending community meetings and collective action as well as to assesses the degree to which unmeasured confounding may influence the relationship.

Results:  Individuals who had attended a community meeting were fourteen and a half times more likely to engage in collective action as compared to their matched counterparts who had not attended a community meeting (OR=14.46, 95% CI: 12.59, 16.61).  That is more than seventeen percent lower than the odds ratio for a similar logistic regression without propensity score matching (OR=17.43, 95% CI: 15.88, 19.14).  Sensitivity analysis suggests that if unmeasured confounding is on the order of magnitude of the variables included in the model, little additional confounding is present.

Conclusion and Implications:  While individuals who attend community meetings are more likely to engage in collective action, the use of a standard logistic regression to analyze this relationship is potentially biased by a wide range of socio-demographic confounders.  The use of propensity score matching allows for a more accurate estimate of this relationship and strengthens the ability to make a causal inference based on the association.  Researchers may be drawing overly broad conclusions about this association based on standard regression methods, while this study corrects for these methodological shortcomings.