Association for Public Policy Analysis & Management

We explore the existence of social interactions in program participation within small neighborhood networks. Our population of interest is pregnant women in California and their participation in Medicaid during pregnancy. Causal estimation of social interactions faces a number of threats to identification with the leading concern being endogenous sorting into networks, represented by a census block in our analysis. To deal with this we employ the empirical methodology developed by Bayer et al. (2008). Each distinct observation in our estimation sample is a matched pair of a currently pregnant mother in a broad neighborhood, defined as an agglomeration of nearby census blocks or a census block group, and a recent mother who has successfully given birth in the same neighborhood. The relevant comparison of interest then becomes between mothers that reside on the same census block as opposed to those that reside in the same census block group but not on the exact same census block. As expected, we document substantial unconditional correlations in individual and census block characteristics but it falls close to zero once we condition on our census block group fixed effect. Therefore, conditional on this fixed effect, defining informal networks as other pregnant mothers who reside on your specific census block likely provides as good as random sorting into peer groups. In addition, we deal with the reflection problem by defining the peer group of a currently pregnant mother as all mothers residing on the same census block who gave birth 6 months prior to the conception of our currently pregnant mother. Results show that a currently pregnant mother is around 3 percentage points more likely to be enrolled in Medicaid if a recently pregnant mother on her census block also participated in these programs during her pregnancy. This is equivalent to approximately an 8% increase relative to the baseline average participation in Medicaid. Moreover, we also document substantial heterogeneity in the estimated network effect across race and nativity status. Our results for this subset of the analysis align with conventional wisdom: currently pregnant mothers are more likely to participate in Medicaid if their matched pair belongs to the same ethnic or racial group. Similarly, foreign born Hispanic mothers are more likely to participate if they observe a previously pregnant participating native Hispanic mother, implying potential information flows. Furthermore, our estimated network effect is substantially stronger for neighborhoods which are likely to have lower knowledge of welfare participation rules, implying potential use of targeted information dissemination policy options for reducing gaps between eligibility and participation in welfare programs. Finally, we also provide suggestive evidence that increased Medicaid participation is likely to translate into healthier behavior among pregnant women with earlier and more intensive participation in prenatal care.