Motivated by the evaluation of the effect of GATT, we investigate the role of network information in propensity score matching. Under the assumption of strong ignorability, propensity score matching (PSM) is a widely used technique in causal inference studies to adjust for bias arising from an unbalanced distribution of observed confounders between a treatment and a control group. Both theoretical and applied works has recently considered the PSM for structured data, but the analysis of interlinked ...
Motivated by the evaluation of the effect of GATT, we investigate the role of network information in propensity score matching. Under the assumption of strong ignorability, propensity score matching (PSM) is a widely used technique in causal inference studies to adjust for bias arising from an unbalanced distribution of observed confounders between a treatment and a control group. Both theoretical and applied works has recently considered the PSM for structured data, but the analysis of interlinked data is still missing. In this paper we consider the implementation of PSM in the context of network data. In our application, together with individual unit characteristics, also features of the social network in which units are embedded are considered as confounders (i.e., variables that impact on both the probability of receiving the treatment and the outcome). We study the sensibility of causal inference with respect to the presence of characteristics of the network in the set of confounders conditional on which strong ignorability is assumed to hold. We find that estimates of the average causal effect are sensitive to the presence of network information in the set of confounders, therefore we argue that estimates may suffer from omitted variable bias when network data are ignored, at least in our application.
+