Omitted Variable Bias and Wartime Legacies. A Reply to Malesky and Nguyen (Journal of Comments and Replications in Economics, 2024)

Malesky and Nguyen (2024, Journal of Comments and Replications in Economics, MN) reassess a study on the effects of wartime violence on civic engagement in Vietnam, attributing discrepancies in previous findings to coding and historical errors. They argue that prewar party strength, rather than wartime exposure, drives contemporary civic engagement. However, their reanalysis omits key geographic covariates, which are essential for the validity of the original study's quasi-experimental design and were included in both Barceló (2021) and Miguel and Roland (2011). MN claim that the control variables South and Latitude are collinear and therefore drop both from their models. This is an elementary mistake: collinearity between control variables does not affect the treatment effect (Distance) and should not impact its estimation. Even if collinearity were an issue, the appropriate response would be to remove only one of the collinear variables, not both. This misspecification leads to omitted variable bias and a violation of the exclusion restriction of the instrumental variable model. Once this omission is corrected and either South or Latitude, or both, are properly accounted for, the originally reported results withstand all of MN's other proposed modifications, including controlling for prewar party strength, excluding zone IV provinces, and adjusting standard error clustering, among others. This study confirms that wartime exposure enhanced long-term civic engagement in Vietnam and highlights how improper model specifications can produce biased estimates that may obscure, rather than clarify, our understanding of the legacy of historical events.