Chris Liao

I conduct a large-scale empirical study of open source software organizations to examine which organizational practices affect organizational resilience. Measuring resilience by the impact of a key member’s exit on software development and release activity, I find that on average, organizational outcomes drop permanently by up to 0.8 standard deviations post-departure. I then develop a novel method combining flexible machine learning models with event studies that separates organizations into groups that differ markedly in measured resilience. Leveraging interpretable machine learning methods, I show that practices highlighted in the existing literature do not enhance resilience on their own, but do enhance resilience through complementarities with other practices.

This note introduces a procedure for estimating heterogeneous dynamic treatment effects in event studies with staggered adoption, allowing heterogeneity to vary by treatment adoption date and with observation-level time-invariant covariates. The approach reformulates the problem as estimating a conditional linear model with binary regressors and estimates all cohort–event time conditional treatment effects jointly using a single generalized random forest, with propensity scores adapted to staggered treatment adoption in panel settings.