State Paid Family Leave and Maternal Employment: A Cautionary Tale of Parallel Trends

Abstract

I examine whether state-level paid family leave (PFL) programs increase employment among women who recently gave birth. Using a staggered difference-in-differences design that exploits variation from five states adopting PFL between 2004 and 2020, I estimate effects using both traditional two-way fixed effects and heterogeneity-robust estimators. Naive TWFE estimates suggest PFL increases maternal employment by 1.7 percentage points. However, Callaway-Sant'Anna estimates show essentially zero effect ($-0.18$ pp, SE $= 0.98$ pp), and a formal pre-test strongly rejects parallel trends (p $< 0.001$). Pre-treatment event study coefficients are significantly different from zero across most periods, indicating systematic differences between treated and control states that predate policy adoption. This finding illustrates the importance of heterogeneity-robust estimators and formal pre-trend tests in staggered difference-in-differences designs. The inability to identify credible causal effects despite large sample sizes underscores that data availability does not guarantee causal identification.