Parallel trends and stable coefficients rule out omitted-variable bias in the design.

Evidence

We further rule out any differential trends in support for the Nazi Party before the banking crisis erupted: Danat exposure does not predict support for the Nazi movement or its predecessor parties in any federal election prior to the banking crisis, not even during the early years of Germany’s Great Depression in 1930. A difference-in-differences analysis shows parallel trends in exposed vs. unexposed cities before Danat’s failure, but a highly significant differential in each election thereafter. We also find that including city controls and region fixed effects in our regressions leads to no material change in coefficients, while the R2 increases by over 50 p.p. Unobservables are hence unlikely to explain our finding, reducing potential concerns about self-selection and omitted variable bias (Altonji et al., 2005; Oster, 2019). — Sebastian Doerr et al., Financial Crises and Political…, 34

Categorization

• QUESTION: The R2 jump of 50 p.p. with region fixed effects and controls merits scrutiny—does it reflect inflation from many fixed effects rather than genuine control success?