42.1 The Philosophy of Robustness

Before diving into specific techniques, it’s worth understanding why robustness checks matter. In empirical research, we rarely have perfect certainty about the “correct” specification. Our choices about which controls to include, how to measure variables, which functional forms to use, and how to address endogeneity can all influence our results. Sensitivity analysis helps us understand whether our main conclusions depend critically on these choices, or whether they remain stable across reasonable alternatives.

A robust finding is one that persists across multiple plausible specifications. This doesn’t necessarily mean the coefficient estimate must be identical across all specifications, some variation is expected and often informative. Rather, robustness means that the key substantive conclusion (e.g., the sign, statistical significance, or economic magnitude of an effect) remains consistent despite reasonable variation in modeling choices.