Panel Data Stories Without Heroic Claims

Panel dataApplied microReporting

Fixed effects soak up time-invariant confounders, but they do not grant moral authority. In panel labs, we ask participants to describe what FE actually removed in words a plant manager understands. If the answer relies on jargon, we rewrite until the manager could repeat it. We also spend time on economic magnitude. A statistically precise effect can be operationally irrelevant. Participants practice pairing coefficients with back-of-envelope costs implied by the estimate, even when those costs are rough. The point is to keep conversations honest about levers. Dynamic panels receive scepticism proportional to instrument count. When instruments multiply, we require a plain-language story about why exclusion is plausible and what happens if it is not. That story belongs near the estimates, not in a footnote on page nineteen. Finally, we encourage documenting attrition. Unbalanced panels are reality; hiding imbalance behind a sentence about “unbalanced FE” is not. Cohorts produce a short attrition appendix even when reviewers might skip it, because downstream teams often do not.