Economics Research Seminar – Professor David A. Jaeger
Title: Robustness? Range Tests for Equality and Equivalence Across Specifications
Date: Wednesday 28 October 2026
Time: 13:30 to 14:30
Venue: NUBS.2.08
If you would like to attend, please register using the following link
Robustness? Range Tests for Equality and Equivalence Across Specifications
Speaker: Professor David A. Jaeger (University of St Andrews)
Professor Jaeger is am the Director of the Master of Public Policy program at the University of St Andrews and currently serve as the President of the European Society for Population Economics.
His research lies in applied microeconomics, with a particular focus on labor economics, education, (im)migration, causal inference, and applied econometrics.
Abstract:
Applied economists routinely report estimates from several specifications, observe that they are “similar,” and conclude that their results are “robust.” These claims are implicitly about the range of estimates but are rarely supported by formal inference. When several specifications are estimated on the same data, the estimates are likely to be correlated, and similarity can reflect imprecision as easily as stability. I formalize this practice as inference on the range of estimates across specifications. The minimum equivalence bound, R(.95), is the smallest tolerance within which the data support the claim that the estimates agree, and so bounds the disagreement the data allow. The equality p-value, p_R, is a range-based test of whether the estimates are statistically distinguishable from one another. Noise makes equivalence harder to establish even as it makes equality harder to reject. Simulations show that the bound attains its nominal coverage in regular designs, while popular “eyeball” criteria frequently provide false comfort. Applications to five prominent papers illustrate the framework, confirming some published robustness claims and reversing others. In a pre-registered survey of CEPR and NBER affiliates, expert ratings agree with the framework when estimates are obviously similar or obviously different but diverge in intermediate cases. I suggest that R(.95) and p_R be reported whenever multiple specifications are offered as evidence of robustness.