Sample Composition Fragility in Demographic-Macro Research: A Diagnostic Framework

Demographic variables are widely used in crosscountry panel regressions to explain capital flows, interest rates, exchange rate regimes, and fiscal dynamics. Yet the literature reports strikingly contradictory findings: demographic effects on the current account are positive in some studies, negative in others, and null in several OECD-focused panels. We show that these contradictions are the expected consequence of a multi-dimensional coefficient surface. Using a 141-country panel covering 97% of world population and 99% of world GDP, we estimate demographic effects across 18 papers and 45 major findings, classifying results as robust (20 findings, 44%), conditional (16 findings, 36%), or spurious (9 findings, 20%). The coefficient on the leading demographic variable (Z₁) ranges from-112 to +98 on the current account alone, depending on a country's income level, institutional configuration, and monetary regime. A pooled linear estimate averages over this surface, and the result depends entirely on which countries are in the sample. We identify four mechanisms that determine where on the surface a given sample sits-OECD composition weight, GDP/capita confounding, influential observation clusters, and the KAOPEN-income confound-and propose a nine-item diagnostic toolkit for researchers working with demographic variables in crosscountry panels.