Many factors inhibiting and facilitating economic growth have been suggested. Will international income data tell which matter when all are treated symmetrically a priori? We find that growth determinants emerging from agnostic Bayesian model averaging and classical model selection procedures are sensitive to income differences across datasets. For example, many of the 1975-1996 growth determinants according to World Bank income data turn out to be irrelevant when using Penn World Table data instead (the WB adjusts for purchasing power using a slightly different methodology). And each revision of the 1960-1996 PWT income data brings substantial changes regarding growth determinants. We show that research based on stronger priors about potential growth determinants is more robust to imperfect international income data.