The Anarchy of Numbers: Aid, Development, and Cross-country Empirics—Revised July 2004

Abstract

Contemporary econometric literature has generated a large number of stories of the relationship between how much foreign aid a countries receives and how it grows. All the stories hinge on the statistical significance in cross-country regressions of a quadratic term involving aid. Among the stories are that aid raises growth (on aver-age) 1) in countries where economic policies are good; 2) in countries where policies are good and a civil war recently ended; 3) in all countries, but with diminishing returns; 4) in countries outside the tropics; 5) in countries with difficult economic environments, characterized by declining or volatile terms of trade, natural disasters, or low population; or 6) when aid increases in countries experiencing negative export price shocks. The diversity of results prima facie suggests that many are fragile. Easterly et al. (2004) find the aid-policy story (Burnside and Dollar, 2000) to be fragile in the face of an expansion of the data set in years and countries. The present study expands that analysis by applying more tests, and to more studies. Each test involves altering just one aspect of the regressions. All 19 tests are derived from sources of variation that are minimally arbitrary. Twelve derive from specification differences between studies, what Leamer (1983) calls “whimsy.” Three derive from doubts about the appropriateness of the definition of one variable in one study. The remaining four derive from the passage of time, which allows sample expansion. This design allows an examination of the role of “whimsy” in the results that are tested while minimizing “whimsy” in the testing itself. Among the stories examined, the aid-policy link proves weakest, while the aid-tropics link is most robust.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 22 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.