Abstract:
The scientific community rewards priority: the researcher who establishes a novel finding first, claims credit that later entrants cannot. This reward may increase the incentive to reach statistical significance. Using 43,140 tests published in the AER, JPE, and QJE between 2001 and 2010, together with text-based and bibliographic measures of novelty, I examine whether more novel articles are more prone to selective reporting. Running caliper regressions around conventional significance thresholds, I find suggestive evidence that novel articles are more likely to report just-significant results at the 10 percent level, an association concentrated around the threshold and driven by textual novelty.
