Publication bias in two political behavior literatures

Publication bias occurs when the probability that a paper enters the scholarly literature is a function of the magnitude or significance levels of the coefficient estimates. We investigate publication bias in two large literatures in political behavior: economic voting and the effects of negative advertising. We find that the pattern of published estimates is consistent with the presence of publication bias and that bias is more prevalent in the most influential and highly cited outlets. We consider the possible causes and find some evidence that papers systematically employ one-sided hypothesis tests in response to failure to meet the more demanding critical values associated with two-tailed tests, a practice that leads to misleading reports of the probability of Type I errors.