There are a couple of interesting posts at Economist's View and Marginal Revolution about publication bias in political science and economics journals. Publication bias exists if journal referees and editors tend to reject papers that exhibit statistically insignificant results while the same quality papers with statistically significant results tend to get rejected less often (where statistically significant is defined as a 95% chance that your regression coefficient is not equal to zero). In addition to publication bias, there is evidence of a researcher response. Here is a picture of spike in t-statistics a little greater than 1.96 and a dearth a little below in empirical papers published in political science:
Source: Kevin Drum
In other words, researchers massage their data to just ... get ... the p-value ... to 0.05 because they know their chances for rejection are greater than at t=1.92. As for me? I'm perfectly content with p=0.10 so you should scour my publications for a spike at t > 1.645.
My favorite line of the abstract posted at MR is this one:
Note that [t-stat] Inflation is larger in articles where stars are used in order to highlight statistical significance ...
