Imagine a study finds labeling paid family leave as “supported by the Republican Party” increases support by 10 percentage points. What would be the effect if the issue were climate policy? Or if the cue were that Democrats support it? What about in an entirely different country? In short, how much does one estimate tell us about future estimates? In two papers Clifford, Leeper and Rainey examine how much we can infer about the general population of estimates using a single estimate.
May 08, 2025
Publication bias occurs when the distribution of observed effects differs from the distribution of all such effects. Usually, this means there is some filter based on the magnitude or direction of the effect (positive or negative). Crucially, we do not observe all effects, so we must infer whether publication bias has occurred.
December 05, 2024
Cohen’s $d$ is commonly used as the standardized effect size in meta-analysis. Meta-analyses now also tend to include tests of publication bias. Both of these are fine on their own, but when you mix Cohen’s $d$ and publication bias methods, you may run into problems.
December 03, 2024
I was reading O’Rourke’s history of meta-analysis paper and he mentions what is possibly the first paper to systematically look at publication bias based on significance. This is Sterling’s 1959 paper “Publication Decisions and Their Possible Effects on Inferences Drawn from Tests of Significance–Or Vice Versa”.
November 26, 2024
