Publication bias is common in social sciences because journals favor statistically significant results. However, publication bias depends on a variable's role—key or control—in regression models. This study shows that when democracy is a key variable (rather than a statistical control) it faces stronger publication bias.
New Meta-Regression Analysis Strategy:
The authors created an approach to detect this conditional effect by analyzing 229 model estimates from FDI research on democracies.
Key Finding:
Our analysis reveals that democracy appears in about twice as many studies when treated as a key variable compared to being controlled for, even though controlling variables are often more relevant for causal inference.
Why It Matters:
This demonstrates how publication bias varies depending on a variable's analytical role. Variables crucial to an argument (like democracy) may be selectively published while controls remain consistently present.
| The Conditional Nature of Publication Bias: A Meta-Regression Analysis was authored by Erica Owen and Quan Li. It was published by Cambridge in PSR&M in 2021. |
