Figure 2
From: Meta-analyses triggered by previous (false-)significant findings: problems and solutions
Bias in effect estimates from meta-analysis due to inclusion of false-positive studies. The dashed horizontal line indicates the bias over all simulated meta-analyses, which is equal to zero. The triangles indicate the bias in relation to the number of false-significant studies (we assumed no treatment effect) in the meta-analysis when only large studies showed a false-significant effect. The crosses show the bias when all false-significant studies were small.