Each study measures n people and runs a one-sample t-test of H0: μ = μ0 (think μ0 = 37 °C). In a simulation we choose the truth: the effect size d = (μ − μ0)/σ says how far the real mean is from μ0, in standard deviations (d = 0.5 with σ = 0.41 °C means a real mean about 0.2 °C away). The curve is the null distribution of the test statistic; the shaded tail(s) are the rejection region of size α. Bars show the statistics the studies actually produced, pink where H0 was rejected.
When H0 is true (d = 0) every rejection is a Type I error, and they happen at rate α; the P-values are spread evenly over (0, 1). When H0 is false, the rejection rate is the power, and every non-rejection is a Type II error.