Fig 1.
p-curves in the presence of p-hacking for different sample sizes.
The y-axis depicts the share of statistically significant p-values. nmax denotes the maximum sample size drawn from a uniform distribution with a minimum of 50 and p < 0.05 denotes the share of statistically significant p-values from 500,000 iterations. The dashed line represents a hypothetical uniform distribution of p-values.
Fig 2.
Vibration plot for the effect of malaria prevalence on economic growth.
The vibration plot shows estimates of the effect of malaria prevalence in 1966 on the annualized average growth rate of real GDP per capita (1960–1996) on the x-axis. The y-axis shows transformed p-values of these estimates. The plot is based on 100 random samples of countries drawn from a uniform distribution with sample size between 50 and 99. For each sample of countries all 5,005 regression models are estimated resulting in 500,500 estimates of β. The dashed lines represent the 1, 50, and 99 quantiles of the distribution of transformed p-values and of the distribution of β, respectively. The solid line represents p = 0.05. Note that due to the transformation of p-values estimates above the line are statistically significant and below the line estimates are insignificant.
Fig 3.
p-curve and histogram of estimates for the effect of malaria prevalence on economic growth.
The p-curve of the estimated β of Eq (6) is shown in the left graph. The corresponding histogram of the estimated β is shown in the right graph. The y-axis displays the share of significant p-values. The graphs are based on the p-values of 100,000 statistically significant and negative estimates of β.