Fig. 4
From: Estimating risk ratio from any standard epidemiological design by doubling the cases
Ratio of the estimated and true RR (panel A) and the coverage (panel B), type I error (panel C) and power (panel D) of the RR estimated by naive and expanded data logistic regression analysis of case-control samples in simulation studies. 2000 simulation iterations were repeated in each simulation scenario. The results from log-binomial regression of the simulated cohort data are displayed for comparison: for a prevalence of 40%, this model failed to converge for 2 simulation cycles for RR=1.5 and 1432 cycles for RR=2, and these were excluded. Panel B excludes scenarios with RR > 1 where the naive/weighted logistic regression model had coverage lower than 50%