Estimating risk ratio from any standard epidemiological design by doubling the cases

Table 5 Bias, empirical SE (Emp. SE), mean SE, coverage, type I error and power of lnRR from log-binomial and expanded data logistic regression analysis of simulated cohort, with 2000 simulation iterations in each scenario

Prevalence	True lnRR	Method	Bias	Emp.	Mean	Coverage	Type I/
				SE	SE		Power¹
0.1	0	Log-binomial	-0.001	0.213	0.211	95.4	4.6
		Expanded data logistic	-0.002	0.214	0.211	95.4	4.6
	0.223	Log-binomial	-0.001	0.202	0.201	95.2	21.4
		Expanded data logistic	-0.001	0.202	0.201	95.2	21.7
	0.405	Log-binomial	0.002	0.194	0.194	95.0	56.5
		Expanded data logistic	0.002	0.194	0.194	95.2	56.4
	0.693	Log-binomial	0.000	0.185	0.185	95.8	96.0
		Expanded data logistic	0.000	0.185	0.185	95.6	96.0
0.2	0	Log-binomial	-0.005	0.136	0.139	95.8	4.2
		Expanded data logistic	-0.006	0.137	0.139	95.6	4.4
	0.223	Log-binomial	-0.002	0.131	0.131	95.6	39.1
		Expanded data logistic	-0.002	0.132	0.132	95.3	39.1
	0.405	Log-binomial	0.004	0.126	0.126	95.3	89.3
		Expanded data logistic	0.004	0.126	0.126	95.3	89.2
	0.693	Log-binomial	0.002	0.122	0.120	94.9	100
		Expanded data logistic	0.002	0.122	0.120	94.6	100
0.3	0	Log-binomial	-0.004	0.102	0.104	95.3	4.7
		Expanded data logistic	-0.004	0.103	0.105	95.3	4.6
	0.223	Log-binomial	-0.001	0.096	0.098	94.8	63.1
		Expanded data logistic	-0.001	0.098	0.099	95.0	61.3
	0.405	Log-binomial	-0.003	0.094	0.094	95.2	98.9
		Expanded data logistic	-0.003	0.095	0.095	94.7	98.7
	0.693	Log-binomial	-0.002	0.092	0.088	93.4	100
		Expanded data logistic	-0.002	0.093	0.089	93.7	100
0.4	0	Log-binomial	-0.001	0.081	0.082	95.6	4.4
		Expanded data logistic	-0.002	0.082	0.084	95.3	4.7
	0.223	Log-binomial	-0.005	0.074	0.076	95.8	81.6
		Expanded data logistic	-0.005	0.077	0.078	95.2	79.4
	0.405	Log-binomial²	-0.002	0.074	0.072	93.7	100
		Expanded data logistic	-0.002	0.076	0.073	93.8	100
	0.693	Log-binomial³	-0.017	0.067	0.067	95.2	100
		Expanded data logistic	0.002	0.068	0.068	95.3	100

¹Values reported in this column are the type I error when true lnRR=0 and power otherwise.
²Based on 1998 simulation cycles where the log-binomial regression converged.
³Based on 568 simulation cycles where the log-binomial regression converged

ISSN: 1471-2288