Table 3.

Simulation Results: Effect of Sample Size on ARR Precision^*: Regression Risk Analysis (RRA), Poisson, and Mantel–Haenszel (M–H) Estimates

Sample Size	LRA ARR	CI	Poisson	CI	M–H	CI
100,000	3.97	[3.87,4.07]	3.88	[3.79,3.98]	4.01	[3.77,4.26]
50,000	4.04	[3.89,4.18]	3.96	[3.83,4.08]	4.31	[3.94,4.71]
5,000	4.11	[3.62,4.60]	4.06	[3.62,4.50]	4.31	[3.22,5.77]
2,500	4.02	[3.36,4.67]	3.87	[3.28,4.46]	3.56	[2.51,5.05]
500	4.09	[2.35,5.83]	3.89	[2.33,5.44]	4.01	[1.67,9.63]
250	3.68	[1.13,6.22]	3.49	[1.46,5.52]	3.00	[0.86,10.41]

^*Precision is manifest as the 95% confidence interval (95% CI): All were calculated using STATA. RRA confidence interval was calculated using bias corrected bootstrapping in STATA.

ARR, adjusted risk ratio; LRA, logistic risk analysis.