Table 2.

Type I error rate and power based on simulation results comparing the conventional maximum likelihood method (Wald) and the PML method

	All outcome phenotypes			Excluding complete separation
	Wald	PML	# phenotypesa	Wald	PML	# phenotypesa
Null hypothesis (Type I error)
OR = 1	0.04	0.04	720	0.05	0.04	526
Alternative hypotheses (Power)
OR = 2	0.62	0.61	11	0.65	0.65	10
OR = 3	0.83	0.83	11	0.86	0.86	11

Note: Logistic regressions performed with PS adjustment, using PS generated with elastic-net penalty of α = 0.1.

^aAverage number of outcome phenotypes remaining in the simulated datasets.