Table 5.

95% confidence intervals for mean ratio/difference for 6 statistical models under the alternative hypothesis for 2 population distributions, controlling for confounders

			Statistical Model
Population distribution	Direction of confounding	n2/n1	Poisson	QP	NB	ZINB	Linear	HC
Poisson	Inflated	1/9	95.1	95.0	95.4	94.7	31.0	55.2
		1/2	95.4	95.3	95.5	95.6	75.8	81.0
		1	95.2	95.1	95.3	96.0	95.6	94.9
		2	95.0	95.0	95.1	96.1	72.5	61.8
		9	95.5	95.4	95.5	96.5	43.5	19.5
	Deflated	1/9	95.1	94.9	95.2	95.8	50.1	60.7
		1/2	95.2	95.2	95.4	96.6	75.1	79.0
		1	95.3	95.2	95.5	96.5	93.6	94.8
		2	95.3	95.2	95.4	96.9	88.9	89.0
		9	95.2	95.1	95.3	96.8	72.9	66.0
NB	Inflated	1/9	75.0	89.4	94.8	93.6	54.8	85.2
		1/2	77.9	93.9	94.7	94.4	87.3	90.6
		1	79.1	95.7	94.7	95.0	96.8	94.3
		2	80.9	97.3	95.0	95.7	92.9	81.3
		9	83.0	98.6	95.2	95.9	90.7	53.2
	Deflated	1/9	77.2	92.5	95.0	94.4	68.6	78.3
		1/2	77.5	93.3	94.6	95.2	83.2	87.9
		1	76.5	93.6	94.7	95.3	92.2	94.2
		2	78.1	94.6	94.7	95.9	91.8	93.1
		9	79.0	96.3	95.0	96.5	90.4	82.6

Abbreviations: HC, heteroskedasticity consistent; NB, negative binomial; QP, quasi‐Poisson; ZINB, zero‐inflated negative binomial.

95% confidence intervals for the mean ratio are given for Poisson, NB, and QP, for the mean difference for linear regression.

Distributions under the alternative hypothesis, m1/m2 = 2, with sample size n1 + n2 = 600, estimation from 10 000 simulations.

Confounders were age and gender and were associated with a mean inflation of 40%. For inflated mean, group 2 was older (mean age 45) and with more men (70%) than the first group (respectively, 35% and 30%), and for deflated mean, group 1 was older with more men.

Valid models (95% IC between 93% and 97%) are flagged in bold.