Table 3.

Power (%) of the tests of interaction effects^a

Effects	Independenceb			Dependencec
	0		log(1.2)	0		log(1.2)
Retrospective likelihood, unspecified distribution of GM, X (function CCMO.na)
X 1 × G M	6.0		56.2	4.8		56.0
X 1 × G C	5.2		85.0	6.0		84.6
Retrospective likelihood, double-additive model for Pr(GM | X) (function CCMO.dep)
X 1 × G M	6.2		61.2	5.2		61.2
X 1 × G C	4.8		84.8	6.0		85.0
Prospective likelihood (function Spmlficmcm)e
X 1 × G M	6.2		40.8	7.0		38.8
X 1 × G C	5.4		84.1	6.0		82.4
Logistic regression (function glm)
X 1 × G M	6.2		49.0	6.0		44.8
X 1 × G C	5.0		49.2	4.4		48.8
Stratified retrospective likelihood (function gxe from Haplin)
X 1 × G M	3.6		52.6	71.8		–f
X 1 × G C	5.0		49.4	53.2		–

^aWald tests of the interaction terms except for the stratified retrospective likelihood where a trend test is performed. Other model coefficients were set to β₁ = log(1.8), β₂ = log(1.5) and β₃ = log(1.8). Type I error level was set to α = 0.05.

5% of the offspring genotypes were set to missing completely at random.

^bIndependence between environmental exposure X₁ and maternal genotype G^M.

^cLinear dependence between environmental exposure X₁ and maternal genotype G^M.

^dPercentage of child genotypes missing completely at random.

^eProportions of replicates where likelihood maximization failed were 3.2% and 17.2% for the two columns under dependance and 3.4% and 20.6% for the two columns under independance.

^fPower not displayed because of inflated Type I error.