Table 1.

rs10938397: Comparison of interaction terms chosen by the five methods. The dedicated boosting algorithm took 92 steps. Cells that are labeled “-” mean that a particular approach did not select that variable. Each approach first fits (the same) main effects; “Full” refers to fitting all interaction terms using a linear model; “GlobalB” is the globallboosttest algorithm; “Boosting” is the dedicated boosting algorithm.

	Main Effects			Interaction Effects
	Estimate	Std. Error	p-value	Full	AIC	BIC	Lasso	GlobalB	Boosting
(Intercept)	40.597	1.928	< 0.001
rs10938397	0.209	0.082	0.011
Age	−0.195	0.008	< 0.001	−0.016	−0.018	-	-	-	−0.014
Amount of exercise	−0.066	0.005	< 0.001	−0.013	−0.013	-	−0.009	-	−0.010
Exercise at 18	1.387	0.138	< 0.001	−0.358	−0.318	-	−0.227	-	−0.215
Exercise at 35	0.345	0.147	0.019	0.074	-	-	-	-	-
Exercise at 50	−0.518	0.134	< 0.001	−0.067	-	-	−0.005	-	-
% Calories from carbo.	−0.007	0.017	0.665	−0.002	-	-	-	-	-
% Calories from protein	0.183	0.024	< 0.001	0.031	-	-	-	-	0.016
% Calories from fat	0.096	0.019	< 0.001	−0.005	-	-	-	-	-
Education level	−0.359	0.030	< 0.001	0.093	0.091	-	0.041	-	0.060
Ever smoking	0.401	0.121	0.001	0.278	0.261	-	0.191	-	0.164
Current smoking	−3.153	0.218	< 0.001	−0.093	-	-	-	-	-
Alcohol	−0.612	0.055	< 0.001	−0.007	-	-	-	-	-
Hispanic	−0.329	0.216	0.127	0.263	-	-	0.143	-	0.019
African American	2.532	0.160	< 0.001	0.525	0.469	-	0.467	0.030	0.362
Asian/Pacific Islander	−3.936	0.275	< 0.001	−0.389	-	-	−0.269	-	−0.229
American Indian	−0.603	0.565	0.286	1.336	1.308	-	0.991	-	0.816
Region middle	−0.315	0.144	0.029	−0.080	-	-	-	-	-
Region south	−0.361	0.137	0.008	−0.069	-	-	-	-	-