Consider the data from the Myopia Study described in Section 1.6.6 whose variables are described in Table 1.10. The binary outcome variable is MYOPIC (0 = Yes, 1 = No). Consider as independent...

Consider the data from the Myopia Study described in Section 1.6.6 whose variables are described in Table 1.10. The binary outcome variable is MYOPIC (0 = Yes, 1 = No). Consider as independent variables all others in Table 1.10 except spherical equivalent refraction (SPHEQ) as it is used to define the outcome variable, the composite of near-work hours (DIOPTERHR) and study year (STUDYYEAR).

(a) Use purposeful selection to obtain what you feel is the best model for estimating the effect of the risk factors on myopia. This analysis must include identification of the scale in the logit of all continuous covariates and selection of interactions. Assume that all possible interactions among your main effects are clinically reasonable.

(b) Repeat problem 6(a) using stepwise selection of covariates (main effects and then interactions among main effects forcing in the main effects).

(c) Repeat problem 6(a) using best subset selection of covariates with Mallows’ Cq (main effects and then interactions among main effects forcing in the main effects).

(d) Repeat problem 6(a) using multivariable fractional polynomial selection of main effects followed by purposeful selection of interactions.