In the modeling of the GLOW500 data using purposeful selection age was modeled as linear in the logit. We noted that the estimated coefficients for the quartile design variables for age in Table 4.10...

In the modeling of the GLOW500 data using purposeful selection age was modeled as linear in the logit. We noted that the estimated coefficients for the quartile design variables for age in Table 4.10 suggested an alternative parameterization: using the design variable for the fourth quartile AGE_4. This parameterization of age was not pursued further. Proceed with purposeful selection using AGE_4. To save time, assume that your main effects model is the one in Table 4.9 but with AGE replaced by AGE_4. Compare your model to the one in Table 4.15 that resulted when age was modeled as linear in the logit. Which model do you think is the better one for estimating risk factors for fracture?