In a multivariable regression model with a nominal scale variable that has three categories, how many indicator terms would need to be included? In general, for a variable with n categories, what is the expression for the number of terms that would need to be included in the model?
The analysis depicted in Figure 12–4 is more efficient than a stratified analysis but also more biased. Why is it more biased?
Why is an exponential curve, such as the one in Figure 12–6, not a reasonable model for the shape of a dose-response trend? What would be the biologic implication of a dose-response curve that had the shape of the curve in Figure 12–6?
If the age term is removed from the model shown in Table 12–2, what would happen to the coefficients for blood pressure? Why?
In a regression model with a continuous exposure variable, why is it desirable to have a single exposure term in the model when evaluating confounding?
If we have a continuous exposure variable and use a single exposure term to evaluate confounding, the shape of the dose-response curve for that term will be implied by the model. That imposition can be avoided by factoring the exposure into several terms defined by categories of the exposure. The use of several exposure terms, however, will make it difficult to evaluate confounding. How can we evaluate confounding and also avoid the model-imposed restrictions on the shape of the dose-response curve?