LR modeling

I have a series of primary exposure variables (ordered, categorical
coded as dummy variables) that I will use in logistic regression. I
plan to use each primary variable separately and to adjust each time
for several potential confounders such as age, gender, BMI, hours/wk
etc.


When I look at example of this kind of risk factor modeling in the
literature however, it seems as if most of the studies just go down the

list and adjust each variable for the same two or three covariates.
There is no mention of whether each has a linear relationship with the
log odds of outcome or whether any of them are effect modifiers instead

of confounders.


The model building steps seem well defined for predictive modeling but
not risk factor modeling. Any thoughts? Frank Harrell's book was
recommended and it is on order.


Thanks!
Marc

.