EPI 204 Homework4
Part I
You are interested in modeling the effect of smoking (binary, 1 if smoker, 0 if non-smoker) on
subsequent BMI (continuous). You believe age (in years, continuous), sex (binary, 1 if male, 0 if
female), race (categorical, 1 if white, 2 if black, 3 if other), and baseline BMI (continuous) are
confounders. You decide to fit a GLM.
1.What is the link function you plan to use for this model? Explain why you chose this link function.
2.
Write out the model you plan to fit, using mathematical notation.
E[BMI] = b
0
+ b
1
*smoking + b
2
*age + b
3
*sex + b
4
*black + b
5
*other + b
6
*baseline_BMI
3.
Interpret the beta coefficient for sex.
Holding all other variables constant, males have an increment of b
3
in expected BMI compared to the
expected BMI of females, over the study period.
4.Here is some partial SAS code you plan to use to execute this approach. How would you appropriately fill in the question marks? (NOTE, not all question marks necessarily need to be filled)
5.
Suppose you are interested in effect modification of the smoking-BMI relationship by sex.
Write out a single model you can fit to assess whether effect modification is significant or not.
E[BMI] = b
0
+ b
1
*smoking + b
2
*age + b
3
*sex + b
4
*black + b
5
*other + b
6
*baseline_BMI +
b
7
*smoking*sex
Interpret any new coefficient(s) that appear in this model but do not appear in the model you
wrote in Question#2.
Holding all other variables constant, b
7
is the additional increment in expected BMI in male smokers
above and beyond what we would expect, given the effect of being male in smokers and the effect of
smoking among males, over the study period.

#### You've reached the end of your free preview.

Want to read all 4 pages?

- Spring '14
- Hernandez-Diaz