writing to clarify what you mean by this question.
6. Specify a model (linear equation) for the risk of lung cancer as a function of sex, colgrad, age, smoking, snp, and snp*smoking.
the outcome is binary (case, 0,1), so it is not possible to use linear regression. I assumed the phrase "linear equation" has nothing to do with linear regression.
is that correct?
if so, then I don't understand how the logistic regression models differ between question 6 and question 8
thank you
Hi Laura,
Yes, linear equation doesn't mean that you have to do a linear regression. Remember that the OR of an outcome can be expressed as a linear function on a logarithmic scale. The equation in question 8 is an example of a linear equation for a binary outcome (Logit{P(D = 1|G= g; E = e)} = γ0 + γ1g + γ2e + γ3eg). However, we are asking you to specify or replace each variable in the equation using the provided predictors.
You may use the slides on Effect modification Part 2b as a guide in answering the question. Specifically, at the end of part 2b, Dr. Chan talks about similarities between logistic and linear models.
I hope this clarifies the question
Hi,
I'm still a bit confused as to how the equations differ between Q6 and Q8. The logit command will give us the coefficients - which when exponentiated - yield the OR that we get with the logistic command. Or is Q8 just asking for the explanation of results from the prior regression.
Thank you
Hi Sukhmani,
Question 8 does relate back to q6. In q6 we ask you to write the equation only; in question 7 we ask you RR11, RR10, and RR01; in q8, we provide some formal terminology about the equation and ask you to program it in STATA and give us the multiplicative interaction with 95% CI. Recognize the difference in what we are asking you for in q6 and q8, vs. q7.
I hope this helps.
