Reference no: EM132496290
Question: Read in the problem3 data set from the course website. We have a binary outcome Yit for = i = 1,....., n subjects with data measured at times t = 1,.....,4. We also observe one Covariate, Xi, which is subject-specific and does not vary across time within a subject. We decide to fit two different amulets.
(i) Fit model 1 using a GEE with an independence working correlation structure. Provide an estimate and 95% confidence interval for /3.
(ii) Using only your output from the GEE fit in part (b) do you think that the GEE was needed in this case or would a standard GLM have been valid? Explain why or why not.
(iii) Fit model 2 using a GLMM. What is the estimate of the random effects variance, ^σu2?
(iv) Using GLMMs, perform a likelihood ratio test of H0 : β* = 0.
(v) Create a plot that has the predicted probability of success as a function of X from both model 1 and model 2. Your plot should look s' lar to slide 50 of the week 12 lecture slides in that there should be individual lines for each subject when using the GLMM, and one population line from the GEE.
(vi) Other than the difference in interpretations, can you think of at least one reason why the GEE would be preferred over a GLMM? I'm looking for you to find at least one assumption that the GLMM model relies on for correct inference that the GEE does not rely on.
Attachment:- problem3.rar