The data set multivisitsbp extends the HERS SBP data to a series of up to six visits and borrows the set-up used in Problems 11.6–11.9 to simulate missing data through a missing data indicator miss...

The data set multivisitsbp extends the HERS SBP data to a series of up to six visits and borrows the set-up used in Problems 11.6–11.9 to simulate missing data through a missing data indicator miss mar.

(a) To mimic an analysis on complete data, examine a series of models (ignoring the missing data indicators). Fit a GEE model with terms for time (visit) and BMI (bmi). Then, fit a series of mixed models with fixed effects terms for time and BMI but with varying variance/covariance structures. You might try a random slopes model along with first-, second-, and third-order autoregressive (AR1–AR3). Do you reach similar conclusions about changes in SBP over time (given by the coefficient for visit) in these models?

(b) Repeat the model fits in (a) restricted to available data (miss mar equal to 0) under simulated missingness. Do you reach similar conclusions about changes in SBP over time across these models? How do they compare to the corresponding complete data results in Problem 11.10? Discuss how this might affect choice of variance–covariance structure for mixed models with missing data. Would you prefer a more parsimonious structure (like random intercepts) or a richer one (like third-order autoregressive)? Explain.