Modelling binary longitudinal data with ALR: the cerebro data

  As a final example, we turn to the cerebrovascular deficiency crossover trial discussed in Diggle et al. (1994), Section 8.3. In this trial, 67 patients were given an active drug (A) and a placebo (B) in the order A,B (the AB group) for 34 of the patients (allocated at random), and in the order B,A (the BA group) for the remaining 33. For each patient and each treatment there is a binary response: 0 for an abnormal ECG reading, and 1 for a normal reading. The data are available in the cerebro balanced data structure, and tabulated (after first combining the period 1 and 2 responses) below:

We wish to determine whether the active drug increases the probability of a normal ECG reading.