The use of replicates in logistic measurement error modelling

The use of replicates is essential in logistic measurement error models with random, additive error, as it gives a consistent estimate of the variance of the measurement error. In this paper, we investigate two different aspects of the use of replicates. We look into the efficiency gain in using more than two replicates in situations where the replicates are correlated. We find that, unlike the situation with independent replicates, the efficiency gain is an almost linear function of the number of replicates. Further, we study the allocation of replicates among study subjects, and find that the optimal way of doing this is to allocate few replicates to as many subjects as possible, rather than allocating many replicates to fewer subjects.