Background. Within the field of infertility, hierarchically structured data are not hard to find. For example, in the assisted reproductive technology whereby a crossover design is conducted, observations will be clustered within couples. The random-effects logistic regression model is a very popular choice for the analysis of multilevel data. The purpose of this article is to compare different statistical software implementations of random-effects logistic regression models using the multilevel dataset from the crossover trial in infertility.

Methods. Sixty-two couples with primary or secondary infertility due to male factor entered the study. The sixty-two couples were randomly equally divided into two groups. Each group began one of the two treatment modalities (controlled ovarian hyperstimulation in conjunction with timed intercourse or intrauterine insemination) for three consecutive cycles and then switched to the alternative treatment after one rest cycle, if pregnancy was not achieved. Random effects logistic models were fit to a dataset of couples undergoing assisted reproductive technology. The estimates obtained were compared among the four statistical packages.

Results. The parameter estimates (protocol) obtained from all the four statistical packages were not very much dissimilar. The software differs considerably in computing time. SAS^© and R^© took few seconds to compute the estimates.

Conclusions. In comparing the four statistical packages, it was found that the estimates were not very much dissimilar. Thus, there seems to be no explicit preference for either a frequentist or Bayesian approach. The choice for a particular implementation may largely depend on the computing time.

crossover designs, logistic regression, random effects, multilevel, conditional estimates.