Random effects in paired-comparison experiments using the Bradley-Terry model

The Bradley-Terry model for paired-comparison experiments assumes that the parameters for treatment ratings are the same for each comparison. We generalize this model by envisioning a population of judges from which a sample is taken to perform the experiment. The preference parameters can then be treated as random variables, and this induces a correlation between outcomes involving the same judge. Such a model may be helpful in isolating variation due to judges, from true variation in the treatments, as well as in accounting for the nonindependence of judgments.