A quantitative examination of the sources of speech discrimination test score variability

Several sources of variability inherent in any speech discrimination measurement are outlined. Limitations of the use of the binomial theorem to predict intrasubject (test-retest) variability are examined. First, the underestimation of variability that is caused by the inclusion of items of different degrees of difficulty is quantified by a reanalysis of published data. Second, it is shown that variability will be larger than expected if the subject's ability is different during test and retest sessions. Fortuitously, these two deviations from the model have opposing effects on the score variability. The estimates provided by the binomial theorem are thus better than if only one effect was present. The need for a clear distinction to be maintained between inter- and intra-subject variability and the effect of list differences on test-retest variability are also discussed.