Most neuropsychological tests consist of multiple items, and a subject's test score is the sum of the item scores. The test results for each subject thus comprise multiple data-points, and any data-set with test results from more than one subject has at least a two-level structure, with the test item as the first level and the subject as the second level. This structure may be exploited to yield more nuanced statistical analyses than those that treat each subject's test score as a single data-point. Exploiting this structure allows us to take into account the effect of test length and dispersion on score variance and may enhance statistical power. Focusing on tests for which the score can be regarded as a binomial random variable, and using the binomial general linear model, we describe appropriate statistical methods for exploiting test structure in analysing a case series, comparing a case with a control sample, and testing for dissociation. These methods also allow multiple predictors, both categorical and continuous, to be taken into account, thereby enhancing the capacity of researchers to test hypotheses in a case series and to investigate other explanatory factors, in addition to case-control status.
- Binomial general linear model
- Case series
- Test length