Neuropsychological tests are essential for assessing cognitive function, but floor and ceiling effects, summary score limitations, and longitudinal censoring reduce accuracy. This thesis applies censored normal (Type 1 Tobit) models to longitudinal data from the Boston Naming Test (ceiling effects) and Word List Delayed Recall (floor effects). Simulations show that ignoring these effects misestimates change points and population trends. Tobit models improved classification of Mild Cognitive Impairment and score prediction compared to standard models, with higher ROC values and lower mean squared error. Accounting for ceiling and floor effects enhances cognitive decline detection and supports earlier intervention strategies.
