|
Abstract:
|
Measurement in mental health and psychosocial research hinges critically on the reliability of instruments (questionnaires) designed to capture the often latent constructs that define a disorder or a condition. As such instruments are rated by independent evaluators or by subjects themselves, instrument/rater reliability has been a major concern for deriving scientific evidence from such a "soft" outcome based research discipline, affecting scientific and research integrity in many psychosocial interventions. One of the key indices for assessing outcome reliability is the kappa class of statistics. However, existing methods on such a class of statistics do not sufficiently address data clustering and missing data, two most common issues in longitudinal data analysis. In this talk, I introduce a new class of U-statistics based estimates to facilitate inference under clustered study designs.
|
