J. Najera-Zuloaga, D. Lee, I. Arostegui
The beta-binomial (BB) distribution has been proposed to analyse data in several fields. The adequacy of the BB distribution to fit discrete, bounded and skewed data has been previously documented. The fact that it is not an exponential family member makes classical regression methodology inappropriate, especially in mixed-effects framework. Our goal was to develop a BB mixed-effects model to analyze longitudinal discrete and bounded outcomes.
In this work, we extended the marginal approach for mixed-effects modelling, assuming that the outcome follows a BB distribution and including Gaussian random effects in the linear predictor. We performed a simulation study in order to compare our proposal to others commonly used in this framework. Finally, we applied our modelling proposal to real data on health-related quality of life.
We conclude that this modelling approach is very convenient, especially for skewed outcomes, which is quite often for patient-reported outcomes.
Palabras clave / Keywords: beta-binomial, mixed-effects, patient-reported outcomes
Programado
Sesión V10 Modelos Estadísticos II
1 de junio de 2018 17:20
Sala 3
