Abstract:
|
This paper studies the need of using longitudinal survey weights in longitudinal multilevel modeling. Most large scale surveys provide a set of weights including design weights, cross-sectional and longitudinal weights along with their data. For cross-sectional data the computation and use of weights in multilevel data analysis is well-elaborated. However, the performance of longitudinal weights computed as the product of a nonresponse adjusted design weight and the inverse of estimated wave-specific response propensities is not. We address this shortcoming by conducting a simulation study on a synthetic population of students in schools that evolves over time. For this purpose, we draw samples, include nonresponse at the school and student level and apply "standard" weighting procedures. We examine the performance of longitudinal weights in statistical analysis by estimating population totals and longitudinal models, such as growth curve or fixed effects panel models.
|