Chapter 3: Sampling Weight Considerations for Multilevel Modeling of Panel Data

National longitudinal panel surveys in education, such as the Early Childhood Longitudinal Study (ECLS), face the problem of respondent attrition. On public-release data files for these studies, sets of panel weights are provided that allow the analyst to appropriately weight the observations in the study to account for such attrition. Several sets of weights are provided depending on which waves of data are intended to be used in the analysis. However, these weight adjustments are based typically on sampling information from the base year; information obtained from early waves in the survey program is not utilized in the adjustment model. Other methods to address potential bias from nonresponse include full information maximum likelihood (FIML) and multiple imputation (MI); these methods could make use of responses obtained during all data collection waves. In this chapter, we first describe common approaches to sampling weight adjustment for nonresponse and attrition, including cell adjustment and response propensity models; present arguments for using FIML and MI with auxiliary variables when undertaking longitudinal analyses; and demonstrate, using data from the kindergarten cohort of the Early Childhood Longitudinal Study, the differences in estimates from analyses when nonresponse adjusted panel weights are used as compared with FIML and MI utilizing the base year sampling weights only but with auxiliary information.