Online Program

Estimation of stationary parameters using dynamic sample weights

*Duncan Ermini Leaf, University of Southern California Schaeffer Center for Health Policy and Economics 

Keywords: estimation, weights, weighted mean

Longitudinal panel survey data often include cross-sectional weights for each wave. These weights specify the number of people in the population represented by each member of the sample and are typically necessary when estimating a population parameter from the sample data. If a parameter remains constant over some time interval, a sensible estimate of that parameter should use data from every sample member present within the interval. This presents a challenge because each sample member can have multiple weights: one weight for each survey wave in the interval. One solution is to use only a single weight for each sample member. Two such approaches are considered here. First, a new method is introduced that is based upon a weighted mean. The second method is based upon a recommendation in the Health and Retirement Study documentation. The two methods are compared using simulated data and survey data from the Health and Retirement Study.