|
Activity Number:
|
284
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #305723 |
|
Title:
|
Optimal Probability Weighting Methods in Longitudinal Models for Data with Nonignorable Unequal Selection and Nonignorable Wave Nonresponse
|
|
Author(s):
|
Sharon L. Christ*+
|
|
Companies:
|
The University of North Carolina at Chapel Hill
|
|
Address:
|
100 East Franklin Street, CB# 8115, Chapel Hill, NC, 27599,
|
|
Keywords:
|
weighting ; longitudinal analysis ; missing data ; nonresponse ; growth curve model
|
|
Abstract:
|
Panel studies that include unequal selection of individuals often also suffer from intermittent nonresponse. Probability weighting may be used to correct bias due to unequal inclusion of individuals and wave-specific nonresponse. Several weighting approaches were applied to models of change for repeated continuous outcomes with nonignorable missingness. Estimator properties for the weighting methods for models of change in the mixed (random effects) and marginal modeling frameworks were compared empirically using simulated data. Probability weighting that accounts for both the unequal selection of individuals into the study and intermittent dropout over time performs the best in a marginal model. Traditional panel weighting used with the sample of individuals observed at every time point also performs well with larger sample sizes compared to weighting observations in the mixed model.
|
