Missing dependent variables in fixed-effects models

• Main Author: Abrevaya, J. (Author)
• Format: Article
• Language: English
• Published: Elsevier Ltd 2019
• Subjects: Classical minimum distance; Dependent variables; Estimation approaches; Fixed effects; Fixed effects models; Intelligent systems; Linear projection; Linear projections; Minimum distance; Missing at randoms; Missing data; Monte Carlo methods
• Online Access: View Fulltext in Publisher

 This paper considers estimation of linear fixed-effects models in which the dependent variable may be missing. For cross-sectional units with dependent variables missing, use of covariate information from all time periods can provide efficiency gains relative to complete-data methods. A classical minimum distance (CMD) estimator based upon Chamberlain (1982, 1984), which is consistent under a missing-at-random (MAR) type assumption, is proposed for the static fixed-effects model. In certain circumstances, it is shown that “within” variation in the dependent variable is not even required for identification of the model parameters. The CMD estimation approach is extended to the case of (autoregressive) fixed-effects models with lagged dependent variables. Monte Carlo simulations investigate the performance of the CMD approach relative to existing methods. Extensions to models with sequential exogeneity and missing covariates are also discussed. © 2018 Elsevier B.V.