558 Panel Data Methods
12.3.2 Health outcomes 579
12.3.2.1 Self-reported data 579
12.3.2.2 Anthropometric measures 581
12.3.2.3 Biomarkers 582
12.3.3 Modeling costs and expenditure 584
12.4 Methods for dealing with unobserved heterogeneity
and dependence 585
12.4.1 Deviations and conditional estimates 585
12.4.1.1 Dynamic models 586
12.4.2 Numerical integration and classical simulation-based inference 587
12.4.3 Bayesian MCMC 588
12.4.4 Finite mixture models 589
12.4.4.1 Latent class models 589
12.4.4.2 Finite density estimators and discrete factor models 591
12.4.5 Copulas 592
12.5 Models for longitudinal data 593
12.5.1 Applications of linear models 593
12.5.1.1 Models for longitudinal and spatial panels 593
12.5.1.2 Dynamic panel data models: GMM estimators 594
12.5.2 Applications with categorical outcomes 595
12.5.2.1 Pooled and random effects specifications 595
12.5.2.2 GMM estimators 596
12.5.2.3 Finite mixture models 597
12.5.3 Applications with count data 597
12.5.3.1 Poisson/log-normal mixtures 597
12.5.3.2 Finite mixtures 599
12.5.4 Applications of quantile regression and other
semiparametric methods 601
12.6 Multiple equation models 602
12.6.1 Applications using MSL 602
12.6.2 Applications using Bayesian MCMC 603
12.6.3 Applications using finite mixtures 605
12.6.4 Applications using copulas 606
12.7 Evaluation of treatment effects 607
12.7.1 Matching 607
12.7.2 Regression discontinuity 609
12.7.3 Difference-in-differences 609
12.7.4 Instrumental variables 614
12.8 Future prospects 619
12.1 Introduction
A common thread that runs through this chapter is the “evaluation problem”: is
it possible to identify the impact of policies from empirical data? The focus of the
chapter is on individual-level longitudinal data, so consider an “outcome”yit, for
individualiat timet. The treatment effect of interest is:
TEit=it=yit^1 −y^0 it, (12.1)