562 Panel Data Methods
the VR program is reversed when selection bias is taken into account and there is
evidence of perverse cream-skimming, with those most likely to benefit being the
least likely to be selected by the program administrators.
A note on the scope of the chapter
This chapter takes the identification of treatment effects as its starting point and
concentrates on microeconometric methods that can be used with longitudinal
and other complex and multilevel datasets. Although the methods described in
the chapter are widely used throughout applied econometrics, the applications
reviewed here all relate to one specific area: health economics. The chapter follows
an earlier review of the literature on “health econometrics” (Jones, 2000) and con-
centrates on studies that have appeared as peer-reviewed publications from 2000
onwards. The emphasis is on applications that use health and health care as out-
comes. Less attention is devoted to the large number of studies of health-related
behaviors, such as diet, smoking, drinking and illicit drugs (see, e.g., Cowell,
2006; Deeet al.,2005; Forster and Jones, 2001; Harriset al.,2006; Terza, 2002;
Van Ours, 2006) and to those that investigate the impact of health, health care
and health insurance on labor market outcomes (see, e.g., Askildsenet al.,2005;
Auet al.,2005; Auld, 2002; Bradleyet al.,2005; Contoyannis and Rice, 2001;
Disneyet al.,2006; French, 2005; Hogelund and Holm, 2006; Morris, 2006, 2007;
Royalty and Abraham, 2006; Stewart, 2001; Van Ours, 2004) or labor outcomes for
health care professionals (see, e.g., Arulampalamet al.,2004; Frijterset al.,2006;
Holmas, 2002). The scope does not include studies that use econometric tech-
niques in the context of contingent valuation and discrete choice experiments,
where random effects models are often applied (see, e.g., Ryanet al.,2006); multi-
nomial models of the choice of insurance plans or health care providers (see, e.g.,
Deb and Trivedi, 2006; Ho, 2006; Sahnet al.,2003); productivity analysis based
on models of cost and production functions and estimation of stochastic frontier
models (see, e.g., Bradfordet al.,2001; Burgess, 2006; Dranove and Lindrooth,
2003; Smith and Street, 2005; Wilson and Carey, 2004); and in the context of cost-
benefit and cost-effectiveness analysis, where econometric methods are starting to
be used alongside methods from biostatistics and epidemiology (see, e.g., Briggs,
2006; Hochet al.,2002; Willanet al.,2004).
The focus is primarily on studies that use micro-level data derived from longi-
tudinal, multilevel and other complex data structures. Relatively few cross-section
studies are discussed and the chapter does not attempt to review studies that use
aggregate time series or panels and that apply pure time series methods (see, e.g.,
Aakvik and Holmas, 2006; Abadie and Gay, 2006; Chou, 2007; García-Ferreret al.,
2007; Leigh and Jencks, 2007; Oret al.,2005; Paton, 2002; Ruhm, 2003; Wang and
Rettenmaier, 2007). Analysis of longitudinal data often makes use of the meth-
ods of survival analysis (see, e.g., Arulampalamet al.,2004; Chou, 2002; Disney
et al.,2006; Farsi and Ridder, 2006; Forster and Jones, 2001; Frijterset al.,2006;
Harrison, 2007; Holmas, 2002; Kyle, 2007; Piconeet al.,2003a; Stewart, 2001; Van
Ours, 2004, 2006), but these methods are not discussed in detail here.