Andrew M. Jones 619is then:
LIV=[
∂E(y|x,p(x,z))
∂p(x,z)]1 −p(x,z)=ud. (12.41)
This can be used to test for heterogeneity in the treatment effect and to construct
estimates of the other treatment effects of interest, such as the ATE and ATET.
12.8 Future prospects
Like other areas that are rooted in applied microeconomics, such as develop-
ment, environmental and labor economics, modern empirical analysis in health
economics is dominated by the tools of microeconometrics and the use of
individual-level data drawn from social surveys and administrative sources. The
trend is towards complex longitudinal and multilevel data structures, with increas-
ing reliance on linkage of a variety of sources. Health economists have exploited the
full range of microeconometric techniques, and applications to health data have
driven methodological innovations in the context of variables with skewed and
heavy-tailed distributions, multinomial choices, count data and mixture models.
Much of the empirical analysis done by health economists seeks to estimate
causal effects and fits within the treatment–outcomes framework. A challenge for
successful applied work is to find appropriate sources of variation to identify the
treatment effects of interest. Estimation of causal effects can be prone to selec-
tion bias, when the assignment to treatments is associated with the potential
outcomes of the treatment. Overcoming this selection bias requires variation in
the assignment of treatments that is independent of the outcomes. One source of
independent variation comes from randomized controlled experiments. These are
the norm in the evaluation of new clinical therapies, but their use for the evaluation
of broader health and social programs remains relatively rare.
Applied researchers in health economics face a twin challenge. The first is to make
the best possible use of the available non-experimental data by combining robust
econometric methods, such as those presented in this chapter, with imaginative
and convincing sources of identification. The second is to seek opportunities to
encourage the agencies responsible for designing and funding health and social
programs to collect new and comprehensive longitudinal datasets, to facilitate the
linkage of different datasets and to make greater use of randomized designs in the
evaluation of new initiatives.
Acknowledgments
This chapter draws on joint work with Teresa Bago d’Uva, Silvia Balia, Paul Contoyannis,
Cristina Hernández Quevedo, Xander Koolman, José Labeaga, Miaw-chwen Lee, Roberto
Leon Gonzalez, Nigel Rice, Stefanie Schurer, Eddy van Doorslaer and John Wildman, and by
Casey Quinn. I am grateful for comments from Giorgia Marini, Edward Norton, Pedro Rosa
Dias and João Santos Silva.