Andrew M. Jones 563The empirical findings of many of the studies are discussed, but no attempt is
made to provide a systematic synthesis of the empirical results. It is notable that
meta-analyses of regression results are beginning to appear in the health economics
literature. For example, Gallet and List (2003) present a meta-analysis of the tobacco
price elasticity and Gemmillet al.(2007) carry out a meta-regression of estimates
of the price elasticity of prescription drugs.
On the whole, the original sources for the econometric methods are not cited.
These are reviewed in Volume 1 of this Handbook (particularly the chapters by
Badi Baltagi, William Greene and Lung-fei Lee) and in other chapters in this second
volume, in particular those by Colin Cameron, William Greene, and David Jacho-
Chávez and Pravin Trivedi.
12.2 Identification strategies: finding relevant variation
The success of applied work depends on finding appropriate sources of variation
to identify the effects of interest. Estimation of treatment effects can be prone to
selection bias, where the assignment to treatments is associated with the poten-
tial 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. While
these are the norm in the evaluation of new clinical therapies, their use for the eval-
uation of social programs remains rare (Gertler, 2004; Kremer, 2003; Miguel and
Kremer, 2004). Most economic studies have to draw on non-experimental, or obser-
vational, data. This section presents a series of case studies from the recent literature
and describes how these have sought out relevant identifying information.
12.2.1 Randomized experiments
The “gold standard” methodology that is used to identify the efficacy and effec-
tiveness of new medical technologies is the randomized clinical trial (RCT). Much
of the work done by health economists to measure the cost-effectiveness of these
technologies draws on data collected within RCTs to perform statistical analyses
(Briggs, 2006) or to calibrate decision analytic models (Claxtonet al.,2006). Econo-
metric methods are sometimes used in secondary analysis of such data to model
costs and outcomes as functions of observable covariates (Willanet al.,2004).
Broader randomized social experiments are far less prevalent. One exception,
which has played a highly influential role in the development of health economics
and has driven many of the early developments in the use of econometrics in the
field, is the RAND Health Insurance Experiment (Manninget al.,1987). The RAND
experiment was designed to address the problem of self-selection in the choice
of insurance plans. Participants were randomized between a health maintenance
organization (HMO) and reimbursement plans and across plans with different
levels of co-payments and a plan with deductibles. The RAND study has had a
strong influence, especially in the US. It has focused attention on the use of two-
part or multi-part models to model health care utilization and expenditures and on
the choice of functional form to deal with heavily skewed data and the consequent