Andrew M. Jones 575The idea of an anti-test may provide a useful strategy as part of a robust-
ness/sensitivity analysis. A good example of this is Galianiet al.’s (2005) evaluation
of the impact of the privatization of local water services on child mortality in
Argentina. They adopt two strategies for assessing the reliability of their difference-
in-differences approach that can both be interpreted as anti- or placebo tests. The
first, which is a good practice to adopt in any difference-in-differences analysis,
is to estimate a placebo regression: the model of interest is estimated using only
data from the pre-treatment period, but including an indicator of those cases that
will go on to be treated. If this indicator of hypothetical treatment is significant
it is a sign that the treated and controls are not comparable and that the “parallel
trends” assumption required for difference-in-differences analysis is not valid. The
second strategy adopted by Galianiet al.(2005) is that, as well as measuring deaths
from infectious and parasitic diseases, they include measures of deaths from causes
unrelated to water quality. The fact that they detect a reduction for the former but
not for the latter creates confidence in their difference-in-differences identification
strategy.
12.3 Data and measurement issues
12.3.1 Administrative data or sample surveys
Much of the applied work done by health economists uses social surveys. These
are often designed to provide representative random samples of the underlying
population. Most often the sampling follows a multi-stage design with clustered
and/or stratified sampling (see, e.g., Joneset al.,2007b). Data may be collected
by face-to-face interviews or postal, telephone or web-based questionnaires, and
in health surveys this is often supplemented by clinical tests and measurements.
Many surveys are one-off cross-sections, but increasingly researchers have turned
to longitudinal, or panel, surveys which give repeated observations on the units of
interest, whether they be individuals, households or organizations. Sample surveys
are the mainstay of microeconometric research and some of the more popular
datasets are summarized in Table 12.1.
In health economics, administrative datasets often prove more useful and reliable
than social surveys. Administrative datasets include sources, such as tax records,
reimbursement and claims databases, and population registers of births, deaths,
cancer cases, HIV/AIDS cases, unemployment, etc. (see, e.g., Aakviket al.,2003;
Aakviket al.,2005; Atellaet al.,2006; Blacket al.,2007; Chalkley and Tilley, 2006;
Dano, 2005; Dranoveet al.,2003; Dusheikoet al.,2004, 2006, 2007; Farsi and
Ridder, 2006; Gravelleet al.,2003; Ho, 2002; Lee and Jones, 2004, 2006; Marini
et al.,2008; Martinet al.,2007; Propperet al.,2002, 2004, 2005; Riceet al.,2000;
Seshamani and Gray, 2004). These datasets are collected primarily for administra-
tive purposes and are made available to researchers for secondary analysis. Some
countries allow comprehensive linkage of different sources of administrative data
based on personal identification numbers (see, e.g., Blacket al.,2007). Adminis-
trative datasets are typically large, often with millions rather than thousands of