Andrew M. Jones 581A similar identification strategy to Lindeboom and Van Doorslaer (2004), that
relies on objective measures capturing all of the genuine variation in health, is
adopted by Etilé and Milcent (2006), who estimate generalized ordered probit
models with a four-category measure of SAH. They construct synthetic measures of
objective health from a latent class analysis of a set of self-reported indicators, such
as activities of daily living (ADLs) and body mass index (BMI). The latent class anal-
ysis is used to condense the sample into six classes and indicators for these classes
are included in the ordered probit model. This works by constructing classes such
that the underlying health indicators are independent of each other, conditional
on class membership. Latent class models are presented as an alternative to the
grade of membership approach that has been used in some earlier work (see, e.g.,
Lindeboomet al.,2002). Estimates on a sample of 2,956 individuals aged under
65 from the French Enquête Permanente sur les Conditions de Vie des Menages
(EPCV) survey show evidence of reporting bias and, unlike Lindeboom and Van
Doorslaer (2004), there is evidence that this is related to income. Etilé and Milcent’s
(2006) findings suggest a concave relationship between health and income in terms
of health production and convexity with respect to reporting, with overoptimism
among the rich and overpessimism among the poor. They conclude that the prob-
lems of reporting bias can be minimized by collapsing the four-point scale into a
binary measure of poor health.
Jurges (2007) focuses on cross-country differences in reporting of SAH as mea-
sured for those aged over 50 in the ten countries covered by the first wave of
the Survey of Health, Ageing and Retirement in Europe (SHARE). Generalized
ordered probit models are used to regress SAH on a set of objective measures,
such as grip strength, walking speed and BMI, to get a set of disability weights.
The average thresholds across the SHARE countries are then used to reclassify the
reported data, assuming that the disability weights are constant across countries.
The variation in SAH is decomposed into the component that is explained by the
objective measures and the component attributed to reporting bias. The findings
suggest that those in the Danish and Swedish samples overrate their health, while
those in Germany underrate their health. For Austria and Greece there is little
bias.
12.3.2.2 Anthropometric measures
Anthropometric measures have long played a role in studies of developing coun-
tries, especially those focused on child health issues. With the growing problem
of adult and childhood obesity in more affluent nations, they are increasingly
being used in that context as well. Typical anthropometric measures are height
and weight, which may be self-reported or measured by a professional; infant
length for children aged under two; demi-span, which is based on the length of an
outstretched arm and is used among older populations who may have difficulties
standing straight; the waist-to-hip ratio; and BMI, which is the most commonly
used indicator of obesity. BMI is calculated as weight in kilograms divided by height
in meters squared, with a BMI of 30 or greater indicating obesity and 25–30 indi-
cating overweight. The confounding effect of levels of muscle development means