19
rie consumption (expenditures or intakes) with
respect to percentage changes in income de-
pends very much on what the original level of
calorie consumption is, and what populations
are included in the sample. How the sample is
segmented is, of course, critical. That is, for
most food security policy, researchers' only
interest is in the elasticity estimates for specific
groups (i.e., those which are [most] calorie defi-
cit). Establishing exactly who are included in
these groups, however, is very difficult.
One result of this problem is that valid com-
parisons among data sets, or generalizations of
findings, are not possible unless specific infor-
mation identifying a household's income level,
landholding size, place of residence (esp. urban
versus rural), or other factors that explain the
varying relationship between income and con-
sumption, is available and controlled for. At the
very least, the initial income or calorie adequacy
levels of households need to be known and
accounted for before meaningful interhousehold
or intersample comparisons regarding expendi-
ture habits and consumption linkages can be
inferred from elasticity estimates. Using the
examples from the previous section, for ex-
ample, rather than implying that access and
consumption are more closely linked in Ghana,
Rwanda, and The Gambia, than in Kenya, the
lower consumption elasticity in Kenya is more
likely to be the result of households in the
Kenyan sample having higher incomes and calo-
rie adequacy than those in the Ghanaian,
Rwandan, and Gambian surveys.
Aggegating and averaging data is also a prob-
lem. Often these "elasticity" studies draw infer-
ences from comparisons of elasticities estimated
from mean levels of income, caloric intake, farm
size, etc. An obvious limitation of this approach
is that such studies tell little about those at the
lowest income (or food consumption) levels. In
fact, depending on the income levels of the
households in the sample, it would not be sur-
prising to find an income elasticity of calorie
consumption at the mean income level of the
sample to be nearly zero, while the elasticity for
the poorest households might be nearly one. As
Ravallion (1990) argues, "it is clear that in de-
veloping countries we are far more concerned
about changes in calorie intake for people whom
we deem to be undernourished than for those
who are not. And for those who are poorly
nourished, one can rightly be more concerned
about those who are a long way from an ad-
equate intake than those who are quite close to
it."
One solution to this problem has been to
divide households according to income levels--
for instance, into income quintiles. Another
possibly useful approach would be to only in-
clude households with intakes below some cer-
tain consumption cutoff point. Also, other dis-
aggregations of data, besides income, may be
very important for correctly understanding and
interpreting elasticities of food demand. For
instance, results may differ by location. In par-
ticular, income elasticities of calorie intake have
been found to be greater in rural areas than in
urban areas (Ravallion 1990). And the rate at
which these elasticities decline as incomes grow
is greater in urban areas than in rural areas (von
Braun et al. 1993).
An even more fundamental and important
criticism of the use of elasticity of food demand
estimates to assess the linkages between in-
come and food consumption is raised in
Ravallion (1990) and Anand and Ravallion
(1993). They point out that the responsiveness
of food intake to changes in income, and the
responsiveness of food adequacy to changes in
income, are not the same. For example, a low
income elasticity of nutrient intake does not
necessarily imply that aggregate undernutrition
(as measured by a "headcount" index) is unre-
sponsive to income.
This distinction between the responsiveness
of food intakes and food adequacy to income
changes would be especially evident in cases
where a large proportion of the sample popula-
tion is consuming food at or near the minimum
requirement levels. As Ravallion (1990) de-
scribes, "the marginal effect of a change in the