Methodologies now commonly used for the construction of poverty maps assume a substantial
degree of homogeneity within geographical areas in the relationship between income and its predictors.
However, local labor and rental markets and other local environmental differences are likely
to generate heterogeneity in such relationships, at least to some extent. The purpose of this paper
is to argue that useful if only indirect and suggestive evidence on the extent of area heterogeneity
is readily available in virtually any census. Such indirect evidence is provided by non-monetary
indicators—such as literacy, asset ownership or access to sanitation—which are routinely included
in censuses. These indicators can be used to perform validation exercises to gauge the extent of
heterogeneity in their distribution conditional on predictors analogous to those commonly used in
poverty mapping. We argue that the same factors which are likely to generate area heterogeneity
in poverty mapping are also likely to generate heterogeneity in such kind of validation exercises.
We construct a very simple model to illustrate this point formally. Finally, we evaluate empirically
the argument using data from Mexico. In our empirical illustrations, the performance of
imputation methodologies to construct maps of indicators typically feasible with census data alone
is indeed informative about how effectively such methodologies can produce correct inference in
poverty mapping