We introduce a spatially hierarchical factor model, or a vulnerability index, to measure dengue fever in Uruguay. Our proposal combines spatial information among different municipalities across the region (large scale information) with census tracts information at the municipality level (small scale information). Aedes aegypti, the main dengue fever transmitting vector, was reintroduced in Uruguay in 1997 with no cases of the disease been registered up to this point in time. It is of great importance to point the regions of the country which are vulnerable to the reintroduction of the disease. It is common to observe, in social studies, social indices built based on sets of indicators observed on census tract level of municipalities across the countries. In our sample the number of census tracts vary significantly, ranging from as low as 16 (Bella Union) up to 1012 (Montevideo) tracts. A simple comparison with a benchmark procedure, one which aggregates observations at the municipal level before building the index, suggests that our vulnerability index is more sensitive to local characteristics and, therefore, more capable of capturing subtle differences across regions of the country. Our factor model entertains point referenced data at the country level and areal data within municipalities. We expect that within a municipality, census tracts which are close together, tend to have similar values of the variables, and behaving on a more independently fashion if the tracts are far apart. On the other hand, in the country level, one expects that index values vary smoothly across the municipalities. More specifically, we combine the information available on p variables measured at n_i, (i=1,...,n) census tracts across n municipalities. The municipality size (number of tracts) is taken into account and provide a tool of weighing the contribution of a variable (according to its location) to the overall vulnerability index. Moreover, different from standard procedures, independence across locations is not imposed.
This is joint work with Hedibert F. Lopes (GSB-UC), Esther Salazar (IM-UFRJ), Mariana Gómez (Universidad de la Republica, Uruguay) and Marcel Achkar (Universidad de la Republica, Uruguay).