Abstract

In surveys, samples are chosen to be representative for a given population. In case we are interested in a particular subpopulation, in general it is not valid to work with the subsample obtained by limiting the original sample to this subpopulation as there is no guarantee that the subsample will be representative for the subpopulation of interest.

In this work, we present a novel approach to the problem of estimating population
variables of a survey for non-representative groups. Our approach relies on learning a distribution regression model from the representative distribution (of the group) of some variables in another related survey to the variable of interest. Using the aforementioned model, a rough estimate of the variable of interest is obtained for different subgroups of the sample. We apply our methodology to estimate the well-being of the Mexican rural and urban population in a separate way using data from the Household Income and Expenditure survey. In 2021, the National Institute of Statistics and Geography (INEGI) conducted the National of Survey of Self-Reported Well-being
(ENBIARE), but the results are only representative at the National and State level. Using the distribution of specific variables from the Household Income and Expenditure survey (ENIGH), a distribution regression model to the subjective well-being is estimated. The model is used to estimate and compare the well-being of the rural and urban population