A nonparametric approach to the sample selection problem in survey data

Responses to economic surveys are usually noisy. Item non-response, as a particular type of censored data, is a common problem for key economic variables such as income and earnings, consumption or accumulated assets. If such non-response is non-random, the consequence can be a bias in the results of studies that try to understand the economic behavior of the population. This doctoral thesis models the problem of survey non-response by means of nonparametric techniques which allow for much weaker assumption than the parametric and semiparametric alternatives. The basis for these techniques consist on estimating a bounding interval around the parameter of interest, where the distance between bounds is determined by the selection problem (the non-response). Although the approach is flexible in that it allows for any type of selective response, the drawback is that it leads to an increase of the uncertainty of estimates of the parameters of interest. Nevertheless, and as illustrated by the empirical sections in this thesis, the use of intervals that identify a possible region for the unknown parameter may also be very useful at either studying the characteristics of the population or to test for economic hypotheses of interest.