A Methodology to Answer to Individual Queries: Finding Relevant and Robust Controls

Abstract

In this paper we present the guidelines for an estimation and inference method aimed to answer questions to databases about new potential entries. The idea is to classify the data and find entries similar to that about which the query is posed, and use it to generate control observations as to find the results in terms of them, instead of the entire database. This amounts to a trade-off between robustness and relevance. This allows inferring an adequate answer to a query. While a solution exists,it is not constructive and requires to explore possible alternative definitions of similarity as well as distributions that fit to the selected data. We illustrate these ideas with six cases in which the answer obtains in a linear regression.