Estimation by kernel weighting of parameters related to employment in the confinement period

Abstract

During the period of COVID-19's confinement, new working methods that were not normally used began to be relevant. This is the case of teleworking or the use of new techniques for conducting surveys. The gold standard for carrying out surveys is probability sampling based on face-to-face interviews, but due to this situation of social isolation, non-probabilistic methods, such as online or web surveys, began to be used. However, in order to make reliable estimates from non-probability samples we must use special techniques to reduce the bias that appears in them. In this paper we will study a technique for bias reduction in non-probabilistic surveys that stands out for its promising results, known as Kernel Weighting. It requires a probabilistic sample as auxiliary information, and its performance can be improved using Machine Learning techniques, such as regularised logistic regression. We will use a non-probabilistic survey focused on studying the employment situation of the Spanish population during COVID-19, and as probabilistic survey the CIS Barometer of May 2020. We will compare the new estimates with those obtained in the original survey, observing important differences.