Use of machine learning techniques in non-probabilistic samples

Abstract

Non-probabilistic surveys are increasingly used because they are easy and cheap to carry out. Even official statistical agencies are starting to use this type of surveys in their research, due to the difficulty and the amount of resources needed to carry out probabilistic surveys, which are currently the best option due to their reliability. When non-probabilistic surveys are used, the classical estimation methods cannot be used since the initial conditions for carrying them out are not met, so over the years new estimation techniques have been emerging in this type of sampling. Some of the most relevant estimation techniques currently being used are those related to machine learning techniques.In this work we focus on the estimation technique for non-probabilistic samples statistical matching, which can be enhanced and improved if we complement it with a machine learning technique known as XGBoost. We are going to study a variable of interest extracted from a real non-probabilistic survey carried out during the COVID-19 pandemic, and check if by applying such estimations we obtain better results than without applying this type of techniques.