Calibration and XGBoost reweighting to reduce coverage and non-response biases in overlapping panel surveys: Application to the Healthcare and Social Survey

Abstract

Healthcare statistical services worldwide have used probability surveys to provide information on the social, economic and health impact of COVID-19, or its seroprevalence and evolution, or the characteristics of the infected population. The Healthcare and Social Survey (ESSA, Spanish acronym) arises from the need to provide data on the evolution of the COVID-19 impact which can be considered when making decisions, so as to prepare and deliver an effective Public Health response in the different populations concerned. This survey has an overlapping panel survey design with measurements throughout 2020 and 2021, and random samplings stratified by province and degree of urbanization. Each ESSA measurement comprises two samples: a longitudinal sample taken from previous measurements and a new sample taken at each measurement. This design allows longitudinal estimates and more accurate cross-sectional estimates to be obtained thanks to the larger sample size. However, the problem of non-response is particularly aggravated in the case of panel surveys due to population fatigue with repeated surveys. The objective of this research article is to develop a new reweighting method for overlapping panel surveys affected by non-response. Considering the design, timing and objectives of this survey, our reweighting methodological approach produces suitable estimators for both cross-sectional and longitudinal samples. The weights are the result of a two-step process: the original sampling design weights are corrected by modelling non-response with respect to the longitudinal sample obtained in a previous measurement using machine learning techniques, followed by calibration using the auxiliary information available at the population level. The proposed method is applied to the estimation of totals, proportions, ratios, and differences between measurements, and to gender gaps in the variable of self-perceived general health. For addressing future health crises such as COVID-19, it is therefore necessary to reduce potential coverage and non-response biases in surveys by means of utilizing reweighting techniques as proposed in this study.