Unit nonresponse biases in estimates of SARS-CoV-2 prevalence

Julia Post; Fabian Class; Ulrich Kohler

doi:10.18148/srm/2020.v14i2.7755

Julia C. Post Faculty for Economic and Social Research, University of Potsdam
Fabian Class Faculty for Economic and Social Research, University of Potsdam
Ulrich Kohler Faculty for Economic and Social Research, University of Potsdam

DOI: https://doi.org/10.18148/srm/2020.v14i2.7755

Keywords: COVID-19, prevalence, probability samples, unit nonresponse, conservative confidence limits, nonresponse bias

Abstract

Since COVID-19 became a pandemic, many studies are being conducted to get a better understanding of the disease itself and its spread. One crucial indicator is the prevalence of SARS-CoV-2 infections. Since this measure is an important foundation for political decisions, its estimate must be reliable and unbiased. This paper presents reasons for biases in prevalence estimates due to unit nonresponse in typical studies. Since it is difficult to avoid bias in situations with mostly unknown nonresponse mechanisms, we propose the maximum amount of bias as one measure to assess the uncertainty due to nonresponse. An interactive web application is presented that calculates the limits of such a conservative unit nonresponse confidence interval (CUNCI).