Accurate Covid-19 prevalence measurement in the field

Georganas, S.; Velias, A.; Vandoros, S.

Accurate epidemic prevalence measurement is a necessary condition for informed policy decision-making. In the Covid-19 pandemic especially, wrong prevalence measurement can lead to tremendous waste, be that in life years or economic output. A number of countries offer random Covid-19 tests to estimate the prevalence of the virus in the population, and report daily positivity rates. However, since virus testing has to be voluntary, all tests done in the field, even if supposedly random, suffer from selection bias. This bias, unlike standard biases in polling, is not limited to having a representative sample, and thus cannot be corrected by the usual methods (quota sampling etc). The issue is that people who feel they have symptoms (or other reasons to suspect they are carrying the virus), are up to 38 times more likely to volunteer to get tested, and testing stations cannot readily correct this by oversampling (i.e. selecting people without symptoms to test). Using controlled, incentivized online experiments with over 500 subjects of all ages in a European country, we show that this difference in testing propensities leads to sizeable bias; 'random' tests in the field inflate infection figures by up to five times. We suggest ways to correct the bias of the testing stations, but even better, a cleaner way to sample the population to avoid the bias altogether. Our methodology is relevant for covid-19, but also any other epidemic where carriers can have informative beliefs about their own carrier status.