Abstract

Survey non-response is a relevant but under-researched topic in a pandemic setting, which merits investigation for the unexpected consequences it can produce on the reliability of the survey results. In this paper we investigate non-response error, using data collected at the turn of the pandemic on a known sample of 5,605 individuals first involved in the Italian Lives panel (ITA.LI) and then invited to join an ad hoc survey on the health crisis (ITA.LI Covid-19). We compare respondents and non-respondents to ITA.LI Covid-19, using the multivariate logistic regression technique to estimate variations in the probability of taking part in the survey attributable to specific characteristics of the sample members. Our results highlight that highly-educated people, those who live alone, and people living in municipalities experiencing a higher growth of mortality rates after the pandemic outbreak are more likely to take part in the survey, while full-time workers are still less likely to participate. Based on these results, we formulate general recommendations on how to minimise non-response error when sampling frame data are available, both at the survey design stage and at the data management stage