Authors
Abstract
To minimise non-response bias in survey-based estimates, both bias prevention measures during data collection and bias adjustment measures post-data collection are employed. We evaluate such measures using the UKHLS Covid-19 Study as a case study. The Covid-19 Study is a primarily web-based derivative of the annual UK Household Longitudinal Study. We find that telephone follow-ups of web non-respondents and non-response weighting helped to increase dataset quality, but inviting non-regular internet users to the web survey was of little benefit. We develop a statistical test of non-response weight performance. Inverse-probability non-response weights outperform simple calibration weights.