An increasing proportion of web surveys are completed on tablets and smartphones, even if the questionnaires are not designed for mobile completion. Earlier research has shown that differences in data quality between answers completed on different devices are often small. Most of these studies have however relied on experimentation on volunteer samples to compare differences in data quality across different devices. In this study, we use rich register data available from the sampling frame available to Statistics Netherlands, to look at data quality for general population surveys. Specifically, we will look at both selection and measurement effects across different devices for a number of Statistics Netherlands’ social surveys. In this paper, we compare desktop/laptop, tablet and smartphone respondents in terms of background variables, unit and item non-response, and a range of data quality indicators like percentage breakoffs, average completion times, question order effects, social desirability, straightlining, the length and rounding of open answers and percentage don’t know answers. We give recommendations for optimization of the designs over devices.