Coverage and nonresponse errors in an individual register frame based Swiss telephone election study

In this paper we compare coverage bias with nonresponse bias in a telephone election survey. The sample was drawn from a register of individuals in Switzerland, which included basic socio demographic characteristics. Telephone numbers were subsequently matched using three steps: An automatic match by the Federal Statistical Office using listed numbers, a match by the survey agency using commercial sources, and numbers delivered by respondents following a postcard request. In addition, weanalyse coverage and nonresponse bias in associations between variables for the matched sample as well as respondents only and reasons for nonresponse. In the automatically matched sample the probability of being matched ranged between 70.5% (the divorced) and 93.3% (those aged 73 or over). Coverage bias can be slightly improved by investing in more matching efforts. Nonresponse bias has small effects on top of coverage bias, with the exception of older people, for whom telephone numbers were more easily matched, but who were much less likely to respond. Bias in associations from undercoverage and nonresponse depend on the variables analysed. Reasons given for nonresponse by different person-groups follow expected patterns.