Several authors have claimed that turnout differences between social groups are generally small – at least outside the USA. The evidence they cite in support is, however, not unproblematic since turnout differences are assessed on the basis of surveys in which people recall their own behavior. In this paper, we analyze a survey for the German general election of 2013, which includes information not only about actual respondents but also about non-respondents. For each randomly drawn address, interviewers had to provide an assessment of the surrounding neighborhood and to rate the quality of the neighborhood – even if they failed to realize an interview. This new data allows us to assess the response rate in high- and low-turnout neighborhoods, which helps to estimate real turnout differences.
