Interviewers‘ scope for influence on random sampling. A comparative analysis of ALLBUS surveys

Abstract

Looking for hidden bias in random samples, we re-analyzed 12 ALLBUS Surveys (Germany‘s GSS, 1980-2000). In these surveys people had been selected by varying procedures: the interviewers had been allowed some amount of scope for own activities and thereby (hypothesis) induced bias of differing severeness.

The sampling bias is measured by means of internal criteria. By definition, the total population as well as the random sample are restricted to the end that known population parameters arise, in this case a total population of couples (one men, one women, same household, at least one child up to 14 years old), where the male person is working full-time outside the household and the female is not. As a result of these restrictions, an unbiased random sample is expected to consist of 50% women exactly, in spite of them being more easily accessed than their male partners.

Comparing the results of existing random samples (ALLBUS) to the expected values, the relative frequency of women differs systematically: in cases of sampling procedures prescribing the interviewers behavior in detail, we found a significant higher number of males (difficult to access) compared to samples which allowed more freedom