Survey data and official statistics should reflect the distribution of demographic and socio-structural characteristics of a population as correctly as possible. However, due to the selective dropout of specific population groups, biases occur for survey data that, in the past, could often be demonstrated for various surveys by univariate comparisons with official data. In particular, selective dropouts were proven for small households and for members of a low social stratum (so-called middle-class bias). These deviations could be reproduced for the ALLBUS social survey (1986, 88, 90) in comparison to the microcensus (1985, 87, 89). The bias with regard to household size has especially significant consequences, as it prevents a representative weighting in terms of individuals for the ALLBUS household sample. In the context of social stratification, multivariate analysis managed to show that dropouts were primarily explicable as effects of educational qualification and participation in employment. Since the distribution differences between the ALLBUS and microcensus in terms of occupational status as employed and unemployed can be arranged on the basis of a dichotomy without any substantial information loss, the so-called middle-class bias proves to be a bias in education.
Hartmann,P. & Schimpl-Neimanns, B., 1992: Zur Repräsentativität soziodemographischer Merkmale des Allbus - Multivariate Analysen zum Mittelschichtbias der Umfrageforschung. ZUMA Arbeitsbericht 92/01 [.pdf] (2.13 MB).
Hartmann, P. & Schimpl-Neimanns, B., 1992: Sind Sozialstrukturanalysen mit Umfragedaten möglich? Analysen zur Repräsentativität einer Sozialforschungsumfrage. Kölner Zeitschrift für Soziologie und Sozialpsychologie 44 (2): 315-340 [.pdf].
Hartmann, P. & Schimpl-Neimanns, B., 1993: Affirmative Repräsentativitäts"beweise" oder Test konkreter Hypothesen zu Verteilungsabweichungen? Kölner Zeitschrift für Soziologie und Sozialpsychologie 45 (2): 359-365.
Hartmann, P. & Schimpl-Neimanns, B., 1994: Zum Sinn von Verteilungsvergleichen. Allgemeines Statistisches Archiv 78 (3): 338-339.
