AN ANALYSIS OF BIAS IN SURVEY RESEARCH

In: Public opinion quarterly: journal of the American Association for Public Opinion Research, Band 26, Heft 1, S. 102-111

Abstract

Some form of bias is present in most survey res. The presence of such bias, however, while strongly affecting the marginal distribution of responses may not have any effect upon the interrelationship of variables. In an attempt to examine the effect of biased sampling upon interrelationships, comparisons are made between R's & non-R's on panel surveys. While the R's are found to differ from the non-R's on such variables as educ & interest, a comparison of the interrelationship between educ, interest & other variables is found to be the same for the R & non-R samples. In general, where variables are independently related to the source of bias, the use of the biased or unbiased data will show the same interrelationships between these variables & other variables. However, if the bias is differentially distributed,eg over-or-under-representation of any 1 cell or any pair of diagonal cells, then the test of the hypothesis may be affected. A similar analysis can also be made for question bias; whereas a biased question will affect the marginal distribution of responses, it may not affect the direction of the relationship of this question to other questions. The suggested model for the analysis of bias offered is based upon the general model for the analysis of intervening variables, with the bias being viewed as a manifest index of some meaningful underlying variable. The use of this model permits one to analyze biased samples, instruments, etc by controlling the bias as an intervening variable. It even suggests the possibility of making use of deliberately biased samples & instruments in special cases. AA.