ASSESSING AND COMPENSATING FOR SELF-SELECTION BIAS (NONREPRESENTATIVENESS) OF THE FAMILY RESEARCH SAMPLE

Even probability samples may not be "representative," because self-selection bias, especially with regard to non-locatable subjects and refusals, may distort the statistical findings from being unbiased estimates of the targeted population's values. To help assess the extent of this bias, family researchers can obtain and make use of a range of "plausibly correlated characteristics" (PCCs), such as marriage and divorce public records, that are typically available on nonrespondents and respondents alike. If analyses show that the respondent and nonrespondent groups differ on one or more of these PCCs, they can be used again to statistically "compensate " for the bias found in a recalculation of the study's findings. The compensation techniques most useful are weighting and hazard rate models. The article provides an extensive case example of these analyses.