Observation Errors in Sociological Surveys: a Model and a Method

Abstract

In sociological surveys, respondents occasionally give false answers to factual questions which inquire if the respondent has some attribute. The social scientist may wish to correct estimates of the proportion of respondents having the attribute or he may wish to know the amount of error which is occurring in his survey. These two objectives can be achieved by a probabilistic model which assumes (1) that respondents have some probability of giving false information, and (2) that if respondents are queried about the attribute on two or more occasions, among those with the attribute, there is statistical independence in their responses, as is also the case for those without the attribute. With further simplifying assumptions, the social scientist can estimate the probability ofgiving false information and the true proportion of respondents with the attribute.