Several theoretical hypotheses are developed concerning the relation of question and respondent characteristics to the reliability of survey attitude measurement. To test these hypotheses, reliability is estimated for 96 survey attitude measures using data from five, 3-wave national reinterview surveys-three Michigan Election Panel Surveys and two reinterview studies conducted by the General Social Survey. As hypothesized, a number of attributes of questions are linked to estimated reliability. Attitude questions with more response options tended to have higher reliabilities, although there are some important exceptions. More extensive verbal labeling of numbered response options was found to be associated with higher reliability, but questions explicitly offering a “don't know” alternative were not found to be more reliable. Question characteristics were confounded to an unknown degree with topic differences of questions, which were significantly linked to reliability, leaving the influence of question characteristics on reliability somewhat ambiguous. Characteristics of respondents were also found to be related to levels of reliability. Older respondents and those with less schooling provided the least reliable attitude reports. These results are discussed within a general framework for the consideration of survey errors and their sources. © 1991, SAGE PUBLICATIONS. All rights reserved.
