Middle Alternatives, Acquiescence, and the Quality of Questionnaire Data

Some scholars have suggested that offering a middle alternative on a rating scale is necessary to measure opinions accurately, whereas other scholars have suggested that middle alternatives offer "easy outs" to respondents who want to avoid taking sides on an issue. In this paper, we evaluate these competing hypotheses using data from an experiment conducted in the 1992 Euro-Barometer Survey. Via structural equation modeling of responses to agree/disagree items measuring attitudes toward science and technology, we found that offering a middle alternative reduces the amount of random measurement error in the responses, thereby increasing reliability, while not affecting the validity of attitude measurements. This suggests that middle alternatives should be included in rating scales in order to maximize data quality. We also found evidence of acquiescence response bias in answers to the agree/disagree items; while unrelated to the presence of a middle alternative, this bias was stronger among older, less educated, and female respondents. And controlling for this bias greatly improved the apparent validity of attitude items.