Middle alternatives and measurement validity: a recommendation for survey researchers

Abstract

ABSTRACTQuestionnaires routinely measure unipolar and bipolar constructs using rating scales. Such rating scales can offer odd numbers of points, meaning that they have explicit middle alternatives, or they can offer even numbers of points, omitting the middle alternative. By examining four types of questions in six national or regional telephone surveys, this paper found that omitting the middle alternative and forcing respondents to pick a side compromised data quality, for questions presenting two conditions with middle option ‘about equally,’ and for quasi-bipolar questions with middle option ‘fair.’ In addition, omitting the middle alternative never helped improve data quality for a variety of unipolar questions. Lastly, there is a marginally significant improvement in terms of data quality for bipolar questions with middle option ‘neither good nor bad’ when offering a follow-up asking whether they lean one way or the other or whether they are genuinely in the middle.