Effects of Rating Scale Polarity on Attitude Measurements with Latent Variables

Beside the item content, rating scales represent a central data base for survey measures. Attitudes are often measured by a certain number of items thought to present a latent concept. This paper focuses on the polarity of rating scales and its impact on the attitude measurements with latent variables. With respect to scale polarity, bipolar and unipolar rating scales are distinguished. Bipolar scales reflect two opposing alternatives (e.g. agree-disagree) with a conceptual zero as midpoint (neither/nor). Unipolar scales (e.g. level of importance; do not agree –agree) reflect varying levels of the same dimension with the middle category presenting the conceptual dimension’s midpoint (e.g. moderately). In surveys so called “mixed” rating scales are often used, as for instance a unipolar rating scale with a midpoint that reflects a conceptual zero point.

The present study addresses the question how attitude measures with latent variables are affected by unipolar, bipolar and mixed rating scales. A 2x2 randomized experimental design was implemented varying polarity (unipolar vs. bipolar) and middle category (matching vs. non-matching to scale polarity). Different attitude constructs were measured using these rating scales. The effect of rating scale formats on item loadings, error terms and dimensionality of items were assessed using Confirmatory Factor Analyses (CFAs). The effects were also modeled and tested with help of structural equation modeling (SEM) taking response sets and certain personal characteristics into account. The participants were 522 members of GESIS online access panel representing a probability sample of German residents. The results show an effect of experimental manipulation on item loadings and error term variances. The effect of experimental manipulation on attitude measurements was mediated by response sets. The results are discussed in terms of their applicability for surveyors and researchers.