Measurement bias due to response styles: a structural equation model assessing the effects of modes of data-collection

Abstract

This paper validates measures of response styles as latent variables using structural equation modeling. Next to measurement validation the main objective is to assess whether different modes of data collection bring along measurement bias due to response styles. Results indicate that Internet panel and telephone survey respondents do not show a higher yeah-saying tendency than do people responding to a postal mail survey. Participants in web panel surveys also use the range of rating scales similarly compared to postal mail participants. Telephone survey respondents used a wider range of rating scale options. This may be due to primacy and recency effects of the response options. Internet pop-up surveys seem to lead to more yeah-saying, while respondents also use a narrower range of the rating scale.