Subscale Distance and Item Clustering E ects in Surveys: A New Metric

We explore the e ect of a form of question context on responses to a computer-mediated marketing research survey. As an increasing proportion of marketing research is conducted through computer interfaces, the pool of potential context e ects is rapidly expanding. We conduct an experiment using a multi-item scale that consists of ve dimensions and manipulate three such context e ects: whether an item is explicitly labeled according to subscale (labeled), whether an item is presented in isolation or with other items on the screen (grouped), and, of principal interest, whether all the items from a subscale are presented contiguously or not (clustered). An initial analysis using analysis of variance (ANOVA) indicates a signi cant reduction in variation of item scores within a subscale (subscale variance), and hence an increase in reliability, due to each of labeling, grouping, and clustering. In a more re ned analysis, we use a special onedimensional case of the spatial and attribute based distance metric proposed in Hoch, Bradlow, and Wansink (1999) to explain subscale variance, replacing the indicator variable for clustering used in the ANOVA. This metric provides a scalar measure of how much variation exists in the order of presentation of items within a subscale (the subscale distance). This analysis indicates a signi cant decrease in subscale variance (increased reliability) with decreasing subscale distance, but no longer signi cant e ects due to labeling and grouping. Implications for researchers conducting surveys in computer-mediated environments are discussed.