Test of the effect of scale labels on response linearity

served that many researchers spend a lot of time constructing questionnaire items that ensure maximal content and face validity while they overlook the fact that the labeling of the rating scale could cause invalid responses. The central issue of the present study is whether labels should be attached to all points of a scale or only to the extremes. Frisbie and Brandenburg (1979) found small but significantly different means obtained using fully labeled scales and endpoints-only-labeled scales. When comparing two different formats of Likert-type scales Dixon, Bobo, and Stevick (1984) found that means obtained using these two kinds of la-beling did not differ significantly. They found greater variability in ratings obtained with endpoints-only-labeled scales. Weng (2004) found that test-retest reliability is higher for fully labeled scales than for scales where not all response categories are labeled. These results show that labeling affects the mean and variability of responses. However, does labeling also affect the linearity of responses? Several studies have pointed out that when subjects are asked to judge combinations of two or more stimuli, they use simple algebraic rules to in-tegrate the subjective values of stimuli (Anderson 1981, 1982, 1996). These rules are revealed by the pattern of curves obtained from a factorial ex-perimental design. Specific patterns of factorial curves occur when the rat-ing scale is linear. Thus, to assess whether a particular scale is linear one needs only to test whether a pattern of factorial curves found previously for a specific integration rule also occurs when labels are attached to all points of the scale or only to the extremes. In this study we used this method to assess the interval properties of two differently labeled rating scales