{"id":574,"date":"2020-01-22T04:45:51","date_gmt":"2020-01-22T04:45:51","guid":{"rendered":"http:\/\/measuringu.com\/labeling-scales\/"},"modified":"2022-03-21T16:36:47","modified_gmt":"2022-03-21T22:36:47","slug":"labeling-scales","status":"publish","type":"post","link":"https:\/\/measuringu.com\/labeling-scales\/","title":{"rendered":"Comparing Fully vs. Partially Labeled Five- and Seven-Point Scales"},"content":{"rendered":"
In our earlier article<\/a>, Jim Lewis and I reviewed the published literature on labeling scales.<\/p>\n Despite some recommendations and \u201cbest practice\u201d wisdom, we didn\u2019t find that fully labeled scales were measurably superior to partially labeled scales across the 17 published studies that we read.<\/p>\n In reviewing the studies in more detail, we found many had confounding effects when comparing between full labeling and partial labeling\u2014meaning it was hard to isolate the effects of labeling.<\/p>\n When comparisons were well controlled (no confounding), labeling differences weren\u2019t statistically significant. In cases where there were effects of labeling, the scales used were more similar to rubrics (e.g., rating faculty performance or clinical judgments) rather than measures of sentiment such as satisfaction<\/a> or agreement.<\/p>\n We felt there was need for more data, so we conducted two new studies that better isolated the effects of labeling on more commonly used scales of sentiment in UX and customer research.<\/p>\n <\/p>\n In September 2019, we asked 202 U.S.-based online panel participants to rate how satisfied they were with their smartphone and respond to three versions of a five-point satisfaction item.<\/p>\n The first version of the satisfaction item had only the endpoints labeled (Figure 1); the second version was labeled with the end and middle points (Figure 2); and the third version was fully labeled (Figure 3). All variations included numbers.<\/p>\n <\/a><\/p>\n <\/p>\n <\/a><\/p>\n <\/p>\n <\/a><\/p>\n <\/p>\n We used a within-subjects study to allow us to detect small differences in scores. In this within-subjects study, participants saw all versions of the satisfaction scale. Only participants who reported having a smartphone were asked to participate in the study.<\/p>\n These three questions were part of a larger survey, and the three variants were randomly shown at different points in the middle of the survey. Between the three satisfaction questions were other questions regarding attitudes toward website design elements and\u00a0 other measures of brand attitude, and their intent to recommend products.<\/p>\n Figure 4 shows an overall slight and statistically significant pattern of satisfaction scores lower as more labels are added (using a paired t-test). When only the endpoints are labeled, the mean score is 4.34, 1% higher than when a neutral label is added (mean = 4.29; p = .029) and 2% higher than the fully labeled variant (mean = 4.26; p = .002).<\/p>\n <\/a><\/p>\n <\/p>\n We also examined the effects on top-box scores<\/a>, shown in Figure 5, and saw a similar, albeit more attenuated, pattern. Slightly more selected the highest score when only the endpoints were labeled (49% vs 48% and 45% for the other two conditions). Similar to our earlier study on the effects of the neutral score<\/a>, when only the endpoints are labeled, they tend to attract respondents.<\/p>\nStudy 1: Fully vs. Partially Labeled Five-Point Scale<\/h2>\n
Study 1 Results<\/h2>\n