Fig. 2

Random-Parameters Logit Model Estimates: Preference Weights. A US Adolescents (N = 197). B US Caregivers (N = 207). C UK Adolescents (N = 100). D UK Caregivers (N = 200). CI Confidence interval, DCE Discrete-choice experiment, IV Intravenous. Note: Attributes are presented in the order in which they appeared in the DCE questions. The vertical bars around each mean preference weight represent the 95% CI around the point estimate. Because all attribute levels are effects coded, the sum of preference weights across levels of an attribute equals 0. Within each attribute, a higher preference weight indicates that a level is more preferred. The change in utility associated with a change in the levels of each attribute is represented by the vertical distance between the preference weights for any 2 levels of that attribute. Larger differences between preference weights indicate that respondents viewed the change as having a relatively greater effect on overall utility. For example, looking at Fig. 2A, a change in improvement in symptom control from “very poor to poor” to “very poor to fair” yields a utility increase of 0.745 (− 0.310 − [− 1.054]), whereas an increase in the time until next flare-up from 1 to 5 months yields a smaller utility increase of 0.547 (− 0.091 − [− 0.456]). Although both changes yield positive utility increases, the change from “very poor to poor” to “very poor to fair” yields a change 1.36 times more important than the change from 1 month until next flare-up to 5 months until next flare-up (0.745 ÷ 0.547). Alternatively, a change from no stomachache, nausea, or vomiting to stomachache (but with no feelings of throwing up) yields a negative utility change of − 0.703 (− 0.011 − 0.692), whereas a change from no headaches caused by the medicine to having headaches caused from the medicine yields a negative utility change of − 0.666 (− 0.333 − 0.333)