Jeroen van de Pol

144 Chapter 5.2 Data analysis Data gathered with the CBC choice task were analyzed using Sawtooth Software (Lighthouse Studio version 9.8.0). This results in utilities which represent the relative attractiveness of an attribute and level. Positive values reflect a preference and higher values mean a greater attractiveness. Latent class analysis [25] was used to determine if subgroups (classes) were present based on preferences provided. With latent class analysis, one tries to identify an underlying (latent) variable by means of researching observable variables. The number of classes was based on the goodness-of-fit for the model. Multiple goodness-of-fit statistics are available within latent class analysis. The LogLikelihood, McFadden’s pseudo p 2 , Akaike and Bayesian information criteria [19,25] and were used to determine the number of classes present within the study sample. Participants were assigned to the class on which they had the highest probability. Parametric (student T-test) and non-parametric tests (Mann-Whitney U and Kruskal-Wallis) were used, depending on the distribution of data, to determine differences between classes on background characteristics. For all analyses, p-values were considered statistically significant when < 0.05. Microsoft Excel 2016 and SPSS version 23.0 were used for descriptive analysis of basic characteristics. Results Respondents In total, a random selection of 9,025 panel members were invited to participate in the study between November 2019 and February 2020. The questionnaire was accessed by 3,697 (41.0%) invited panel members. Of these, 2533 (28.1%) panel members completed the questionnaire (see supplementary material for information regarding drop-out rate per step of the total questionnaire). No statistical significant differences were found between participants who fully completed the questionnaire and those who did not with respect to gender and age (n = 1,067). After quality check of the data on potential unreliability, an additional 71 participants (0.8%) were excluded, resulting in a study population of 2462 (27.3%) participants, see figure 2. Table 1 shows participants’ basic characteristics.