Jeroen van de Pol

146 Chapter 5.2 Latent class analysis The latent class analysis resulted in a four-class model showing a McFadden’s pseudo ρ 2 of 0.25 (0.2 -0.4 indicates a good fit). With 5 classes or more, fit statistics improved only slightly and yielded smaller classes with unclear differentiation fromother classes (see table 2). The average maximum membership probability was 83.5%. Part-worth utilities of each class per attribute level are shown in figure 3 and table 3. Table 2: Latent class results regarding number of classes and best fit. # Classes Log-likelihood McFadden’s pseudo ρ 2 AIC BIC 2 -13721.9 0.20 27469.8 27575.3 3 -13193.3 0.23 26426.7 26588.9 4 -12814.8 0.25 25683.6 25902.6 5 -12666.6 0.26 25401.2 25677.0 Figure 3: Part-worth utilities of each attribute only showing the extended service level per class. Table 3: Latent class analysis results showing part-worth utilities and relative importance of attributes. Class 1 n=822 Class 2 n=351 Attribute Level Utility (SE) RI Utility (SE) RI Drugs for minor ailments Also prescription drugs 0.23 (0.01) 6.1% -0.31 (0.014) 35.6% Pharmacogenetics Advice regarding pharmacogenetics 0.49 (0.009) 13.0% -0.16 (0.01) 18.4% Point-of-care-testing Offers tests 0.4 (0.008) 10.6% -0.07 (0.011) 8.0% Track & Trace Provides track & trace 0.47 (0.004) 12.5% 0.11 (0.005) 12.6% Medication record On paper and online 0.59 (0.01) 15.5% 0.62 (0.01) 18.0%