Fig. 3.

The relative importance of individual variables in explaining the membership of respondents in one of the four response types (cf. Table 4). Importance values were determined by means of the machine learning algorithm Random Forest, and indicate the percent increase in mean squared error (MSE) in the modelled data when values for that particular predictor were randomly permuted and all other predictors were left unaltered. Higher values indicate a higher importance of the variable (for details see Breiman 2001; Cutler et al. 2007)