• *The second step: Building of the intermediate model.*

The result of the first phase (i.e., ICC estimation) allowed the identification of the two-level model as the most suitable for the analysis. In particular, Equation (1) represents the model in which intercepts casually vary among groups while a general model considers fixed predictors, both at an individual and group level [19]. Using a stepwise procedure [24], several models were identified and the one which showed the lowest AIC was selected. The AIC calculates the likelihood of a model for future estimations and in particular, a smaller AIC means that the corresponding model shows a better prediction performance [19,35].

Moreover, the likelihood ratio test (LRT) was applied to compare the simplest two level model and intermediate ones and according to this, the best model was the intermediate one (Tables 2 and 3).

The intermediate model was described as:

$$\begin{array}{l} \Pr(\text{Will} = \text{Yes} | \text{x}) = \gamma 0 + \gamma 1 \text{ Prov} + \gamma 2 \text{Origin} + \gamma 3 \text{Family} + \gamma 4 \text{Cons} + \gamma 5 \text{Will}\\_q + \gamma \text{Gurosity} + \gamma \text{Gurifft} \\ \gamma \text{Gurosity} + \gamma \text{Tæregettic} + \gamma 8 \text{Age} + \text{u0j} + \text{rij.} \end{array} \tag{2}$$


Note: The AIC (Akaike information criterion) and the BIC (Bayesian information criterion) are the well-known model fit indices.

Source: Our elaboration.


**Table 3.** Parametric estimation results for the intermediate model Equation (2).

Note: n.s. means that variable is not significant. The AIC (Akaike information criterion) and the BIC (Bayesian information criterion) are the well-known model fit indices.

Source: Our elaboration.
