*3.2. Econometric Analysis*

Next, a linear regression model relating students' decisions in the MTG experiment to some attitudinal and sociodemographic characteristics was performed. A main reference is Glaeser et al. [35], together with other related works [36,37]. Our MTG differed from that of Glaeser et al. [35] in the endowment and the multiplier: Their game only endows the trustor, creating asymmetry in the endowment, which is a violation of our egalitarian condition and uses a multiplier of 2. Regarding the trust and fairness questions, these authors used the ones collected by the Global Social Survey. Many scholars have studied behavior or personality differences between business degree students and other degree students. Some studies argue that business degree students tend to show a more selfish behavior [38,39]. In particular, Sautter et al. [40] found a significant difference between finance students and other students and conclude that the former behave in a less ethical manner.

In our model specification, we included the following independent variables:


O

To select explanatory variables and models, we followed a stepwise method by applying Akaike's information criterion (AIC) and Bayesian information criterion (BIC) alongside the significance of variable coe fficients. In a first step, we regressed a full model to identify those non-significant variables. In a second step, these were removed to regress a second model and compare the performance of both models according to AIC and BIC. This procedure was repeated to find the model that fits the data best.

Concerning trustor students, we estimated an OLS model, reported in Table 5. This model explained 36.32% of the variability of observed trust (λ = sent amount/initial endowment). Observe that some variables have a positive coe fficient: studies, gender, and self-declared trust at the medium level have a positive e ffect on observed trust. Specifically, both BA students and female students showed more trust than Tourism students or male students. Therefore, there was a gender e ffect but in the opposite direction to H3. Moreover, the variable 'Employee' exhibited a significant and negative coe fficient. This indicates that students that have a job show lower trust. Similarly, those trustors that in the questionnaire self-state a low trustiness in others behaved accordingly in the game, as indicated by a negative coe fficient. Finally, the dummy 'Information treatment' had a negative and significant coe fficient. In T1 trustors showed lower trust than in T0. This last result allows us to conclude that the fact of having information about the other's accumulated earnings during the game does have an effect on the trustor's decision, but we cannot prove its direction.


**Table 5.** Percentage λ sent to the trustees as a function of trustor's characteristics.

\* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001.

Regarding trustee-students, we considered the return rate (*r*) as a proxy of observed reciprocity. The decision to return an amount may be a ffected by the trustee's empathy with others and the self-concept of solidarity, or by other personal characteristics, such as academic studies, housing, economic autonomy, and etcetera. We perform two OLS regression models to study the trustee's observed reciprocity, presented in Table 6. In model 1, we found that housing and treatment significantly affected the trustee's return decision. We also found a negative coe fficient related to the level of studies, meaning that BA students, on average, returned lower amounts than students enrolled in Tourism. The variable 'housing' showed an overall positive e ffect on the trustee's return decision. In model 2, regarding question 2 on inter-personal intelligence (see Figure 3), the answers corresponding to the faces (Happy) - and Ⴧ (Machiavellian) were statistically significant but with opposite e ffects. Thinking that the trustor was happy had a positive e ffect on the trustee's decision, whereas thinking that the trustor was Machiavellian had a negative e ffect on the trustee's decision. Interestingly, all labels related to the trustee's self-concept about solidarity showed a positive and significant e ffect. Finally, the variable treatment exhibited a negative and significant coe fficient in both models, showing that having information about the other's accumulated pay-o ff during the MTG had a negative e ffect on the trustee's return decision. We can prove that there is an e ffect, on average, such that having information about their partner's accumulated earnings each period a ffects the return decision of trustees. However, we cannot prove the direction of such e ffect. In both models, the variable gender was not statistically significant, and therefore, H4 cannot be confirmed.


**Table 6.** Return rate of trustees as a function of personal characteristics/information.

\* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001.

Whereas Glaeser et al. [35] did not find a significant relationship between the attitudinal questions and the behavior of subjects in the laboratory, the opposite was found by Capra and Lanier [36] and Aksoy et al. [37]. Altruism was introduced by Capra and Lanier [36] as a control variable able to explain the decision in the TG. Aksoy et al. [37] replicated Glaeser et al.'s [35] work by using the original version of Berg, Dickhaut, and McCabe [22], concluding that both the attitudinal questions and altruism are good predictors of the behavior of subjects. Our results are in line with these findings.
