Programme Impact on Experimental Trust

The next step in understanding the impact of the Activate! programme on trust is to investigate the impact it had on trusting behaviour. This is measured by the percentage of the initial endowment that Player A sent in the trust game. It is important to note that the number of participants in this regression is lower than the other regressions, as only half of the treatment and control groups are assigned to being Player A. Regressions are run using the % sent as the dependent variable, with the independent variable being the treatment dummy variable. Three control variables are then added, which are the gender variable, baseline CES-D and race variable.

Controls are also added for risk, altruistic behaviour and expected return. Table 8 shows the regressions with these control variables added individually while Table A.4 shows various combinations of these mea- sures being added. The key result from these regressions is that the programme increased trusting behaviour, as measured by the percentage sent by Player A when controlling for Player A’s expected return as well as when controlling for altruistic behaviour.

The first regression (1) in Table 8 is a linear regression model which uses the treatment dummy as the independent variable. Additionally, baseline depression, gender and race are added as controls. The result shows that there is a positive impact of the treatment on the amount sent by Player A but that this is not significant.

and demographic factors. The coefficient on baseline CES-D is positive but is not significant. Interestingly, there is a significant difference between the amount sent by the men and women, with men sending a higher amount than the women. The gender variable is significant at the 5% level and shows that men offered approximately 6.8% more than the women in the trust game. Finally, the race variable is positive but not significant. With these controls added, the treatment variable is still not significant at the 10% level.

Table 8: Regressions showing the Impact of the Activate! Programme on % Sent of Player A (trust game)

(1) (2) (3) (4) (5)

% Sent % Sent % Sent % Sent % Sent

Treatment 0.0109 0.0348 0.0327 0.0443 0.0515* (0.0296) (0.0310) (0.0324) (0.0331) (0.0307) Base CES-D 0.00460 0.00490 0.00627 0.00604 (0.00680) (0.00691) (0.00731) (0.00660) Gender (1=Male) 0.0677** 0.0676** 0.0753** 0.0611* (0.0313) (0.0317) (0.0334) (0.0312)

Race (1=black African) 0.0956 0.0943 0.0759 0.0885

(0.0754) (0.0760) (0.0805) (0.0743)

Mean Row (Risk) 0.00357

(0.00950)

Dictator Game Offer 0.0598

(0.0802) Expected Return -0.0779 (0.0698) Constant 0.445*** 0.277*** 0.262*** 0.268*** 0.341*** (0.0223) (0.0836) (0.0930) (0.0934) (0.0899) Observations 289 256 253 228 230

Standard errors in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01 CES-D: a short, self reported scale used as an indicator of depression % Sent: the % that Player A sends in the trust game

Mean Row: mean switching row in coin flip game, higher values indicate lower risk preferences Dictator Game Offer: the % that A offers to B in the dictator game

Expected Return: the % Player A expects to be sent back in the trust game

Additional controls are included in regression (3) to (5) in Table 8, with various combinations of these controls used in regressions (2) to (8) in Table A.4. As described in Fehr (2009), risk preferences, social preferences (e.g. altruism) and expected return all have an influence on trust. Therefore, these three variables are used to control for the effect of these on the amount sent by Player A. The first is mean switching row in the risk game, with later switching indicating a higher degree of risk aversion. The second is the dictator game offer, which can be used as a measure of altruistic behaviour. The third control is Player A’s expected return, which is measured as the percentage that Player A expects Player B to return.

The control that is introduced in regression (3) is the risk measure (mean row). The coefficient on it is negative but is not significant at the 10% level. As a result of adding the risk measure, the magnitude of the treatment coefficient increases but remained insignificant. The magnitude and significance of the other control variables remain consistent with the previous regression.

on it is positive but not significant at the 10% level. Adding this control treatment does not change the significance of the treatment variable or the other variables, though the magnitudes change slightly.

The final regression (5) in Table 7 introduces the expected return of Player A as a control. The intro- duction of the expected return variable results in the coefficient on the treatment variable being significant at the 10% level. The variable is positive and shows that the treatment group sent approximately 5.2% more than the control group, once expected return is included. The other variables have similar coefficients and significance levels as the other regressions. The expected return variable is not significant but is negative.

In all the regressions in Table A.4 the treatment variable is positive, but in (4) and (7) it is significant at the 0.10 level. In both of these regressions expected return is used as a control variable. This indicates that the programme had a positive impact on the behavioural trust measure of the participants when controlling for their expected return.