**Causing a loss to the charity**
We were also interested in the possibility that participants ceased to take bribes when the total earnings for charity were equal to zero. That is, participants might have been unwilling to bring a loss for the charity. Since the zero threshold possibility is confounded with the charity earnings at a given point and with the trial order, we first compared three models, each including only one of these predictors. The model including charity earnings resulted in the best fit, so we next added trial order and zero threshold in separate models and compared them. Including trial order led to a better fit, but improvement of the model was not significant for either the trial order, *Χ<sup>2</sup>*(1) = 3.13, *p* = .07, or a threshold at zero, *Χ<sup>2</sup>*(1) = 1.00, *p* = .32. We therefore did not find any evidence that participants were specifically unwilling to decrease the earnings for charity below zero (see also Figure A1).
![Figure A1][1]
**Figure A1. The development of earnings for charity during the experiment.**
The figure displays the development of total earnings for charity for each participant (black lines). It is possible to see that many of the participants classified all the objects correctly according to their color resulting in the earnings of 2000 points. Incorrect classifications were also relatively infrequent and most of the participants has final earnings for charity higher than 1000 points. While we hypothesized that participants might be less likely to incur losses to charity than to reduce its gains, there is no discernible threshold at 0 points. The blue line representing the development of average earnings across all participants is very slightly convex, which shows the order effect found in the analysis – participants are somewhat less likely to classify the object incorrectly in later trials. Note that the abscissa is cut below -2000 points and that a jitter was applied to lines in both coordinates so that the frequencies were easier to judge. The earnings for charity took in fact only discrete values that were multiples of 200.
[1]: https://mfr.osf.io/export?url=https://osf.io/64fws/?action=download&direct&mode=render&initialWidth=760&childId=mfrIframe&format=1200x1200.jpeg
