RESULTS OF THE TRUST EXPERIMENT

In the Sept 15 lecture we ran a small Trust Game experiment. Here we show a summary of the (aggregate) results.

During the lecture, I argued that the behavior of player A (the trustor) can usually be predicted by the index (P-S)/(R-S). Likewise, the behavior of player B can be predicted by the index (T-R)/(T-S). When we aggregate our results per Trust Game, we can see that our results are no exception to that general rule.

Here are the results for player A, the trustor:

And here we show the results for player B:

Both graphs show a strong connection between the relevant index and the percentage of players A trusting and the percentage of players B honoring trust.

Another finding that is consistent with the literature, is that the temptation value is not a strong predictor of trust. That is counterintuitive, and goes against one of the standard assumptions of game theory. For players A it should matter quite a lot what player B is likely to do, and that in turn can be adequately predicted by the temptation value. Following that logic, the temptation value should be a strong predictor of player A's behavior. It is not. Players A are a bit more shortsighted than is good for them.

A finding that is inconsistent with previous findings is that a student's choice as player B is not a good predictor for behavior as player A.

Note1 We cannot emphasize enough that a real experiment should not be conducted in such a classroom setting, for -we hope- obvious reasons.

Note2 The more statistically inclined obviously note that using the data in this aggregate way is not optimal; multi-level logistic regression can/should be used here. That would allow additional and stronger tests (for instance allowing to test for effects of gender or study type) and would make more efficient use of the data. Interested in how that is done? Let me (CS) know. Contrary to what is usual, we do find rather large individual differences, possibly caused by study type of the students.