We experimentally investigated trusting behaviour in patients, first-degree relatives and a healthy control group by means of a modified version of a multi-round trust game (Berg et al., 1995; King-Casas et al., 2005, 2008). Participants in the experimental groups were investors throughout the game. The role of trustees was taken on by students. At the beginning of each round of the trust game, both the investor and the trustee received €10 to ensure that investors did not make their initial investment choice to avoid inequality. Investors were then asked to transfer any (integer) amount between €0 and €10 to the anonymous trustee. The transferred amount was tripled by the experimenter. After having received that amount, the trustee decided whether to honour the investor’s trust and send all or some of the received money back or behave untrustworthy and keep all or most of the received money.

In order to examine whether social dysfunction is driven by aberrant bottom–up feedback learning or top–down processing of socially relevant information, we implemented three conditions. First, the ‘baseline condition’ (five rounds) was used to get an unbiased estimate of basic trust. Therefore, the investor was not informed about how much the trustee reciprocated in a round and feedback about the reciprocated amounts was given only at the end of the whole experiment. The investors were aware of this when making their decisions. Second, the ‘context condition’ (five rounds) was used to investigate the effects of a priori information about the partners’ trustworthiness. The investor was matched with a new trustee and, before making the investments, shown truthful information about this trustees’ trustworthiness during the baseline condition. Specifically, the investor was informed whether the trustee, on an average, returned less, as much as, or more than the amount invested by the investor. To ensure no confounds in the investigation of the contextual effect, the investor received no feedback about the trustees’ actual trustworthiness in the current condition. Third, the ‘feedback condition’ (10 rounds) was used to obtain a measure of the effect of bottom–up feedback learning on prior beliefs and the development of trust and reciprocity in consecutive interactions. In this condition, the investor received direct feedback about the amount of money that was reciprocated by the trustee in response to each investment. Investors interacted with the same trustee within each condition, but with different trustees across the three conditions to minimize carry-over effects from one condition to another. The same set of trustees played with two different investors in the baseline and context condition, because the lack of feedback in the former precluded carry-over effects between the conditions. As information about the average trustworthiness in the baseline condition was given in the context condition, all investors were paired up with a new trustee in the feedback condition. All participants were informed about this before making their decisions in the respective condition.

Specifically, we hypothesized that psychosis risk would be associated with a loss of basic trust, a diminished sensitivity to contextual information and an inability to engage in trusting interactions. This would be reflected in smaller invested amounts in the baseline condition, a reduced effect of information about the counterparts’ trustworthiness on investment behaviour and reduced reciprocal trust in response to observed trustworthy behaviour by the partner in the feedback condition.

For more details, please see the linked article: To trust or not to trust: the dynamics of social interaction in psychosis