The Trust Game Does Not (Only) Measure Trust: The Risk-Trust Confound Revisited

Abstract

The trust game has become the behavioral measure of choice in social science investigations of trust. This measure is often used uncritically to compare levels of trust across people and cultures even though those levels may be affected by people's attitudes to risk. We evaluate an incentive-compatible experiment designed to investigate the potential for a risk-trust confound using a sample of 202 students at the University of Cape Town in 2016. We depart from the earlier risk-trust literature by using a risk preference task that incorporates a wide range of prizes and probabilities. This allows us to investigate whether earlier ambiguous results concerning risk-trust interactions simply reflect weak measurement instruments, and facilitates the estimation of structural econometric risk preference models. We find that amounts sent in the trust game are indeed associated with attitudes to risk, that the magnitude of this relationship is economically significant, and that it is robust across statistical models. In addition, we find that most previous studies of the risk-trust confound use preference elicitation mechanisms that are underpowered for identifying risk-trust relationships. Our results caution against the widespread use of the trust game to measure and compare levels of trust without careful adjustments for risk attitudes.