It seems that we have two different methods to measure trust. The trust game is commonly said to represent a behavioral measure, whereas survey questions are attitudinal measures. The questions is, are these two methods measuring the same thing?
Among a small group of people, trust has sometimes been measured by a practice where one of the group members is supposed to fall down, whereas the rest of the group is supposed to catch the one who falls. The one who falls shows trust and the others show trustworthiness by capturing her or him.
The practice can also be used as a thought experiment to consider what factors would influence our propensity to trust. Would I be more confident that others will catch me if they were family members, colleagues, friends, compatriots? Would my trust be affected by morally arbitrary factors like skin color, gender, or ethnic background?
While available for a group of friends or colleagues, the above method could hardly be used for scientific purposes. And yet we would need a reliable and practical method for measuring trust in order to study questions such as how trust varies between contexts and cultures, how trust influences well-being, or how trust is influenced by various social factors.
In decision making experiments, trust has been measured by a so called trust game where two players, randomly assigned into the roles of a sender and a responder, interact in a setting where the sender can show trust and the responder trustworthiness.
In the trust game, the sender is endowed with a certain sum of money, usually around ten euros. The task of the sender is to decide how much of this endowment he or she passes on to the responder. If the sender decides to pass on an amount of money, the sum is multiplied, usually tripled, and transferred to the responder.
In the next stage of the game, the responder decides how much he or she returns to the sender. By sending money to the responder, the sender shows trust, and by returning money the responder shows trustworthiness.
An important characteristic of the setting is that the players have a possibility to earn more money by showing trust and trustworthiness because the amount sent is multiplied. However, at the same time, a rational payoff-maximizing player would not send anything because the responder’s dominant strategy is to return nothing.
The trust game is seen as an analogy of a number of situations where trust and trustworthiness are relevant, that is, a large variety of economic or political interaction.
In the standard version of the trust game, the sender and the responder remain anonymous throughout the experiment, and choices are made via computers.
The usual result over a series of experiments with slight variations to the basic design is that, contrary to the rational payoff-maximizing assumption, senders show trust and responders show trustworthiness.
However, amounts sent and amounts returned vary to a great extent depending on the specific framing and context of the experiment. For example, giving subjects a chance to play both roles in the trust game and ensuring anonymity reduces contributions, whereas student subjects tend to be less trustworthy than participants from other groups, and it also seems that on average, less trust is shown in experiments that are conducted in Africa.
What do we learn from trust game experiments? Critics argue that not much because the use of students as subjects reduces external validity and abstract choice situations lack an apparent counterpart in the real world.
Another standard way of measuring trust is via surveys where respondents come from a representative sample of the whole population, and where questions bear more similarity with real world choice situations. However, the questions measuring trust in surveys are by no means unproblematic.
The standard question used in the American General Social Survey (GSS) and European Social Survey (ESS) is formulated as follows: “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” The World Values Survey uses two separate questions “Most people can be trusted” and “Do you think most people try to take advantage of you”.
The main critique of survey questions such as these is that they are imprecise and hard to interpret. For example, what does “most people” refer to? Does it mean acquaintances one normally interacts with, or rather strangers that are met only occasionally?
It therefore seems that we have two different, and more or less imperfect, methods to measure trust. The trust game is commonly said to represent a behavioral measure, whereas survey questions are attitudinal measures. The questions is, are these two methods measuring the same thing? Is there a correlation between the behavioral and attitudinal measures of trust?
This question was first studied by Glaeser et al (2011) who focused on the GSS trust item and asked whether people who reported to trust strangers in a survey also showed trust to another player in the trust game. Interestingly, what they found was that attitudinal trust correlated with trustworthiness in the trust game but not with trust. This is somewhat counterintuitive because the GSS-trust question is supposed to measure trust in other people and not the trustworthiness of the respondent.
Indeed, Glaeser et al’s original result has later been tested by comparing behavior in the trust game with a variety of attitudinal survey questions measuring trust. Existing results from a series of experiments are not particularly robust as some studies found correlation and others do not.
However, recent contributions suggest that correlation between attitudinal and behavioral measures is found when the survey item measures specifically trust in strangers and the experiment is designed to test trust in an anonymous counterpart.
Furthermore, it seems that experimental trust is not correlated with trust in institutions nor with trust in known others giving further support for the conclusion that it particularly measures trust in strangers.
It is evident that more research is required to establish robust results on the relationship between survey and behavioral measures. However, at this point it seems promising that we already have a bit more detailed answer to the question of what we are measuring exactly when we think that we are measuring trust.
Edward L. Glaeser, David I. Laibson, Jose A. Scheinkman and Christine L. Soutter (2000). Measuring Trust. The Quarterly Journal of Economics 115(3): 811–846.
Kaisa Herne is a professor of political science at the University of Tampere. Her research focuses on questions of justice and fairness, the concepts of impartiality, trust and reciprocity, deliberative democracy and voting rules. Her publications range from political philosophy to empirical testing of theories related to political science, economics, and social psychology. In empirical work, she has mainly used the experimental method.