Social norms as anchor points for trust

ABSTRACT

We investigate whether a social norm communication can be used to nudge decision makers towards increased trust. An experiment was conducted using a modified trust game, where subjects played two rounds of the game. Prior to the second round, a social norm of sending behaviour was communicated to participants. The impact of the social norm communication on trusting behaviour varied with initial transfers: For those who initially transferred amounts below this point, transfers (indicative of trust) increased. However, this impact was almost entirely offset by decreased transfers among those who had initially transferred amounts above this point, suggesting that the social norm communication acted as an anchor point towards which subsequent transfers were drawn. The result in our study was that mean trusting behaviour did not change significantly with the social norm communication. The net impact of an anchor point would be expected to vary with the distribution of behaviour prior to the social norm communication, risking a negative net impact in some cases. Our findings therefore suggest a need for caution in using broadly targeted social norm communications.

KEYWORDS:

• Trust game
• Anchoring
• Experiment
• Social norms

Highlights

• We investigate a social norm communication to increase trust.

• We compare sent amounts in a trust game before and after revealing the social norm.

• Low initial trust increases but high initial trust decreases after the social norm.

• The net impact of the social norm would vary with the distribution of initial trust.

• Our results suggest caution in using broadly targeted social norm messages.

Notes

¹ There are a number of studies where high transfers (average above 75% of the endowment) are seen in trust games, such that the deliberately vague statement we included as our social norm would in fact be true of specific experiments (see, for example, the seminal work by Glaeser et al. (2000) with an average transfer of 83%; as well as papers by Swope et al. (2008) with an average transfer of 89% and Lazzarini et al. (2005) with transfers of 86% on average, inter alia). We selected a social norm greater than the typical average transfer seen in trust games, as we wanted to see whether transfers would increase with a higher social norm.

² Prior to the main game, a pilot game had been tested with a smaller group of participants to get a better idea about initial transfer amounts, from which we could select an anchor point to use for our social norm communication. For the purposes of our experiment, we were interested in whether a positive social norm would increase trust (measured as the sent amount in the game). To that end, we wanted to select a point above the existing levels of trust (sent amounts). The pilot game had an average transfer amount of 60%, leading to our selecting 75% as the anchor point for the main experiment.

³ In July 2018, when the experiment was run, ZAR100 was equivalent to approximately 7.50 US dollars or 6.50 Euros.

⁴ It would be theoretically possible for the random draw to result in fewer than 10 unique subjects being paid since random draws of five sender numbers could see a subject being paid first based on their own sender number and second by being the receiver of another selected sender player’s transfer. In our game, the random draw resulted in 10 unique subjects being paid.

⁵ The pilot study to establish an appropriate level for the social norm communication was conducted with a smaller group (n=50) of honours economics students.

⁶ Seventy-six percent of return decisions show (weakly) monotonic increases in return amount as sent amount increases.

⁷ The mean percentage returned does not vary a lot by the sent amount: The smallest mean percentage returned for a transfer amount was 39.4% (for R10 sent amount) and the maximum mean percentage returned for a transfer amount was 42.6% (for R100 sent amount). Although the minimum and maximum percentages returned align with the minimum and maximum positive sent amounts, changes in the mean percentage returned do not increase monotonically. For example, the second-highest mean percentage returned was 42.3% (for R60 sent amount). A reviewer suggested taking an “expected value” approach to calculating the return percentage by weighting return percentages according to the distribution of senders’ decisions. All results are robust to this alternative specification.

⁸ Recall that the transfer cannot exactly equal the R75 social norm point since transfers were reported in multiples of R10.