Studying voluntary contributions to a public good: Four clearly distinct behavioural types account for over 90% of participants

Behavioural types in public goods games: A re-analysis by hierarchical clustering. Francesco Fallucchi & R. Andrew Luccasen. http://www.gambit-project.org/turocy

Abstract: We re-analyse participant behaviour in standard economics experiments studying voluntary contributions to a public good. Previous approaches were based in part on a priori models of decision-making, such as maximising personal earnings, or reciprocating the behaviour of others.  Many participants however do not conform to one of these models exactly, requiring ad hoc adjustments to the theoretical baselines to identify them as belonging to a given behavioural type.  We construct a typology of behaviour based on a similarity measure between strategies using hierarchical clustering analysis.  We identify four clearly distinct behavioural types which together account for over 90% of participants in six experimental studies.  The resulting type classification distinguishes behaviour across groups more consistently than previous approaches.

Keywords: behavioral economics, cluster analysis, cooperation, public goods
JEL Classifications: C65, C71, H41.

---
Own-maximisers (OWN, 25.8% of participants) allocate zero or very few tokens to the group account in all contingencies, named because the modal allocation is zero. The modal behaviour among strong conditional cooperators (SCC, 38.8%) is to match average allocations on a one-to-one basis. The classification contrasts these with weak conditional cooperators (WCC, 18.9%), who have allocation strategies that are increasing in the allocations of other group members, contributing on average about half as much as the other group members. There is a small but distinct group of unconditional cooperators (UNC, 4.7%), corresponding to contributions which are at or near full contribution of the tokens  to the group account. (Recall that full contribution to the group account maximises the group’s total earnings from those tokens.) This labeling parallels that of Kurzban and Houser (2005), where, in a setting in which participants play the game repeatedly, those whose always contribute more than the average of the others in their group are called (unconditional) “cooperators”. The final cluster (11.8%) is labeled various (VAR), and contribute on average about one-half of the tokens. This group is the most diverse; in addition to a high frequency of contributions exactly equal to 10, there are also “negative conditional contributors” (Burton-Chellew et al., 2016) who decrease their contribution in response to higher contributions by others.