Homo Æqualis: A cross-society evaluation of four models of other-regarding preferences
by Abigail Barr, Ian Crawford, Chris Wallace, Jean Ensminger, Joseph Henrich, Clark Barrett, Alexander Bolyanatz, Juan Camilo Cardenas, Michael Gurven, Edwins Gwako, Carolyn Lesorogol, Frank Marlowe, Richard McElreath, David Tracer, and John Ziker
Extended Abstract:
A useful model concurs with observations, generates valid predictions and, if the goal is to extend insight, is interpretable and tractable. Since the early 1980s, experimental economists have repeatedly shown that models assuming only self-regarding preferences are limited in the extent to which they concur with observed human behaviour. These findings led to the development of models that incorporate other-regarding preferences and, subsequently, to evaluations of the new models’ usefulness. To date, all such evaluations have relied on data pertaining to the “standard” population of experimental subjects, i.e., to students studying in universities located in high income countries. This is an entirely appropriate starting point. However, it limits the extent of the heterogeneity that the models are required to capture, explain, and predict. In this paper we use a behavioural dataset pertaining to the most diverse sample of experimental subjects ever worked with to evaluate the relative usefulness of four models of other-regarding behaviour. The dataset was generated by Henrich et al (2006, 2010, 2012). It includes standard dictator game (DG), ultimatum game (UG), and third-party punishment game (TPG) offers, UG acceptance strategies, and TPG fining strategies for representative samples of adults drawn from fifteen very distinct populations, including several from small-scale societies in Amazonia, the Arctic and sub-Saharan Africa. Three of the societies are partially or entirely nomadic, three rely partially or entirely on hunting and foraging, six engage to varying degrees in waged work, ten engage to varying degrees in farming, one is comprised of students at a US university.
Henrich et al (2006, 2010, 2012) report significant variations in behaviour across the societies and show that members of more market integrated societies tend to make higher offers and members of larger societies are more inclined to punish low offers. Mean DG offers range from 0.26 to 0.47 of the stake across the societies, mean UG offers from 0.26 to 0.48, mean TPG offers from 0.20 to 0.43, mean minimum acceptable offers in the UG from 0.06 to 0.38, and mean minimum unfined offers in the TPG from 0.04 to 0.41. These marked behavioural variations across societies provide a unique evaluative opportunity. Building on the work of Roth et al (1991), they allow for the incorporation of cross-society variations in expected rejecting and fining behaviour into the evaluations of the models’ concurrence with observed UG and TPG offers. They also allow for the incorporation of cross-society variations in preference parameter distributions into model evaluations. Finally, they facilitate an investigation into whether model performance varies systematically depending on the characteristics of the societies from which subjects are drawn. We investigate whether and to what extent four interpretable and tractable models incorporating preferences for equality and efficiency, i.e., for a larger sum of payoffs across all game participants, concur with or fit the behaviours observed in this dataset, generate valid within-society predictions, and provide a useful framework within which to think about cross-society behavioural differences.
Each of the models we evaluate is comprised of a well-behaved, parameterised utility function and an assumption about individuals’ expectations regarding the actions of other members of their own society. All of the utility functions incorporate a positive preference for own final payoff and an aversion to inequality. In two, inequality aversion is captured by a single preference parameter. In two, following Fehr and Schmidt (1999), separate preference parameters are assigned to disadvantageous inequality and advantageous inequality. In one of each of these pairs of utility functions, following Engelmann and Strobel (2004), we also incorporate a preference for efficiency. All four models assume that expectations about the behaviour of playing partners are society-specific and correct.
Where possible we adopt a within-subject evaluative approach similar to Andreoni, Castillo and Petrie (2003) and Blanco et al (2011). Elsewhere, our evaluative approach is similar to the aggregate-level approach of Fehr and Schmidt (2006), although in contrast to Fehr and Schmidt (2006) we can make use of the cross-society component of our dataset to do what might be referred to as a within-society evaluation.
We find that the models that distinguish between aversion to disadvantageous inequality and aversion to advantageous inequality concur with a significantly greater proportion of the UG and TPG offers and strategies compared to the models containing only one preference parameter pertaining to inequality. The models that incorporate a preference for efficiency concur with an only marginally greater proportion of the UG and TPG offers and strategies compared to the models incorporating no such preference. However, incorporating a preference for efficiency significantly increases the rate of concurrence across the DG and UG offers within subjects and improves predictive performance.