Social scientists have shown a great interest in fairness preferences, most commonly in situations in which the decision maker has a stake in the distribution and has no uncertainty about her own position. In the laboratory, this situation is mimicked by the classical dictator game (Kahneman et al. 1986), in which a decision maker decides on how to split a pie between herself and one other person. This game has had a large impact through the insight that people often care about the income of others. These findings led to the development of social preference models (Fehr and Schmidt 1999; Bolton and Ockenfels 2000) that not only emphasize the perceived (distributional) fairness of outcomes.Footnote 3
Besides the recent interest in perceived fairness of outcomes, there is a long tradition of studying impartial fairness from a normative point of view in philosophy, economics, and political science. In this tradition, researchers are interested in how a decision maker should choose between different income distributions. In their seminal works, Harsanyi (1953) and Rawls (1971) famously propose that the decision maker needs to be behind the veil of ignorance to achieve a just decision. That is, to make a fair decision, the decision maker should be ignorant of her own income, social status, wealth, abilities etc., and thus ignorant of her own stake in the outcome. In this original position, decision makers are then able to make judgments about income distributions which are free of self-interest. Hence, the veil of ignorance can be seen as a thought experiment aimed at reaching impartiality of decision makers, which should lead to fair and just outcomes.
There are many ways to define and measure fairness ideals. We focus on an abstract spectator setting that is both easy to implement and context free. This provides a complementary approach to the more specific instruments that measure preferences given certain contextual factors like production or risk (e.g. Konow 2000; Cappelen et al. 2013; Almås et al. 2019). All of these experiments involving fairness likely capture related behavioral traits. Through the abstract nature of the elicitation, we try to elicit preferences that transfer to different contexts more easily and are thus a stable aspect of preferences. The significant correlations found with policy preferences, political preferences, and charitable donations seem to indicate that they are indeed stable and travel across contexts.
Empirical literature
The empirical literature on fairness is split into several methodologies, with some studies using a set-up inspired by the veil of ignorance and others using spectator judgments. Most empirical papers based on the veil of ignorance are versions of dictator games with uncertainty about the decision maker’s position in the final allocation (e.g. Bosmans and Schokkaert 2004; Schildberg-Hörisch 2010). In contrast, the majority of empirical papers that focus on spectator judgments are so-called vignette studies. These studies present hypothetical situations to participants, who are then asked to make a choice or judgment. The choice made reveals some aspect of participants’ values or social norms applicable to the situation.Footnote 4 The incentivized impartial spectator we use in our experiments, was used by Konow (2000) to test the accountability principle as a basis of perceived fairness and by Cappelen et al. (2013) to study the impact of luck and risk on fairness preferences in spectators and in stakeholders and spectators, respectively.
Traub et al. (2005) highlight the importance of the distinction between spectator and veil of ignorance judgments, as well as the importance of the amount of information possessed by their subjects when making a moral judgment. They find a statistically significant difference between the choices made by subjects when they switch from a self-concern (veil of ignorance) to an umpire mode (impartial observer). In their experiments, subjects are asked to judge lotteries. Those who are informed about the associated probabilities of the lotteries are more inequality averse as umpires than as decision makers behind the veil. Subjects who do not know the associated probabilities are less inequality averse as umpires than as involved decision makers behind the veil. Similarly, Croson and Konow (2009) compare behavior behind the veil of ignorance with impartial spectators and find that “stakeholders are less inclined to respond to the generosity of others than are spectators”. More recently, Becchetti et al. (2011) show that impartial spectators and, to a smaller extent, stakeholders behind-the-veil both reward talent more strongly than informed stakeholders.
Konow (2009) presents interesting results regarding the choices of impartial spectators. His vignette study shows that spectators are more likely to agree on what is fair when they possess more information. This finding suggests that the impartial spectator is an attractive approach to elicit fairness preferences. One might hope to find agreement when all relevant information is known. Amiel et al. (2009), using a vignette survey with students in several countries, present evidence suggesting that impartial spectators behavior is impacted more by social concerns (i.e. they transfer more money) than involved participants. Aguiar et al. (2013) compare fairness views of planners behind the veil of ignorance, impartial spectators and ideal observers (who are assumed to be omniscient) to find out which of those procedures is most useful in ensuring impartiality. They find that the ideal observers choose significantly more equal distributions than stakeholders behind the veil or impartial spectators. A benevolent dictator (a third party observer) is also used in Cettolin and Riedl (2016) as a device to elicit fairness ideals under conditions of uncertainty. Their result indicates that the fairness of uncertain allocations, like their certain counterparts, is judged according to a variety of different ideals.
In a study related to ours, Almås et al. (2019) compare the fairness preferences of Scandinavians and Americans using a large-scale online experiment. They show that Americans and Scandinavians differ significantly in their fairness views. While they also use a spectator design, there are several differences to our study. Specifically, Almås et al. (2019) (i) employ a different experimental design, including different treatments and different roles (‘workers’ versus ‘spectators’); (ii) introduce luck and merit; (iii) use American and Scandinavian participants, not German ones; (iv) have a significantly smaller set of socio-demographic background information; (v) and do not correlate behavior in the experiment with redistributive preferences and charitable giving. Differences (i) and (ii) have a strong impact on the design, as they result in a different set of fairness types.
In this paper, we document fairness types in a large, heterogeneous sample of the German population. Such samples exhibit a larger variation in demographic characteristics than standard lab samples, allowing for a more accurate picture of preferences in the population. In addition, the panel vehicle offers rich information about subjects collected in many waves over several years. Consequently, we can correlate fairness ideals to a large range of demographic variables as well as to political and policy preferences, and charitable behavior. Moreover, we show that our parsimonious and abstract approach to classify respondents into fairness types captures important aspects of individual preferences, that help to predict and understand voting behavior beyond what is contained in demographic variables.
Normative fairness ideals
Normative theory has carved out several ideal types of distributive justice. In our empirical analysis, we exploit the fact that these different normative theories predict different choice patterns and assign a fairness type to our participants based on the criterion that is revealed to be most important to the decision maker. This paper focuses on four types, egalitarian, maxi-min, efficiency, and maxi-max.
First, egalitarianism proposes a simple comparison rule for allocations (e.g., Roemer 1996). In this philosophy, equality is considered the only moral or fair criterion by which to judge outcomes. Hence, in this view, differences in outcomes (e.g., income) should be minimized. Although there is some discussion about what should be equalized, egalitarians are expected to redistribute windfall gains, such as any earnings from the experiment, equally. In our experiment, egalitarianism violates the Pareto principle, as the egalitarian allocation is dominated by the maxi-min allocation. Intriguingly, we find that egalitarianism is the most popular ideal type.
Second, Rawls (1971) suggested that if people were to be placed behind the veil of ignorance, they would choose distributions according to the difference principle. This principle selects the distribution that maximizes the minimum income. Based on the assumed general acceptance and impartiality of the participant behind the veil, he argues that this maxi-min rule is the only moral or fair selection criterion. However, from an empirical point of view, support for this principle seems somewhat less compelling. Frohlich and Oppenheimer (1990) find little support for the maxi-min principle, instead many participants endorse the efficiency principle with a floor constraint, i.e. maximize the sum as long as a minimum income is guaranteed.Footnote 5 In contrast, Mitchell et al. (1993) find considerable support for Rawls’ maxi-min principle in an experiment with a hypothetical society and decision makers behind the veil of ignorance. Similarly, Michelbach et al. (2003) find that a “considerable minority” uses maxi-min as a decision criterion.
Third, the (weak) Pareto principle allows another simple approach to distributive justice. If in allocation \(\mathcal {A}\) at least one person is better off and no one is worse off than in allocation \(\mathcal {A}'\), then \(\mathcal {A}\) Pareto-dominates \(\mathcal {A}'\) and \(\mathcal {A}\) should be preferred to \(\mathcal {A}'\). While this principle is compelling, it does not provide a complete ranking of all potential allocations, i.e. it does not compare losses of one individual with the gains of another. If these losses and gains are treated equally, this boils down to the simple efficiency ranking as, for example, proposed by Posner (1983). In the monetary allocations in our experiments, efficiency yields the same ranking of distributions as Harsanyi’s Harsanyi (1953), Harsanyi (1955) utilitarianism, as long as one assumes that utility is (close to) linear in the sums of money involved. Hence, to the extent that utility functions are linear over the stakes in our experiment, utilitarianism does not require separate attention in our set-up. Efficiency concerns have been shown to play an important role in settings with personal stakes in the distribution of money (Charness and Rabin 2002; Engelmann and Strobel 2004), but less is known in situations involving impartial spectators.
Finally, one can also obtain a (partial) ranking of allocations based on the maximization of the highest income. Admittedly, this maxi-max criterion seems like a rather theoretical possibility. The position of the most advantaged individual does receive some attention in liberal thought in the Rawlsian tradition, but then mostly as something that might be envied (Green 2013). The maxi-max criterion gives a complete and transitive ranking of the alternatives (Brafman and Tennenholtz 1997). We included this criterion to test the reliability of our data. If it had been chosen by a similar number of participants as the other criteria, this could have been a signal of random choices. As it turns out, maxi-max preferences are empirically irrelevant (\(n=14\)). We will therefore not discuss any empirical results relating to maxi-max preferences. For completeness, we will however display them in the figures and tables.