Some are just lucky - Questioning merit-based justifications of Inequality through ABM

Dominik Klein, Johannes Marx and Simon Scheller

Economic inequality constitutes a significant and persistent attribute of modern societies (Piketty, 2014) that has been shown to effect, inter alia, individual health (O’Donnell et al., 2015), well-being (Ferrer-i Carbonell and Ramos, 2014) and social participation (Mitchell and Shillington, 2002). The question of when and to what extent inequality can be justified has, hence, been the subject of extensive debate in political philosophy. One family of approaches (Nozick, 1974; Hayek, 2012; Rothbard, 1998) is based on the assumption that persons deserve economic benefits that are derived from their individual merits, actions or specific skills. Consequentially, so the argument, even grave existing inequalities can be justified when they result from individual efforts and abilities and differences therein. Nozick, for example, argues that inequality can be justified as long as processes of acquisition and transfer of goods adhere to certain basic principles of justice (Nozick, 1974, p. 158-9).

The common core of such arguments is the implied link between one’s success and the quality of individual actions. In other words, inequality is justified as long it is caused by free, individual choices (merit-based justification of inequality). Commonly and subliminally, many such argument assert merit as the default explanation for existing inequalities. When encountering inequalities, so the implicit argument, the default cause to be assumed are differences in skill or effort. This places the burden of proof onto the shoulders of those who claim that inequality is unjustified. In this presentation, we attack this implicit inference. By means of an Agent Based Model, we show that severe inequality can emerge even in an idealized setting, where all individuals share the same starting conditions. In this simulation, all agents are fully rational, equally equipped and hold identical initial belief. Any emerging inequality, hence, cannot be traced back to differences in skills or effort. The only assumption needed for inequality to arise is that one of the agents involved makes a single, minor mistake once in the beginning of a simulation run. The implicit inference involved in the merit based justification hence turns out invalid: inequality can emerge without differences in skills or effort. A fortiori, the emerging inequality is, on average, not to the detriment of the agent making the mistake. That mistake, hence, cannot be employed to justify the emerging inequality either. A basic assumption of our model is that agents can produce an economic surplus through pairwise economic interaction. Such interaction may, for instance, represent a joint production process or an employer hiring an employee for a certain job. Before being able to start production, however, both agents involved need to agree on how to distribute any potential surplus generated. This agreement process is costly, as any time spent on bargaining cannot be used for production. In the simulation, we operationalize this by limiting each pair’s overall interaction time.

The core strategic problem faced by agents, hence, is their choice of bargaining strategy. Tough bargaining policies can grant a high share of all benefits produced, but risk overall production to be low, as much time can be lost on bargaining. Softer bargaining approaches, in contrast, foster early agreement and leave more time for production. On the downside, these strategies may only result in a fraction of the possible benefits produced. A central parameter in the trade-off is the agents’ toughness, i.e. how many rounds of incompatible demands they are willing to endure before switching to a more agreeable offer.

Within the simulation, agents pick their bargaining strategy by a utility maximization rule. All agents are equipped with probabilistic beliefs about other agents’ bargaining strategies. Based on these beliefs, they compare their available bargaining policies in terms of expected utility and opt for the best choice in this respect. Moreover, agents update their beliefs. Every interaction provides a new data point on other players’ strategies. Once a bargaining process is completed, this information is used to update the players’ beliefs about the distribution of strategies. As we will argue, this updating policy suffers from a crucial asymmetry. Briefly speaking, agents can acquire exact information about softer agents, who are the first to depart from a tough bargaining stance. However, when paired against a tougher opponent, the player himself willl be the first to give in. In this case, he only receives incomplete information about how long the opponent was prepared to maintain her tough stance. As we show, this asymmetry has intricate effects on the long term quality of beliefs and expected utility considerations based thereon.

In the simulation, we assume one of these agents to make a slight mistake once. In her very first encounter, this agent does not play a utility maximizing strategy, but is one round off in when to soften up her bargaining stance. As it turns out, such a minor mistake is sufficient to trigger significant inequality in terms of long term accumulated wealth. Moreover, the emerging inequality is not to the detriment of the agent making the mistake: Averaging over many simulation runs, the mean income of the mistaken agent equals the entire population’s mean income. Hence, we conclude that the merit based justification of inequalities, inferring from the existence of inequality differences in skill or effort, is not valid: Even under equal starting conditions for all agents, severe inequality can emerge. As all agents are equally equipped and fully rational, this inequality cannot be attributed to any differences in skill or effort.

Beyond the insights gained on inequality and its normative status, we argue that the present model illustrates methodological aspects of agent based modeling for normative issues. The current model, crucially, does not aim for a realistic description of agents’ capacities. Rather, it represents an idealized even if scenario where agents are equally endowed and fully rational. It is exactly through such idealized models where all potential confounders are eliminated, that the underlying normative argument can be evaluated in full clarity. Within our normative context, idealization, hence is not to be seen as unavoidable defect, but as integral strength of the model developed.


  • Ferrer-i Carbonell, A. and Ramos, X. (2014). Inequality and happiness. Journal of Economic Surveys, 28(5):1016–1027.
  • Hayek, F. A. (2012). Law, legislation and liberty: a new statement of the liberal principles of justice and political economy. Routledge.
  • Mitchell, A. and Shillington, E. R. (2002). Poverty, inequality and social inclusion. Laidlaw Foundation Toronto.
  • Nozick, R. (1974). Anarchy, State, and Utopia. New York: Basic Books.
  • O’Donnell, O., Van Doorslaer, E., and Van Ourti, T. (2015). Health and inequality. In Handbook of Income Distribution, volume 2, pages 1419–1533. Elsevier.
  • Piketty, T. (2014). Capital in the 21st century. Cambridge: Harvard Uni.
  • Rothbard, M. N. (1998). The ethics of liberty. New York University.

Author: Research Group for Non-Monotonic Logics and Formal Argumentation

Created: 2019-02-26 Tue 09:37