The Income Cliff in Households: Insights from Agent-Based Computational Modelling

Thursday, 19 July 2018: 15:30
Oral Presentation
André GROW, University of Leuven (KU Leuven), Belgium
Jan VAN BAVEL, KU Leuven, Belgium
In Western countries, the distribution of relative incomes within marriages tends to be skewed in a remarkable way. Husbands usually do not only earn more than their wives, but there is a striking discontinuity in their relative contributions to household income at the 50/50 demarcation line: many wives contribute less than or almost as much as their husbands, but few contribute more. This ‘cliff’ has been interpreted as evidence that men and women avoid situations where the wife would earn more than her husband, since this would run against traditional gender norms. In this paper, we use agent-based computational modelling to demonstrate that the cliff in the relative-income distribution can also emerge without such avoidance. We start from a simple economic matching model that has frequently been used in earlier marriage market research. We gradually relax some of the simplifying assumptions of this model, to create a series of increasingly realistic models. We feed each of these models with empirical income data from 27 European countries and submit them to systematic computational simulation experiments. Our results show that in each model a cliff can emerge from differences in men’s and women’s average incomes, even if they do not attach special meaning to a situation in which a wife earns more than her husband. For this, the specific assumptions of the different models do not matter. However, the fit between the empirical data and the simulation outcomes is closest for the model that makes the fewest simplifying assumptions. Taken together, our results shed light on some of the possible mechanisms that might have generated observed distributions of relative income within households, but also caution against inferring people’s partner preferences and social norms from aggregate-level mating patterns.