Split-the-Difference Offers: Theory and Widespread Empirical Evidence from the Field

This paper uses novel, detailed data on sequential offers from seven vastly different real-world negotiated-price settings to document a pattern that leaps out of each dataset: Bargaining agents favor offers that split the difference between the two most recent offers. Our settings include negotiations for used cars, auto insurance injury claims, TV game show questions, auto rickshaw rides, housing, international trade tariffs, and online retail. Our sample sizes range from several hundred bargaining sequences to tens of millions, depending on the dataset. We offer a theoretical model demonstrating that split-the-difference behavior is compatible with game theory as a perfect Bayes equilibrium, but it is far from predicted by it, and that 50/50 splits of offers do not yield 50/50 splits of actual surplus. We present a number of patterns in the data that point to split-the-difference offers being a strong behavioral norm in bargaining, whether high-stakes or low-stakes negotiations.