A buyer makes a final offer for a small firm and will walk away if it is refused. A union and an employer exchange proposals while a strike consumes the profits they are dividing. How a surplus is split depends on who can commit, who can refuse, and what delay costs each side.
The three variants test backward induction against concerns for fairness. In the ultimatum game the question is whether responders accept small positive amounts and whether proposers anticipate their refusal. The dictator game removes the responder's veto, so that a comparison of the two separates generosity from the fear of rejection. Alternating offers over a shrinking pie test whether the cost of delay shapes the division as the theory predicts.
In the ultimatum game, studied first by Güth, Schmittberger and Schwarze (1982), a responder who cares only about money accepts any positive amount, so the proposer offers the smallest unit and keeps the rest, and in the dictator game the proposer keeps everything. Neither prediction depends on the size of the pie.
With alternating offers, the last stage is an ultimatum game over what remains of the pie, and each earlier proposer offers the other side what that side would obtain by refusing, so that agreement is reached at once. The finite case goes back to Ståhl (1972), and Rubinstein (1982) showed that the game without a final stage has a unique subgame perfect equilibrium. On this site a common discount factor states the share of the pie that survives each rejection. With two stages the first proposer keeps one minus the discount factor, and with three stages and a factor of 0.8 the first proposer keeps 84 percent. Without a final stage the share is one divided by one plus the factor, about 56 percent. Fehr and Schmidt (1999) show how a dislike of unequal outcomes changes these predictions.
In the meta-analysis of Oosterbeek, Sloof and van de Kuilen (2004), covering 75 ultimatum results, proposers offered 40 percent of the pie on average and 16 percent of offers were rejected. Offers below a fifth of the pie are refused about half the time in student populations. Behavior varies across societies, and Henrich et al. (2001) report mean offers between 26 and 58 percent in fifteen small-scale societies.
Dictators give less than proposers in the ultimatum game, and Engel (2011) finds an average transfer of 28 percent across more than six hundred treatments, with 36 percent of dictators giving nothing and 17 percent giving half. The difference from ultimatum offers led Forsythe et al. (1994) to conclude that fairness alone does not explain what proposers do.
In alternating offers, opening proposals lie between the equal split and the equilibrium. Binmore, Shaked and Sutton (1985) found that the most common opening offer in a two-stage game was half of the pie, and that participants who then took the proposer's role demanded close to the equilibrium share. Ochs and Roth (1989) found the equilibrium to have little predictive power, and they observed that most responders who rejected an offer went on to propose a division that gave them less money than the offer they had refused.