Bargaining in Legislatures: An Experimental Investigation of Open versus Closed Amendment Rules

We investigate the Baron and Ferejohn (1989) noncooperative game theoretic bargaining model of legislative equilibrium. Legislative outcomes are sensitive to formal rules specifying who may make proposals and how they will be voted on. With a random proposal recognition rule and a closed amendment rule (proposals are voted up or down with no room for amendments) the model predicts no delays in benefit allocation, that benefits will be allocated to a minimal winning coalition, and that benefits within the coalition will be strongly skewed in favor of the proposer. In contrast, with a random proposal recognition rule and an open amendment rule (proposals may be amended before they are voted on) the model predicts delays in benefit allocation, that benefits will be more evenly spread among winning coalition members, and that coalitions need not be restricted to a minimal majority. With experience we find strong qualitative support for the model's predictions: All proposals are passed without delay with the closed rule versus 81% of all proposals with the open rule. Minimal winning coalitions are effectively proposed in 67% of all cases with the closed rule versus 4% with the open rule, and benefits are more evenly distributed with open rule. Quantitative predictions of the model fail however: Most importantly, proposers consistently fail to allocate themselves anything close to what the theory predicts. Further, the probability of immediate acceptance is much higher than predicted in the open rule as proposers consistently expand the winning coalition beyond the model's prediction in attempts to limit amendments. The evolution of play over time is reported (outcomes under both treatments are much more similar early on then later). Tests show that subjects' votes in favor of a proposed allocation are significantly affected by their own share (in the expected direction) but that the distribution of shares across all voters has no significant effect.