Effect of learning on stakeholder negotiation outcomes : modeling and analysis of game-generated data

Abstract

A design negotiation game based on a stakeholder salience framework was created by the APACE research team to explore negotiation dynamics between stakeholders with individual attributes and agendas. Experimental data was collected anonymously during games played by groups of human participants through a web interface. It was found that the negotiation process takes a non-zero number of iterations even under conditions that strongly favor agreement. A realistic scenario was created based on extensive interviews with the major stakeholders involved in a real negotiation of a plan for a new government information technology system. Solution space exploration of this scenario demonstrated that the experimentally obtained solutions lie far from the optimality frontier. Performance differed significantly in two groups of participants with dissimilar professional experience; games played by interns achieved higher scores than those played by senior staff. An agent-based model was built to simulate multi-stage design negotiation. Utility functions of individual players were based on their private agendas. Players voted for a design according to the relative attractiveness of the design as established by the individual utility function. The negotiation process helps players discover other players' agendas. It was hypothesized that knowledge of each other's private objectives would enable groups of players to achieve design solutions that are closer to optimal. Effects of learning were introduced into the model by adding a fraction of the sum of all players' utility function to each individual utility function. Simulated games with learning effects yielded solutions with higher total player scores than simulated games without learning did. Results of simulated games with a substantial level of learning effects were similar to average experimental results from groups of interns. Results of simulated games without learning were close to the average results of games played by senior staff.