Résumé
Role-playing games (RPG) are decision-making supports, developed in the field, and are used to help stakeholders' "point of view wording" in resources management situations. A particular RPG (KatAware) was built in the Kat River watershed (in South Africa), according to the ComMod approach, in order to help people in designing a shared water management plan. Only two sessions of KatAware were played, in which the players seemed to behave in a cooperative way. Unfortunately, the results obtained after having run only two sessions are ambiguous, as stationary replications of these sessions are difficult to implement. The experimental method allows such a replication by controlling all the parameters. The objective of the experimental method put in place is to assess game sessions outcomes and to reflect about general design rules and dynamics followed when building role-playing games. Among all the parameters, the context of the game influences the outcomes. After having simplified KatAware, our study shows how the context, defined as the level of information carried by the game, could be decomposed and re-composed according to its main dimensions, i.e.: illustration of the instructions; communication; repetition of periods; players' experience. The impact on players' choices of different levels of these dimensions were experimentally tested with students (at the "Experimental Economics Lab in Montpellier", France). We assessed the impact of the two first dimensions: "illustration" and "communication", by comparing the outcomes obtained after having varied their levels with the results obtained with the referential treatment, in which neither illustration nor communication were introduced in the protocol. We showed that the addition of watery illustrative elements within the instructions, first by only introducing the sentence "this experience is based on a water management situation", and then by describing such a situation through a story telling, increases noise in behaviours. However, this watery context does not impact players' decisions in the same way as another context chosen - in our protocol: employees in a firm -, which provided more familiar cultural references to the players. Finally, players' argumentation of their choices - through an ad-hoc controlled communication treatment - improved learning outcomes: while the observed choices converge towards theoretical equilibrium faster when communication among players is allowed, the final equilibriums are not significantly different in presence or in absence of communication. (Résumé d'auteur)