Keywords

participatory modelling; parameterization; Bayesian network; game approach; Madagascar

Location

Session D1: Tools And Methods of Participatory Modelling

Start Date

12-7-2016 10:30 AM

End Date

12-7-2016 10:50 AM

Abstract

In participatory modeling, actors are a main source of information and at the same time users of the model outputs. Many different approaches have been applied for incorporating actors into modeling processes. While games have proven to be valuable as a mean for parameterizing agent- based models, they have never been used to parameterize Bayesian networks even though Bayesian networks have been extensively used in participatory approaches.

The reasons for the absence of gaming approaches in the parameterization of Bayesian networks are potentially two-fold: (1) assessing of the conditional probabilities in the network is complex as, both, probability and conditions are hard to grasp for human beings; (2) usually there are many conditional probabilities that are needed for one conditional probability table.

In this contribution, we present a game and its application in a case-study in north-eastern Madagascar that was used to help parameterize a Bayesian network to model land-use decisions. In the game, participants had to take decisions concerning their labor input for their different land-use types given certain game-inherent conditions. Thus, participants were implicitly providing parametrization data without having to bother about conditional probabilities. The game was played with farmers and representatives of other land use-related institutions (e.g., consultants, civil servants). Results show how the qualitative findings of a game can transparently be used for model parameterization. Furthermore, this contribution discusses lessons-learned needed to be taken into account for further applications.

 
Jul 12th, 10:30 AM Jul 12th, 10:50 AM

Parameterizing Bayesian networks with a game: a case-study from north-eastern Madagascar

Session D1: Tools And Methods of Participatory Modelling

In participatory modeling, actors are a main source of information and at the same time users of the model outputs. Many different approaches have been applied for incorporating actors into modeling processes. While games have proven to be valuable as a mean for parameterizing agent- based models, they have never been used to parameterize Bayesian networks even though Bayesian networks have been extensively used in participatory approaches.

The reasons for the absence of gaming approaches in the parameterization of Bayesian networks are potentially two-fold: (1) assessing of the conditional probabilities in the network is complex as, both, probability and conditions are hard to grasp for human beings; (2) usually there are many conditional probabilities that are needed for one conditional probability table.

In this contribution, we present a game and its application in a case-study in north-eastern Madagascar that was used to help parameterize a Bayesian network to model land-use decisions. In the game, participants had to take decisions concerning their labor input for their different land-use types given certain game-inherent conditions. Thus, participants were implicitly providing parametrization data without having to bother about conditional probabilities. The game was played with farmers and representatives of other land use-related institutions (e.g., consultants, civil servants). Results show how the qualitative findings of a game can transparently be used for model parameterization. Furthermore, this contribution discusses lessons-learned needed to be taken into account for further applications.