Study On Neural Network Modeling Of Cooperation In Public Goods Dilemma

Based on a laboratory experiment of cooperation in public goods dilemma, the paper establishes a BP neural network model. This model has high forecast accuracy, and could be used to forecast people's behavior. The paper applies experiment and BP model to prove research hypothesis and makes conclusions as follows:a) Cooperation in four information display usage are significant different. When investor number & name are displayed, people are most inclined to cooperate, the next are display usage of investor number and non-investor number & name. People are least inclined to cooperate relatively if nothing is displayed. This result illuminates that "number" and "name" are both contributing for cooperation. The neural network model shows that their contribution is two-sided: positive & negative.b) Logistic regression indicates that, two dimension of others' behavior have significant influence on people's behavior. Investor/non-investor number has a negative influence on investment probability, while the past experience has a positive one. Import these two significant statistical variables to BP neural network, which creates a highly accurate model. The model shows that, the influence of input on output is not monotone function, and people's behavior is not always conditional cooperation either.c) The paper also finds out that, there are "Start-up Effect" and "Fifty-percent Decline". "Start-up Effect" means people would like to invest so as to push others, when experiential value is few. "Fifty-percent Decline" means that break-even point will activate the motivation of greed, urging people not to cooperate.