Modeling Analysis Of The Impact Of Reputation On Game Strategy Selection

Generalized linear model and generalized linear mixed models are two types ofstatistical model, which, in recent years, have found their wider uses and moreimportant position in practical applications. It has a great significance when applyingthe two types of statistical model into the study of the effects of credibility on gamestrategy choice.This paper, based on the principle of repeated two-person prisoner’s dilemmagame, designs some related experiments on the basis of Dreber et al.(2008). Therandom two-person prisoner’s dilemma game can be realized through computersimulation. And strategy selection and return information of the participants in thegame can be collected with Z-tree software. Participants in the experiment can choosecooperation strategy, non-cooperation strategy, or punishment strategy. Before theexperiment, participants will be informed that they will be on to the next round of thegame with the probability of75%after the current round of game with the opponent,but they don’t know how many rounds of interaction game they will take with theopponent eventually. A total of five times experiment are carried out with134participants in the paper. The five experiments are divided into anonymous group T1and non-anonymous group T2by the different initial conditions that the participantsare anonymous or not in the experiment.Based on the experimental data, firstly, the Mann-Whitney U testing method isused for preliminary analysis of the experimental results. Secondly, the correspondinggeneralized linear model and generalized linear mixed models are establishedrespectively by the cooperation strategy usage frequency, non-cooperation strategyusage frequency, punishment strategy usage frequency. Finally, a reasonable model isconfirmed according to the AIC criterion.Study has shown that generalized linear mixed models have a better fitting effecton the modeling analysis of the impact on game strategy choice of credit compared to the generalized linear model. From the results of the established generalized linearmixed models, we can get the conclusions that significant differences are existed inthe choice of three strategies between participants in the two groups. Compared withthe anonymous experiment, when the participants’ name are displayed, they will bemuch more inclined to use cooperation strategy, while less use non-cooperationstrategy or punishment strategy. Existence of the credibility factor can be better topromote the cooperation between groups, and inhibit the use of punishment strategy.