Structure And Application Of Weight3-Index Model Of Causal Induction

There are some drawbacks in psychological research of causal induction, include a):in themajority of study, it is not clear that the effect of background causes were hold constant betweensituations in which the candidate cause was present vs. absent; b): many researchers proposed manymodels to explain subject’s causal induction, but these models interpret all kinds of subject’sperformance by only one index; c): the cognitive load of different methods that subjects choose toinduce causal relation may affect subject’s performance, but the effect of cognitive load haven’tbeen detected drastically; d): the discrepancy between judgment of causal strength and causalstructure should be studied in depth.Base on exiting studies, the paper propose a weight three indexes (W-3I) model for causalstrength and structure judgment respectively, and test W-3I model’s predictive ability by sevenexperiments, which adopt a covariation present format that ensure the effect of background causeshold constant between situations in which the candidate cause was present vs. Absent. In order tocomparison model’s superiority, correlation coefficient and mean square error were computed.The three indexes of causal strength judgment are ΔP, PPC and P(E/C)(P(E/C) for generationcontingency and P(-E/C) for prevention contingency), the weight is0.5186,0.3653,and0.0936insummary format, is0.4225,0.0018,0.5229in sequential format. The three indexes of causalstructure judgment are ΔP, PPC and CCP, the weight is0.3029,0.4799and0.2235in summaryformat. The W-3I model of causal structure judgment in sequential format haven’t been establishedbecause there has no such study.Experiment1and2test the predictive ability of W-3I model of causal strength judgment insummary format. Experiment1considered the impact of ΔP and P(E/C) on strength judgment,results showed that ΔP has much bigger impact than P(E/C) on subject strength judgment, just asW-SI model’s prediction. Experiment2investigated cause density effect (P(E) effect) in causalstrength judgment, results revealed that there are no any cause density effect in4designs of ΔP:ΔP=0,.25,.50and.75. The prediction of seven models, W-3I, ΔP, PPC, P(E/C), wpCI, EI, H, SS,were compared. For data of experiment1or2, the prediction of W-3I get the highest correlationcoefficient, the prediction of SS model get the lowest mean square error with data. The mean squareerror between prediction of W-3I model and experiment1data second to the mean square errorbetween prediction of SS model and experiment1data, the mean square error between prediction ofW-3I model and the data of experiment2very close to the mean square error between prediction ofSS model and the data of experiment2.Experiment3and4test the predictive ability of W-3I model of causal strength judgment insequential format. More information is presented to eliminate cognitive load discrepancy betweenget ΔP and P(E/C). It can be draw from experiment’s data that when P(E/C) is high(P(E/C)=0.75),subject ’s causal strength judgment changed along with the change of ΔP, either in experiment3(use extent question) or experiment4(use counterfactual question). The model compare betweenseven models just mentioned in previous paragraph show that the prediction of W-3I got the best fitwith the data of experiment3or4: W-3I model’s prediction get the highest correlation coefficientand the lowest mean square error with data.The W-3I model of causal structure judgment based on experiment5,6,7. Experiment5and6 probe into the influence of CCP index. Results showed that CCP has enormous implication onpeople’s causal structure judgment, when ΔP hold constant, the effect of CCP can offset the effectof PPC. Experiment7detect the effect of virtual sample on structure judgment. Results indicatedthat virtual sample has remarkable effect on subject’s judgment: higher virtual sample induce lowerstructure judgment for design of PPC=0, but induce higher structure judgment for design ofPPC=0.5and1. The prediction of seven models, W-3I, ΔP, PPC, CCP, Support, SS+andχ2, werecompared. For data of experiment5and7, the prediction of W-3I get the best fit with experimentdata: the highest correlation coefficient and the lowest mean square error; but for data of experiment6, the correlation coefficient between prediction of W-3I and data second to the correlationcoefficient between prediction of χ~2and data, the mean square error between prediction of W-3Iand data is higher than the mean square error between two model’s(Support and χ2) predictionand data.In short, the prediction of W-3I model got the best fit with experiment data: in sevenexperiments, the prediction of W-3I model got the highest correlation coefficient for sixexperiments and the lowest mean square error for five experiments. The superiority of W-3I modelmay come from the composition of model: ΔP, PPC, P(E/C) for strength judgment and ΔP, PPC,CCP for structure judgment. Three index make W-3I model can interpret many study findinginduced by one or more indexes, and induced by complex causal situation. Further research shouldfocus on context that contain two or more candidate causes, that model every subject’s performanceon every contingency, that causal structure judgment use sequential format, and so on.