Improvement And Application Of Abnormal Answer Data Detection Index Based On Cumulative Sum Method And Change Point Analysis Metho

In today’s society,with the rapid development of the Internet and information technology,the cost of data collection is becoming increasingly low.Researchers can use online surveys or online tests to collect user data,and summarize the laws of user behavior from the perspective of big data,thereby inferring their psychological needs.In theory,a user’s response pattern should be able to completely reflect their true level of psychological characteristics.However,in practice,users may exhibit abnormal response patterns due to various reasons.The existence of abnormal response patterns can contaminate data,leading to a decrease in the quality of test data and a threat to the effectiveness of the test.Therefore,effective detection of abnormal responses is crucial both in theory and in practice.Researchers commonly use Person Fit Statistics(PFS)to identify exceptional subjects.Compared with traditional PFS,PFS based on Statistical Process Control(SPC)can retain the response information of each item at the test level and make use of it.Currently,there are two types of PFS based on statistical process control,one based on Cumulative Sum(CUSUM)and the other based on Change Point Analysis(CPA).CUSUM presents the potential abnormal behavior of subjects in a visual way and can intuitively identify exceptional respondents and corresponding items,but the location of exceptional items can only be determined through human observation.In contrast,CPA can overcome the weakness of CUSUM and can directly locate exceptional items.This identification of change points is useful for researchers to achieve effective data cleaning.However,in some situations,the detection performance of CPA is not as good as CUSUM.Therefore,this study proposes to combine CPA and CUSUM to develop a new type of PFS,aiming to combine the advantages of the two indicators and detect abnormal responses in psychological tests,while comparing the detection performance of PFS for exceptional behavior of subjects in different psychological tests.Firstly,this study analyzed the advantages and disadvantages of CUSUM and CPA,and combined the information from both to construct a new PFS indicator.In order to verify the performance of the new indicator in detecting the random response behavior of subjects in the later stage,this study compared the detection performance of the new indicator with the two existing indicators under different simulation scenarios through Monte Carlo simulation.At the same time,this study compared the actual detection utility of different PFS in psychological tests using real data.The results of the study showed that:(1)The statistical power of each method is mainly affected by the test length,change point position,and misfit ratio,and the type I error rate is mainly affected by the misfit ratio of the subjects.(2)In the two-level scoring test,the statistical power of the new indicator is higher than CUSUM and CPA method under all simulation conditions.In the multi-level scoring test,the new indicator is more suitable for scenarios where the misfit ratio of subjects is higher.(3)In the empirical study,the effects of CUSUM,CPA,and the new indicator on the reliability and validity of the test were examined.It was found that after deleting the non-fitting subjects,the reliability and validity of the test were significantly improved.Among them,the indicators belonging to CPA showed a better improvement effect.The new indicator proposed in this study combines the advantages of CUSUM and CPA,and can further improve the detection performance of random response behavior of subjects,which promotes the application and development of personal fit statistics.In addition,it also provides important reference and guidance for the practice of psychological tests.