The Effectiveness Of Person-fit Statistics In Personality Questionnaire

Content: Person-fit statistics (PFS) refer to the consistency of an individual's response pattern, ether with a given measurement model or with other examinees'response patterns within a given sample. They were a series of new kinds of method to discriminate the aberrant responses behaviors in psychological tests which based on Guttman model and Item response theory. If we can distinguish the aberrant item score patterns, the test score will be more interpretable, and the reliability and validity will be heightened. Nowadays most of the researches in person-fit statistics were concentrated on ability tests and computer adaptive tests, but less in personality questionnaires. Looking for appropriate PFS of personality questionnaire and raising the effectiveness of PFS in personality questionnaire will make sense to the development of personality tests.The first research had a special design to get the positively and negatively worded EPQ questionnaire for the same group of participants. Choose the number of items which the response vector in positively worded items is unequal to that in the negatively worded items (called unequal index Y) as the criterion of estimating the person-fit statistics. Four person-fit statistics such as C*,U3,lz and ECI4z was chosen to investigate influences to the reliability and validity of the tests, and analyze the correlation between the four person-fit statistics and unequal index Y. Then the effectiveness of PFS in personality questionnaire were investigated .and the power of the four person-fit statistics was compared. As shown in the result, after deleting some percentage of aberrant individuals, the reliability, criterion-related validity and the construct validity were heightened; the reliability, criterion-related validity and the construct validity of the half of the good-fit participants were higher significant than that of the bad-fit participants; the four PFS had different correlation levels with Y, the correlation coefficient between lz and Y was the highest; comparing the four person-fit statistics, lz and C* were the most sensitive to the reliability and the validity. So we concluded that PFS can rectify the reliability and validity. It meant that PFS were suitable to personality tests and lz based on IRT and C* based on CTT were the most ideal.In the second research, we simulated a large scale of data by Monte carlo simulation method to investigate the adaptability of lz in the likert scale. As shown in the result lz was normal distribution。-1.65 could be seen as the criterion to discriminate normal response patterns and the aberrant response patterns. It was good in all kinds of aberrant response patterns. The distribution of lz was influenced by the category of response patterns, the proportion of aberrant individuals and the proportion of aberrant items. The validity of lz was influenced by the category of response patterns and the proportion of aberrant items.Though the content and the method were different of these two researches, they proved the same conclusion that the PFS was effective in personality questionnaires. This research was just an explorative process. There were many limitations in it. For example the effectiveness of PFS was influenced by many factors such as the large of sample, the number of items, the proportion of aberrant items, the position of aberrant items, the parameters estimation method of IRT and so on, all of these must be investigated in further study.