关键词: EPQ, 信度概化, 层次多元回归分析

Abstract:

EPQ is widely used all around the world, and there is no exception in China. In China, five revised versions of EPQ are available, which are all popularly used. Caruso(2001) has conducted a reliability generalization analysis on the EPQ used in other countries and has gained some significant results. However, no such analysis has been made so far in the Chinese context. This paper attempted to study the reliability generalization on the EPQ in China. Methods: The analysis was done from the perspectives of descriptive statistics and hierarchical multiple regression with data from seven Chinese major psychology journals from 1998 to 2008. What’s more, our findings were compared with those of the analogous study on EPQ in other countries, which was conducted by Caruso et al. Comparing the results of the Reliability Generalization on EPQ at home and abroad, the following similarities can be found: 1) the sampling characteristics do affect the reliability coefficients. 2) A low proportion of articles report reliability coefficient or its range from the data at hand, 5.80% and 6.20% respectively; 3) P scale owns a low reliability coefficient, probably because its unidimension condition is not met; 4) standard deviation of scores in subscales is the main predictor variable for the reliability coefficients of P, N, E and L subscale. However, differences in the studies are also discovered: 1) 84.82% of researches in China do not report reliability coefficient from the data at hand, comparing with 62.51% of researches in the other countries, which indicates that more scale users in China ignore the reliability from the data at hand; 2) Variables such as the number of items, the mean of scores, standard deviation of age, sample type have different predictive functions on the reliability for EPQ in China and in other countries. In China, the prediction from mean of scores to P, N and L subscale as well as the prediction from number of item to P, E and N subscale are found to be statistically significant, which is not found in other countries. On the other hand, the prediction from standard deviation of age to P, and L subscale and the prediction from sample type to P subscale are not statistically significant in China, but are all significant in other countries. In summary, the analysis and comparison demonstrate that 1) “reliability induction” is inappropriate when using questionnaires available; besides reporting the background and size of samples, it is of great necessity to report the reliability coefficient of the samples at hand, which can make the research complete by providing the information mentioned above; 2) a certain heterogeneity of samples may enhance the reliability in the use of EPQ; 3) the increase of item numbers which does not follow the psychometric rules will not improve the reliability of scores necessarily.

Key words: Eysenck Personality Questionnaire (EPQ), reliability generalization (RG), hierarchical multiple regression analyses