Choice of optimal item response model for analysis of self-report questionnaire
- VernacularTitle:用项目反应理论分析自陈量表时最佳模型的选择
- Author:
Jing ZHOU
;
Qingke GUO
- Publication Type:Journal Article
- From:
Chinese Journal of Tissue Engineering Research
2005;9(44):187-189
- CountryChina
- Language:Chinese
-
Abstract:
BACKGROUND: Item response theory (IRT) is widely used in the westem world, but its use for non-cognitive measurement has been very limited in China. It is necessary to explore whether IRT is applicable in non-cognitive settings and find the optimal model.OBJECTIVE: To find the optimal IRT model for self-report questionnaire analysis with the instance of emotional competence scale (ECS) and compare the effectiveness of different models in light of model data fitting and the volume of information provision.DESIGN: A comparative study of different psychometric models tested with the same data set.SETTING: Shandong Laiwu Vocational and Technical College; Department of Psychology, Liaoning Normal University.PARTICIPANTS: Totally 617 college students of Shandong Normal University (311 males and 306 females)and 564 senior high school students of the 17 Laiwu High School (283 males and 281females) participated in the study during May 2004.METHODS: Emotional competence scale (ECS) was adopted for psychometric measurement for 5-point Likert-type items. The scale was divided into 9 subscales, namely impulse control, empathy, persistency, interpersonal intimacy, social skills, emotional regulation, emotional stability, sense of responsibility, and self-confidence. One-parameter, two-parameter and three-parameter logistic model and grade response model were used for analysis of the 9 subscales, followed by goodness of fit test (analogous to Chi-square test) and comparison of the measurement precision (reliability)of the three models.MAIN OUTCOME MEASURES: Log-likelihood ratio, average χ2 statistics and average volume of information when one-parameter, two-parameter and three-parameter Logistic model.RESULTS: Totally 1 220 questionnaires were handed out and 1 181 with valid answers were obtained. Comparison of the 3 logistic model showed that the two-parameter logistic model had the least log-likelihood with the least items whose mean square residual error were greater than 2, and the volume of test information was greater than that provided by one-parameter model and no less than the three-parameter model. Therefore, the two-parameter logistic model was the best for 2-point scoring model. But the measurement precision of two-parameter Logistic model was lower than that of multi-grade response model.CONCLUSION: When 2-point items are adopted in self-report questionnaire, 2-parameter logistic model can be applied but not 1- or 3-parameter Logistic models. But when the questionnaire uses items that have more than 2 response grades, the measurement precision can be better than that of 2-point data. Merge of the options for the items may result in lowered measurement precision.
