关键词: 识字能力, 单维性检验, 探索性结构方程建模, 非参数项目反应理论

Abstract: Comparability of test scores are affected by the test's dimensionality, before estimating student's scores, we need to estimate literacy's dimensionality. The study aims to analyze dimensionality of literacy ability based on dichotomous items scale for compulsory education phase students. We assume that literacy is unidimensional. The tool is Chinese character quantity test for compulsory education stage students, the test is constructed of two editions; the first edition is used for 1 to 2 grade students, and is consisted of 36 Chinese characters. The second edition is used for 3 to 9 grade students, and is consisted of 45 Chinese characters. And the test's reliability and validity are good, can be used to analyze dimensionality of literacy ability. More than 1300 students, from one to eight grades students, are selected randomly. Using exploratory structural equation modeling (ESEM) and nonparametric item response theory (NIRT), including the MSP and DETECT, to analyze literacy dimensionality. At first, do exploratory structural equation modeling analysis, using Mplus6.0. The model fit analysis results show, the model fit index for unidimensional model and multidimensional model are all acceptable, and the multidimensional model is better than unidimensional model, but considering of terseness of model, the unidimensional model is more appropriate. The eigenvalue analysis results show, the largest eigenvalue divide the second eigenvalue is 2.62 for 1~2 edition, and 5.35 for 3~9 edition test. According to the criterion, the 3~9 edition test is one-dimensional test, and the 1~2 edition is not. For factor loading analysis, in 1~2 edition test, in single factor model, the factor loading are above than 0.4 for all, except the 8 item. In double factors model, in despite of the 23 item is less than 0.4, the 1 item to 11item and 14 item are all belong to the factor 1, others are belong to the factor 2. Because items are assigned according to the word frequency from high to low, the results indicate, word frequency influenced literacy, but it is just a difficult dimensionality rather than an ability dimensionality. The 3~9 edition test shows similar results. In a word, the unidimensional model is superior to the multidimensional model, and in multidimensional model, didn't find differences between sound and meaning of literacy. The multidimensional results reflect a difficult dimensionality characteristics, is the role of word frequency. Second, nonparametric item response theory analysis is conducted, using Mokken scale to test dimensionality with MSP 5 procedure, the results find, 1~2 test and 3~9 test show unidimensional scale features as c increasing. At last, using DETECT to test dimensionality with DETECT v.2.1, the results show, D<0.2 for 1~2 test and 3~9 test, that is to say, 1~2 test and 3~9 test show unidimensional scale features. For three methods, the literacy ability tends to accept unidimensional hypothesis. We can draw the conclusion, the literacy ability tend to be unidimensional ability, but would be affected by word frequency. So teachers should pay attention to recreate literacy teaching, let students read more, improve students' familiarity of literacy, in order to help students to increase quantity of literacy. But there are some aspects need to improve, for example, at selecting participants, this study selects participants only from eight grades, from nine grades would be better; and the number of 1to 2 grade students is only 253, increasing the number would be better.

Key words: literacy ability, unidimensional assessment, exploratory structural equation modeling (ESEM), nonparametric item response theory (NIRT)