An in-house placement test for Japanese university students was constructed based on neural test theory (NTT; Shojima, 2008), a latent rank theory for analyzing test data using the mechanism of a self-organizing map (Kohonen, 1995). The test consists of three different types of questions: vocabulary and grammatical questions (Vg), listening comprehension with dialogue (Dig), and listening comprehension with monologue (Mlg). First, 80 Vg, 47 Dig and 35 Mlg questions were answered by 119 to 222 students. Then misfit items were eliminated using both misfit statistics and item reference profile. Ultimately, 32 Vg, 13 Dig, and 19 Mlg questions were used for the placement test. The KR-20 of the test after the misfit elimination was .706-.863. Seventy-five students who answered all items used for the test were divided into 5 groups based on the latent rank estimated by NTT (R). The results were compared with latent abilities estimated by one parameter logistic model (θ). The correlation coefficients between R and θ were .90-.96. In addition, the results were compared with other English proficiency test scores: CASEC and TOEIC Bridge the students took a few weeks after the placement test. The correlation coefficient between R and CASEC was.80. The correlation coefficient between R and TOEIC Bridge was.89. Although the cases are limited in number, the present study suggests that the reliability and validity of NTT are enough for an in-house English placement test. Also, test data analysis utilizing NTT is comprehensive and easier to interpret than analyses using other test theories.
抄録全体を表示