An Application of Full Information Item (Bi)Factor Analysis to the Reading Comprehension of a TOEIC Practice Test

영어 독해의 요인 구성에 대한 학계의 일반적인 견해는 독해능력은 읽기능력이라는 주요인과 읽기 외의 비본질적인 요소를 반영하는 부차적 요인으로 구성되었다는 위계 모형이 주류이다. 그러나 TOEIC 시험의 독해 부분에 관한 Educational Testing Service (ETS)의 연구 보고서(Woodford, 1982; Wilson, 2000)에 따르면, 영어 독해 능력을 비위계적 이차원 구조로 파악하여 그 두 차원을 독해 문항의 형식에 맞추고 있다. 본 연구는 독해 능력의 요인은 시험의 문항 형식이 갖는 영향과는 분리되어야 한다는 전제하에, 국내의 한 대학에서 시행되었던 토익 모의고사 결과를 full information item (bi)factor analysis를 이용하여 분석하고, 토익 독해 시험의 요인을 비위계적 이차원 구조로 파악하는 것은 문항 형식의 차이에 따른 변량을 요인으로 해석하는 오류일 가능성을 제시하였다.

While the majority of the studies on the factor structure of English reading comprehension test supports the hierarchical model with one dominant reading ability and a minor factor for nuisance effects, the existing literature on the factor structure of the TOEIC reading section supports two non-hierarchical factors. To examine the factor structure of the TOEIC test, the current study applied full information item (bi)factor model to a set of reading comprehension data from a TOEIC practice test administered in a university. Unlike the traditional factor analysis of test items based on the inter-item correlations, this new methods use the item response data directly by fitting Thurston's multiple-factor model (a multi-dimensional item response theory model for dichotomous items) in its estimation. Then the IRT parameter estimates are translated into factor analysis parameters. The bi-factor model is a specialized version of the full-information item factor analysis by enforcing each item to have non-zero loadings on a general factor and at most one group factor. The results by these methods showed that the factor structure of the reading section can be more efficiently explained by one primary factor (reading ability) loaded across all the parts of the test and a minor factor (test method effects) for residual variance, which is consistent with the current factor studies. This could imply that one of the two factors in Wilson (2000) is likely to be related to the test method effects (item group effects), not to another aspect of reading ability. Further implications from the studies are discussed