Open AccessAN ANALYSIS OF THE DIMENSIONALITY OF TOEFL READING COMPREHENSION ITEMS
Author(s) -
Schedl Mary,
Gordon Ann,
Carey Patricia A.,
Tang K. Linda
Publication year - 1995
Publication title -
ets research report series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.235
H-Index - 5
ISSN - 2330-8516
DOI - 10.1002/j.2333-8504.1995.tb01662.x
Subject(s) - test of english as a foreign language , reading (process) , test (biology) , reading comprehension , psychology , mathematics education , construct (python library) , curse of dimensionality , item analysis , exploratory factor analysis , construct validity , language proficiency , language assessment , linguistics , computer science , psychometrics , artificial intelligence , developmental psychology , paleontology , philosophy , biology , programming language
The issue of exactly what is measured by different types of reading items has been a matter of interest in the field of reading research for many years. Language teaching and testing specialists have raised the question of whether a reading test for foreign students wishing to enter university in the United States should include questions testing abilities beyond linguistic and very general discourse competencies, or indeed whether it is possible to separate these language competencies from other competencies. The purpose of this study was to investigate the dimensionality of the TOEFL ® reading test, based on the specifications in use as of April 1991. Of particular interest was whether four item types identified in the test specifications as “reasoning items” could be shown to measure, in addition to general reading ability, any abilities not measured by the other item types in the TOEFL reading test. Two techniques, Stout's procedure and NOHARM analyses, were employed to investigate the hypothesized two‐factor model. In both cases the data failed to fit the model, indicating that TOEFL “reasoning items” cannot be shown to measure a unique construct. However, the follow‐up exploratory analyses indicated that all 10 test forms used in the study violated the assumption of essential unidimensionality, and all of the forms appeared to fit a two‐factor model where the second factor may be related to passage content or position.