Abstract:
Local independence of test items is an assumption in all Item Response
Theory (IRT) models. That is, the items in a test should not be related to
each other. Sharing a common passage, which is prevalent in reading
comprehension tests, cloze tests and C-Tests, can be a potential source
of local item dependence (LID). It is argued in the literature that LID
results in biased parameter estimation and affects the unidimensionality
of the test. In this study the effects of the violation of the local
independence assumption on the person measures in a C-Test are
studied. A C-Test battery comprising four passages, each containing 25
blanks, was analysed twice. Firstly, each gap was treated as an
independent item and Rasch’s (1960) dichotomous model was
employed. In the second analysis, each passage was treated as a super
item and Andrich’s (1978) rating scale model was used. For each
person, two ability measures were estimated, one on the basis of the
dichotomous analysis and one on the basis of the polytomous analysis.
The differences between the two measures, after being brought onto the
same scale, are compared and the implications are discussed.
Machine summary:
"The probability that the first item is wrong when the second is right is: (View the image of this page) The important issue about these equations is that they contain no person parameters, which means that the relative difficulties of the items can be estimated without assuming anything about the ability of the persons who happen to have taken the test.
Table 1 Measure order and fit statistics for reading items (View the image of this page) Data Collection and Analysis To collect the data, the participants were firstly asked if they would volunteer to take part in a study on "language profcicency and C-Test".
Figure 3 Plot of person measures from the two analyses with the hypothetical cut-off score (View the image of this page) To corroborate the results more and to see whether the contextual clues vary with the proficiency levels of the testees, along with the C- Test battery, the subjects took a reading test which was intended to be a test of text-level skills.
Table 3 Correlations between the C-Test and the reading test for the 3 ability groups Low-Ability Group Mid-Ability Group High-Ability Group Raw Score Range (View the image of this page)Conclusions The results of the study showed that the low-ability students have lower measures on the dichotomous analysis, apparently because they cannot take advantage of the context clues because of their low proficiency and even what we call context clues here work against them because they have lower measures on the dichotomous analysis."