University of Oslo Workshops June 29-30 Conference July 1-3 ...

Exploring Local Item Dependency for Items Clustered around Common
Reading Passage in PIRLS Data
Valerie Quittre, University of Liege, Belgium
Christian Monseur, University of Liege, Belgium
Since the IEA 1991 Reading Literacy Study (Elley, 1994; Wolf, 1995), cognitive data of
international assessments are usually scaled according to IRT models. One of the
cornerstones of IRT models is the assumption of Local Item Independence (LII). In PIRLS
assessment, as well as in other international assessments in reading literacy such as PISA
(Organisation for Economic Cooperation and Development - OECD), testing material is
hierarchically structured, i.e. several items are related to a common passage. Monseur & al
(submitted) noted that “this test format may be viewed as the most appropriated to assess a
complex process such as reading comprehension. In real life situation, students have to use
different cognitive processes to understand various components of a same text”. This
embedded structure might therefore violate the assumption of LII. As stated by Embretson
and Reise (2000, p. 188), “Practically, local independence is violated when item responses
are linked”. As noted by Balazs and De Boeck (2006, p. 2) “ignoring Local Item Dependence can
have serious consequences for the goodness of fit of a model, for the parameter estimates and for
confidence intervals.” This study was aimed at detecting passage-related local item dependencies in
PIRLS 2006 cognitive data and estimating its relationship with the country performance.
Keywords: local item dependency; IRT; PIRLS
Differential Item Functioning in the Aspect of Gender Differences in Reading
Literacy
Antra Ozola, University of Latvia, Latvia
Andrejs Geske, University of Latvia, Latvia
The purpose of the study is to analyze the IEA PIRLS 2006 set of items in order to check if
a differential item functioning (DIF) can be noticed. The groups of comparison are based
on gender since PIRLS 2006 study has showed that Latvia has the 6th largest gender gap in
reading literacy scores. So the DIF analysis is expected to show if an instrument could be
one of the reasons of the gender gap in achievement. Items easiest and hardest for each sex
are found as well as items with the biggest and smallest DIF values to check if an item
format gives any advantages or disadvantages to any of the gender groups of Latvian
students. The results show that the set of PIRLS 2006 items does not have any gender bias
on Latvian students since there are items with no DIF at all and there is almost equal
number of items with DIF in favor of each of the sexes. The length of the item stem does
not have any correlation with the item difficulty or differential item functioning between
sexes. There is no difference in DIF when the items of literary and informative texts are
compared and the same is true for multiple choice and constructed response items. The
main conclusion is that the achievement gap between boys and girls in Latvia is caused by
the differences in their skills/abilities and not by a differential item functioning.
Keywords: secondary analysis; differential item functioning; PIRLS; reading literacy;
gender differences
46