Page No.: 
James Dean Brown

This study explores the link between some of the linguistic characteristics
of cloze test items and· the corresponding item difficulty estimates.1 Five
reading passages were randomly selected from an American public library
and made into 3O-item cloze tests by deleting every 12th word. EFL students
(N=179) at the post-secondary level in Japan each took one of the resulting
30-item cloze tests. The five cloze tests were randomly administered across
all of the subjects. Any differences between the cloze tests or the individual
test items were therefore assumed to be due to other than sampling
differences. The result was a set of 150 item difficulty estimates (5 tests
times 30 items), which served as the dependent variable: cloze item
difficulty. Each item was also analyzed for linguistic characteristics, which
served as the independent variables, e.g., the content/function word distinction,
passage readability, number of words per sentence, frequencies of
occurrence in passage(s) and many others. Multiple-regression analysis of
the linguistic characteristics as predictors of the item difficulty estimates
showed that characteristics such as frequency of occurrence, number of
charactel'$ per word, and number of syllables per sentence account for up
to 32 percent of the variation in item difficulties. These results are discussed
in tenns of their implications for language testing research and plans for
future research on a larger scale.