This paper explains how to develop and evaluate a computerized
adaptive test for measuring second language proficiency.
Based on the logistic model of Rash, the author developed a software
to test the proficiency level of students in their first year of
English as a second language (placement test). The program "adapts"
the difficulty of a test item to the estimated level of a student's
proficiency. In other words, if a student is able to answer an item
correctly, the following item will be more difficult -- and vice
versa. The test is "tailored" to the student's ability
level. This prodecure is being repeated until the program is "quite
sure" about the testee's level of proficiency, i.e. until certain
test criteria have been met.
In order to show the differences between traditional (paper-and-pen)
tests and computerized adaptive tests, and in order to show that
the software actually "tests" correctly, volunteers were
tested in both ways (i.e. they first took part in a traditional
test, then in a computerized adaptive test). The comparison of the
test results shows that a computerized adaptive test can judge the
students' level of proficiency in about half of the time and with
about half of the number of items needed in a traditional test.
In other words, the computerized adaptive test is more efficient
in several ways:
- Test administration is facilitated since the software "tailors"
each test individually.
- The software needs less items (approx. 50 %) and less time (approx.
50 %) than a traditional test (and still comes to the same estimation
of a testee's level of proficiency).
- The factor "computer anxiety" does not play a role
if the testees are allowed to practice the computerized test.
- As opposed to traditional tests, the results of different testees
in a computerized adaptive test can easily be compared.
|