A Theoretical Contribution to ESL Diagnostic Test Construction

This paper considers the results of an experimental 40-item diagnostic test dealing with 10 grammatical mistakes typically made by Chinese students; the analysis focuses on the scores of these 10 mini-tests. The purpose of the experiment was to calculate the reliability of the mini-tests and then to determine how many items are needed to establish 'good' reliability. Two forms (A & B) were administered a week apart to 298 ESL students. Validity of the mini-tests was checked by constructing a composition with the same grammatical mistakes and asking the students to identify them. Reliability coefficients (K-R #20) ranged from .67 to .91. The data were then analyzed as if each mini-test in Form A had only 3 items, and then only 2 items; r ranged from .61 to .87, and from .28 to .82 respectively. From a different point of view, the optimum number of items may be suggested by asking how much useful information is lost if a decision is made on the basis of 2 items rather than 4. If the criterion is the student's consistently good, or poor, performance from A to B, the degree of such consistent performance is very stable whether based on 4, 3, or 2 items per subtest

Diagnostic test · Language assessment · Linguistics · Mathematics education · Test (biology) · Educational Technology and Assessment · Geology · Medicine · Philosophy · Psychology