This paper reports the results of an experiment conducted to compare the reliability and validity of the doze test and the C-test as a device for measuring global English language proficiency. Three different passages with approximately the same levels of readability were chosen from an ESL textbook for the construction of doze tests A, B, and C and C-tests A, B, and C. For each doze test every seventh word was deleted up to a total of 24 blanks. For the C-tests the second half of every second word was deleted until the same number of mutilations was reached. A group of 83 Japanese university students of English participated in the experiment.
Standardized structure and listening subtests of CELT were used for the criterion measures against which the two reduced redundancy tests were evaluated. Reliability coefficients computed by the Kuder Richardson formula 20 were obtained for each test. Pearson product-moment correlations were also calculated between the tests and the criterion tests.
The following research questions were addressed: 1) Are the C-tests more likely to be affected by the selection of passage? 2) To what extent are the doze and C-tests different in reliability? 3) Is there any difference between the tests in predicting the student's sucess in learning?
The statistics computed in the experiment showed no significant differences between the tests except that the doze tests are slightly more affected by the different passages.