NotesFAQContact Us
Search Tips
ERIC Number: ED542021
Record Type: Non-Journal
Publication Date: 2012-Dec
Pages: 44
Abstractor: As Provided
Reference Count: 33
Statistical Considerations in Choosing a Test Reliability Coefficient. ACT Research Report Series, 2012 (10)
Woodruff, David; Wu, Yi-Fang
ACT, Inc.
The purpose of this paper is to illustrate alpha's robustness and usefulness, using actual and simulated educational test data. The sampling properties of alpha are compared with the sampling properties of several other reliability coefficients: Guttman's lambda[subscript 2], lambda[subscript 4], and lambda[subscript 6]; test-retest reliability; as well as congeneric reliability. The comparisons are based on different sample sizes and test models comprising dichotomous item and polytomous item tests. It is concluded that alpha is indeed a lower bound to reliability except under the assumption of essential-tau-equivalence; however, it is robust to violations of this condition, and its values are competitive with other coefficients' values based on splitting a test into parallel halves or repeating a test to estimate test-retest reliability. Because it is not always possible to construct parallel half-tests or obtain test-retest data, it is very useful to have reliability coefficients, such as alpha, which are free of these constraints. Tables and figures are appended. (Contains 8 tables and 6 figures.)
ACT, Inc. 500 ACT Drive, P.O. Box 168, Iowa City, IA 52243-0168. Tel: 319-337-1270; Web site:
Publication Type: Reports - Research
Education Level: Higher Education
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: ACT, Inc.