Answers are grounded in this paper and cite their sources.
When a test is composed of a series of clustered items (testlets), authorities recommend that testlet scores be used in estimating internal consistency reliability. However, if the testlets vary in length or employ a variety of item formats, the scores will probably fail to satisfy the requirements of the tau equivalent model. In such a case, coefficient alpha will be negatively biased and a congeneric coefficient is preferable. The degree of bias in testlet-based alpha is considered via hypothetical examples and real test data. A simple formula for computing a testlet-based congeneric coefficient is presented.
No takes yet. Share an insight, caveat, or question.
Leonard S. Feldt (2002) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: