ERIC Number: ED230557
Record Type: Non-Journal
Publication Date: 1980-Jul-15
Reference Count: N/A
Studies of Domain Referenced Item Models. Final Report. Volume 1: Sections I and II.
Harris, Chester W.; And Others
The first section of this two-volume report describes a point of view about testing in connection with classroom instruction. A model was developed, based on this point of view, postulating a generic task (such as addition of two-digit numbers). A random sampling principle was used to generate sample items, the performance on which gives an unbiased estimate of the proportion of items in the universe that can be answered correctly by the examinee. For generic tasks that involve completion and/or short answers that are constructed by the examinee, an item model is specified that relates performance on a particular item to a two-category latent class variable. Item difficulty is seen to be a product of two terms, one describing the "easiness" of the generic task, and the other describing the easiness of the particular content embedded in the item. Four methods of estimation of these parameters are described in Section II. On the basis of a Monte Carlo study, Goodman's maximum likelihood method is recommended. (PN)
Publication Type: Reports - Research; Numerical/Quantitative Data
Education Level: N/A
Sponsor: National Inst. of Education (ED), Washington, DC. Teaching and Learning Program.
Authoring Institution: California Univ., Santa Barbara.