NotesFAQContact Us
Search Tips
ERIC Number: ED191909
Record Type: RIE
Publication Date: 1980-Apr
Pages: 25
Abstractor: N/A
Reference Count: 0
Some Approaches to Determining the Stability of Estimated Variance Components.
Smith, Philip L.
Accurate estimation of variance components used in generalizability theory is essential for the theory to be viewed as an efficacious mechanism for studying the reliability and validity of a measurement procedure. This paper explores two alternatives for dealing with the apparent instability of small sample size used in determining the accuracy of the variance component estimates resulting from a G study design. First, the accuracy of a confidence interval approach as an indicator of the magnitude of error in variance component estimates was explored. Monte Carlo methods were used to test the adequacy of this approach. Generally, the specific confidence interval procedure used resulted in moderately accurate confidence intervals. Second, the use of sequential designs to provide variance component estimates for multifaceted G studies was explored. The use of sequential designs requires that the investigator set up a series of small G studies designed to produce required variance component estimates and then combine the results of these studies in setting up a more complex D study. This approach was shown to be particularly useful in reducing dispersion effects which can significantly affect least squares estimates of variance components in designs that contain several facets. (Author/RL)
Publication Type: Speeches/Meeting Papers; Reports - Research
Education Level: N/A
Audience: N/A
Language: English
Sponsor: National Inst. of Education (DHEW), Washington, DC.
Authoring Institution: Wisconsin Univ., Madison. Dept. of Educational Psychology.
Identifiers: Confidence Intervals (Statistics); G Analysis; Generalizability Theory; Jackknifing Technique; Variance (Statistical)
Note: Paper presented at the Annual Meeting of the American Educational Research Association (64th, Boston, MA, April 7-11, 1980).