Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.

Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

A comparison of Kernel Equating to the Test Characteristic Curve method

Rebecca L Norman Dvorak, University of Nebraska - Lincoln

Abstract

This study examines the accuracy of Kernel Equating (KE) and the Test Characteristic Curve (TCC) method under different conditions. Through a Monte Carlo simulation study simulees’ equated scores on form Y are compared to parametric true scores based on item and person parameters. The study uses the non-equivalent anchor test (NEAT) equating design. For KE both the chain equating (CE) and post-stratification equating (PSE) techniques are examined. The 2 parameter logistic (2PL) model is used for TCC equating and for calculating parametric true scores. The effects of four independent variables (sample size, test length, the percent of anchor items, and average factor loading) on equating method accuracy are investigated. ^ Results suggest that both equating methods perform fairly well under the varying levels of independent variables. Test length, the percent of anchor items, average factor loading, and sample size do affect which method is more accurate. Root mean square difference (RMSD) values indicate that there is an interaction between these four variables. For example, on the long tests (75 items) KE is always more accurate when the average loading is .62 regardless of the percent of anchor items or sample size. When the average loading is .50 then TCC equating is more accurate when used with 30% anchor items and for the large sample size condition. For the other .50 loading conditions and 75-item test length KE is either more accurate or the two methods are indistinguishable. ^ A graphical examination of the average mean difference between parametric true scores and each equating method reveals that the accuracy of each equating method varies along the score range of form X. Specifically, KE produces more accurate expected scores on Y for individuals who score in the middle to upper-middle range of scores on X. This corresponds to the range in which the majority of individuals fell. In contrast, TCC equating is more consistently accurate across the entire range of scores. Overall, this study suggests that KE tends to perform well in comparison to TCC equating.^ Key Words: Kernel Equating, Test Characteristic Curve Equating^

Subject Area

Education, Tests and Measurements

Recommended Citation

Norman Dvorak, Rebecca L, "A comparison of Kernel Equating to the Test Characteristic Curve method" (2009). ETD collection for University of Nebraska - Lincoln. AAI3350452.
http://digitalcommons.unl.edu/dissertations/AAI3350452

Share

COinS