Optimal characteristics of anchor tests in vertical scaling: a special case of non equivalent groups with anchor test (NEAT) design in vertical scaling

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Gilbert Njiru Ngerano (Creator)
The University of North Carolina at Greensboro (UNCG )
Web Site: http://library.uncg.edu/
Richard Luecht

Abstract: There are multiple empirical issues and complications associated with vertical scaling methods that have not been sufficiently explicated even though there has been scanty research conducted within the general framework of the nonequivalent group with anchor test (NEAT) design. Germane to any vertical scale study is the issue of optimal characteristics of anchor tests whenever the preferred data collection design is NEAT. The main focal point of this research study is to explore some of practical problems as well as complexities that frequently emerge in the context of vertical scaling methods under NEAT design. Specifically, the study investigated various study conditions and comparison of their performance with different equating methods. This study used both real and simulated data. The real data were from a large-scale testing program for professionals. The simulated study was carried out using 162 conditions, where the major factors included: (1) total test length; (2) item a-discrimination parameter; (3) between-grade mean ability difference; (4) distribution of ability difference; and (5) anchor test mean difficulty difference. The results of the simulation indicate that small between-grade mean ability difficult when considered together with a short test length, a moderate item a-discrimination parameter, below average distribution of ability difference, and below average anchor test mean ability difference produce most reasonable results. In addition, the results revealed that equating error somewhat depended on satisfaction of the underlying equating assumptions that are related to a specific equating method under each study condition. For instance, Braun/Holland, Frequency Estimation Equating, keNEATPSE linear, and keNEATPSE equipercentile methods performed almost similarly under all study conditions; however, a closer examination of the above equating methods corroborate that when the equating relationship was linear, keNEATPSE linear outperformed all linear-related equating methods considered in this study. Similarly, when the equating relationship was non-linear, keNEATPSE equipercentile was more accurate in terms of total error, because it produced the smallest RMSE values than all non-linear equating methods. Other results are summarized in greater depth in Chapter V.

Additional Information

Language: English
Date: 2019
Equating, Non-equivalent groups with anchor test (NEAT) design, Vertical scaling
Educational tests and measurements
Examinations $x Design and construction
Education $x Research $x Methodology

Email this document to