A Comparison of Traditional and Kernel Equating Methods

Abstract

Abstract:In this study, the equated score results of the kernel equating (KE) method compared with the results of traditional equating methodsequipercentile and linear equating and 9th grade 2009 ÖBBS Form B of Social Sciences and 2009 ÖBBS Form D of Social Sciences was used under an equivalent groups (EG) design. Study sample consists of 16.249 students taking booklets B and another 16.327 students taking D in that test. The analysis of the test forms was carried out in four steps. First, descriptive statistics were calculated for the data and then it was checked whether the data obtained from the two booklets satisfy the equating conditions. In the second step, the booklets were equated according to methods. Lastly, the errors for each equating methods were calculated. Kernel equating results were nearly same to the results from the corresponding traditional equating methods. In Kernel equating, when parameter h was selected as optimal, equated scores provided almost identical results as traditional equipercentile equating. When it was selected large, this time the equated scores provided results almost identical to traditional linear equating. It is concluded that Kernel equating methods are relatively more the most appropriate equating method method than traditional equating methods. ARTICLE HISTORY

The Impact of Covariate Variables on Kernel Equating under the Non-equivalent Groups

“…Another finding of the research is that standard errors of linear equating are lower than those of equipercentile equating. This finding is consistent with research by Choi (2009), Liou, Cheng, andJohnson (1997), Mao (2006), Akın-Arıkan, andGelbal (2018). The main reason for this is that the large h parameter value reduces the standard error.…”

A Comparison of IRT Observed Score Kernel Equating and Several Equating Methods

“…According to von Davier et al (2004), KE differences between -2SEED and 2SEED could be regarded as mainly coming from sample uncertainty than functions themselves. Attributing to its advantages of pre-smoothing and continuization of score distributions, KE has been testified and shown equivalent to or better than other equating methods, especially traditional ones, in the aspect of equating accuracy and stability (Chen, 2012;von Davier and Chen, 2013;Kim, 2014;Leôncio and Wiberg, 2017;Wedman, 2017;Arıkan and Gelbal, 2018;De Ayala et al, 2018).…”

Harmonizing the CBCL and SDQ ADHD scores by using linear equating, kernel equating, item response theory and machine learning methods

“…It also includes the possibility to use more information that is available when the scores are coming from the same individuals. In kernel equating, the scores are first converted from discrete to continuous using for example a Gaussian kernel distribution ( von Davier et al, 2006 ; Liu and Low, 2008 ; Arikan and Gelbal, 2018 ). Kernel equating can be used in such a way that one exploits the single group design: the information of what CBCL scores go together with which SDQ scores in the same children.…”