Naslov
Item parameter estimation as a function of item context: A comparison of 2-parameter logistic model and classical test theory

Autori
Ljubotina, Damir

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
17. Dani Ramira i Zorana Bujasa: Sažeci priopćenja / Kamenov, Željka ; Jelić, Margareta ; Jokić Begić, Nataša - Zagreb : Odsjek za psihologiju Filozofskog fakulteta Sveučilišta u Zagrebu, 2005, 26-26

Skup
17. Dani Ramira i Zorana Bujasa

Mjesto i datum
Zagreb, Hrvatska, 15.12.2005. - 17.12.2005

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Domaća recenzija

Ključne riječi
item parameter estimates; item context; IRT; CTT

Sažetak
The goal of this study was to examine the influence of item context on the bias and the standard error in item parameter estimation. A discriminative item with known true parameters validated in the context of items giving a poor projection of the true score will likely get a biased estimation of its discriminating power in relation to its true value. In the group of items where we are estimating some items with known characteristics (&laquo ; estimated&raquo ; items) the context is defined by the characteristics of other &laquo ; contextual&raquo ; items. A simulation study with a factorial 3 x 3 x 3 experimental design was conducted. The influence of three main effects associated with the characteristics of an item group was examined: a) similarity/difference of the &laquo ; contextual&raquo ; and &laquo ; estimated&raquo ; items according to their discriminating power, b) the ratio of &laquo ; contextual&raquo ; and &laquo ; estimated&raquo ; items, and c) total number of items in the composite. The known true parameter values are compared with the estimates obtained in different experimental conditions. Item parameter estimations were conducted under the 2-parameter logistic model of item response theory and under the assumptions of the classical test theory. The simulation study emphasizes a significant influence of context in which the item parameter estimation is done, both for the classical test theory and the item response theory. As far as the classical discrimination indices are concerned, this experiment showed that their estimation in the context of &laquo ; contextual&raquo ; items with lower discriminating power tends to underestimate the true value of the parameter. This effect can, to a certain extent, be compensated for by adding more items in the composite. When we deal with discrimination parameters under item response theory, it is harder to predict this contextual effect, but in this case it caused an overestimation of the true parameter value. In this case the contextual effects significantly increased the standard error, while this wasn't the case when classical indicators were used. Item parameter estimation in the context of similar items in a long composite, or amongst highly discriminating items can reduce these effects.

Izvorni jezik
Engleski

Znanstvena područja
Psihologija

POVEZANOST RADA