Academic journal article Kuram ve Uygulamada Egitim Bilimleri

Comparing Differential Item Functioning Based on Manifest Groups and Latent Classes *

Academic journal article Kuram ve Uygulamada Egitim Bilimleri

Comparing Differential Item Functioning Based on Manifest Groups and Latent Classes *

Article excerpt

(ProQuest: ... denotes formulae omitted.)

A test item should be able to measure ability without involving characteristics of subgroups that consist of individuals. This is because individuals with equal abilities should be able to correctly answer an item at the same rate even though they are in different subgroups. If items included in the test provide more advantages for one group over another, the item is considered to be biased (Camili & Shepard, 1994; Mellor, 1995; Zumbo, 1999). Therefore, when developing a test, items should be examined in terms of item bias.

Item bias determination processes are carried out in two stages. The first stage is a statistical process during which item response distributions are examined in reference groups and focal groups established by considering observed variables (gender, country etc.) under equal ability levels (Cohen & Bolt, 2005; Steinberg & Thissen, 2006). In this distribution, differentiation in the probability of correct answers given provides the differential item functioning (DIF) of an item. Statistical properties of an item with DIF also vary among groups (Angoff, 1993; Clauser & Mazor, 1998). For this reason, experts should reveal in qualitative studies whether items with DIF are biased.

DIF analyses are usually carried out over reference and focal groups established based on manifest groups. In these studies, it is assumed that the characteristics of all participants are similar in the manifest group (De Ayala, Stapleton, & Dayton, 2002). In line with this assumption, an item with DIF is considered advantageous or disadvantageous for all individuals in a manifest group. Indeed, according to Samuelsen (2005), the reliability of results in DIF methods for the manifest group is affected by the assumption that a group consists of homogeneous communities. This is so in terms of its measured ability, and from the lack of consideration of the possibility that an item may contain DIF in the same group. This is because individuals in different subgroups (including gender, socioeconomic level, or culture) can be divided into latent classes that may be homogeneous with respect to ability (De Ayala et al., 2002; Samuelsen, 2005). A high level of overlap between these latent groups and manifest groups is low in probability. In other words, if members of a manifest group are also included in a single latent class, a 100% level of overlap can be considered in the manifest group and latent classes. This rate shows similarity between distinctive properties of the latent class and the manifest group. However, individuals of a group may also be members of another latent group. Accordingly, in these cases, especially when the ratio of overlap is less than 70%, obtained DIF results can be biased using only manifest group variables (Bilir, 2009; Samuelsen, 2005).

Hu and Dorans (1989) pointed out that girls tend to achieve lower scores than boys in the event of item removal. However, removal of an item also resulted in an increase in Asian American girls' scores compared with those of Latin and Asian American boys. Accordingly, though it seemed that girls' scores declined, scores in Latin and Asian American latent class, within the subgroup of girls, increased. Cohen and Bolt (2002) pointed out that items showed DIF as per gender in their DIF method study on the manifest variable. However, latent class analysis results found out that about 50% of women and 40% of men were included in different classes. De Ayala et al. (2002), in a DIF study applied on classes determined through latent class analysis that three items showed DIF in the black race in the latent class but did not show DIF in the black race from the other class. These results suggest that the manifest variable method might determine an item as an item with DIF for all members even if this was not the case in reality.

Previous studies indicate that group homogeneity assumptions are not always met. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.