Differential Bundle Functioning Using the DFIT Framework: Procedures for Identifying Possible Sources of Differential Functioning

Are Inferential Reading Items More Susceptible to Cultural Bias Than Literal Reading Items?

“…A two-tailed test was conducted for the first two hypotheses given that it was not explicitly stated which group would be favored/disadvantaged by the bundles after controlling for overall reading ability. The process is consistent with exploratory DBF research (see Einarsdottir & Rounds, 2009;Kim & Huynh, 2010;Kim & Jang, 2009;McCarty, Oshima, & Raju, 2007;Oshima, Raju, Flowers, & Slinde, 1998). By contrast, a one-tailed test was conducted for the last four hypotheses given that it was explicitly stated which group would be favored/disadvantaged by the bundles after controlling for overall reading ability.…”

A Review of Recent Developments in Differential Item Functioning

“…The first method is DFIT, which is an IRT-based framework for assessing differential item functioning of items and tests (Oshima et al, 1997;Oshima, Raju, Flowers, & Slinde, 1998;Raju, van der Linden, & Fleer, 1995). It estimates the expected between-group squared difference in true scores after conditioning on ability and can be used with binary or polytomous data (Bolt, 2002;Flowers, Oshima, & Raju, 1999).…”

DFIT: An R Package for Raju's Differential Functioning of Items and Tests Framework

“…Originally proposed to identify DIF in dichotomous items and to be able to analyze differential functioning at the test level, it has subsequently been expanded to be used with bundles of items (Oshima, Raju, Flowers, and Slinde 1998), polytomous data (Raju, Fortmann-Johnson, Kim, Morris, Nering, and Oshima 2009), multidimensional models (Oshima, Raju, and Flowers 1997), and calculation of conditional DIF statistics (Oshima and Morris 2008). Within the DFIT framework differential functioning is analyzed between two groups of respondents: the reference group, and the focal group.…”

“…Comparisons with the program DFIT8 ) and with the SAS (SAS Institute Inc. 2013) macro DIFCUT (Nanda, Oshima, and Gagné 2006), which implement the current approach, have been conducted by taking published item parameters (Raju et al 1995Oshima et al 1998Oshima et al , 2009Wright and Oshima 2015) and verifying against the results for all three DFIT indices; the same testing procedure was employed for Raju's area measures (Raju 1988), the Mantel-Haenszel statistic (Wright 2011), and the implementation of asymptotic variance-covariance matrices calculations (Li and Lissitz 2004); additionally, the simulation procedure relies on the package mvtnorm (Genz, Bretz, Miwa, Mi, and Hothorn 2016), whose accuracy was studied by Mi, Miwa, and Hothorn (2009). Although the accuracy for the IPR algorithms follows from the previous, cut-off points for NCDIF were also compared with published results; differences were consistent with differences between repeated runs of the algorithm such as those shown in Table 4.…”