I am considering issues of bias using the NEO-PI-R. In South Africa we often expect differences between groups due to language proficiency so that first language English speakers will respond differently by virtue of their familiarity with the language. I want to investigate if we see systematic differences between first and second language English speakers on the NEO-PI-R and to consider these arguments within the context of potential bias in the test.I am not looking at item-level analysis, I am looking across scales so at the domain and facet scores.
I can establish differences between groups using mean differences but will this be evidence of bias? If I calculate congruence coefficients and there is no evidence of agreement, would this be considered sufficient evidence of potential bias? |Should I be considering testing for a significant difference between the cronbach values obtained from both groups? Or am I totally missing the boat and should I actually be doing something else?
The best method is using differential item functioning (DIF) through multidimensional item response theory (mIRT). You should consider item by item! Cronbach's alpha is not helpful. IRTPro 2.1 software can help in your analisys.
- Badges
- Science topic
- Similar topics
- Linguistics
- Philology
- Language