Applications of item response theory to the examination of the psychometric properties and differential item functioning of the comprehensive assessment and referral evaluation dementia diagnostic scale among samples of Latino, African American, and white non-Latino elderly

Item response theory was used to examine the psychometric properties of a cognitive screening measure used in several epidemiological surveys among Latino, African American, and White non-latino elderly. Estimates of precision (reliability) examined across several values of theta (the estimate of degree of cognitive impairment) were good .70s to,90s) in the range representing most respondents, Overall reliability, although adequate, was lower among the White non-latino and high-education subgroups relative to other racial/ethnic and education subgroups, Differential item functioning (DIF) was examined using several area- and model-based tests. Tests of the magnitude of DIF showed the measure to be relatively free of DIF for the racial/ethnic subgroup and education subgroups examined. However, one item related to remembering the telephone number was more difficult for Latinos than for other racial/ethnic subgroups. Several other items evidenced mild DIF, and one (difficulty remembering words or names) was a poorly discriminating item. The subjective memory items, intended to provide additional information at the earlier, "borderzone" stages of cognitive impairment, did provide more information at the mild to moderate levels of impairment, although maximum information was not provided at these borderzone ranges.