Specific Aims: Aim 1. To compare different methodological approaches in order to detect biased test items by socio-cultural factors among Spanish-and English-speaking older adults using the SENAS dataset. Hypothesis 1. Evidence of DIF due to language and socio-cultural factors will be observed. Aim 2. To determine if accounting for DIF strengthens the association between estimated ability and dementia diagnosis. Hypothesis 2. The ability estimates that account for DIF will have a stronger relationship with dementia diagnosis.

Background: This RCMAR Collaborative Pilot Research Project provides a unique opportunity for junior investigators from four RCMAR sites (Columbia, MUSC, UCLA and UCSF) to work with expert faculty to directly address two of the National Institute on Aging (NIA) RCMAR aims: 1) Enhancing the diversity in the professional workforce by mentoring minority academic researchers for careers in minority aging and health research; and 2) Creating culturally-sensitive health measures that assess the health status of minority elders with greater precision. This proposal is strengthened by the collaboration of these four investigators from different RCMAR sites due to the expertise that each brings in methodology in minority aging research, theories of aging, and clinical knowledge. The principal investigator, Dr. Frances M. Yang (Columbia RCMAR), a gerontologist, instructor in psychiatry and clinical epidemiology, and UCSF RCMAR scholar and epidemiologist, Dr. Kala Mehta, have had the opportunity to meet at the Friday Harbor Psychometrics (FHP) Workshop, where RCMAR faculty (Drs. Teresi, Jones, Mungas, and Ocepek-Wilikson) played a key role in laying the theoretical and methodological foundation for this proposal. A major focus of the workshop was to introduce a psychometric theory known as item response theory (IRT) (Baker, 1985; Hambleton & Swaminathan, 1985), and its associated statistical methods in modern psychometric test development. Another focus was to encourage collaboration among participants to use IRT techniques to evaluate neuropsychological instruments for evidence of DIF attributable to ethnicity. This current RCMAR request for proposals is a vehicle for ongoing collaboration among junior minority investigators, and the application of IRT methods for developing and evaluating measures of cognitive functioning in minority aging research. Therefore, the purpose of this study is to identify DIF and assess the impact on scale level scores using three different DIF methods—PARSCALE/DIFdetect, MULTILOG/DFIT, and Mplus/MIMIC—across socio-cultural characteristics; in turn the sensitivity of these measures to dementia will be assessed. Based upon the knowledge obtained from the FHP Workshop, we are developing psychometric methods and skills for cognitive assessment in culturally diverse populations. Two of the investigators (Drs. Yang and Mehta) are currently working on studies and manuscripts using the knowledge and SENAS dataset obtained from the FHP Workshop. Dr. Yang (Columbia RCMAR) is taking the lead in profiling quantitative magnetic brain images that will be used to externally validate neuropsychological test results in a multiethnic sample. She is also working with her mentor, Dr. Jones, using MIMIC to detect DIF in depression scales for her NIA T32 post-doctoral fellowship. The second proposed investigator, Dr. Mehta, used DIF DETECT to determine if DIF was attributable to race, education and language for the object-naming test for participants represented in this dataset in the SENAS. The third investigator is Mr. Chengwu Yang (MUSC RCMAR), who is currently a biostatistics doctoral student to Dr. Barbara C. Tilley (MUSC RCMAR). Their work has encompassed detecting DIF in the SF-36 and SF-12 in diverse populations using MIMIC modeling in Mplus to conduct these analyses. For this proposal, Mr. Yang will gain further training in Mplus/MIMIC and MULTILOG/DFIT to detect DIF in the SENAS. A fourth investigator (TBD) is a research student of Dr. Leo Morales (UCLA RCMAR), who will be conducting DIF detection using MULTILOG/DFIT. DFIT Results Table