Improving Measurement in Longitudinal Studies of Aging

This paper provides a how-to guide for planning, implementing, and evaluating the MIMIC (Multiple Indicator and MultIple Causes) method in diverse populations – in this case, large samples of Black and white respondents from the MADICS study. The MIMIC approach can also probe for differential item functioning (DIF) in longitudinal measures (i.e., temporal invariance). MIMIC models afford powerful claims about measurement – particularly in terms of biased items – and are relatively simple to specify and test. Therefore, we argue that MIMIC models are sorely underutilized and serve important roles in ensuring sound and fair measurement. To increase their use, this tutorial carefully explains how to specify, interpret, and evaluate MIMIC models, as well as provides sample code in R (lavaan) and MPlus. MIMIC models are explained in accessible “plain English” and Greek (notation), with an OSF folder providing annotated code and output.

This study examined the mode effect on the respondent performance of cognitive tests where the mode (web vs. telephone) was randomly assigned. Those assigned to the Web mode scored higher than those to the telephone mode, particularly in the Serial 7 task and numeracy items. It recommends a mode-dependent scoring system to indicate cognitively impaired but not demented status.

Cognitive impairment is a major health issue impacting many older adults, making it imperative to have indicators of cognitive functioning. Web and phone surveys have shown to be useful in measuring cognitive functioning, allowing for the development of a cognitive impairment score.

This study examined whether participants in the U.K. Understanding Society Innovation Panel were willing to provide a blood sample in different interview settings (randomly assigned to 1. Nurse collection of dried blood spot (DBS), 2. Nurse collection of blood sample by venepuncture, or 3. Self-collection of DBS), and how resulting cardiovascular risk biomarkers compared. Although the willingness was lowest in the self-collection DBS, demographic characteristics of participants in self-collection mode were not different to those in nurse collection mode. Further, clinical biomarker information relevant to cardiovascular disease risk did not differ between the venepuncture blood sample and the DBS. This demonstrates that DBS collection offers acceptable measures of clinically relevant biomarkers, enabling the calculation of population levels of cardiovascular disease risk.

This study compares the performance of IQCODE in understanding objective cognition across the U.S., England and India. It reports strong associations between IQCODE and objective cognition at low cognitive functioning in the U.S. and England. In India, however, the IQCODE was less sensitive to impairments at the lowest levels of cognitive functioning, particularly among those with no education. Informant characteristics may differentially impact informant reports across countries. Researchers should consider the use of country-specific adjustments to the IQCDOE scoring based on informant characteristics to improve cross-national comparability.

Regression modeling or confirmatory factor analysis (CFA) can be used to incorporate information from substudies with comprehensive neuropsychological testing into measures of cognition in broader aging surveys. Compared to a gold standard measure based on comprehensive neuropsychological testing, both approaches had lower mean squared error than existing comparison approaches. Associations with example risks were similar across all approaches, though estimated standard errors were most accurate for CFA models. The similarity across approaches may be due to the brevity of available cognitive assessments in the example dataset.

This paper presents key lessons on cognitive assessment from the international collaboration of aging studies. It reports challenges associated with administering cognitive tests across populations from different cultural and linguistic backgrounds and importance of maintaining consistency across time and studies, test implementation feasibility across both high-income and low- and middle-income countries, and comprehensive cognitive batteries for improved measurement precision.

This study examined the relationship between response times on cognitive testing in computerized in-person interviews and cognitive performance. Nonlinear associations between response time and cognitive functioning was reported, after adjusting for traditional cognitive test scores. Results indicate that response times from cognitive testing may contain important information on cognition not captured in traditional scoring. Incorporation of this information has the potential to improve existing estimates of cognitive functioning.

Using the data on response time of standard cognition questions in the National Social Life, Health, and Aging Survey, this study examined the relationship between the response time and the Montreal Cognitive Assessment (MoCA). The results show that response time predicted current as well as future MoCA. This predictive power varied by race and age but not by gender.

This study examined the utility of response time on online surveys for discriminating respondents’ cognitive health. Specifically, it used response time from 1053 items across 37 online surveys administered over 6.5 years and cognitive health measured at the end of this time in a multilevel location-scale model. Average response time as well as fluctuations in response time were associated with subsequent cognitive health. This suggests that response time on survey items may be a potential indicator of cognitive impairment.

This examined the relationship between errors respondents make in completing survey interviews (e.g., implausible responses, skipped questions) and subsequent dementia incidents using the Health and Retirement Study data. All response error variables showed an independent relationship with dementia, where the relationship was stronger for those who were younger and cognitively normal at baseline.

Bluetooth and Wi-Fi networks can be very useful in collecting information about social networks within a specific location, and can be utilized to make connections within aging populations residing within assisted living facilities. This social network data can also be connected to relevant health data.