Education shapes episodic memory measurement via test specifications: Evidence from the China Health and Retirement...

Many studies have examined how education influences cognitive decline trajectories, often reflected through episodic memory deficits measured by word recall tests. However, little is known about how education affects episodic memory measurement in longitudinal studies where word-list complexity and test form vary. Our study aims to explore whether education influences episodic memory measurement via test specifications in a Chinese context.

23,951 respondents aged over 45 (78,364 person-years) from five waves (2011-2020) of the China Health and Retirement Longitudinal Study (CHARLS) were included. We fitted two random-intercept models-one for immediate and one for delayed test scores-to examine how education influences episodic memory measurement across varying test formats and word list complexities. Based on these results, we applied a hybrid frequency-estimation equating approach to facilitate longitudinal studies in CHARLS, accounting for education's impact when word recall tests use varying specifications.

Respondents with higher education scored better on immediate and delayed word recall tests, but all education groups were negatively affected by increased word-list complexity, with lower-educated individuals more vulnerable. Higher-educated respondents also gained more improvement in word recall outcome from extra practice trials when complexity remained constant. After equating, the predicted trajectories reflected more accurate cognitive decline over time, enhancing the measurement's validity.

These findings suggest that education strongly influences episodic memory assessment, as test specifications-word-list complexity and test form-interact with participants' education and shape performance gaps between higher- and lower-educated groups. For equating techniques in longitudinal studies, frequency estimation suits waves with similar complexity, whereas equipercentile equating better addresses substantial complexity differences, thereby enhancing measurement validity.