Cohort Changes and Sex Differences After Age 50 in Cognitive Variables in the English Longitudinal Study of Ageing.

OBJECTIVES: This paper models cognitive aging, across mid and late life, and estimates birth cohort and sex differences in both initial levels and aging trajectories over time in a sample with multiple cohorts and a wide span of ages. METHODS: The data used in this study came from the first 9 waves of the English Longitudinal Study of Ageing, spanning 2002-2019. There were n = 76,014 observations (proportion male 45%). Dependent measures were verbal fluency, immediate recall, delayed recall, and orientation. Data were modeled using a Bayesian logistic growth curve model. RESULTS: Cognitive aging was substantial in 3 of the 4 variables examined. For verbal fluency and immediate recall, males and females could expect to lose about 30% of their initial ability between the ages of 52 and 89. Delayed recall showed a steeper decline, with males losing 40% and females losing 50% of their delayed recall ability between ages 52 and 89 (although females had a higher initial level of delayed recall). Orientation alone was not particularly affected by aging, with less than a 10% change for either males or females. Furthermore, we found cohort effects for initial ability level, with particularly steep increases for cohorts born between approximately 1930 and 1950. DISCUSSION: These cohort effects generally favored later-born cohorts. Implications and future directions are discussed.

Reformulating and simplifying the DF analysis model.

DeFries-Fulker (DF) Analysis for unselected populations is reformulated as a no-intercept model with centered variables and only two independent variables. The reformulation serves three purposes. First, the original formulation implicitly estimated two different values for c2 and two values for h2. The new formulation resolves this ambiguity. Second, because the original formulation estimated h2 with the coefficient from a regression-interaction term, whether to center the interaction variables was unclear. The new formulation explicitly resolves this issue. Finally, the new formulation estimates fewer parameters, and therefore improves estimation efficiency and statistical power.