Fig. 1. Illustration of guiding prediction. Aging individuals with greater moment-to-moment process fluctuations at a given point in time are expected to show greater subsequent longitudinal decline in mean levels of functioning than individuals who fluctuate less. Adapted from Lindenberger et al. (2006).

Fig. 2. Graphical representation of a univariate dual change score model (McArdle & Hamagami, 2001) as implemented here. Observed variables are depicted as squares, latent variables as circles, regressions as one-headed arrows, and variances and covariances as two-headed arrows. Means and intercepts are indicated by the triangle. Unlabeled paths are fixed to 1.

Fig. 3. Is high trial-to-trial variability (coefficient of variation, CoV) preceding and predicting decline in categories performance? Model implied means from a bivariate dual change score model (full coupling model; McArdle & Hamagami, 2001) for categories performance (A) and trial-to-trial variability (B) as function of time of assessment and varied initial (t1) means (−0.5SD, estimated mean, and +0.5SD) of trial-to-trial variability (A) and categories (B). The figures show that lower initial trial-to-trial variability is associated with a less negative development of categories performance (A), whereas the initial levels of categories performance have little impact on development of trial-to-trial variability (B).

Descriptive statistics for measures included in the models

Note: CoV = coefficient of variation. The second time of assessment (t2) was conducted on average 1.95, t3, 3.76, t4, 5.53, t5, 8.94, and t6, 13.00 years after t1, respectively. Suspected dementia is coded as 10 and no suspected dementia as −10. All other variables are standardized to the T-metric (M = 50; SD = 10) with the t1 assessment providing the reference values. Time-to-death and chronological age are additionally centered.

Trivariate latent growth curve model: estimates (S.E) of means and variances for trial-to-trial variability (CoV), categories, and digit letter, after statistically controlling for the influence of time-to-death, chronological age, and suspected dementia

Note. CoV = coefficient of variation.

Are changes in trial-to-trial variability (CoV) associated with changes in mean cognitive performance?

Correlations among intercepts (IC) and slopes (SL) of trial-to-trial variability, categories, and digit letter, after statistically controlling for the influence of time-to-death, chronological age, and suspected dementia. Note: CoV = coefficient of variation; IC = intercept (i.e., scores at t1). SL = slope (i.e., 2-year linear change scores). Variability = trial-to-trial variability. Significance testing refers to the underlying covariances.

Is high trial-to-trial variability (CoV) preceding and predicting decline in mean cognitive performance?

Parameter estimates for bivariate dual change score models including statistical control for the influence of time-to-death, chronological age, and suspected dementia. Note: CoV = coefficient of variation. For the estimates of the γ parameters, significance testing refers to significance of the χ² (df = 1) in nested model comparisons comparing a model fixing this parameter to 0 against the full coupling model (see Section 1.3 and Section 2). For the estimates of the β parameters, significance was determined from their standard errors.