Elsevier

Intelligence

Volume 40, Issue 4, July–August 2012, Pages 352-361
Intelligence

Does the direction and magnitude of cognitive change depend on initial level of ability?

https://doi.org/10.1016/j.intell.2012.02.006Get rights and content

Abstract

Longitudinal change in five cognitive abilities was investigated to determine if the direction or magnitude of change was related to the individual's ability level. Adults between 18 and 97 years of age performed three versions of 16 cognitive tests on two occasions separated by an average of 2.7 years. In order to control for influences associated with regression toward the mean, level of ability was determined from scores on the first version of the cognitive tests on the first occasion, and across-occasion change was examined on the second and third versions. Change in every cognitive ability was significantly more negative with increased age. However, there was little indication of ability-dependent change in any of the five cognitive abilities, either in differences between composite scores, or in estimates of latent change. Although there are reasons to expect cognitive change to be less negative at either high or low levels of ability, these data suggest that neither the direction nor magnitude of change is related to initial ability when influences of regression toward the mean are controlled.

Highlights

► Does the rate of cognitive change vary according to one's initial level of ability? ► Prior research on this topic has been limited to a few cognitive variables. ► Furthermore, few studies have directly controlled problems of regression to the mean. ► A novel method was used to control for regression to the mean. ► The results revealed little evidence of ability-differential cognitive change.

Introduction

The primary questions in the current report were whether longitudinal change on cognitive tests operates in the same manner throughout the entire distribution of ability, and whether ability–change relations were similar at different ages. Less negative change might be expected among individuals with high levels of ability, either because they have a greater magnitude of some type of cognitive reserve (Stern, 2009), or because they have a lifestyle involving physical or cognitive activity conducive to the preservation of cognitive ability. It is also possible that individuals with low levels of ability have more negative change because they are in the early stages of neurodegenerative disease and the influences associated with the disease are superimposed on normal age-related influences.

However, there are at least two reasons why differential change might be expected in the opposite direction. First, individuals with low levels of ability may be close to a functional floor, in which case there may not be much room for negative change to be detected. And second, if the highest level of performance can only be achieved if all processes are functioning near optimum, individuals at higher ability levels might be more vulnerable to decline because disruption in any of several processes could lead to a reduction in performance, whereas lower levels of performance may be possible with many different combinations of processes.

Because average cognitive change has been found to differ as a function of age (e.g., Salthouse, 2010a, Salthouse, 2010b), ability–change relations might also vary with age. For example, if different factors are contributing to change at different ages, and to the relations between initial ability and change, divergence of change according to initial ability might only occur at older ages, with more uniform change occurring at younger ages.

In a recent review of research on intelligence, Nisbett and colleagues stated: “Intelligence declines in old age, but do brighter individuals decline less or more? We do not have longitudinal data on the question …(Nisbett et al., 2012).” In fact, however, a number of studies have been published with longitudinal data, including four in which the phrase “Is age kinder to the initially more able” was included in the title. Much of this research was recently reviewed by Gow et al. (2012). They reported that the existing results were inconsistent, and suggested that small sample sizes, different ages at baseline and when change was assessed, and differences in the interval between assessments and in the nature of the cognitive tests could have contributed to the variability in results.

The Gow et al. (2012) review focused on studies in which the baseline measurement occurred in childhood or young adulthood because the authors suggested that studies with baselines in middle or late adulthood are difficult to interpret because cognitive declines may have already occurred. However, a somewhat different, but equally meaningful, question concerns the relation between an individual's current ability level and his or her subsequent cognitive decline, and for this question a short interval between baseline assessment and change is desirable. Furthermore, if the sample includes participants across a wide age range it is possible to determine if the ability–change relations vary as a function of age as one might expect if progressively more individuals are experiencing appreciable decline with increased age.

Several studies have examined the relation between level and change in ability when both the initial and change assessments have been in middle- or old-age. For example, Wilson, Beckett, Bennett, Albert, and Evans (1999) reported that older adults with lower levels on a composite measure of cognition at baseline had greater declines than older adults with higher initial levels. However, Reynolds, Finkel, Gatz, and Pedersen (2002) found different patterns of ability–change relations across different methods of analyzing change, and Wilson et al. (2002) found that the ability–change correlations ranged from .02 to .53 across composite scores representing different cognitive domains. Because the studies differed in the number and breadth of the cognitive domains evaluated as well as in the types of analytical procedures, it is difficult to determine reasons for the discrepancies in ability–change relations at this time.

It is useful to consider methods that can be used to examine the relation between initial ability and change. Some information about differential change can be obtained from comparisons of the ratio of the variance of scores on the first and second occasions, and from inspection of the correlation of the scores across the two occasions. The rationale is that if people differ in the direction and magnitude of longitudinal change, the variance on the second occasion should differ from that on the first occasion because the variance in change is superimposed on the initial variance, and correlations across the longitudinal interval (i.e., stability coefficients) should be less than 1.0 if the ordering of individuals differs on the two occasions. Although information of this type indicates whether the individual differences are greater in one occasion or the other, or whether the individual differences are stable across occasions, it is not directly relevant to the question of whether the direction and magnitude of change varies according to the individual's level of ability.

More direct information is available by comparing change in individuals selected from different portions of the ability distribution. However, it is not always recognized that the results from these types of analyses could be distorted by regression toward the mean. The problem is that if some of the high values on the first assessment are due to measurement error in the positive direction and some of the low values are due to measurement error in the negative direction, then the scores at the second assessment will tend to be less extreme than those on the first assessment for statistical, rather than substantive, reasons. In other words, individuals with high initial levels may exhibit some artifactual decline whereas individuals at low initial levels may exhibit artifactual gain. Unless influences of regression toward the mean are taken into consideration, therefore, the observed change will likely be composed of an unknown mixture of true change and of influences of regression toward the mean.

Several attempts have been made to either examine, or control, influences of regression toward the mean in analyses of cognitive change. For example, Baltes, Nesselroade, Schaie, and Labouvie (1972) used a time-reversed design to investigate the existence of regression toward the mean in longitudinal change in cognitive abilities. As noted by Alder, Adam, and Arenberg (1990), however, the time-reversed strategy merely indicates whether regression toward the mean is operating, and neither eliminates it, nor provides an estimate of its magnitude. Alder et al. (1990) used growth curve models with adjustments for measurement error to examine the relation between initial level and change in measures of vocabulary and visual memory. However, it is not clear how effective this method is in actually reducing influences of regression toward the mean, and the results were inconsistent because the estimated relations differed in direction for the two measures, with a positive ability–change relation for the visual memory measure but a negative relation for the vocabulary measure.

Because most regression occurs from the selection assessment to the next assessment, Nesselroade, Stigler, and Baltes (1980) proposed that effects associated with regression toward the mean could be minimized by using scores at the first assessment for selection, and then examining change beginning from the second assessment. Implementation of this staggered selection-assessment strategy therefore requires a minimum of three assessments, one for selection and two to assess change. In a traditional longitudinal study with only a single assessment at each occasion, application of this strategy will result in loss of change information originating from the first occasion, which is unfortunate because early change could be different than later change.

An alternative implementation of the staggered selection-assessment procedure can be used with a measurement burst design, such as that incorporated in the Virginia Cognitive Aging Project (VCAP; Salthouse, 2007, Salthouse, 2010a). VCAP is an ongoing cross-sectional and longitudinal study of cognitive aging that began in 2001. A unique feature of the project is a measurement burst design in which three assessments on parallel versions of 16 tests are administered at each longitudinal occasion. Because multiple assessments are available at each occasion, individuals at different levels of ability can be selected on the basis of the scores on the first assessment at the first occasion, and then change examined from the first to the second occasion on subsequent assessments. That is, assume that the three assessments at Time 1 are designated T11, T12, and T13, and those at Time 2 are designated T21, T22, and T23. The staggered selection-assessment procedure can therefore be implemented in a measurement burst design by determining initial ability on the basis of T11 scores, and then examining change from T12 to T22 and from T13 to T23.

Section snippets

Participants

All participants in VCAP who returned for a second occasion were administered the three-session measurement burst. However, on the first occasion some of the participants were administered the VCAP battery only on the first session and were administered different cognitive tests on the second and third sessions, and therefore these participants had only one measurement burst. Characteristics of the longitudinal participants with one and two measurement bursts, and of the participants with only

Results

As noted earlier, scores on the second and third versions of each test were adjusted with regression equations to minimize across-version differences in mean performance, and all scores converted to z-scores based on the mean and standard deviation of the T11 scores. Composite scores were then created at each session by averaging z-scores for the tests representing each ability.

The initial analyses were conducted on the T11 and T21 composite scores from all 1898 longitudinal participants. The

Discussion

Fig. 1 dramatically illustrates the regression-toward-the-mean phenomenon as the change from the first (T11) to the second (T12) assessments was negative among individuals in the top quartiles of initial ability and was positive among individuals from the bottom quartile. These results are intriguing because regression toward the mean is often assumed to be small when the measurements are reliable, and yet the high alternate-form correlations in Table 3 indicate that the composite scores in

Acknowledgments

This research was supported by Award Number R37AG024270 from the National Institute on Aging. The content is solely the responsibility of the author and does not necessarily represent the official views of the National Institute on Aging or the National Institutes of Health. There are no conflicts of interest.

I would like to thank John R. Nesselroade for valuable comments on an earlier draft of this manuscript.

References (30)

  • I.J. Deary et al.

    An “instantaneous” estimate of a lifetime's cognitive change

    Intelligence

    (2004)
  • A.J. Gow et al.

    Is age kinder to the initially more able?: Yes, and no

    Intelligence

    (2012)
  • T.A. Salthouse et al.

    Contextual analysis of fluid intelligence

    Intelligence

    (2008)
  • Y. Stern

    Cognitive reserve

    Neuropsychologia

    (2009)
  • A.G. Alder et al.

    Individual-differences assessment of the relationship between change in and initial level of adult cognitive functioning

    Psychology and Aging

    (1990)
  • J.L. Arbuckle

    AMOS (version 7), [computer program]

    (2007)
  • P.B. Baltes et al.

    On the dilemma of regression effects in examining ability-level-related differentials in ontogenetic patterns of intelligence

    Developmental Psychology

    (1972)
  • A.A.M. Bielak et al.

    Depressive symptoms predict decline in perceptual speed in older adulthood

    Psychology and Aging

    (2011)
  • E. Ferrer et al.

    Longitudinal modeling of developmental changes in psychological research

    Current Directions in Psychological Science

    (2010)
  • E. Ferrer et al.

    Multivariate modeling of age and retest in longitudinal studies of cognitive abilities

    Psychology and Aging

    (2005)
  • A.J. Gow et al.

    Mental ability in childhood and cognitive aging

    Gerontology

    (2008)
  • F.J. Infurna et al.

    Dynamic links between memory and functional limitations in old age: Longitudinal evidence for age-based structural dynamics from the AHEAD study

    Psychology and Aging

    (2011)
  • A.R. Jensen

    The g factor: The science of mental ability

    (1998)
  • J.R. Nesselroade et al.

    Regression toward the mean and the study of change

    Psychological Bulletin

    (1980)
  • R.E. Nisbett et al.

    Intelligence: New findings and theoretical developments

    American Psychologist

    (2002)
  • Cited by (21)

    • Resilience resources may buffer some middle-aged and older Black Americans from memory decline despite experiencing discrimination

      2023, Social Science and Medicine
      Citation Excerpt :

      In the current study, 1862 participants were combined across these three time points to increase the sample size of Black Americans. Studies have shown that longitudinal research is highly susceptible to regression-to-the-mean effects when only two time points are included, which can greatly affect the modeled longitudinal trajectory (Gross et al., 2015; Salthouse, 2012). With only two time points, most people that start higher will have a negative trajectory and most people starting lower will have a positive trajectory.

    • Midlife level and 15-year changes in general cognitive ability in a sample of men: The role of education, early adult ability, BMI, and pulse pressure

      2017, Intelligence
      Citation Excerpt :

      A comparison across five levels of initial performance revealed no differences in cognitive change. Whereas several subsequent studies confirmed a lack of relationship between initial ability level and rate of cognitive change in advanced age (e.g. Christensen et al., 2001; Eisdorfer, 1963; Salthouse, 2012), others instead suggested that a higher initial ability level attenuates decline, for example in Raven's matrices (e.g. Bourne, Fox, Deary, & Whalley, 2007) and verbal ability (Deary, McLennan, & Starr, 1998). Thus, the evidence concerning the relation between initial ability level and rate of cognitive change in late midlife or old age is mixed.

    • Age-related brain activation during forward and backward verbal memory tasks

      2014, Neurology Psychiatry and Brain Research
      Citation Excerpt :

      Furthermore, the neural activity and PSC in bilateral MFG decreased with age. We suggest that the decrease of activity across the four age groups might be related directly to differences in brain processes underlying executive functions, although specific processes remain to be determined; early structural or volumetric deterioration within the frontal areas has been suggested to be responsible for executive function decline.42–44 It is difficult at this time to interpret precisely the brain changes underlying a decrease or omission of significant brain activity to a particular area as reflecting some sort of decline or degeneration, versus something more related to changes in efficiency which also might be possible (although notably there were also some performance declines on the BRT in older participants that also must be accounted for).

    • Cognitive reserve moderates the association between hippocampal volume and episodic memory in middle age

      2013, Neuropsychologia
      Citation Excerpt :

      Another reason for some studies' failure to detect effects of reserve might be their wide age range. For example level of education among 49–82 year olds or baseline cognitive ability among 18–97 year olds did not affect the rate of cognitive change in studies of Van Gerven, Van Boxtel, Ausems, Bekers, and Jolles (2012), Salthouse (2012), respectively. The narrow age range in our study has the advantage that the sample is more homogeneous.

    View all citing articles on Scopus
    View full text