Although there have been many reports over the last 100 years of age-related differences in cognitive functioning, there is still considerable controversy about the age at which cognitive decline begins. This lack of consensus is unfortunate because the question is important for both practical and theoretical reasons. For example, the age at which cognitive decline begins is relevant to the optimum time to implement interventions designed to prevent or reverse age-related declines. Many interventions currently target adults 60 years of age and older. However, if people start to decline when they are in their 20s and 30s, a large amount of change will likely have already occurred by the time they are in their 60s and 70s. This may affect the likelihood that interventions at that age will be successful because the changes might have accumulated to such an extent that they may be difficult to overcome.
The question of when decline begins is also relevant to the theoretical investigation of potential causes of declines in cognitive functioning because declines that begin early are unlikely to be attributable to conditions specific to later life, such as menopause, retirement from paid employment, or certain age-related diseases. The answer to the question of when decline begins may also indicate which period in adulthood is likely to be most informative for learning about causes of age-related cognitive decline because, for example, studies restricted to samples of older adults might have limited value for discovering the causes of a phenomenon that originated decades earlier.
One type of evidence suggesting that age-related cognitive declines begin relatively early in adulthood are the age trends in a variety of neurobiological variables that can be assumed to be related to cognitive functioning. Among the variables that have been found to exhibit nearly continuous age-related declines in cross-sectional comparisons beginning when adults are in their 20s are measures of regional brain volume (
Allen, et al., 2005;
Fotenos, et al., 2005;
Kruggel, 2006;
Pieperhoff, et al., 2008;
Sowell, et al., 2003), myelin integrity (
Hsu, Leemans, et al., 2008;
Sullivan & Pfefferbaum, 2006), cortical thickness (
Magnotta, et al., 1999;
Salat, et al., 2004), serotonin receptor binding (
Sheline, et al., 2002), striatal dopamine binding (
Erixon-Lindroth, et al., 2005;
Volkow, et al., 2000), accumulation of neurofibrillary tangles (
Del Tredici & Braak, 2008), and concentrations of various brain metabolites (
Kadota, et al., 2001).
Furthermore, cross-sectional declines in comparisons of cognitive functioning based on samples of 250 or more adults across a wide age range have been reported since the 1930s (
Jones & Conrad, 1933), and have been described in numerous recent publications (
Salthouse, 1998;
Salthouse, 2005;
Salthouse, et al., 2003;
Schroeder & Salthouse, 2004;
Schaie, 2005). In virtually every case, the age trends in these studies have revealed nearly monotonic declines in average level of cognitive performance starting in early adulthood.
It might appear on basis of these well-replicated results with neurobiological and cognitive variables that there is a simple answer to the question of when cognitive decline begins. That is, because cross-sectional age comparisons have consistently revealed nearly continuous age-related decreases in presumably relevant neurobiological variables and in various measures of cognitive performance that appear to begin when adults are in their 20s or early 30s, one might conclude that cognitive decline begins shortly after individuals reach maturity. However, in striking contrast to these empirical results are numerous assertions that cognitive decline begins late in life:
- “Cognitive decline may begin after midlife, but most often occurs at higher ages (70 or higher).” (Aartsen, et al., 2002)
- “…relatively little decline in performance occurs until people are about 50 years old.” (Albert & Heaton, 1988).
- “…cognitive abilities generally remain stable throughout adult life until around age sixty.” (Plassman, et al., 1995)
- “…no or little drop in performance before age 55…” (Ronnlund, et al., 2005)
- “…most abilities tend to peak in early midlife, plateau until the late fifties or sixties, and then show decline, initially at a slow pace, but accelerating as the late seventies are reached.” (Schaie, 1989).
A dramatic discrepancy therefore exists between a substantial body of empirical results on one hand, and frequent claims about the time course of cognitive aging on the other hand. Because one cannot hope to explain a phenomenon until its nature, including its trajectory, is accurately described, it is essential to understand the reasons for this discrepancy.
Some of the differences between the evidence just mentioned and the cited assertions may be attributable to variations in how the same findings are interpreted, or to emphases on different types of cognitive variables. However, it is likely that a major reason for the discrepancy is that different patterns of age-cognition relations have been found with longitudinal, or within-person, comparisons, and with cross-sectional, or between-person, comparisons. One of the first reports of a longitudinal comparison with cognitive variables was described in a 1928 book (
Thorndike, et al., 1928). Although other researchers at about the same time reported cross-sectional declines between 18 and 50 years of age on the Army Alpha test, these authors described a study in which the scores for people between 16 and 45 years of age increased over a 5-to-9 year interval. Rather than revealing decline, therefore, these results suggested that there were improvements in cognitive functioning with increased age when the comparisons were based on observations of the same people at different ages. Subsequent longitudinal studies have replicated the finding of relatively preserved, or even enhanced, levels of cognitive functioning with increased age in longitudinal comparisons involving adults up to about 60 years of age.
illustrates these patterns with cross-sectional and longitudinal age trends on two tests from the Seattle Longitudinal Study (
Schaie, 2005). The top two panels illustrate that there are nearly monotonic age-related declines in the cross-sectional comparisons (dotted lines), but that longitudinal comparisons (solid lines) reveal either stable or increasing age trends. The bottom two panels portray the same data in a different format. In these figures the vertical axis corresponds to standard deviation units rather than T-scores, and the bars represent the cross-sectional difference (black bars) or the longitudinal changes (gray bars). In order to maximize comparability with the results of the current project in which the average retest interval was 2.5 years, the 7-year differences and changes in these figures have been converted to a 2.5 year interval by algebraic substitution (i.e., X = 2.5 * [Score/7]). Despite the different formats, the upper and lower panels reveal the same pattern of moderately large negative age trends in the cross-sectional comparisons (dotted lines and black bars), but little or no age decline in the longitudinal comparisons (solid lines and gray bars).
It is not surprising that divergent results such as those portrayed in have led some researchers to the conclusion that little or no cognitive decline occurs before about age 60. However, a critical assumption of this interpretation is that the results of longitudinal comparisons are more accurate or valid than cross-sectional comparisons with respect to “true” age relations, and it is important to consider what might be responsible for the different patterns of results in the two types of comparisons before accepting this assumption. Only after this issue is resolved can a definitive conclusion be reached about when cognitive decline begins because decline may begin late if cross-sectional results are misleading, but decline may begin early if longitudinal results were found to be influenced by a variety of non-maturational determinants in addition to the maturation effects of primary interest.
What is likely the dominant interpretation of the different age trends found in cross-sectional and longitudinal comparisons of cognitive functioning attributes the discrepancy to characteristics other than age confounding cross-sectional comparisons.
Kuhlen (1940) may have been the first to describe what are now commonly referred to as cohort effects, which include a variety of influences on cognitive functioning associated with changes in the social and cultural environment, such as quantity and quality of education, nature of health care, etc.
Although the cohort interpretation is widely accepted, it is currently somewhat underspecified. For example, a critical expectation of the cohort interpretation is that statistical control of cohort-defining variables should eliminate the cross-sectional age trends. Some variables, such as years of education, are easily assessed, but the prediction that cross-sectional age trends would be eliminated after adjusting for these other variables cannot be adequately tested until all of the cohort-relevant variables are identified and measured. Another limitation of the cohort interpretation is that little is currently known about the time course of cohort influences relative to the age range at which cross-sectional age differences are apparent. That is, cohort influences are sometimes referred to as generational effects, as though they occur over intervals of 25 years or more, but they would have to operate over periods as short as 5 or 10 years to account for the cross-sectional age differences found in some cognitive variables.
Another factor that has been mentioned as a possible contributor to the different cross-sectional and longitudinal age trends is retest effects associated with prior testing. Retest effects refer to influences on the difference in performance between the first and a subsequent measurement occasion that are attributable to the previous assessment. That is, the mere fact that an individual has already been evaluated could change his or her performance on a successive measurement occasion, in which case the age trends inferred from longitudinal comparisons may be misleading with respect to “true” age effects. Cross-sectional comparisons do not involve testing the same individuals again, and therefore retest effects could be contributing to the discrepancy between cross-sectional and longitudinal results by distorting the age trends in longitudinal comparisons. Although seldom mentioned in discussions of the discrepancy between cross-sectional and longitudinal age trends, several findings appear more consistent with the retest interpretation than with the cohort interpretation.
First, because non-human laboratory animals are typically raised in nearly constant environments, age comparisons in non-human animals can be assumed to be free of cohort contaminations attributable to changing environments. To the extent that cohort differences distort cross-sectional comparisons, therefore, little or no age differences in cognitive functioning might be expected in comparisons of non-human animals. However, there are numerous reports of cross-sectional age-related declines in measures of memory and cognition in species ranging from non-human primates (
Herndon, et al., 1997) to fruit flies (
LeBourg, 2004). Second, although relatively few longitudinal studies have been conducted with non-human animals, it is noteworthy that several studies with rats have reported smaller longitudinal age changes than cross-sectional age differences in measures of maze learning (
Caprioli, et al., 1991;
Dellu, et al., 1997;
Markowska & Savonenko, 2002). Because the different cross-sectional and longitudinal age trends cannot be attributed to cohort differences distorting the cross-sectional results in organisms raised in constant environments, the discrepancies in these studies are likely attributable to retest effects distorting the longitudinal comparisons. And third, several studies examining regional brain volume, which is a variable presumably related to cognitive functioning but not susceptible to practice effects, have reported longitudinal age declines that are at least as large as the cross-sectional differences (
Fotenos, et al., 2005;
Raz, et al., 2005;
Scahill, et al., 2003).
Although the results just described are consistent with the interpretation that longitudinal age trends in cognitive functioning are distorted by the presence of retest effects, they are all indirect. The primary prediction from the retest interpretation is that estimates of retest effects should be moderately large and positive, such that they offset any negative effects of aging or maturation. Several methods have been proposed to estimate the magnitude of retest effects, but only a few have been applied to adults under the age of 60, which is the period most relevant to the question of when age-related decline begins. One method of estimating retest effects is based on a comparison of the cognitive performance in samples of people tested once with those tested twice (
Ronnlund, et al., 2005;
Schaie, 2005). This difference, after adjusting for any differences in initial level of performance, has been used as an estimate of the benefit of prior test experience. Most of the estimates derived from this method have been positive, and considerably larger than the annual cross-sectional age differences. A second method of assessing retest effects has relied on variability across research participants in the retest intervals to decompose the observed change into maturational effects and retest effects. Because this latter method requires a special type of longitudinal design in which people vary in the interval between successive assessments, such that there is not a perfect confounding of the increase in age and the increase in test experience, it has only rarely been used. Nevertheless, both
McArdle, et al., (2002) and
Salthouse, et al., (2004) found that in adults under the age of 60, the retest estimates derived from this method were positive and moderately large.
Three different methods of estimating retest effects in longitudinal studies were examined in the current project. As just mentioned, two of the methods, comparing performance of people of the same age tested twice with those tested once, and capitalizing on variability in the retest intervals to distinguish maturation and retest components of longitudinal change, have been used in earlier research. A new method relied on a comparison of the magnitude of change in a longitudinal study with the change in a short-term retest study in which the test-retest interval ranged from 1 to 14 days as the primary basis for distinguishing maturation and retest effects. The rationale is that it is very unlikely that maturation influences are operating over such a short interval, and thus the results of short-term retest studies provide a relatively pure estimate of the potential impact of retest influences that can be compared with estimates of longitudinal change. Furthermore, because the retest interval in the longitudinal study varied from 1 to 7 years, the effect of the retest interval on the magnitude of longitudinal change can also be examined to determine the time course of these retest influences.
The publisher's final edited version of this article is available at 
