What and When of CognitiveAgingTimothy A. SalthouseUniversity of VirginiaABSTRACT—Adult age differences have been documented on awide variety of cognitive variables, but the reasons for thesedifferences are still poorly understood. In this article, I describeseveral findings that will need to be incorporated into eventualexplanations of the phenomenon of cognitive aging. Despitecommon assumptions to the contrary, age-related declines inmeasures of cognitive functioning (a) are relatively large, (b)begin in early adulthood, (c) are evident in several differenttypes of cognitive abilities, and (d) are not always accompaniedby increases in between-person variability.KEYWORDS—aging; cognition; reasoning; memory; speedThe phenomenon of cognitive aging has been noticed almost as long asthe phenomenon of physical aging, but it is still not well understood.This is unfortunate because cognitive functioning can affect one’squality of life, and even the ability to live independently. Furthermore,cognitive functioning in early adulthood may be related to the devel-opment of pathologies such as Alzheimer’s disease in later adulthood.One way to conceptualize understanding is that it is equivalent toknowing answers to the questions of what, when, why, where, and how.In this article, I summarize some of the progress that has been achievedin describing the phenomenon of cognitive aging in terms of thequestions of what and when. Although not much is yet known aboutwhy (what is ultimately responsible), where (in the nervous system),and how (via what mechanisms) age-related cognitive changes occur, akey assumption of my research is that answering these other questionswill be easier as the characterization of what and when becomes moreprecise.WHAT AND WHENIt is often assumed that age-related effects on cognitive functioningare small, are limited to aspects of memory, begin relatively late inadulthood, and possibly affect only some people, so that any age-re-lated declines are accompanied by increases in between-person var-iability. However, recent research in my laboratory and elsewheresuggests that these assumptions may all be incorrect. Evidence rele-vant to these issues can be illustrated with data aggregated acrossseveral recent studies in my laboratory (Salthouse, 2001a, 2001b;Salthouse, Atkinson, & Berish, 2003; Salthouse & Ferrer-Caja, 2003;Salthouse, Hambrick, & McGuthry, 1998; Salthouse et al., 2000).Participants in these studies were recruited through newspaper ad-vertisements, appeals to community groups, and referrals from otherparticipants. Nearly all of the participants reported themselves to bein good to excellent health, and they averaged approximately 16 yearsof education.Four tests were common to most of these studies. A vocabulary testinvolved the examinee selecting the best synonyms of target words, ineach case from a set of five alternatives. A speed test required theparticipant to classify pairs of line patterns as the same or different asrapidly as possible. Reasoning was assessed with the Raven’s Pro-gressive Matrices, in which each test item consists of a matrix ofgeometric patterns with one missing cell, and the task for the par-ticipant is to select the best completion of the missing cell from a setof alternatives. Finally, a memory test involved three auditory pre-sentations of the same list of unrelated words, with the participantinstructed to recall as many words as possible after each presentation.Data for the vocabulary, speed, and reasoning tests are based on 1,424adults, and those for the memory test are based on 997 adults.Because the raw scores for the four tests are in different units, all ofthe scores have been converted to z scores (by subtracting each scorefrom the mean for that test and then dividing by the standard deviation)so that the age trends can be directly compared. The means for the zscores are plotted as a function of age in Figure 1. The bars above andbelow each point are standard errors, which represent the precision ofthe estimate (i.e., the smaller the bars, the more precise the estimate).Six important observations about the data in this figure can be noted.First, scores on the vocabulary test were higher with increased ageuntil about the mid-50s, after which they either remained stable ordeclined slightly. Findings such as these have been interpreted asindicating that knowledge accumulates with increased age, but com-pelling explanations for why this age function is curvilinear are not yetavailable (Salthouse, 2003).Second, similar negative age trends are evident in the measures ofspeed, reasoning, and memory. Although not represented in the figure,the correlations between age and these variables were also similar, asthey were .47, .48, and .43, respectively, for the speed, rea-soning, and memory variables.Third, the age-related effects on the speed, reasoning, and memoryvariables are fairly large. Not only are the age correlations forthese variables greater than most correlations involving individualdifferences reported in the behavioral sciences, but the averageAddress correspondence to Timothy A. Salthouse, Department of Psy-chology, P.O. Box 400400, University of Virginia, Charlottesville,VA 22904-4400; e-mail: [email protected] DIRECTIONS IN PSYCHOLOGICAL SCIENCE140 Volume 13—Number 4Copyright r 2004 American Psychological Societyperformance for adults in their early 20s was near the 75th percentilein the population, whereas the average for adults in their early 70s wasnear the 20th percentile.Fourth, the relations between age and the speed, reasoning, andmemory variables are primarily linear. This observation is relevant topotential interpretations of the effects because the absence of obviousdiscontinuities in the functions suggests that transitions such as re-tirement, or menopause for women, are probably not responsible formuch, if any, of the effects.Fifth, the data in the figure indicate that age-related effects areclearly apparent before age 50. For some variables, there may be anacceleration of the influences at older ages, but age-related differ-ences are evident in early adulthood for each variable.And sixth, the age-related declines in these samples are not ac-companied by increases in between-person variability. One way toexpress the relation between age and between-person variability is interms of the correlation between age and the between-person standarddeviation for the individuals in each