WAIS-III IQs, Horn's theory, and generational changes from young adulthood to old age.
It has long been known that scores on Wechsler's adult scales are generally lower for the older age groups than the younger ones, with Wechsler (1958, p. 142) noting, "What is definitely established is ... that the abilities by which intelligence is measured do in fact decline with age." However, Wechsler and other early clinicians and researchers emphasized age-by-age differences observed in cross-sectional data without considering the impact of contaminating cohort variables such as educational attainment. When education is controlled statistically, cross-sectional analyses of the Wechsler Adult Intelligence Scale (WAIS; Wechsler, 1955) and Wechsler Adult Intelligence Scale-Revised (WAIS-R; Wechsler, 1981) indicate that the lower scores on Verbal subtests by older age groups was an artifact that did not persist when education was equated across ages; declines in mean Performance IQs, however, remained even after removing the contaminating influence of education (Birren & Morrison, 1961; Kaufman, Reynolds, & McLean, 1989). These education-controlled results conform closely to Horn and Cattell's (1966, 1967) predictions for crystallized (Gc) and fluid intelligence (Gf). Verbal IQ, interpreted as a good measure of Gc (Matarazzo, 1972), was a "maintained" ability that increased into the decade of the 60s before declining slightly as people passed their 70th birthdays. Performance IQ, composed of novel problem-solving tasks not dependent on formal education or acculturation and defined as a measure of Gf (Matarazzo, 1972), behaved as a "vulnerable" ability, peaking in the early 20s and then declining steadily and dramatically between ages 25 and 74.
Yet, age-related findings on Wechsler's adult scales, and their interpretation from Horn's theory, are not seen as universal truths about the relationship between aging and intelligence. Indeed, changes with IQ from early to late adulthood have been the subject not only of considerable research (Botwinick, 1977; Horn & Hofer, 1992; Kaufman, 1990, Chapter 7; Lawton & Salthouse, 1998; McArdle, Prescott, Hamagami, & Horn, 1998), but also of notable controversy (Baltes & Schaie, 1976; Horn & Donaldson, 1976); both the investigations and the disputes center around the maintenance of intellectual abilities across the adult life span. Schaie (1983) and his colleagues (e.g., Schaie & Hertzog, 1986) have argued for the maintenance of adult intelligence across virtually the entire life span. In contrast, Horn (1985, 1989, 1991) and his colleagues (e.g., Horn & Hofer, 1992; Horn & Noll, 1997; McArdle et al., 1998) believe that crystallized (school-related) abilities and a few other abilities such as quantitative thinking maintain through old age but fluid (novel problem-solving) skills, visual-spatial abilities, speed, and short-term memory are "vulnerable," peaking relatively early (in some cases late adolescence) before declining during the remainder of the life span.
That dispute is not reconcilable by the present set of studies, which focused on the Wechsler Adult Intelligence Scale-Third Edition (WAIS-III; The Psychological Corporation, 1997; Wechsler, 1997) and its predecessor, the WAIS-R (Wechsler, 1981). Horn and Schaie used different methodologies and different instruments. Horn's data were obtained primarily from cross-sectional and longitudinal studies and "[m]uch of the evidence on vulnerable and maintained abilities has been based on the use of the WAIS" (Horn & Hofer, 1992, p. 71). In contrast, Schaie's data were obtained from cross-sequential investigations, a blend of cross-sectional and longitudinal methodologies, and his conclusions are largely based on the group-administered Primary Mental Abilities (PMA) test (Thurstone & Thurstone, 1949), which includes no tasks that require visual-motor coordination and is not normed above age 18. Because the present studies define intelligence in terms of Wechsler's IQs, and evaluate data obtained from cross-sectional (Study 1) and longitudinal (Study 2) investigations, the results are mainly interpretable within the context of Horn's fluid-crystallized (Gf-Gc) theory, specifically the distinction between maintained and vulnerable abilities.
Nonetheless, the differences between Schaie's and Horn's findings are not as profound as the researchers themselves have sometimes claimed (Kaufman, 1990). For example, scores on PMA subtests that resemble Wechsler's Performance tasks -- Space (a measure of broad visualization or Gv) and Reasoning (a measure of Gf) -- decline more dramatically in old age than do scores on a Gc subtest (verbal meaning), akin to Wechsler's Verbal subtests (Kaufman, 1990, Fig. 7.7; Schaie, 1983, Table 4.18). In Schaie's (1983) data, abilities on both kinds of tasks maintain into the 60s before declining, but when these declines do occur, they are sometimes precipitous (e.g., between ages 67 and 74), especially for Performance-like subtests. Although, as noted, the use of Wechsler's scales in the present studies makes the results more interpretable in the context of Horn's than Schaie's research results, the integration of data from the present cross-sectional (Study 1) and longitudinal (Study 2) investigations renders the present findings relevant to Schaie's cross-sequential findings as well.
The present studies also have theoretical implications. As Horn (1985, 1989, 1991) revised and expanded Gf-Gc theory, it became apparent that Verbal IQ is not a pure measure of Gc, but that two Verbal subtests (arithmetic and digit span) are mainly measures of short-term apprehension and retrieval (SAR), i.e., short-term memory. Similarly, Performance IQ probably measures a blend of Gf, Gv, and Gs (broad speediness) (Horn & Hofer, 1992; Kaufman, 1994), with some researchers insisting that the quartet of subtests associated with the Perceptual Organization factor measures only Gv, and not Gf at all (McGrew, 1997; Woodcock, 1990).
These studies, therefore, continue the work of previous investigators, who examined age-by-age changes on the WAIS and WAIS-R, by addressing the relationship of several of Horn's constructs within his expanded notion of Gf-Gc theory to age changes across the adult life span. The current analyses offer additional insights into the IQ-aging relationship of Wechsler's IQs, over and above the theoretical and neuropsychological insights yielded by prior analyses of the WAIS and WAIS-R, because: (a) the WAIS-III was standardized on a carefully selected, Census-representative sample of adults ages 16-89, in contrast to the 16-64 year range for the WAIS and the 16-74 year range for the WAIS-R; (b) the WAIS-III includes as part of the regular five-subtest Performance Scale the new matrix reasoning task, an untimed prototypical measure of Horn's Gf that does not require visual-motor coordination (in contrast to traditional Performance subtests that depend on speeded responding, visual-spatial ability more so than fluid reasoning, and, in most instances, coordination); (c) in addition to three IQs, the WAIS-III yields standard scores (mean = 100, S.D. = 15) on four Factor Indexes known its verbal comprehension index (VCI), perceptual organization index (POI), working memory index (WMI), and processing speed index (PSI); and (d) the WAIS-III was normed in 1995-1996, permitting an analysis of possible changes from generation to generation in the relationship of IQ to age.
The first point regarding the extended age range permits an extension of knowledge about aging on Wechsler's IQ scales that goes beyond the decade of the 70s and into the 80s. Whereas Ryan and his colleagues have analyzed WAIS-R aging data for an old-age sample that includes adults in their 80s, their sample was not representative of the U.S. and was relatively small in size (Ryan, personal communication, March 1998; Ryan, Paolo, & Brungardt, 1990; Ryan, Paolo, & Dunn, 1995). This extended age range is especially important in view of findings that WAIS-R Verbal IQ, when controlled for education, does not vary significantly with chronological age across the 20-74-year age range, but that for ages 75 and above there is evidence of a decline in Verbal IQ on Wechsler's scales (Schmitz-Scherzer & Thomae, 1983) and on other measures of Gc as well (Kaufman & Horn, 1996), including PMA tasks studies by Schaie (1983).
The second point concerning the inclusion of matrix reasoning in the WAIS-III is primarily of theoretical rather than clinical interest, as it allows a more complete understanding of what Horn abilities are, in fact, measured by Wechsler's Performance IQ -- Gf (Matarazzo, 1972), Gv (McGrew, 1997; Woodcock, 1990), or a blend of Gf and Gv (Horn & Hofer, 1992; Kaufman, 1994). The third point, too, involving the four additional standard scores yielded by the WAIS-III, is primarily of theoretical interest. From Horn et al.'s (Horn, 1985, 1989, 1991; Horn & Hofer, 1992; Horn Noll, 1997) writings and research, there is an apparent correspondence between WAIS-III Indexes and Horn factors: VCI=Gc, POI = Gf/Gv, WMI = SAR, PSI = Gs. In addition, there is a built-in relationship between the WAIS-III IQ scales and Indexes: The VCI and WMI are composed only of Verbal subtests; the POI and PSI are composed only of Performance subtests. Therefore, the theoretical interpretation of the relationship of Verbal IQ to age will be better understood by juxtaposing the pattern of age-by-age means on Verbal IQ alongside the patterns for VCI and WMI. Similarly, age changes on Performance IQ will be understood more clearly from Horn's theory by comparing age-by-age changes on Performance IQ to comparable changes on POI and PSI.
The fourth point concerning generational changes provides potentially the most important information yielded by the present studies. For example, there is some question about the peak age for Gc, as estimated by Wechsler's Verbal IQ. As noted, the peak was in the early 60s for previous editions of Wechsler's adult scales, which corresponds to the peak performance noted by Horn (1985, 1989, 1991) in his own studies. However, other studies with more recent data have suggested a Gc peak in the early 50s (Kaufman & Horn, 1996; Kaufman, Kaufman, Chen, & Kaufman, 1996; Wang & Kaufman, 1993). Also, there is uncertainty about the age-by-age pattern of mean scores on measures of Gf across the life span. WAIS-R data (Kaufman et al., 1989) and Horn's (1985, 1989, 1991) data indicate a steady decline from the early 20s to early 70s; data from the Kaufman Adolescent and Adult Intelligence Test (KAIT; Kaufman & Kaufman, 1993) and Kaufman Brief Intelligence Test (K-BIT; Kaufman & Kaufman, 1990) suggest a significant dip in the 20s, a plateau from the mid-20s to mid-50s, and then a dramatic drop from the mid-50s through old age (Kaufman & Horn, 1996; Wang & Kaufman, 1993). Previous versions of Wechsler's adult scales, normed in the mid-1950s and late 1970s, present one picture of Gf and Gc changes with increasing age. In contrast, the Kaufman scales normed for adults in the late 1980s and early 1990s present a different picture. The differences in the age-by-age patterns of mean scores for Gc and Gf may be due to the difference in the intelligence test studied (Wechsler versus Kaufman), or they may be due to real generational shifts in the age-IQ relationship. The availability of mid-1990s data for a Wechsler adult test permits these competing hypotheses to be evaluated.
The results of two studies are presented in this paper. In Study 1, the WAIS-III standardization data are examined cross-sectionally, with educational attainment controlled, replicating the study conducted with the WAIS-R by Kaufman et al. (1989). Study 2 is a longitudinal investigation with independent samples that follows the performance of seven age cohorts on the WAIS-R and WAIS-III, using standardization data for these two instruments. This study represents a replication of a similar study conducted with the WAIS and WAIS-R (Kaufman, 1990, Chapter 7).
2. Study 1 -- cross-sectional analysis of WAIS-III
The WAIS-III (The Psychological Corporation, 1997; Wechsler, 1997), for adults ages 16 to 89 years, was formatted to be similar to the WISC-III (Wechsler, 1991): It includes Verbal, Performance, and Full Scale IQs; Indexes on four factors (VCI, POI, WMI, and PSI, as previously described), all with mean of 100 and S.D. of 15; and scaled scores on 14 subtests (mean = 10, S.D. = 3), seven Verbal and seven Performance. The WAIS-III was standardized on 2450 adult subjects, selected according to 1995 U.S. Census data, and stratified according to age, gender, race/ethnicity, geographic region, and education level. Subjects were divided into 13 age groups between 16-17 and 85-89 with each age group including 100 to 200 people. Average split-half reliability coefficients, across the 13 age groups, are as follows: .97 for Verbal IQ, .94 for Performance IQ, and .98 for Full Scale IQ. The average value for PSI was .87 (a test-retest coefficient because split-half is not applicable for highly speeded tasks), with the split-half coefficients averaging .93 to .96 for the other three Indexes. The average individual subtest reliabilities ranged from .93 (vocabulary) to .70 (object assembly), with a median coefficient of .85. Stability coefficients based on 394 adults from four broad age groups tested twice (interval averaging about 5 weeks) were as follows: Verbal IQ (.94-.97), Performance IQ (.88-.92), Full Scale IQ (.95-.97), vcI (.92-.96), POI (.83-.92), WMI (.87-.93), and PSI (.84-.90). Average test-retest coefficients for subtests ranged from .69 for picture arrangement to .94 for information; median coefficients, across age groups, were .83 for the seven Verbal subtests and .79 for the seven Performance subtests. The WAIS-III technical manual (The Psychological Corporation, 1997) reports data from numerous factor analytic studies (exploratory and confirmatory) that support the underlying four-factor structure of the WAIS-III for ages 16 to 74 years, thereby offering evidence of construct validity for these ages. For ages 75-89, however, the Perceptual Organization factor was meager (only matrix reasoning had a loading above .40), with most Performance subtests joining the PSI.
The WAIS-III standardization sample comprised 2450 individuals ages 16 to 89. The group was subdivided into 13 separate subsamples between 16-17 and 85-89. Each of the 11 subsamples from 16-17 through 75-79 was composed of 200 individuals; ages 80-84 had n = 150 and ages 85-89 had n = 100. The number of males and females was equal through age group 55-64, but matched Census proportions at ages 65+, when females are more numerous. The sample was also stratified on the variables of race/ethnicity, geographic region, and educational attainment (The Psychological Corporation, 1997, pp. 19-39). When selecting the sample for testing, The Psychological Corporation (1997, Table 2.1) excluded three categories of adults. They excluded individuals with sensory or motor deficits that might compromise the validity of the obtained test scores (color-blindness, uncorrected hearing loss, uncorrected visual impairment, and upper extremity motor disorder). They excluded individuals for reasons concerning alcohol, drugs, and medication -- those undergoing current treatment for alcohol or drug dependency; those who consumed more than three alcoholic beverages more than two nights per week; and those currently taking antidepressants, anti-anxiety medication, or anti-psychotic drugs. Finally, they excluded adults with a known or possible neuropsychological disorder: seeing a doctor or other professional for memory problems or problems with thinking; experiencing any period of unconsciousness of at least 5 min; suffering a head injury that required hospitalization for more than 24 h; having a medical or psychiatric condition that could affect cognitive functioning, such as stroke, epilepsy, Alzheimer's dementia, schizophrenia, or brain surgery.
Table 2 Mean sums of scaled scores on the verbal, performance, and full scales (based on the reference group of 20-34-year-olds) for representative samples of adults ages 16+ on the WAIS, WAIS-R, and WAIS-III Verbal scale Performance scale Age group WAIS WAIS-R WAIS-III WAIS WAIS-R WAIS-III 16-17 54.6 50.7 55.6 48.8 47.4 50.2 18-19 57.3 51.7 58.5 49.4 46.9 50.8 20-24 59.5 58.6 58.1 50.6 51.1 50.4 25-29 62.3 61.4 60.4 51.4 49.9 50.9 30-34 59.3 61.4 62.0 47.9 49.9 49.4 35-44 60.2 57.5 63.4 46.1 45.3 48.1 45-54 58.0 58.1 64.8 41.0 42.1 44.8 55-64 55.8 55.6 59.9 37.1 38.2 40.3 65-69 53.7 53.5 59.2 34.4 33.8 35.9 70-74 47.7 51.5 58.1 29.5 30.6 33.8 75-79 -- 49.1 57.0 -- 27.8 31.2 80-84 -- 48.2 52.1 -- 28.2 27.9 85-89 -- 44.2 49.8 -- 22.9 25.9 Full scale Age group WAIS WAIS-R WAIS-III 16-17 103.4 98.1 105.8 18-19 106.7 98.6 109.3 20-24 110.1 109.8 108.5 25-29 113.9 111.3 111.3 30-34 107.4 111.3 111.4 35-44 106.3 102.7 111.7 45-54 99.1 100.2 109.6 55-64 92.9 93.8 100.2 65-69 88.1 87.4 95.1 70-74 77.2 82.1 91.8 75-79 -- 77.0 88.2 80-84 -- 76.4 80.0 85-89 -- 67.1 75.7 Separate values on the WAIS for ages 25-29 and 30-34 were from data provided by Matarazzo (1972, Table 10.4). Values for the WAIS-R for ages 25-29 and 30-34 are both based on Wechsler's (1981) data for a combined sample of 25-34-year-olds. WAIS data for ages 65-74 are for the stratified elderly sample tested by Doppelt and Wallace (1955). WAIS-R data for ages 75-89 are for the stratified elderly sample tested by Ryan et al. (1990), and were kindly provided by Ryan (personal communication, March 1998) for the age groups shown (n = 60, 75-79; n = 39, 80-84; n = 16, 85-89). Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved.
The standardization sample, therefore, excludes a variety of individuals who make up the adult population. The first two categories of individuals tend to include adults who are just as likely to be young as old. The latter category, however, regarding known or possible cognitive impairment, is age-related; both the number and severity of pathologies accelerate in old age (Rabbitt, Bent, & McInnes, 1997). Therefore, more older than younger individuals will have been excluded from the WAIS-III standardization sample, a fact that needs to be taken into account when interpreting the aging-IQ data. That is to say, the sample of adults ages 75-79, for example, is undoubtedly higher functioning than a random sample of 75- to 79-year-olds in the population. Any observed deficits in intelligence with increasing age might best be thought of as "lower-bound" estimates of the actual deficits within the population. There is also, however, a positive side to the liberal exclusion of adults with possible or known dementia or other impairments that affect thinking: It helps ensure that any observed declines in cognitive function are "real" age-related declines, and not merely artifacts of the inclusion of increasing numbers of subjects with dementia with increase in chronological age.
For the WAIS-R and previous versions of Wechsler's adult scales, scaled scores were obtained for every age group based on reference group norms for ages 20-34 years; then separate IQ tables were developed for each age; group to ensure that every age had mean = 100 and S.D. = 15. The WAIS-III departed from tradition, and followed the method used for Wechsler's children's scales: separate scaled-score norm tables for each age group and a single IQ conversion table for all ages. The method used for the WAIS-III is sensible and has many advantages for the clinician; however, the current method prevents easy comparison of the mean scores earned by different age groups based on a common yardstick (i.e., the reference group norms). The WAIS-III manuals (The Psychological Corporation, 1997; Wechsler, 1997) do not directly provide any age-related data to permit comparisons among age groups. However, The Psychological Corporation generously provided mean scaled scores, for each age, on the 14 WAIS-III subtests, based on the reference group of 400 adults ages 20-34 years. These data permitted direct comparisons across the 16- to 89-year age range on all subtests, Factor Indexes, and IQs.
However, comparisons of mean scores by age, even on the common metric of reference-group scaled scores, is confounded by cohort effects. The one cohort effect that is large and pervasive, and capable of being controlled, is educational attainment. Table 1 shows the percent of subjects at each age with different educational backgrounds; data are presented for the WAIS-III standardization sample, and, for comparison purposes, for the WAIS-R standardization sample as well. As indicated, sizable differences exist for the cross-sectional age groups on educational background. In the WAIS-III sample, for example, about half of the adults in their 80s failed to complete high school ([is less than or equal to] 11 years of schooling) versus about one-third of adults ages 65-79 and about 15% of adults ages 20-34. Analogously, about one-fourth of adults ages 25-34 graduated college (16+ years of schooling) compared to about 14% of those ages 65-79. Because educational attainment is strongly related to IQs on Wechsler's adult IQ scales, correlating about .60-.70 on the WAIS and about .45-.60 on the WAIS-R (Reynolds, Chastain, Kaufman, & McLean, 1987; Wechsler, 1958), it is evident that any mean differences in intelligence displayed by different age groups may merely be a reflection of educational differences from group to group. Therefore, control of educational attainment was essential.
Table 1 Percentage of adults with different amounts of formal education, by age group, for the WAIS-R and WAIS-III standardization samples WAIS-R (normed in 1978) Age 0-11 12 13-15 16+ group years years years years 16-17 97.0 3.0 0.0 0.0 18-19 39.5 48.5 12.0 0.0 20-24 17.0 43.5 29.5 10.0 25-29 17.0 39.3 20.0 23.7 30-34 17.0 39.3 20.0 23.7 35-44 26.0 42.4 13.6 18.0 45-54 34.4 40.0 11.2 14.4 55-64 43.7 36.9 12.5 6.9 65-69 55.0 25.6 8.1 11.2 70-74 60.0 23.8 8.1 8.1 75-79 66.7 23.3 10.0 0.0 80-84 66.7 23.0 10.3 0.0 85-89 68.8 12.4 18.8 0.0 WAIS-III (normed in 1995) Age 0-11 12 13-15 16+ group years years years years 16-17 (14.5) (35.5) (31.0) (19.0) 18-19 (16.0) (32.0) (34.0) (18.0) 20-24 15.0 34.0 40.0 11.0 25-29 14.0 36.5 26.0 23.5 30-34 14.0 33.5 29.5 23.0 35-44 9.5 34.0 27.5 29.0 45-54 17.5 33.5 24.0 25.0 55-64 26.5 38.0 17.5 18.0 65-69 32.0 38.0 16.0 14.0 70-74 31.5 39.5 15.0 14.0 75-79 34.0 37.0 15.5 13.5 80-84 49.3 28.7 10.7 11.3 85-89 50.0 27.0 11.0 12.0 Percentages for the WAIS-R standardization sample ages 16-74 are from Wechsler (1981, Table 5). Percentages for the WAIS-III standardization sample are from The Psychological Corporation (1997, Table 2.6). Values in parentheses denote parents' educational attainment. WAIS-R data for ages 25-29 and 30-34 are based on data provided for ages 25-34. WAIS-R data for ages 75-89 are for the stratified elderly sample tested by Ryan et al. (1990), and were kindly provided by Ryan (personal communication, March 1998) for the age groups shown (n = 60, 75-79; n = 39, 80-84; n = 16, 85-89). Percents for 13-15 years of schooling for the WAIS-R elderly sample are actually for 13+ years (no breakdown was given for 13-15 vs. 16+).
In their cross-sectional analysis of aging and IQ on the WAIS-R, Kaufman et al. (1989) controlled for education by equating all age groups within the 20-74 year range on this key variable. They accomplished the equating of age groups by weighting each age group in accordance with the educational attainment of the most educated WAIS-R age group, ages 25-34 years. First, they computed, separately by age group, the mean test scores earned by adults with different amounts of formal education, i.e., 0-8, 9-11, 12, 13-15, and 16+ years of schooling. Then they weighted each mean by the percent of adults at ages 25-34 who completed the pertinent years of schooling, such that all age groups, in effect, had the same proportions of adults in each education category as the "target" age group of 25-34.
In order to conduct the same type of analysis with the WAIS-III data, it was necessary to know the mean test scores earned by adults in each of the five educational categories, for every WAIS-III age group. Again, these age X education data were kindly provided by The Psychological Corporation. For the WAIS-R, ages 25-34 was selected as the target age group because that group was the most educated. From Table 1, it is evident that the most educated WAIS-III group is ages 35-44, with 29% college graduates and less than 10% high school dropouts. Nonetheless, to be comparable to the procedure used in the WAIS-R study, and permit meaningful comparisons between WAIS-R and WAIS-III data, ages 25-34 were used as the target in the present study.
WAIS-III technical manual (The Psychological Corporation, 1997, Table 2.6), provides percents of adults in each educational category separately for ages 25-29 and 30-34. The midpoint of these percents was used to determine the percents for ages 25-34. These percents are as follows: 0-8 years (4.5%), 9-11 years (9.5%), 12 years (35.0%), 13-15 years (27.75%), and 16+ years (23.25%) of schooling. For each age group between 20-24 and 85-89 years, the pertinent mean score -- whether subtest scaled score, sum of scaled scores for each of the four Indexes, or sum of scaled scores for each of the three IQs -- was weighted by the "target" percent for that educational category. (Scores for ages 16-19 were not equated for educational attainment because only parents' education was provided, and, in addition, because many of these older adolescents had not yet completed their formal education).
To illustrate the weighting procedure, consider ages 20-24: The mean score on each test or scale earned by young adults with 0-8 years of schooling was weighted by 4.5; the mean score earned by those with 9-11 years of schooling was weighted by 9.5; the mean score earned by those with 12 years of schooling was weighted by 35.0; the mean score earned by those with 13-15 years of schooling was weighted by 27.75; and the mean score earned by those with 16+ years of schooling was weighted by 23.25. This education-weighting procedure was applied for each scaled score, Index, and sum of scaled scores at each age, producing a set of scores that equated the mean scores obtained by each age group to the educational attainment of adults ages 25-34. To convert the weighted sums of scaled scores to Indexes or IQs, these sums were entered into the WAIS-III IQ and Index conversion tables provided for all ages between 16 and 89 years (Wechsler, 1997, Tables A.3 through A.9), and interpolated values, to the nearest tenth, were obtained. When two or more sums of scaled scores convert to the same IQ (e.g., sums of 66 and 67 both convert to a Full Scale IQ of 74), then the midpoint of the sums was set equal to the IQ for interpolation purposes (i.e., a sum of 66.5 corresponds to an IQ of 74). These education-controlled scores permitted the comparison of scaled scores, Factor Indexes, and IQs for the 11 age groups between 20-24 and 85-89, removing the confounding due to different levels of educational attainment. The more the educational attainment of an age group deviated from the target age of 25-34 (e.g., ages 80-84 and 85-89), the greater the amount of adjustment to obtained values. Thus, adults ages 85-89, for example, comprised 50% high school dropouts and 12% college graduates (see Table 1); when adjusted for education, however, they earned mean scores "as if'" they were comprised of 14% dropouts and 23.25% college graduates, the same as adults ages 20-34.
The method of equating for education used in this study -- indeed any attempt to equate for education for groups that differ substantially in age -- is an inexact science and must be considered as approximate corrections for society-based inequalities. For example, schooling beyond high school was enjoyed primarily by the elite a half-century ago, whereas now it is commonplace. Indeed, about 50% of each age group between 20-24 and 45-54 had one or more years of college (see Table 1), making post-high school education "average" for young and middle-age adults. The meaning of "some college" or "college graduate," therefore, is not a constant across generations, just as "high school dropout" has a far greater stigma for younger than older adults. Because the meaning of "years of schooling" has changed over time, it is not a perfect yardstick; correcting for it by equating all age groups to ages 25-34 may introduce error. Nonetheless, despite rational and conceptual arguments, there is some empirical evidence that the equating procedure used in this study is sensible, and that "years of schooling" may have a fairly constant meaning across the adult age range. Data presented by Kaufman et al. (1989) for the WAIS-R compared the mean Verbal sums of scaled scores (based on a common reference group) earned by different educational categories within each of seven WAIS-R age groups. Mean scores for those with 0-8 years of schooling averaged about 40 regardless of age, and similar consistency across age was obtained for those with 9-11 years of schooling (mean = 50), 12 years of schooling (mean = 55-60), 13-15 years of schooling (mean = 60-65), and 16 years of schooling (mean = 70-75). Years of schooling was associated with a predictable level of verbal ability on the WAIS-R that was independent of age, a finding that was supported by analysis of variance (Kaufman et al., 1989). Thus, Verbal IQ, consistently shown to be a maintained ability across much of the adult age range (Horn & Hofer, 1992; Kaufman, 1990), was also shown to be maintained within each of five different levels of educational attainment. Although this finding does not negate the possibility that matching on educational attainment favors a specific group (such as the elderly), it does provide at least some empirical justification for the equating procedure used for the present analysis.
In the present study, it was not possible to conduct analyses of variance or multiple regression analyses, similar to the ones conducted by Kaufman et al. (1989) for the WAIS-R, because data were provided in the form of pertinent means and standard deviations, not separately for individual cases. Nonetheless, effect sizes were computed when possible. Additionally, it was possible to examine age-by-age mean scores on the subtests and scales, both with and without a correction for education, and to relate these aging patterns to those observed on the WAIS-R (Kaufman et al., 1989). It was further possible to compare the age-by-age pattern of mean scores on Verbal IQ to the aging patterns for the two Indexes that are composed only of Verbal subtests; (VCI, WMI); and, similarly, it was possible to compare aging patterns for Performance IQ to patterns for the two Indexes composed only of Performance subtests (POI, PSI). The latter two comparisons enable a more fuller understanding of the age changes on the global IQs from the vantage point of Horn's (1989) expanded Gf-Gc theory of intelligence. In the present study, the main focus is on the IQs, the theoretical constructs measured by the WAIS-III, and changes in aging patterns on the IQs from the WAIS-R to the WAIS-III. In a different study, the focus is on the separate subtests and Indexes, with the emphasis on clinical relevance of the data (Kaufman, 2000).
Mean scaled scores based on the reference group of 20- to 34-year-olds were provided by The Psychological Corporation for each of the 14 WAIS-III subtests, separately by age. Mean reference group sums of scaled scores for the three IQs and four Factor Indexes were then computed by summing pertinent mean scaled scores. These mean sums are presented, by age, in Table 2, alongside means for the WAIS and WAIS-R standardization samples (Wechsler, 1955, 1981) and old-age samples (Doppelt & Wallace, 1955; Ryan et al., 1990). Note that the mean sums of scaled scores on the WAIS-R and WAIS-III differ substantially for ages 16-19, a finding that relates to the questionable validity of the WAIS-R norms for older adolescents (see Kaufman, 1990, 2000). On the Verbal Scale, WAIS-R and WAIS-III mean sums are quite similar for ages 20-34, but the WAIS-III mean sums are larger by about 4 to 8 points at ages 35 to 89. These differences of about 0.4 S.D. (sums of scaled scores on the WAIS-R had an average S.D. of 14.87; see Wechsler, 1981, Table 7) probably reflect changes in educational attainment for the age groups 35-44 and older. As revealed in Table 1, the percentage of college graduates was about the same for comparable age groups in the WAIS-R and WAIS-III standardization samples for ages 20-24, 25-29, and 30-34; however, the percentages were substantially higher for the WAIS-III than WAIS-R for each age group from 35-44 to 85-89. These cohort changes play a key role in the longitudinal analysis conducted in Study 2.
Table 3 presents mean IQs for the WAIS-III at ages 20-24 through ages 85-89 (based on reference group norms, ages 20-34), both with and without an adjustment for educational attainment. Figs. 1 and 2 depict the age-by-age patterns of mean scores, with and without educational adjustments, for Verbal IQ and Performance IQ, respectively. In Fig. 3, education-adjusted means by age group are presented for both WAIS-III Verbal and Performance IQs, to demonstrate the extremely different patterns for these IQs across the adult life span. As indicated in Table 3 and Figs. 1 and 3, when adjusted for education, Verbal IQ peaks at ages 45-54; even for adults in their 80s, mean IQs are in the 96 to 98 range, not appreciably different from the mean IQs of 98-99 for ages 20-29 or the unadjusted IQs of about 95-97 for ages 16-19 years (see Fig. 3). From Table 3 and Figs. 2 and 3, it is evident that education-adjusted Performance IQ peaks in the 20s; mean IQs tend to decrease by about 3 or more IQ points (0.2 S.D.) with each older age group, starting with ages 45-54. Adults in their 80s averaged 76 to 78, more than 20 points (1.33 S.D.) lower than the Performance IQs of adults in their 20s. The different age-by-age patterns observed for Verbal and Performance IQ tend to cancel each other out for ages 20 through 54 years; as seen in Table 3 mean education-adjusted and unadjusted Full Scale IQs are nearly identical (99-100) for each group between 20-24 and 45-54.
Table 3 Mean WAIS-III verbal, performance, and full scale IQs for adults ages 20- 89 (based on sums of scaled scores for the reference group ages 20-34 years) -- both with and without an adjustment for educational attainment Verbal IQ Performance IQ Age group Obt. Ed.-Adj. Diff. Obt. Ed.-Adj. Diff 20-24 97.1 98.1 +1.0 99.4 100.0 +0.6 25-29 99.4 99.5 +0.1 99.9 99.9 0.0 30-34 101.0 100.9 -0.1 98.4 98.4 0.0 35-44 102.4 101.2 -1.2 97.3 96.3 -1.0 45-54 103.8 104.4 +0.6 91.8 92.4 +0.6 55-64 98.9 102.1 +3.2 86.3 89.0 +2.7 65-69 98.2 102.9 +4.7 80.9 84.8 +3.9 70-74 97.1 101.1 +4.0 78.8 82.3 +3.5 75-79 96.0 100.3 +4.3 76.2 79.7 +3.5 80-84 91.4 97.8 +6.4 72.9 78.2 +5.3 85-89 89.8 96.3 +6.5 69.9 75.9 +6.0 Full scale IQ Age group Obt. Ed.-Adj. Diff. 20-24 98.5 99.4 +0.9 25-29 99.9 100.0 +0.1 30-34 99.9 99.9 0.0 35-44 100.1 98.9 -1.2 45-54 99.0 99.7 +0.7 55-64 93.8 97.0 +3.2 65-69 90.2 94.9 +4.7 70-74 88.7 92.8 +4.1 75-79 86.8 91.1 +4.3 80-84 81.0 87.4 +6.4 85-89 78.6 85.3 +6.7 Obt.=Obtained; Ed.-Adj.=Education-Adjusted; Diff.=Difference. Obtained IQs are based on sums of scaled scores for ages 20-34. Education-adjusted IQs minus adjusted to match the educational attainment of adults ages 25-34. The "difference" equals mean education-adjusted IQ minus mean obtained IQ. Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved.
Table 4 shows the effect sizes associated with age-by-age changes on the WAIS-III IQ scales. For this table, the age group earning the highest mean education-adjusted IQ (see Table 3) was designated the "Peak Age." The Peak Age was 45-54 for Verbal IQ, 20-24 for Performance IQ, and 25-29 for Full Scale IQ. Then, the mean education-adjusted IQ for each age group was subtracted from the mean earned by the Peak Age group. That difference was divided by 15 to convert it to S.D. units, an accepted measure of effect size. According to McLean (1995), effect sizes based on S.D. units are small if they are [is less than] 0.50, moderate if they are between 0.50 and 1.00 and large if they are [is greater than] 1.00. As shown in Table 3, only the effect size at ages 85-89 was moderate in magnitude for Verbal IQ, with all other values being small. That finding is consistent with previous findings (Kaufman, 1990, Chapter 7) that Wechsler's Verbal IQ is maintained through the 70s before declining substantially at ages 75 and above. Performance IQ, by contrast, yielded moderate or large effect sizes for ages 45-54 and older, again conforming to a wealth of previous research delineating the strong decrease in Performance IQ with increasing age (Birren & Morrison, 1961; Horn & Hofer, 1992; Kaufman, 1990, Chapter 7; Kaufman et al., 1989).
Table 4 Effect sizes (in S.D. units) obtained by subtracting the education-adjusted mean WAIS-III IQ for each age group from the education-adjusted mean IQ earned by the highest-scoring ("peak") age group Verbal IQ Performance IQ Age group Mean diff. Effect size Mean diff. Effect size 20-24 6.3 0.42 Peak Age 25-29 4.9 0.33 0.1 0.01 30-34 3.5 0.23 1.6 0.11 35-44 3.2 0.21 3.7 0.25 45-54 Peak Age 7.6 0.51(a) 55-64 2.3 0.15 11.0 0.73(a) 65-69 1.5 0.10 15.2 1.01(b) 70-74 3.3 0.22 17.7 1.18(b) 75-79 4.1 0.27 20.3 1.35(b) 80-84 6.6 0.44 21.8 1.45(b) 85-89 8.1 0.54(a) 24.1 1.61(b) Full scale IQ Age group Mean diff. Effect size 20-24 0.6 0.04 25-29 Peak Age 30-34 0.1 0.01 35-44 1.1 0.07 45-54 0.3 0.02 55-64 3.0 0.20 65-69 5.1 0.34 70-74 7.2 0.48 75-79 8.9 0.59(a) 80-84 12.6 0.84(a) 85-89 14.7 0.98(a) Diff. = Difference between the mean education-adjusted IQ earned by the highest-scoring age group ("Peak Age") and the mean education-adjusted IQ eamed by each other age group. Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved. (a) Moderate effect size. (b) Large effect size.
Figs. 4 and 5 present education-adjusted mean IQs for the WAIS-R (Kaufman et al., 1989; Ryan, personal communication, March 1998) and WAIS-III on the Verbal and Performance Scales, respectively. The WAIS-III and WAIS-R results appear to be quite similar for Performance IQ across the adult life span, but fairly different for Verbal IQ. In particular, the peak education-adjusted Verbal IQ was at ages 55-64 for the WAIS-R, but at ages 45-54 for the WAIS-III. The notable similarity in the patterns for Performance IQ is quite interesting in view of the fact that the WAIS-III Performance Scale includes a prototypical measure of Horn's Gf (matrix reasoning) and the WAIS-R Performance Scale does not.
Fig. 6 shows education-adjusted mean standard scores (IQ or Index) for Verbal IQ and the two Factor Indexes composed only of verbal subtests (VCI and WMI). Analogously, Fig. 7 presents data for Performance IQ, POI, and PSI. In Fig. 6, the pattern for Verbal IQ is quite similar to the pattern for VCI, as both are maintained abilities. In contrast, WMI presents a vulnerable pattern that is not as extreme as the pattern for Performance IQ, but, nonetheless, reveals a peak in the 20s and decrements through old age. Mean Verbal IQs at each age are a bit lower than the mean Indexes for VCI, but otherwise, the influence of WMI is minimal in affecting the shape of the graph of Verbal IQ relative to VCI. Fig. 7 offers some insight into the vulnerable nature of Performance IQ. Performance IQ resembles closely POI during its period of growth (ages 16-44), but is more similar to PSI during its rapid decline (especially ages 55-74). The implication is that some of the precipitous decline observed for Performance IQ is due to the Gs or speeded component of Performance tasks, more so than the problem-solving (Gf) or visualization (Gv) components. The close association of perceptual speed with cognitive decline is consistent with the finding that sensorimotor variables are quite predictive of intellectual functioning in old age (70-103 years) (Lindenberger & Baltes, 1997). Furthermore, sensory variables (with no motor involvement) may be related to declines on all three Horn factors (Gf, Gv, and Gs) in view of compelling evidence that sensory variables correlate substantially with intellectual variables (Baltes & Lindenberger, 1997; Lindenberger & Baltes, 1994, 1997). In one large-scale study, visual and auditory acuity correlated substantially higher with general intelligence in old age (70-103) than during the earlier adult years (25-69), and fluid abilities correlated substantially with sensory functioning in both age groups (Baltes & Lindenberger, 1997). All Wechsler Performance subtests are heavily dependent on visual acuity for interpreting the pictures and designs, and on auditory acuity for understanding the examiner's instructions. These sensory variables have generally been overlooked when trying to account for declines in Performance IQ across the entire adult life span.
3. Study 2 -- longitudinal analysis of WAIS-R and WAIS-III
The WAIS-R (Wechsler, 1981) and the WAIS-III (The Psychological Corporation, 1997; Wechsler, 1997) were used for this study. Though the scales differ in several key ways, notably the inclusion of three new subtests and the organization of the WAIS-III into four Factor Indexes as well as two IQ scales, these changes do not affect the present study materially. The IQ scales are the only scores analyzed in this longitudinal investigation, so the only modification of interest is the substitution of the new matrix reasoning subtest, a prototypical measure of fluid reasoning, for object assembly when computing Performance IQ and Full Scale IQ. Despite this substitution, WAIS-R and WAIS-III Performance IQs correlated .86 in a counterbalanced study of 192 adults, ages 16-74 years, tested twice (interval 2-12 weeks) (The Psychological Corporation, 1997, pp. 78-79). For the WAIS and WAIS-R Performance Scales (composed of identical subtests) the IQs correlated .84, on the average, based on eight samples totaling 420 individuals (Ryan, Nowak, & Geisser, 1987, Table 2). The WAIS-R and WAIS-III Verbal and Full Scales demonstrated even greater overlap than did the Performance Scale: Verbal IQs correlated .94 and Full Scale IQs correlated .93 for the 192 individuals tested twice on both instruments (The Psychological Corporation, 1997). For the WAIS vs. WAIS-R, Verbal and Full Scale IQs each correlated .92 (Ryan et al., 1987).
Psychometric properties for the WAIS-III were given in Section 2.1. For the WAIS-R, internal consistency coefficients for ages 20-74, the only ages analyzed in this study, averaged .97 for Verbal IQ, .93 for Performance IQ, and .97 for Full Scale IQ (Wechsler, 1981, Table 10). Stability coefficients were .94-.97 for Verbal and Full Scale IQs, and .89-.90 for Performance IQ, for two adult samples (total N = 119) tested twice with about a 1-month interval (Wechsler, 1981, Table 11). Factor analysis produced either two or three factors, but regardless of the number of factors, the two largest dimensions corresponded to the Verbal and Performance Scales, offering solid construct validity for the separate IQ scales (Kaufman, 1990, Chapter 8).
The standardization samples for the WAIS-R and WAIS-III provided the data source for Study 2. Ages 16-19 were eliminated from the WAIS-R because these normative data are of questionable validity; mean scores are virtually identical for ages 16-17 and 18-19 despite the higher educational attainment of the latter group, and both groups scored unusually lower (about 0.5 S.D.) than young adults of 20-24 years (Kaufman, 1983, 1990, Chapter 3). The seven age groups between 20-24 and 70-74 comprised the WAIS-R sample for Study 2. The age groups included the following numbers of adults, equally divided among males and females for each subsample: 20-24 (n = 200), 25-34 (n = 300), 35-44 (n = 250), 45-54, (n = 250), 55-64 (n = 160), 65-69 (n = 160), and 70-74 (n = 160). The total sample for the study was 1480, stratified on the variables of race, geographic region, occupation, education, and urban-rural residence, in addition to age and gender (Wechsler, 1981, Chapter 3). During the collection of WAIS-R standardization data, potential subjects were excluded for the following reasons (Wechsler, 1981, p. 18): institutionalized "mental defectives;" individuals with known brain damage, or with severe behavioral or emotional problems; and individuals with physical defects that would hinder their ability to respond to test items. These exclusionary criteria are less restrictive than the criteria for the WAIS-III, which, for example, excluded adults with alcohol problems and those who take antidepressants, as described previously. Nonetheless, the WAIS-R criteria are consistent with one key intent of the WAIS-III criteria, namely to exclude anyone with neurological or neuropsychological disorders. The result of that decision is to eliminate many of the poorer functioning adults, with larger numbers of adults eliminated for older than younger age groups. Therefore, for WAIS-R and WAIS-III age-related declines, any observed declines are likely to be lower-bound estimates of the true declines that characterize the actual population of older adults in society. For the purpose of the longitudinal study conducted in Study 2, however, the main exclusionary criteria for the WAIS-R and WAIS-III (i.e., those that are likely to differ for older versus younger age groups) are reasonably similar, permitting meaningful comparisons of the age cohorts over time.
For the WAIS-III, the 11 age groups between 20-24 and 85-89 comprised the sample for Study 2, yielding a total N of 2050. Each of nine age groups between 20-24 and 75-79 comprised 200 adults; there were 150 adults at ages 80-84 and 100 adults at ages 85-89 (The Psychological Corporation, 1997, Chapter 2). The standardization sample was described previously for Study 1.
A longitudinal analysis with independent samples was conducted using normative data from the WAIS-R, standardized between 1976 and 1980, and the WAIS-III, standardized in 1995-1996. This study evaluated change in intelligence over a 17-year interval -- between 1978 (the midpoint of the WAIS-R years of standardization) and 1995 -- "because most of the members of the [WAIS-III] standardization sample had been selected and tested by the end of that year" (Flynn, 1998, p. 1234).
This study uses a methodology suggested by Parker (1986), who evaluated changes in WAIS/WAIS-R Full Scale IQ for selected age cohorts, and applied by Kaufman (1990, pp. 212-222) to investigate changes in the separate WAIS/WAIS-R Verbal and Performance IQs for four age cohorts. The methodology involves identifying cohorts who were tested at two points in time, in this case in 1978 and 1995. The use of independent samples ("cohort substitution" according to Kausler, 1982), when the samples are random and comparable, makes it "possible to compute age change estimates that are controlled for the effects of testing and experimental mortality" (Schaie, 1983, p. 106). Experimental mortality in longitudinal studies is not random and has been referred to as "selective attrition," a problem that has plagued longitudinal IQ studies that attempt to assess the same sample of individuals over time (Kaufman, 1990, pp. 206-208; Siegler & Botwinick, 1979).
Seven cohorts were identified for analysis in this study, as summarized in Table 5. As shown, these cohorts range from individuals born between 1954 and 1958, who were tested at age 22 and again at 39, to adults born between 1904 and 1908 (who were 72 during the WAIS-R standardization and 89 when the WAIS-III was normed). Additional cohorts were feasible (1959-1960 and 1961-1962), but were excluded because of the questionable validity of WAIS-R norms for ages 16-17 and 18-19, discussed previously.
Table 5 The seven adult cohorts represented in the WAIS-R and WAIS-III standardization samples Age in 1978 Age in 1995 Cohort (WAIS-R (WAIS-III (year of birth) standardization) standardization) 1954-1958 22 (20-24) 39 (37-41) 1944-1953 29.5 (25-34) 46.5 (42-51) 1934-1943 39.5 (35-44) 56.5 (52-61) 1924-1933 49.5 (45-54) 66.5 (62-71) 1914-1923 59.5 (55-64) 76.5 (72-81) 1909-1913 67 (65-69) 84 (82-86) 1904-1908 72 (70-74) 89 (87-91) Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved.
The first requirement for this longitudinal study is to demonstrate that the independent samples of adults from each cohort are reasonably comparable on important background variables. The independent samples within each cohort are matched adequately on gender and geographic region (The Psychological Corporation, 1997, Chapter 2; Wechsler, 1981, Chapter 3). However, these two variables do not produce substantial individual differences on Wechsler's IQ (Kaufman, 1990, Chapter 6) and are of minimal consequence; the key variables that yield large individual differences in Wechsler IQ are socioeconomic status (notably educational attainment) and race/ethnicity. Table 6 reports percentages of individuals within each cohort at two points in time (1978 and 1995) with different educational backgrounds. These data suggest that the independent samples within each cohort have similar degrees of educational attainment. Where there are differences, they uniformly reflect greater education for the 1995 cohort, a change that is consistent with the notion that -- over time -- adults who were dropouts from high school will tend to earn GED degrees and that high school graduates will enroll for a year or more in a 2-year or 4-year college. The largest increment in education is for the 1954-1958 cohort, for whom 56.5% had one or more years of college in 1995 versus 39.5% in 1978. This cohort was the youngest in 1978 (ages 20-24) and is the most likely to have gone to college in the ensuing 17 years. In general, Table 6 indicates a satisfactory match of each pair of independent samples on educational attainment.
Table 6 Comparison of the educational level of independent samples from the same seven cohorts tested in 1978 (WAIS-R standardization sample) and 1995 (WAIS-III standardization sample) Percentage of adults with 0-11 years 12 years education education 1+ years college Cohort 1978 1995 1978 1995 1978 1995 (year of birth) sample sample sample sample sample sample 1954-1958 17.0 9.5 43.5 34.0 39.5 56.5 1944-1953 17.0 15.1 39.3 33.7 43.7 51.2 1934-1943 26.0 23.8 42.4 36.6 31.6 39.6 1924-1933 34.4 32.0 40.0 38.0 25.6 30.0 1914-1923 43.7 34.0 36.9 37.0 19.4 29.0 1909-1913 55.0 49.6 25.6 28.0 19.3 22.4 1904-1908 60.0 50.0 23.8 27.0 16.2 23.0 Percentages for the WAIS-R standardization sample are from Wechsler (1981, Table 5). Percentages for the WAIS-III standardization sample are from The Psychological Corporation (1997, Table 2.6). WAIS-III percentages for the 1954-1958, 1924-1933, 1914-1923, and 1904-1908 cohorts are based on age groups 35-44, 65-69, 75-79, and 85-89, respectively. WAIS-III percentages for the 1944-1953 cohort were interpolated from values for ages 35-44 and 45-54; WAIS-III percentages for the 1934-1943 cohort were interpolated from values for ages 45-54 and 55-64; and WAIS-III percentages for the 1909-1913 cohort were interpolated from values for ages 80-84 and 85-89.
On the variable of race/ethnicity, the WAIS-R used a white--nonwhite distinction whereas the WAIS-III matched U.S. Census percentages for Whites, African Americans, Hispanics, and "Others" (e.g., Asian American, Native American). The WAIS-R manual does not indicate how "nonwhite" was defined or how Hispanics and "Others" were classified. The most pertinent comparison is percent of Whites in each cohort. For the three youngest cohorts, the percent of Whites was about 88 in 1978 and about 78-79 in 1995; for cohort 1924-1933, the corresponding percents were 89 and 85-86, respectively; and for the three oldest cohorts, the percents were about 91 and 90, respectively (The Psychological Corporation, 1997, Chapter 2; Wechsler, 1981, Chapter 3). These comparisons indicate an adequate match on race/ethnicity, indicating that the independent samples within each cohort appear comparable on the variables of gender, geographic region, educational attainment, and race/ethnicity.
The next step is to control for the instrument, since the WAIS-R was administered in 1978 and the WAIS-III in 1995. This control is straightforward. The counterbalanced study with N = 192 described previously indicated that, consistent with Flynn's (1984, 1987, 1998) evidence of generational changes in IQ test performance, the WAIS-III yields lower IQs than the WAIS-R. Differences were 1.2 points on the Verbal Scale, 4.8 points on the Performance Scale, and 2.9 points on the Full Scale (The Psychological Corporation, 1997, Table 4.1). To equate for instrument, constants of 1.2, 4.8, and 2.9 were added to mean WAIS-III Verbal, Performance, and Full Scale IQs, respectively. The coefficients of .86 to .94 between WAIS-R and WAIS-III IQs of the same name attest to the continuity of measurement from the WAIS-R to its successor, suggesting "that the WAIS-III measures essentially the same constructs as does the WAIS-R" (The Psychological Corporation, 1997, p. 78).
Finally, it is necessary to control for cultural change or "time lag" to ensure that any observable changes over the 17-year interval for each cohort are truly due to development and not to other factors. For example, suppose that every adult, due to cultural factors, gained 7 IQ points during the 17-year interval. If one cohort gained 10 IQ points during that same interval, one would be unable to attribute the 10 points to intellectual growth. First the 7 points that everyone gained would have to be subtracted, before concluding that only 3 of the IQ points reflected actual growth. To control for time lag, the method used by Owens (1966) in his longitudinal study of the Army Alpha, and by Kaufman (1990, pp. 212-222) in his WAIS/WAIS-R longitudinal study, was applied to the present data. That is to say, time lag was computed by comparing the IQs earned by each cohort in 1978 with the IQs earned by adults the same age in 1995. (The method for converting the sums of scaled scores in Table 2 to IQs is described in the two paragraphs that follow.) For example, the youngest cohort (1954-1958) was 20-24 in 1978. To correct for time lag, their IQs were compared to adults ages 20-24 in 1995. These comparisons are shown in Table 7 for all seven cohorts (WAIS-III IQs were first converted to be on the same metric as WAIS-R IQs by adding the indicated constants). The time lags were smallest for the two youngest cohorts, averaging about 1 point for Verbal IQ, 4.5 points for Performance IQ, and 2.5 points for Full Scale IQ. For the older five cohorts, time lags averaged about 6 to 9 points for each IQ. The smaller time lags for the youngest two cohorts is undoubtedly related to the fact that the educational backgrounds for ages 20-24 and 25-34 were very similar for both the WAIS-R and WAIS-III normative samples (Table 1 shows that about 10% of 20-24-year-olds, and about 23.5% of 25-34-year-olds graduated college in both sample). In contrast, the larger time lags for the older five cohorts are consistent with the higher educational attainment achieved by comparable age groups for the WAIS-III versus WAIS-R samples (see Table 1).
Table 7 Time-lag effects: comparison of mean Wechsler IQs earned by the seven cohorts in 1978 with mean Wechsler IQs earned in 1995 by adults matched with each cohort on chronological age Verbal IQ Performance IQ Age 1978 1995 Time-lag 1978 1995 Time-lag group mean mean effect mean mean effect 20-24 97.6 98.3 +0.7 100.2 104.2 +4.0 25-34 100.4 101.4 +1.0 98.9 104.0 +5.1 35-44 96.5 103.6 +7.1 92.6 102.1 +9.5 45-54 97.1 105.0 +7.9 89.1 96.6 +7.5 55-64 94.6 100.1 +5.5 84.2 91.1 +6.9 65-69 92.5 99.4 +6.9 78.8 85.7 +6.9 70-74 91.0 98.3 +7.3 75.6 83.6 +8.0 Full scale IQ Age 1978 1995 Time-lag group mean mean effect 20-24 99.2 101.4 +2.2 25-34 99.9 102.8 +2.9 35-44 95.1 103.0 +7.9 45-54 93.8 101.9 +8.1 55-64 89.6 96.7 +7.1 65-69 86.3 93.1 +6.8 70-74 83.1 91.6 +8.5 WAIS-R IQs (1978 means) were obtained by entering WAIS-R sum of scaled scores on each scale (based on reference group norms, ages 20-34) into the IQ conversion tables for all ages for the WAIS-III (Wechsler, 1997, Tables A.3-A.5). WAIS-III IQs (1995 means) were obtained by entering WAIS-III sum of scaled scores on each scale (based on reference group norms, ages 20-34) into the IQ conversion tables for all ages for the WAIS-III (Wechsler, 1997, Tables A.3-A.5); then WAIS-III IQs were converted to estimated WAIS-R IQs by adding 1.2 points to each mean Verbal IQ, 4.8 points to each Performance IQ, and 2.9 points to each Full Scale IQ. Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved.
After controlling for instrument and time lag, it was possible to compare the test performance of each cohort in 1978 and 1995 to determine any growth or decline in intellectual function. Sums of scaled score on each IQ scale, based on the appropriate reference group of 20-24-year-olds, are shown in Table 2 for the WAIS-R and WAIS-III. These sums provide the relevant units for evaluating growth or decline. However, sums of scaled scores are not a convenient unit for comparison because the three scales have different means for the reference samples: Verbal = 60, Performance = 50, Full Scale = 110. To permit easy comparison of gain or loss on each scale and to use a common metric, it was decided to convert each mean sum of scaled score to an IQ. The most pertinent conversion table was the one provided in the WAIS-III manual because it is designed for all adult ages (as opposed to WAIS-R IQ conversion tables which are separate for each age group). Consequently, pertinent mean sums of scaled scores for both the WAIS-R and WAIS-III were entered into the WAIS-III IQ conversion tables (Wechsler, 1997, Tables A.3-A.5), with interpolations performed as necessary, to produce the set of IQs from 1978 and 1995; because they are based on a common metric, they are directly comparable (once they are controlled for instrument and time lag) to evaluate intellectual change over time.
For the WAIS-R, the sums of scaled scores for each group between 20-24 and 70-74 correspond directly to the ages of the seven cohorts in 1978 (with the one exception that data for ages 25-29 and 30-.34 had to be merged). For the WAIS-III, there were not direct matches between the standardization age groups shown in Table 2 and the age ranges of the cohorts in 1995. As indicated in Table 5, the mean age of the youngest cohort was 39 (range of 37-41), the mean age of the next cohort was 46.5 (range of 42-51), and so forth. Some ages in 1995 were very close to the midpoint ages of standardization age groups. Age 39 for cohort 1954-1958 was about the same as the midpoint of 39.5 for ages 35-44, so the mean sums of scaled scores for age group 35-44 (see Table 2) were entered into the WAIS-III IQ conversion tables to yield the 1995 mean IQs for the youngest cohort. Similarly, the means for ages 65-69 (midpoint=67) were used for the 1924-1933 cohort (age 67.5) and the means for ages 75-79 (midpoint = 77) were used for the 1914-1923 cohort (age 77.5). Mean WAIS-III sums of scaled scores from Table 2 were interpolated from pertinent adjacent age groups for cohorts 1944-1953, 1934-1943, and 1924-1933. For the oldest cohort, midpoint= 89 years in 1995, mean IQs were extrapolated from data for the oldest three normative age groups (midpoint ages of 77, 82, and 87). As explained previously, all mean WAIS-III IQs obtained for cohorts in 1995 were subsequently adjusted by adding 1.2 points to Verbal IQ, 4.8 points to Performance IQ, and 2.9 points to Full Scale IQ, to make them comparable to the 1978 WAIS-R IQs.
Table 8 presents IQ changes from 1978 to 1995 for each of the seven cohorts, adjusted for instrument and time lag. Table 9 provides effect sizes (based on S.D. = 15) for each adjusted change. Except for the two youngest cohorts, IQ loss characterized each cohort on each scale. Effect sizes are small for Verbal IQ except for the oldest cohort, with the effect size for the next-oldest cohort falling just short of a moderate effect size. By contrast, the effect sizes for Performance IQ are moderate, and similar in value, for the five oldest cohorts. As adults born before 1944 aged over the 17-year period, they lost 10 points ([+ or -] 1) of IQ, or about 2/3 of a S.D., whether they were about 60, 70, 80, or 90 years of age in 1995;. Results for the Full Scale IQ reflect the approximate midpoint of the Verbal and Performance IQ changes for the five youngest cohorts. For the oldest two cohorts, however, Full Scale IQ losses were larger than either separate loss depicting the more generalized loss of function in both verbal and nonverbal domains that occurs during the decade of the 80s.
Table 8 Mean change in WAIS-R/WAIS-III IQs for seven cohorts over a 17-year span (1978 to 1995), controlling for instrument and time-lag effects Verbal IQ Cohort 1978 1995 Adj. (year of birth) mean mean Change change 1954-1958 97.6 103.6 +7.0 +6.3 (age 22 to 39) 1944-1953 100.4 104.6 +4.2 +3.2 (age 29.5 to 46.5) 1934-1943 96.5 102.6 +6.1 -1.0 (age 39.5 to 56.5) 1924-1933 97.1 99.4 +2.3 -5.6 (age 49.5 to 66.5) 1914-1923 94.6 97.2 +2.6 -2.9 (age 59.5 to 76.5) 1909-1913 92.5 92.0 -0.5 -7.4 (age 67 to 84) 1904-1908 91.0 90.0 -1.0 -8.3 (age 72 to 89) Median adj. change -2.9 Mean adj. change -2.2 Performance IQ Cohort 1978 1995 Adj. (year of birth) mean mean Change change 1954-1958 100.2 102.1 +1.9 -2.1 (age 22 to 39) 1944-1953 98.9 98.2 -0.7 -5.8 (age 29.5 to 46.5) 1934-1943 92.6 92.8 +0.2 -9.3 (age 39.5 to 56.5) 1924-1933 89.1 85.7 -3.4 -10.9 (age 49.5 to 66.5) 1914-1923 84.2 81.0 -3.2 -10.1 (age 59.5 to 76.5) 1909-1913 78.8 76.5 -2.3 -9.2 (age 67 to 84) 1904-1908 75.6 73.5 -2.1 -10.1 (age 72 to 89) Median adj. change -9.3 Mean adj. change -8.2 Full scale IQ Cohort 1978 1995 Adj. (year of birth) mean mean Change change 1954-1958 99.2 103.0 +3.8 +1.6 (age 22 to 39) 1944-1953 99.9 102.2 +2.3 -0.6 (age 29.5 to 46.5) 1934-1943 95.1 98.3 +2.3 -4.7 (age 39.5 to 56.5) 1924-1933 93.8 93.1 +2.3 -8.8 (age 49.5 to 66.5) 1914-1923 89.6 89.7 +2.3 -7.0 (age 59.5 to 76.5) 1909-1913 86.3 82.9 +2.3 -10.2 (age 67 to 84) 1904-1908 83.1 80.2 +2.3 -11.4 (age 72 to 89) Median adj. change -7.0 Mean adj. change -5.9 Adj. = Adjusted. WAIS-III IQs are adjusted for instrument. The Adjusted change reflects an additional adjustment for time lag. Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved. Table 9 Effect sizes (in S.D. units) of the mean change in WAIS-R/WAIS-III IQs for seven cohorts over a 17-year span (1978 to 1995), controlling for instrument and time-lag effects Verbal IQ Performance IQ Cohort Adj. Effect Adj. Effect (year of birth) change size change size 1954-1958 +6.3 .42 -2.1 .14 (age 22 to 39) 1944-1953 +3.2 .21 -5.8 .39 (age 29.5 to 46.5) 1934-1943 -1.0 .07 -9.3 .62(a) (age 39.5 to 56.5) 1924-1933 -5.6 .37 -10.9 .73(a) (age 49.5 to 66.5) 1914-1923 -2.9 .19 -10.1 .67(a) (age 59.5 to 76.5) 1909-1913 -7.4 .49 -9.2 .61(a) (age 67 to 84) 1904-1908 -8.3 .55(a) 10.1 .67(a) (age 72 to 89) Median effect size .37 .62(a) Mean effect size .33 .55(a) Full scale IQ Cohort Adj. Effect (year of birth) change size 1954-1958 +1.6 .11 (age 22 to 39) 1944-1953 -0.6 .04 (age 29.5 to 46.5) 1934-1943 -4.7 .31 (age 39.5 to 56.5) 1924-1933 -8.8 .59(a) (age 49.5 to 66.5) 1914-1923 -7.0 .47 (age 59.5 to 76.5) 1909-1913 -10.2 .68(a) (age 67 to 84) 1904-1908 -11.4 .76(a) (age 72 to 89) Median effect size .47 Mean effect size .42 Adj. = Adjusted. Standardization data of the Wechsler Adult Intelligence Scale: Third Edition. Copyright [C] 1997 by The Psychological Corporation. Used by permission. All rights reserved. (a) Moderate effect size.
4. Discussion -- Studies 1 and 2
Both of the studies described in this paper addressed key theoretical and practical issues regarding aging and intelligence. Study 1 examined cross-sectional data on the WAIS-III for 13 age groups between 16-17 and 85-89 years, controlling for the important cohort variable of educational attainment for ages 20-89. Study 2 followed seven cohorts longitudinally over a 17-year period on Wechsler's adult scales, using comparable independent samples; the ages of these cohorts averaged 22 to 72 years in 1978 and 39 to 89 years in 1995.
4.1. Cross-sectional vs. longitudinal data
The results of Studies 1 and 2 are remarkably consistent with each other. In both investigations, Verbal IQ emerged as a maintained ability and Performance IQ as a vulnerable ability. For both studies, peak performance occurred at ages 45-54 on the Verbal IQ and at ages 20-24 on Performance IQ (Tables 3 and 7). The only notably lower mean Verbal IQs for any of the age groups occurred for adults in their 80s. Table 4 presents effect sizes for the cross-sectional age groups and Table 9 presents effect sizes for the longitudinal cohorts. Only one effect size for Verbal IQ was large enough to be considered "moderate" in magnitude for each study: Ages 85-89 scored 8.1 education-adjusted IQ points (0.54 S.D.) lower than the peak cross-sectional age group (45-54 years); and the 1904-1908 cohort, which aged from 72 to 89 years, lost 8.3 IQ points (0.55 S.D.) over the 17-year interval. In addition, effect sizes in the .40s -- small but close to moderate -- were observed for ages 80-84 in the cross-sectional study and for the 1909-1913 cohort (ages 67 to 84). Meaningful declines in Verbal IQ, and in Gc abilities in general, though usually not evident for adults in their 70s, have been observed in previous investigations for individuals who are about 80 years of age or older (Schmitz-Scherzer & Thomae, 1983), including Schaie's (1983) cross-sequential studies with the PMA (Kaufman, 1990, Fig. 7.7).
The agreement in the results of Studies 1 and 2 are just as striking for Performance IQ. The steady decline in mean education-adjusted Performance IQs in the cross-sectional investigation is evident in Table 3; mean IQs (based on the 20-34 year reference group) are close to 100 for ages 20-34, about 90-95 for ages 35-64, about 80-85 for ages 65-79, and in the upper 70s for ages 80-89. The steady decline is also evident in the longitudinal investigation (Table 8) as the mean Performance IQ for each cohort drops steadily from the 90s to the 80s to the 70s with increasing age. Effect sizes are shown in Tables 4 and 9 for the two studies. When compared to the peak age of 20-24 in the cross-sectional investigation, the education-adjusted mean Performance IQs demonstrate a slight, nonmeaningful decline through age group 35-44 as effect sizes are small (Table 4). Similarly, in the longitudinal investigation, small effect sizes characterize the small declines from ages 22 to 39 and from 29.5 to 46.5 (Table 9). In the cross-sectional study, moderate effect sizes are observed for ages 45-64 and large effect sizes, in excess of 1 S.D., are seen for ages 65-89 (Table 4). Similarly, moderate effect sizes characterize the loss in IQ points for the five oldest cohorts in the longitudinal study (Table 9). The only reason that large effect sizes are found in the cross-sectional study, but not the longitudinal study, concerns the fact that the older age groups in the former study were all compared to young adults of 20-24 (a difference of 60 or more years for the two oldest age groups), whereas Performance IQ losses were limited to 17-year intervals for all cohorts in the longitudinal study.
The five oldest cohorts in the longitudinal study lost 9.2 to 10.9 Performance IQ points (mean loss = 9.9 IQ points), when corrected for instrument and time lag (Table 8), an average of 0.58 IQ points per year. The mean difference in education-adjusted Performance IQ for ages 35-44 (comparable to the age of the 1934-1943 cohort in 1978) versus ages 85-89 is 20.4 IQ points (Table 3); when divided by the difference in their mean ages (47.5), the average loss is 0.43 IQ points per year. Taken together, the results of both studies suggest a loss of about 0.5 points per year in Performance IQ, or 5 points per decade -- which is precisely Horn's (1985) and Horn, Donaldson, and Engstrom's (1981) best estimate of the adult decline in Gf based on their reviews and results of numerous studies.
The findings of maintained Gc abilities and vulnerable Gf abilities -- including the approximate magnitude of the declines in Gf ability with increasing age, and the ultimate decline in Gc in very old age -- are also consistent with age-related changes in intelligence observed in well-designed, large-scale European studies, such as Rabbitt's (11993) investigation of more than 6000 adults ages 50-96 in the United Kingdom and Baltes and Lindenberger's (1997) study of 687 adults ages 25-103 in Germany.
In Kaufman's (1990, pp. 212-222) longitudinal study of the WAIS and WAIS-R for four cohorts over a 25-year span (1953 to 1978), analogous to Study 2 for the WAIS-R and WAIS-III, the four cohorts lost 1.5 to 5.5 Verbal IQ points (small effect sizes of 0.10-0.37 S.D.). These results parallel the results of Study 2 for individuals in the five youngest cohorts (small effect sizes of 0.07-0.42 S.D.; see Table 91), which are the only comparable cohorts; the oldest cohort in Kaufman's study aged from 47 to 72. The more substantial Verbal IQ losses for adults in their 80s in Study 2 augment the previous findings for the WAIS/WAIS-R, but do not contradict it.
On the Performance Scale, Kaufman's (1990) four cohorts lost 11.6 to 13.5 IQ points, corresponding to moderate effect sizes of 0.77-0.90 S.D. Again, these findings are consistent with the results of Study 2: Each cohort lost a constant number of IQ points over time. However, the moderate effect size for the youngest cohort in Kaufman's study (which aged from 24.5 to 49.5 years) was not found in Study 2; the cohort that corresponds most closely to Kaufman's youngest cohort is the 1944-1953 cohort in Study 2 (ages 29.5 to 46.5), and that cohort lost only about 6 Performance IQ points (small effect size). For the oldest three cohorts in Kaufman's study, which lost an average of 13.0 Performance IQ points, the results conform quite closely to the Study 2 findings. The loss is larger in the prior study (13 vs. 10 points), but so is the interval (25 vs. 17 years). In Kaufman's study, the decline in Performance IQ equates to 0.52 IQ points per year, quite similar to the value of 0.58 observed in Study 2.
Fig. 8 offers additional evidence of the similarity in the results of Studies 1 and 2 regarding the decline in Performance IQ during the adult life span. The mean education-adjusted WAIS-III Performance IQs shown in Table 3 for ages 20-24 through 85-89 are plotted to depict the cross-sectional results, and the mean Wechsler (WAIS-R/WAIS-III) IQs included in Table 8 are plotted to depict the longitudinal results (1978 and 1995 means are both plotted, except that the average values are entered for age groups that are closely similar -- e.g., ages 45-54 in 1978 and ages 42-51 in 1995). As seen in Fig. 8, the graphs overlap remarkably for most ages (age 72 is an exception). The same degree of overlap characterized age differences in Performance IQ in Kaufman's longitudinal study when plotted against the education-adjusted mean Performance IQs in Kaufman et al.'s (1989) cross-sectional study of the WAIS-R (Kaufman, 1990, Fig. 7.9). Note, however, that Fig. 8 is only offered for comparison purposes. It is not appropriate to evaluate age changes with cross-sequential data (Schaie & Hertzog, 1986), and no developmental inferences should be inferred from Fig. 8, a point emphasized by Kaufman (1990) regarding his similar graph. Indeed, the primary developmental inferences from the present pair of studies come from the 17-year cohort comparisons in Study 2. The 10-point Performance IQ decrements for the five oldest cohorts reflect a fairly constant loss of nonverbal intellectual function over a 17-year period for individuals born between 1904 and 1943, and the 7- to 8-point Verbal IQ losses for the two oldest cohorts (those aging from 67 to 84 and from 72 to 89) denote diminishing verbal intellectual function for adults in their 80s.
The substantial decline in Verbal IQ for the oldest portion of the WAIS-III sample, and the steady and often striking decline in WAIS-III Performance IQ across most of the adult life span, are remarkable in view of the systematic exclusion of low-functioning adults from the normative sample. As noted previously, not only were adults with known dementia or brain damage excluded from the sample, but so were adults who merely went to a doctor or other professional for memory problems or problems with thinking. The actual adults tested for each subsample of older individuals (e.g., 65-69, 70-74), though stratified with care on key variables such as educational attainment and race/ethnicity, are undoubtedly higher functioning than their respective age groups within society. Furthermore, thinking and memory problems, and cognitive and emotional pathologies in general, occur more frequently for very old versus moderately old samples (Rabbitt et al., 1997). Therefore, all of the declines in mean Wechsler IQ from age group to age group among the older subsamples are most likely underestimates of the "true" declines that would be observed within society, were it feasible to test stratified random samples of adults without regard to their known or possible neurological disorders. These lower-bound estimates of observed declines are especially important to keep in mind for Study 1, the cross-sectional analysis. The exclusion of low-functioning subjects is less of a factor in Study 2, because both the WAIS-R and WAIS-III standardization samples excluded a wide variety of low-functioning adults. Therefore, the comparisons within each cohort over time are for groups that are roughly comparable in their exclusion of subjects, canceling out, to some extent, this variable. The best argument that the exclusion criteria did not have a large effect on the results of the present studies is the close similarity in the results of Studies 1 (cross-sectional) and 2 (longitudinal) for both Verbal IQ and Performance IQ. However, note that the average decline in Performance IQ from the 40s to the 80s was 0.43 IQ point in Study 1 and 0.58 IQ point in Study 2. The smaller decline in the cross-sectional than the longitudinal study may be the direct result of excluding low-functioning subjects from the WAIS-III normative group. That hypothesis needs to be checked out in future research.
4.2. Generational changes
Fig. 4 (Verbal IQ) and Fig. 5 (Performance IQ) compare the results of cross-sectional analyses of the WAIS-R (Kaufman et al., 1989; Ryan, personal communication, March 1998) and WAIS-III (Study 1), adjusting IQs for educational attainment. The WAIS-III and WAIS-R results are quite congruent for PerFormance IQ across the 20-89 year life span, but fairly different for Verbal IQ. The results for Performance IQ, when coupled with the similarities just noted for the WAIS/WAIS-R and WAIS-R/WAIS-III longitudinal studies and the congruence with Horn's (1985) predictions about IQ changes in Gf over the adult life span, suggest a continuity in the observed decline in Performance IQ with increasing age. Current generations seem no different from previous generations. Whereas data from the KAIT (Kaufman & Kaufman, 1993) and K-BIT (Kaufman & Kaufman, 1990) indicated a plateau from the mid-20s to mid-50s in Fluid IQ, and then a dramatic drop from the mid-50s through old age (Kaufman & Horn, 1,996; Wang & Kaufman, 1993), that pattern was not observed for the WAIS-III. Even though other measures of fluid ability, such as those included on the Kaufman tests, may display a plateau during middle age, such a plateau is not seen on the WAIS-III Performance Scale. The difference seems to be one of instrumentation, not generation.
That is not the case for Verbal IQ, however. The two graphs depicted in Fig. 4 do not overlap and have different shapes. The lack of overlap is probably not important, and relates primarily to the use of different IQ conversion tables. For the WAIS-R, the table for ages 25-34 was used to obtain IQs by Kaufman et al. (1989); for Study 1, the WAIS-III conversion table for all ages between 16 and 89 was used (indeed, there are no WAIS-III IQ conversion tables for separate age groups). The different shapes of the Verbal IQ graphs is important. Peak performance on the WAIS-R was for ages 55-64, and peaks in the 60s have been reported by Horn (1989) and colleagues (Horn et al., 1981; Horn & Hofer, 1992; Horn & Noll, 1997). However, an earlier peak is noted for the WAIS-III, a decisive peak at ages 45-54.
There are a few reasons to believe that the present WAIS-III result reflects a generational change. Horn's conclusions are based on WAIS and WAIS-R Verbal IQ and on studies of other measures of Gc, where the data were uniformly collected prior to 1980. More recent data, however, dispute the peaking of verbal skills in the 60s. The clear-cut peak at ages 45-54 on the WAIS-III Verbal IQ is consistent with the results of Gc measures on tests normed very recently. Education-adjusted mean Crystallized IQs on the KAIT peaked at ages 40-44 (Kaufman & Horn, 1996); education-adjusted mean standard scores on the functional reading subtest of the Kaufman Functional Academic Skills Test (K-FAST; Kaufman & Kaufman, 1994) peaked at ages 40-44 (Kaufman et al., 1996); and Vocabulary standard scores on the K-BIT peaked at ages 40-49 (Wang & Kaufman, 1993). These peaks are a bit younger than the peak observed for WAIS-III Verbal IQ, perhaps because of the fluid reasoning component that was deliberately included in the KAIT and K-BIT crystallized subtests (Kaufman & Horn, 1996). The earlier peaks for KAIT and K-BIT Gc measures are, nonetheless, consistent with the results of the present WAIS-III study. There is an apparent generational shift toward an earlier peak in Gc ability, indicating that this maintained ability is not maintained as long as it used to be. Certainly, there is a notable difference in the educational attainment of many age groups when comparing percentages for the WAIS-R versus the WAIS-III (Table 1). On the WAIS-R, the most educated group was ages 25-34 (43.7% with one or more years of college), whereas the most educated group on the WAIS-III is ages 35-44 (56.5% with one or more years of college). When the same age group is compared from the WAIS-R to WAIS-III, the educational advantage for the WAIS-III is evident for most ages. At the peak age for Verbal IQ (ages 45-54), for example, the percentage of adults with at least some college has about doubled (49.0 vs. 25.6). The educational advantage is surely related to the better Verbal test scores for 50-year-olds in 1995 relative to 1978. But why should the peak age have lowered from the 60s to the 50s? That is not obvious at all. The generational trend is evident from several instruments, including the WAIS-III, but explanations require systematic research.
4.3. Theoretical implications
Fig. 3 displays the extremely different patterns of education-adjusted mean Verbal and Performance IQs for the WAIS-III age groups, a difference that is striking during old age. Table 4 indicates the effect sizes for Verbal IQ vs. Performance IQ in the cross-sectional study and Table 9 provides effect sizes for the longitudinal study. Again, the differences between the aging patterns for the separate Wechsler IQs is dramatic and striking. From Horn's (1985, 1989, 1991) theoretical perspective, Verbal IQ qualifies as a maintained ability and Performance IQ as a vulnerable ability. The present results of both studies, therefore, are much more in agreement with Horn's distinction between maintained Gc and vulnerable Gf abilities during the normal aging process than with Schaie's (1983) contentions that intellectual abilities of all types are maintained throughout most of the life span.
Table 3 presents mean IQs, with and without an adjustment for education, based on reference group (20-34 years) norms; Fig. 3 shows the education-adjusted mean IQs for Verbal IQ and Performance IQ, with unadjusted means shown for ages 16-19. Mean Verbal (V) and Performance (P) IQs are comparable to each other (within 5 IQ points) from ages 16-17 to 35-44. From that point on, with or without an education adjustment, adults demonstrate a V [is greater than] P profile by about 12 points (ages 45-64) to about 20 points (ages 65-89). Whereas the "actual" V and P IQs average 100 for each age, because in clinical practice the IQ norms are age based, the present pair of studies -- like previous investigations of the WAIS and WAIS-R (Birren & Morrison, 1961; Horn & Hofer, 1992; Kaufman, 1990; Kaufman et al., 1989) -- underscore the wide disparity in the aging patterns for the two separate Wechsler IQs. Full Scale IQ, the aggregation of V and P IQs, simply does not correspond to any known theoretical construct, but is the artificial midpoint of two very different constructs.
What these constructs are, however, is less clear. They certainly correspond to Gc (V) and Gf (P) from the vantage point of the, original distinction between these measures of general ability (Cattell, 1963; Horn & Cattell, 1966, 1967). That conception articulated broad Gc and broad Gf as the two main components of Gf-Gc theory. But Horn's (1985) elaboration and refinement of the theory abandoned the broad concepts for narrow concepts: Gc was defined in terms of knowledge base and Gf in terms of reasoning ability, with neither "contaminated" by short-term (SAR) or long-term (TSR) memory, visualization (Gv), or speed (Gs).
Fig. 6 reveals that the aging pattern for V-IQ is really composed of two separate patterns: the maintained pattern of VCI and the vulnerable pattern of WMI, although it more closely resembles the maintained pattern of VCI. The VCI, therefore, is the best measure of Gc on the WAIS-III, whereas V-IQ is a theoretical amalgam from Horn's refinement of Gf-Gc theory. The same can be said for P-IQ, shown in Fig. 7 to have an aging pattern that resembles POI during its period of growth mid PSI during its period of decline. P-IQ does not correspond to a unitary ability in Horn's modern Gf-Gc theory, but is a fusion of subtests that require Gf, Gs, and Gv. The array of abilities that is believed to be measured by Wechsler's Performance Scale corresponds closely to the broad "mechanics" component of cognition in Baltes' (1997) two-component (mechanics-pragmatics) lifespan theory of intellectual development. In contrast to the crystallized-like aspect of the pragmatics component, the mechanics component is vulnerable to the effects of normal aging and subsumes reasoning, spatial orientation, and perceptual speed (Baltes, 1997; Baltes, Staudinger, & Lindenberger, 1999). The speeded component of Wechsler's Performance subtests is most evident in the two PSI subtests (digit symbol-coding and symbol search), but also emerges as a factor on the tasks that offer bonus points for quick, perfect solutions to problems (block design and object assembly). POI is by far a more "pure" measure than P-IQ from Horn's theory, but it is unclear whether the primary ability is Gf or Gv. Despite Woodcock's (1990) and McGrew's (1997) insistence that the POI subtests measure Gv and nothing else, that notion is contrary to Horn's (1989) and Horn and Hofer's (1992) writings that give credit to both Gf and Gv. But which is dominant, Gf or Gv? POI on the WAIS-III seems to be composed of one subtest that is primarily Gv (picture completion), one that is primarily Gf (matrix reasoning), and one that is a clear blend of the two (block design). In fact, the education-adjusted aging patterns of these three subtests are virtually identical (Kaufman, 1998, 2000), offering no clues about the abilities measured by POI. Other research has also demonstrated similar patterns of vulnerability for measures of Gv and Gf, even when the measures seem fairly "pure" from Horn's theoretical perspective (Kaufman et al., 1996). The similarity in the patterns for Gv, Gf, and Gv/Gf subtests gives some support to Baltes' (1997) theoretical mechanics component, which lumps spatial (Gv) and reasoning (Gf) abilities into a single lifespan construct of intelligence.
Ultimately, it is the four Indexes, and not the three WAIS-III IQs, that have the most theoretical relevance for Horn's theory. The VCI displays the aging pattern associated with Gc, the POI pattern resembles Gf/Gv; the PSI pattern resembles Gs, and the WMI pattern resembles SAR (Horn, 1985, 1989; Horn & Hofer, 1992). However, purity is missing even in the separate indexes; for example, the separate subtests that compose the WMI display distinct aging patterns from Horn's expanded Gf-Gc theory -- SAR (digit span), Gq (quantitative) or Gc (arithmetic), and Gv/Gf (letter-number sequencing, a novel memory task that may require visualization for success) (Kaufman, 1998, 2000; Kaufman & Lichtenberger, 1999).
These theoretical differences represent more than armchair speculation and may have neurological ramifications. For example, the Gf and Gv abilities, even if they display the same pattern of vulnerability across the adult life span, are different in a neurological sense. Gv abilities and abilities that blend Gv with Gf (such as the ones measured by block design) are susceptible to right-hemispheric dysfunction (Reitan & Wolfson, 1992). In contrast, fluid reasoning abilities are believed to be most vulnerable to dysfunctions associated with the hypothalamus and adjacent areas (Horn & Noll, 1997). If the latter hypothesis about the neurological "seat" of Gf abilities proves to be true, then Baltes' (1997) and Baltes et al.'s (1999) merger of these abilities into a single mechanics component would not be supported, despite the similar aging patterns for Gv and Gf abilities. However, future research needs to address other issues raised by Baltes' research and speculation apart from his two-component theory. The strong empirical support for a substantial relationship between sensory (and sensorimotor) abilities and intelligence is quite provocative (Baltes & Lindenberger, 1997; Lindenberger & Baltes, 1994, 1997), and may yield new interpretations of the decline of Verbal IQ in very old age and the decline of Performance IQ throughout the life span. Horn and his colleagues (e.g., Horn & Noll, 1997) have consistently sought cognitive explanations for the observed declines in IQs. Baltes' research raises the possibility that the declines may be due, at least in part, to sensory loss, or to what Baltes and his colleagues (e.g., Baltes & Lindenberger, 1997) have hypothesized to be a "common cause" of brain aging that affects both sensory and intellectual functioning.
1. Horn's notion of maintained and vulnerable abilities during the adult aging process are supported by the results of both the cross-sectional and longitudinal analysis of WAIS-III data, much more so than Schaie's belief in the maintenance of various intellectual abilities. V-IQ was maintained and P-IQ was vulnerable, consistent with previous research involving the WAIS-R, WAIS, and other measures.
2. None of the WAIS-III IQs correspond to Horn's theoretical constructs. Full Scale IQ is an atheoretical amalgam from the vantage point of Horn's modern Gf-Gc theory. V-IQ and P-IQ correspond to broad Gc and broad Gf, respectively, consistent with the original Horn-Cattell conception of these global abilities, and with Baltes' two-component mechanics-pragmatics lifespan theory of intellectual development. But the separate IQs are too impure to denote pure abilities in Horn's expanded and refined Gf-Gc theory.
3. The results of cross-sectional and longitudinal aging studies of the WAIS-III converged to offer a crisp picture of the relationship of IQ to age from young adulthood to old age. By controlling for the important cohort variable of educational attainment (Study 1), and by following seven age cohorts over a 17-year interval (Study 2), the following results emerged: V-IQ was maintained throughout most of the life span, declining only during the 80s; P-IQ peaked in young adulthood, then declined steadily and dramatically across the age range (especially at ages 45 and above). These results are especially noteworthy in view of the systematic elimination of many low-functioning adults from the standardization sample because of liberal exclusionary criteria. The elimination from the sample of patients with known or suspected dementia, and other related disorders, support the notion that the declines in mean IQ are "real" and not due to the inclusion of increasing numbers of patients with dementia within each successive: age group. At the same time, the observed declines are likely to be lower-bound estimates of the declines that exist within the population at large.
4. Findings for P-IQ conform closely to previous research results, suggesting little changes in the aging process from generation to generation on the abilities measured by this nonverbal scale. V-IQ peaked at about age 50 in these studies, similar to the results of other aging studies that utilized recent data, but different from the peak after age 60 for previous Wechsler adult scales -- suggesting a generational shift of unknown origin in the relationship of V-IQ to age.
5. The age-by-age pattern of education-adjusted mean IQs on the WAIS-III P-IQ is closely similar to the pattern of Indexes on the POI during its period of early stability (ages 16-44), but they overlap with the pattern for PSI during the period of rapid decline with increasing age (especially ages 55-74); the rapid decline of P-IQ with increasing age, therefore, may be, more a function of the highly speeded nature of some of its subtests than of their Gf or Gv component.
The author gratefully acknowledges Dr. Aurelio Prifitera and Dr. David Tulsky of The Psychological Corporation for generously making pertinent WAIS-III data available for this study, Ms. Cindi L. Kreiman, MA, also of The Psychological Corporation, for conducting the requested data analyses with speed and accuracy; Dr. Joseph J. Ryan for kindly making relevant WAIS-R data available for his sample of elderly adults ages 75-89; Dr. Jason Cole for developing the figures; Dr. James Kaufman for his invaluable research assistance; and Dr. Jim McLean, for conducting data analyses needed for an earlier draft of this paper. This paper includes portions of the data presented as part of an invited address (Senior Scientist Award, Division 16) at the 1998 APA Convention in San Francisco.
Baltes, P. B. (1997). On the incomplete architecture of human ontogeny: selection, optimization, and compensation as foundation of developmental theory. American Psychologist, 52, 366-380.
Baltes, P. B., & Lindenberger, U. (1997). Emergence of a powerful connection between sensory and cognitive functions across the adult life span: a new window to the study of cognitive aging? Psychology & Aging, 12, 12-21.
Baltes, P. B., & Schaie, K. W. (1976). On the plasticity of adult and gerontological intelligence: where Horn and Donaldson fail. American Psychologist, 31, 720-725.
Baltes, P. B., Staudinger, U. M., & Lindenberger, U. (1999). Lifespan psychology: theory and application to intellectual functioning. Annual Review of Psychology, 50, 471-507.
Birren, J. E., & Morrison, D. F. (1961). Analysis of the WAIS subtests in relation to age and education. Journal of Gerontology, 16, 363-369.
Botwinick, J. (19771). Intellectual abilities. In: J. E. Birren, & K. W. Schaie (Eds.), Handbook of the psychology of aging (pp. 580-605). New York: Van Nostrand Reinhold.
Cattell, R. B. (1963). Theory of fluid and crystallized intelligence: a critical experiment. Journal of Educational Pyschology, 54, 1-22.
Doppelt, J. E., & Wallace, W. L. (1955). Standardization of the Wechsler Adult Intelligence Scale for older persons. Journal of Abnormal and Social Psychology, 51, 312-330.
Flynn, J. R. (1984). The mean IQ of Americans: massive gains 1932 to 1978. Psychological Bulletin, 95, 29-51.
Flynn, J. R. (1987). Massive gains in 14 nations: what IQ really measures. Psychological Bulletin, 101, 171-191.
Flynn, J. R. (1998). WAIS-III and WISC-111 IQ gains in the United States from 1972 to 1995: how to compensate for obsolete norms. Perceptual and Motor Skills, 86, 1231-1239.
Horn, J. L. 111985). Remodeling old models in intelligence. In: B. B. Wolman (Ed.), Handbook of intelligence: theories, measurements, and applications (pp. 267-300). New York: Wiley.
Horn, J. L. (1989). Cognitive diversity: a framework of learning. In: P. L. Ackerman, R. J. Sternberg, & R. Glaser (Eds.), Learning and individual differences (pp. 61-116). New York: Freeman.
Horn, J. L. (1991). Measurement of intellectual capabilities: a review of theory. In: K. S. McGrew, J. K. Werder, & R. W. Woodcock (Eds.), Woodcock-Johnson technical manual: a reference on theory, and current research (pp. 197-246). Allen, TX: DLM/Teaching Resources.
Horn, J. L., & Cattell, R. B. (1966). Refinement and test of the theory of fluid and crystallized intelligence. Journal of Educational Pyschology, 57, 253-270.
Horn, J. L., & Cattell, R. B. (1967). Age difference in fluid and crystallized intelligence. Acta Psychologica, 26, 107-129.
Horn, J. L., & Donaldson, G. (1976). On the myth of intellectual decline in adulthood. American Psychologist, 31, 701-719.
Horn, J. L., Donaldson, G., & Engstrom, R. (1981). Apprehension, memory, and fluid intelligence decline in adulthood. Research on Aging, 3, 33-84.
Horn, J. L., & Hofer, S. M. (1992). Major abilities and development in the adult period. In: R. J. Steinberg, & C. A. Berg (Eds.), Intellectual development (pp. 44-99). Boston, MA: Cambridge Univ. Press.
Horn, J. L., & Noll, J. (1997). Human cognitive capabilities: Gf-Gc theory. In: D. P. Flanagan, J. L. Genshaft, & P. L. Harrison (Eds.), Beyond traditional intellectual assessment: contemporary and emerging theories, tests, and issues (pp. 53-91). New York: Guilford.
Kaufman, A. S. (1983). Review of Wechsler Adult Intelligence Scale-Revised (WAIS-R). Journal of Psychoeducational Assessment, l, 309-313.
Kaufman, A. S. (1990). Assessing adolescent and adult intelligence. Boston, MA: Allyn & Bacon.
Kaufman, A. S. (1994). Intelligent testing with the W!SC-III. New York: Wiley.
Kaufman, A. S. (1998, August). What happens to our WAIS-III scores as we age from 16 to 89 years and what do these changes mean for theory and clinical practice? Invited Division 16 award address presented at the meeting of the American Psychological Association, San Francisco.
Kaufman, A. S. (2000). Seven questions about the WAIS-III regarding differences in abilities across the 16 to 89 year life span. School Psychology Quarterly, 15, 3-29.
Kaufman, A. S., & Horn, J. L. (1996). Age changes on tests of fluid and crystallized intelligence for females and males on the Kaufman Adolescent and Adult Intelligence Test (KAIT) at ages 17 to 94 years. Archives of Clinical Neuropsychology, 11, 97-121.
Kaufman, A. S., Kaufman, J. C., Chen, T, & Kaufman, N. L. (1996). Differences on six Horn abilities for fourteen age groups between 15-16 and 75-94 years. Psychological Assessment, 8, 161-171.
Kaufman, A. S., & Kaufman, N. L. (1990). Administration and scoring manual for Kaufman Brief Intelligence Test (K-BIT). Circle Pines, MN: American Guidance Service.
Kaufman, A. S., & Kaufman, N. L. (1993). Manual for Kaufman Adolescent and Adult Intelligence Test (KAIT). Circle Pines, MN: American Guidance Service.
Kaufman, A. S., & Kaufman, N. L. (1994). Manual for Kaufman Functional Academic Skills Test (K-FAST). Circle Pines, MN: American Guidance Service.
Kaufman, A. S., & Lichtenberger, E. O. (1999). Essentials of WAIS-III assessment. New York: Wiley.
Kaufman, A. S., Reynolds, C. R., & McLean, J. E. (1989). Age and WAIS-R intelligence in a national sample of adults in the 20- to 74-year age range: a cross sectional analysis with educational level controlled. Intelligence, 13, 235-253.
Kausler, D. H. (1982). Experimental psychology and human aging. New York: Wiley.
Lawton, M. P., & Salthouse, T. A. (Eds.). (1998). Essential papers on the psychology of aging. New York: New York Univ. Press.
Lindenberger, U., & Baltes, P. B. (1994). Sensory functioning and intelligence in old age: a strong connection. Psychology & Aging, 9, 339-355.
Lindenberger, U., & Baltes, P. B. (1997). Intellectual functioning in old and very old age: cross-sectional results from the Berlin Aging Study. Psychology & Aging, 12, 410-432.
Matarazzo, J. D. (1972). Wechsler's measurement and appraisal of adult intelligence (5th ed.). New York: Oxford.
McArdle, J. J., Prescott, C. A., Hamagami, F., & Horn, J. L. (1998). A contemporary method for developmental-genetic analysis of age changes in intellectual abilities. Developmental Neuropsychology, 1, 69-114.
McGrew, K. S. (1997). Analysis of the major intelligence batteries according to a proposed comprehensive Gf-Gc framework. In: D. P. Flanagan, J. L. Genshaft, & P. L. Harrison (Eds.), Beyond traditional intellectual assessment: contemporary and emerging theories, tests, and issues (pp. 151-179). New York: Guilford.
McLean, J. E. (1995). Improving education through action research. Thousand Oaks, CA: Corwin Press.
Owens, W. A. (1966). Age and mental ability: a second adult follow-up. Journal of Educational Psychology, 17, 311-325.
Parker, K. C. H. (1986). Changes with age, year-of-birth cohort, age by year-of-birth interaction, and standardization of the Wechsler adult intelligence tests. Human Development, 29, 209-222.
Rabbitt, P. (1993). Baseline changes in cognitive performance with age. In: R. Levy, & R. Howard (Eds.), Treatment and care in old age psychiatry (pp. 11-30). Petersfield, England, UK: Wrightson Biomedical Publishing.
Rabbitt, P., Bent, N., & McInnes, L. (1997). Health, age and mental ability. Irish Journal of Psychology, 18, 104-131.
Reitan, R. M., & Wolfson, D. (1992). Neuropsychological evaluation of older children. South Tucson, AZ: Neuropsychology Press.
Reynolds, C. R., Chastain, R. L., Kaufman, A. S., & McLean, J. E. (1987). Demographic characteristics and IQ among adults: analysis of the WAIS-R standardization sample as a function of the stratification variables. Journal of School Psychology, 25, 323 - 342.
Ryan, J. J., Nowak, T. J., & Geisser, M. E. (1987). On the comparability of the WAIS and WAIS-R: review of the research and implications for clinical practice. Journal of Psychoeducational Assessment, 5, 15-30.
Ryan, J. J., Paolo, A. M., & Brungardt, T. M. (1990). Standardization of the Wechsler Adult intelligence Scale-Revised for persons 75 years and olden Psychological Assessment, 2, 404-411.
Ryan, J. J., Paolo, A. M., & Dunn, G. (1995). Analysis of WAIS-R old-age normative sample in terms of gender, years of education, and preretirement occupation. Assessment, 2, 225-231.
Schaie, K. W. (1983). The Seattle Longitudinal Study: a 21-year exploration of psychometric intelligence in adulthood. In: K. W. Schaie (Ed.), Longitudinal studies of adult psychological development (pp. 64-135). New York: Guilford.
Schaie, K. W., & Hertzog, C. (1986). Fourteen-year cohort-sequential analyses of adult intellectual development. Developmental Psychology, 19, 531-543.
Schmitz-Scherzer, R., & Thomae, H. (1983). Constancy and change of behavior in old age: findings from the Bonn Longitudinal study on aging. In: K. W. Schaie (Ed.), Longitudinal studies of adult psychological development (pp. 191-221). New York: Guilford.
Siegler, I., & Botwinick, J. (1979). A long-term longitudinal study of intellectual ability of older adults: the matter of selective subject attrition. Journal of Gerontology, 34, 242-245.
The Psychological Corporation (1997). WAIS-III and WMS-III technical manual. San Antonio, TX: The Psychological Corporation.
Thurstone, L. I., & Thurstone, T. G. (1949). Examiner's manual for the SRA Primary Mental Abilities Test. Chicago: Science Research Associates.
Wang, J., & Kaufman, A. S. (1993). Changes in fluid and crystallized intelligence across the 20- to 90-year age range on the K-BIT. Journal of Psychoeducational Assessment, 11, 29-37.
Wechsler, D. (1955). Manual for the Wechsler Adult Intelligence Scale (WAIS). San Antonio, TX: The Psychological Corporation.
Wechsler, D. (1958). Measurement and appraisal of adult intelligence (4th ed.). Baltimore: Williams & Wilkens.
Wechsler, D. (1981). Manual for the Wechsler Adult Intelligence Scale-Revised (WAIS-R). San Antonio, TX: The Psychological Corporation.
Wechsler, D. (1991). Manual for the Wechsler Intelligence Scale for Children-Third Edition (WISC-III). San Antonio, TX: The Psychological Corporation.
Wechsler, D. (1997). Manual for the Wechsler Adult Intelligence Scale-Third Edition (WAIS-III). San Antonio, TX: The Psychological Corporation.
Woodcock, R. W. (1990). Theoretical foundations of the WJ-R measures of cognitive ability. Journal of Psychoeducational Assessment, 8, 231-258.
Alan S. Kaufman, Tel.: +1-516-498-9575; fax: +1-516-487-3723.
E-mail address: email@example.com (A.S. Kaufman).
Child Study Center, Yale University School of Medicine, 230 South Frontage Road, Box 207900, New Haven, CT 06520-7900, USA
Received 11 February 1999; received in revised form 21 September 1999; accepted 5 April 2000
|Printer friendly Cite/link Email Feedback|
|Author:||Kaufman, Alan S.|
|Date:||Jun 1, 2001|
|Previous Article:||IQ is related to the worst performance rule in a memory task involving children.|
|Next Article:||Preattentive processing and cognitive ability.|