J ournol of Gerontolog\ : P SYC H OLOG I C AL SC IENC ES1995, Vol. 508. No. 4. Pl93-P201

C opyright 1995 b '- The Gerontolo{ical Socieu^ of Amerita

Differential Age-Related Influences on Memoryfor Verbal-Symbolic Informationand Visual-Spatial lnformation?

Timothy A. Salthouse

School of Psychology, Georgia Institute of Technology.

Adults from a wide range of ages were administered parallel versions of three memory tasks designed to involveverbal-symbolic information or vkual-spatial information. Several analytical procedures were used to determinewhether there were selective age-rehted effects on measures rfrecting spatial information processing compared tothose r$ecting verbal-symbolic information processing. The results provided little support for this hypothesis andinstead were consistent with the existence of a common age-related factor that contributes to the age dffirences inmanJ cognitive measures.

KEY issue in research on cognitive aging is the exist-ence of selective or differential deficits in which the

age-related effects on some cognitive measures are greaterthan those on other measures. As an example, it is some-times claimed that the age-related effects are larger onmeasures of performance from tasks involving spatial infor-mation than on measures of performance from tasks involv-ing verbal information. If this characterization is valid, thenit would be consistent with the hypothesis (e.g., Goldstein &Shelly, l981 ; Klisz, 1978) that the right hemisphere is moresensitive to age-related effects than the left because the righthemisphere is often postulated to be more specialized forvisual-spatial processing than the left hemisphere. Findingsof this type are potentially important because they mightallow closer linkages between brain and behavior than hasbeen possible in the past.

Although results suggestive of a differential deficit can bequite informative, previous comparisons of age differencesin verbal and spatial information processing have been weakfor at least three reasons. First, several of the comparisonshave involved a confounding of type of information withother factors because the measures have often been derivedfrom tasks that differ in numerous dimensions in addition tothe type of information. For example, among the contraststhat have been cited as indicative of differential age-relatedeffects on verbal and spatial processing are letter span vsperceptual closure tasks (Kinsbourne, 1974), and the WAIS-R Verbal vs the WAIS-R Performance scales (Klisz, 1978).Different patterns of age trends on these tests have led tospeculations of differential or selective aging of verbal andspatial information processing and also of left and righthemispheric functioning (e.g., Klisz, 1978). However, testssuch as these often differ in the role of previous knowledge,in the familiarity or amount of experience with the stimulusmaterials, in the presence of time limits, in the mode ofresponse (e.g., oral or manual), and possibly in the amountof current processing required to produce a response. Be-cause of these confoundings, it is difficult to determine

whether it is the type of information or these other aspects ofthe task that contribute to any differential age relations thatmight exist.

Several studies have been reported in which age differ-ences have been examined with verbal and spatial tasksdesigned to be equivalent except for the type of information.Unfortunately, the results from those studies have beenmixed. For example, there are some reports of equivalentage differences for nearly parallel tasks involving verbal andspatial information (Schear & Nebes, 1980; Shelton, Par-sons, & Leber, I 982), but there is at least one report of largerage differences with a visual-spatial task than with a presum-ably comparable task involving verbal information (Tubi &Calev, 1989).

A second limitation of much of the earlier research inves-tigating age-related effects on tasks involving verbal andspatial information is that the assessment of the constructshas been narrow, and hence construct validity may havebeen weak. That is, many of the contrasts in previous studieshave been at the level of single measures rather than at themore meaningful level of theoretical constructs. Any givenmeasure is only an indirect, and hence confounded, indicatorof the relevant theoretical construct because it is also in-fluenced by the specific methods, materials, and measuresused in the assessment of the construct. Because these otherinfluences cannot be separated from the theoretical constructunless converging operations and multiple indicators areavailable, it is generally desirable to rely on two or moremeasures to attempt to unconfound the assessment of theconstruct from its specific operationalization. Reliance onmultiple measures also has the advantage of broadening theassessment of the construct, which will likely increase thegenerality of the results.

The third limitation of much of the past research compar-ing age-related effects on verbal and spatial tasks is that thecomparisons have relied on a single analytical procedure,namely, analysis of variance focusing on the Age x Taskinteraction. This is unfortunate because numerous concerns

Pl93

Pt94 SALTHOUSE

have been raised about the meaningfulness of statisticalinteractions (e.g., Salthouse, 1991), especially when thereare age differences in the baseline level ofperformance, andwhen little information is available about the form of theconstruct-variable relation or the discriminating power (e.g.,reliability, range in the measurement scale, cf., Chapman &Chapman, 1973) of the variables.

At least two alternative analytical procedures can be usedto investigate differential age-related effects. One consists ofconducting the analysis of variance (ANOVA) on scoresexpressed in standard deviation units derived either from thedata of one of the age groups or from the data of a separatereference sample. In addition to minimizing across-measuredifferences in variability, this form of standardization hasthe advantage ofproviding a meaningful index ofthe degreeof overlap of the scores in a relevant distribution. Anexample described in Salthouse (1991, pp. 298-299) illus-trates that outcomes of the analyses can be quite differentwhen this type of comparison is used instead of the tradi-tional analysis of variance on the original scores.

A second alternative method of investigating differentialor selective age-related effects involves examining theamount of independent age-related variance in each variableto determine if the variables have distinct and unique age-related influences (Salthouse & Coon, 1994). That is, ratherthan focusing on the size of the age differences in eitheroriginal or standardized units, this type of analysis is con-cerned with the extent to which the age-related influences inone variable are independent ofthose in another variable. Inparticular, if there are significant increments in the propor-tion of variance (i.e., R'�) associated with age in the spatialmeasures after the variance in the verbal measures wascontrolled, then one could infer that at least some of the age-related effects in the spatial measures were independent of(i.e., unique and distinct from) those in the verbal measures.An additional advantage of regression-based measures isthat because they provide estimates of both the unique andthe shared age-related variance, the relative amounts ofeachtype of influence can be determined. Information of thisnature is valuable in providing quantitative estimates of thecontribution of each kind of influence, instead of simplyindicating whether either effect is significantly differentfrom zero.

The primary goal of the present study was to investigatethe relation between age and measures of performance inverbal and spatial memory tasks without the problems iden-tified above. Nearly parallel verbal and spatial memory taskswere designed, and the data were analyzed with severaldifferent analytical procedures, including ANOVAs on orig-inal and standardized scores, and analyses ofthe amounts ofage-related variance in each variable that was independent ofthe variance in the other variable. In addition, multivariateanalyses were conducted to examine relations at the level oftheoretical constructs rather than only at the level of observ-able or manifest variables.

A secondary goal of the current project was to examine theinfluence of processing speed on the age differences in taskswith different kinds of information and processing require-ments. Because previous studies have found that a largeproportion of the age-related variance in a variety of memory

and cognitive variables is shared with measures of process-ing speed (e.g., Salthouse, 1992, 1993a, 1994a), the data inthis study were examined to determine whether similarrelations would be evident with new combinations of speedand memory variables.

To summarize, adults from a wide range of ages wereadministered parallel versions of three memory tasks involv-ing verbal-symbolic or visual-spatial information. The pri-mary question was whether, across several different analyti-cal procedures and measures, there is evidence of greaterage-related influences on measures based on visual-spatialinformation processing than on measures based on verbal-symbolic information processing. In order to provide aseparate sample for standardization pulposes, 100 collegestudents also performed the same battery of tests.

Mnruoo

Subjects. - Demographic characteristics of the studentand adult samples, with the latter divided into double-decadecategories, are summarized in Table l. It is apparent that theresearch participants had an average of I to 2.5 years ofeducation beyond high school and that their average healthrating was in the very good range. Although these character-istics are typical of participants in research studies of thistype, both samples are undoubtedly positively biased ineducational level, and possibly health status, relative to thegeneral population.

Procedure. - In addition to the tasks described below,most of the participants in this project also performed twotasks that are not discussed in this report. One was acomputer-administered version of the Trail Making test, andthe other was a computer-administered associative learningtest. Results from these tests will be combined with addi-tional data from other studies and described in separatereports.

The first four tests were administered with paper-and-pencil procedures and were designed to assess sensory-motor speed (Boxes, Digit Copying) and perceptual-compar i son speed (Le t te r Compar i son , Pa t te rnComparison). Each test consisted of a page of instructionsand examples and a single test page. The task in the Boxestest was to create as many boxes as possible by drawing the

Table l. Demographic Characteristics of Samples, by Age Groups

Students(n : 100)

I 8-39(n : 64)

40-59(n : 55)

60-88(n : 54't

Age 2O.l( 1 . 7 )

7o Female 33.0

Education 13.9( 1 . 2 \

Health 2.O(0 .8 )

25.3 4'1 .9(6 .5 ) (5 . I )5 l .6 60 .0

t4.3 14.5( I .9 ) (2 .s )

2 .O 2 .3(0.e) ( l .0)

70.4(6.4)

50.0

1 3 . 1(2. ' t)

2 .2( t . 0 )

Note. Education refers to years of formal education completed, andhealth refers to self-rating on a 5-point scale (l : Excellent, 5 : Poor).

fourth line in three-sided figures. The task in the DigitCopying test was to copy digits from an upper square into alower square. In both of these tests the measure of perfor-mance was the number of items completed in 30 sec. TheLetter Comparison test consisted of 2l pairs of letters whichwere to be classified as same or different by writing an S or aD on the line between the members of the pair. The PatternComparison test consisted of 30 pairs of line patterns whichwere to be classified as same or different by writing an S or aD on the line between the members of the pair. In both ofthese latter tests, in order to adjust for guessing, the measureof performance was the number of correct responses minusthe number of incorrect responses written in 30 sec.

The two reaction time tasks involved physical identityjudgments with respect to a pair of letters or a pair ofunfamiliar symbols. The symbols were similar in size to theletters because both sets of stimuli were constructed from an8 x 8 matrix. The stimulus pair was presented in a verticalarrangement in the middle of the computer display, togetherwith a reminder that the "/" key was to be pressed forsame decisions and the "2" key was to be pressed fordifferent decisions. An initial practice block of 18 trialswas administered with each type of stimuli, followed byblocks of 45 trials each with letters, symbols, symbols, andletters. Both accuracy and reaction time (RT) in msec wererecorded, but because mean accuracy was greater than96Vowith both types of stimuli, only the RT results were analyzedfurther.

Figure I is a schematic illustration of the sequence ofevents in the two versions of the three memory tasks. Each ofthese tasks was administered with two blocks of practicetrials, a block of trials in the verbal condition, two blocks oftrials in the spatial condition, and a final block of trials in theverbal condition.

Trials in the Matrix Memory task were preceded by areminder on the screen concerning the type of information(identities or positions of the target items) that was to beremembered. The matrix stimulus, consisting of 7 shadedletters in an array of25 letters, was then presented for 3 sec.This was followed by the response display, which consistedofseven horizontal lines in the verbal condition, and a blankmatrix in the spatial condition. Responses were entered bytyping the letters in the verbal condition and by using arrowkeys, followed by pressing the space bar, to indicate targetpositions in the spatial condition. In both conditions theparticipants were required to produce seven unique re-sponses on each trial to eliminate biases against guessing.Each experimental block consisted of five trials, and therewere two trials in each practice block.

Trials in the Element Memory task consisted of a 2-secpresentation of a complex stimulus, a .5-sec interval with ablank display, and then the presentation of a probe stimulus.The task was to decide whether the probe stimulus hadpreviously been an element in the complex stimulus. Thecomplex stimuli in the verbal condition consisted of sevenrandomly selected letters, and the probe stimulus was asingle letter. The complex stimuli in the spatial conditionconsisted of seven connected line segments forming a pat-tern, and the probe stimulus was a single line segment. Inboth conditions the probe stimulus was an element of the

VERBAL AND SPATIAL INFORMATION Pl95

Matrix Memory

Element Memory

Keeping Track

Figure l. The stimuli and trial structure for the three pairs of memorytasks.

complex stimulus (i.e., one of the seven letters, or one of theline segments in the same position in the array) on 507o of thetrials. Responses were communciated by pressing the "/"

key for YES, and pressing the "2" key for NO. Eachpractice block contained five trials, and the experimentalblocks each contained 50 trials.

The Keeping Track tasks were based on those describedby Salthouse, Babcock, and Shaw ( I 99 I ). In each condition,a trial began with the presentation of an initial value (i.e., adigit, or an asterisk in a particular spatial location) in each oftwo squares. A series of transformations was then presentedrequiring the subject to update the status of the relevantvariable by carrying out the specified arithmetic operation.or by repositioning the asterisk according to the length anddirection of the displayed arrow. Finally, a question markappeared in one of the squares indicating that the currentvalue of that variable should be reported. The responseconsisted of typing the number in the numeric (verbal-symbolic) condition and using arrow keys to position acursor in the final location of the target in the spatialcondition. The measure of performance was the percentageerror in the verbal-symbolic condition and an index of errormagnitude in the spatial condition (in the form of the meandistance in a Cartesian coordinate system between the cursor

FT;FEfnl r t t r l ̂ l F l r Il Y l r l u l z f N llEfol|i:FfnlTlElrfHTF-l

G B E A F H D

Pl96 SALTHOUSE

position and the correct target position). The stimulus dura-tions, .75 sec per display for numeric information and 1.5sec per display for spatial information, were selected fromthe results of the Salthouse et al. (1991) studies to yieldmoderate levels of accuracy in both young and old adults.Practice blocks consisted ofthree trials each. and there werel5 trials in each experimental block.

RnsuLrs

StudentsMeans, standard deviations, and estimates of reliability

(obtained by boosting the correlation between the measuresfrom the two trial blocks by the Spearman-Brown formula) forthe student data are presented in Table 2. It can be seen that thereliabilities are generally in the moderate range, with thelowest value of .60 for the matrix memory-spatial variable.

Relations among the variables were examined with con-firmatory factor analyses (J<ireskog & Sorbom, 1993).Based on earlier research (Earles & Salthouse, 1995;Salthouse, I 993b) , the speed measures were hypothesized toform three factors: Boxes and Digit Copying comprising aSensory-Motor Speed factor; Letter Comparison and PattemComparison comprising a Perceptual Comparison Speedfactor; and Reaction Time-Letters and Reaction Time-Sym-bols comprising a Reaction Time Speed factor. The con-firmatory factor analysis indicated that the hypothesizedstructure provided a good fit to the data (i.e., 1, (df : 6, l/: 100) : 4 .56; RMR : .026;GFI : .98;AGFI : .95) .Correlations between the factors were .73 (St : .13)between Sensory-Motor Speed and Perceptual ComparisonSpeed; .33 (SE : .l l) between Sensory-Motor Speed andReaction Time Speed; and .48 (SE : . l2) between Percep-tual Comparison Speed and Reaction Time Speed.

Two memory factors corresponding to verbal (i.e., matrixmemory-verbal, element memory-verbal, and keepingtrack-verbal) and spatial (i.e., matrix memory-spatial, ele-ment memory-spatial, and keeping track-spatial) informa-tion were hypothesized for the memory variables. The con-

Table 2. Characteristics of Variables in Student Sample (n : 100)

Variable Mean Est. Rel

firmatory factor analysis indicated that the model providedan adequate fit to the data (i.e., X'(df : 8, N : 100) =20.29; RMR : .057; GFI : .94; AGFI : .85). Thecorrelation between the verbal and soatial factors was .78( S E : . l l ) .

AdultsTable 3 contains means, standard deviations, estimated

reliabilities, and the proportions of variance associated withlinear and quadratic age-related trends for the six speedvariables and the six memory variables. Notice that allreliabilities were greater than .8 and that, although the linearage relations were significant in all variables, none of thequadratic (age-squared) age trends was significantly greaterthan zero.

The memory variables from each set of parallel tasks weresubjected to an Age (18-39, 40-59,60-88) by Condition(verbal, spatial) analysis of variance. Although the age (i.e.,F's > 10.0) and condition (i.e., F's > 143.0) effects weresignificant in each analysis, the Age x Condition interactionwas significant only in the contrast of the keeping track-verbal and keeping track-spatial measures [i.e., F(2,170) :12.42, MS. : 0.711. This interaction in this case wasattributable to a greater increase of errors with age in thespatial measure (i.e., means of 2.16,3.75, and3.92 for theyoung, middle, and old groups, respectively) than in theverbal measure (i.e., means of .50, .59, and .70, for theyoung, middle, and old groups, respectively). Note, how-ever, that the different scales for the two measures make thisinteraction diflicult to interpret.

Scores ofthe adults on each variable were converted to z-scores from the student distribution to provide common unitsfor comparison. The means in each of three age groups areplotted in Figure 2 for the six memory variables. The sametype of Age x Condition analysis of variance conducted onthe original scores was also conducted on these standardized

T"bl" 3. Ch"r".,"d

Proportion ofVariance

Variable Mean Est. Rel . Age Age'SD

BoxesDigit CopyingLetter ComparisonPattern ComparisonReaction Time-LettersReaction Time-SymbolsMMVMMSEMVEMSKTVKTS

52.356.112.020.55545'796.285 . t 2

86.0I t . 70 .382 . 6 1

1 5 . 88 . 42 . 8J . O

86850.490.789.0

12.60 . 1 80.86

- t -)

.88

. t )

.60

.78

.89

.68

.82

BoxesDigit CopyingLetter ComparisonPattern ComparisonReaction Time-LrttersReaction Time-SymbolsMMVMMSEMVEMSKTVKTS

50.8 15 . t5 0 . 9 I 1 . 59 . 5 3 . 6

1 5 . 6 4 . 4748 214'792 2294 .98 I .383 . ' 7 t l . 1 6

79.7 10 .66 8 . 5 t 2 . l0 .59 0 .223.44 1 .37

. I 13* .003

.199* .004

.246+ .001

.335+ .002

.251* .000

.256* .000

.27'7* .001

.402* .001

.087+ .001

.070* .001

. l70x .009

.142* .015

.88

. 9 1

.88

.88

.80

.84

.84

. 9 1

Nores. MMV : matrix memory-verbal; MMS = matrix memory-spatial; EMV : element memory-verbal; EMS : element memory-spatial; KTV : keeping track-verbal; KTS : keeping track-spatial.Maximum scores in the memory tasks were 7.0 in the MMV and MMStasks, 100.0 in the EMV and EMS tasks, and 0.0 (no errors) in the KTV andKTS tasks.

Nores. MMV : matrix memory-verbal; MMS : matrix memory-spatial; EMV = element memory-verbal; EMS : element memory-spatial; KTV : keeping track-verbal; KTS : keeping track-spatial.Maximum scores in the memory tasks were 7.0 in the MMV and MMStasks, 100.0 in the EMV and EMS tasks, and 0.0 (no errors) in the KTV andKTS tasks.

+ p ( . 0 1 .

oooat

X - zcoEf

U)

MMV..€F

MMS- 6 -

EMV

EMS

KTv-D-

KTS

Ghronological Age

Figure 2. Means at three age levels for the six memory measuresexpressed in standard deviations from the distribution of l0O collegestudents. MMV : matrix memory-verbal; MMS : matrix memory-spatial; EMV : element memory-verbal; EMS = element memory-spatial; KTV : keeping track-verbal; and KTS : keeping track-spatial.

scores. The age effects in these analyses were all significant(i.e., F's > 10.30), and the condition effects were signifi-cant in the matrix memory (F' : 3l .65) and element mem-ory (F : 27.07) tasks. Only one Age x Condition interac-tion was significant, and that involved the contrast of the twomatrix memory measures [i.e., F(2,170) : 6.67, MS, :2.14). Examination of Figure 2 reveals that the interactionwas attributable to larger age effects for the verbal measurethan for the spatial measure.

It appears from the results described above and from thedata summarized in Table 3 and in Figure 2 that the verbal andspatial memory variables differ with respect to the magnitudeof their relations with age. However, it is important to notethat the rankings of the measures with respect to degree ofage-sensitivity differ across types of comparisons. For exam-ple, the Age x Condition interaction on the raw scoressuggested that the keeping track-spatial measure was moreage-sensitive than the keeping track-verbal measure, but thesame type of analysis on the standardized scores revealed noevidence of differential effects on those variables: instead. itsuggested that the matrix memory-verbal measure was moreage-sensitive than the matrix memory-spatial measure. Fur-thermore, the age sensitivity of the matrix memory-spatialmeasure was larger than that of the matrix memory-verbalmeasure in the amount of variance associated with age (Table3), but the ranking of the measures was reversed in thecomparison in standard deviation units (Figure 2).

The next set of analyses examined the amount of age-related variance in each memory measure that was indepen-dent of, or distinct from, the age-related variance in the othermeasure from the same type of task. Hierarchical regressionanalyses were used for this purpose by determining theincrement in the R'� associated with age after control of thevariable from the parallel memory task. The residual propor-tions of age-related variance were .l l0 @ <.01) for the

VERBAL AND SPATIAL INFORMATION Pl91

matrix memory-spatial measurel .017 for the matrixmemory-verbal measure; .020 for the element memory-spatial measure; .035 (p < .01) for the element memory-verbal measure; .051 (p <.01) for the keeping track-verbalmeasure; and .029 (p < .01) for the keeping track-spatialmeasure.

Although two of the spatial tasks have a significantamount of unique age-related variance, this is also the casefor two of the verbal tasks. Therefore, while it is true that notall of the age-related variance in these measures is sharedwith the measure from the parallel task, this pattern occursfor measures from tasks involving verbal-symbolic informa-tion as well as for measures from tasks involving visual-spatial information.

The results from the analyses described thus far providelittle support for the hypothesis that age-related deficits aregreater in tasks involving spatial information than in thoseinvolving verbal information. However, as mentioned ear-lier, results from single measures may have limited general-izability because of the large specific influences associatedwith particular methods, materials, and types of assessment.For the next set of analyses, therefore, the variables wereaggregated to examine influences at the construct level ratherthan at the level of single measures.

As with the student data, three speed factors were hypoth-esized, and a confirmatory factor analysis revealed a good fitof the data to this structure (i.e., X'(df : 6, N : 113) :13.65; RMR : .029; GFI : .97; AGFI : .91) .Corre la-tions between the factors were .84 (SE : .05) betweenSensory-Motor Speed and Perceptual Comparison Speed;.44 (SE : .07) between Sensory-Motor Speed and ReactionTime Speed; and .61 (SE : .06) between Perceptual Com-parison Speed and Reaction Time Speed.

Separate verbal and spatial factors were hypothesized forthe memory variables, and the confirmatory factor analysisrevealed that this structure provided a moderate fit to the data( i .e . , X ' (df : 8 , N : 173) : 39.56; RMR : .047;GFI :.94; AGFI : .84). However, the correlation between theverbal and spatial factors was .98 (SE : .04), which was notsignificantly different from 1.0. There is thus no evidencefor the existence of distinct memory factors corresponding toverbal and spatial information in the data from the adultsample.

A complete measurement model was examined with uni-tary speed and memory latent constructs. Correlated resid-uals were allowed between the Boxes and Digit Copyingmeasures and between the two reaction time measures toaccommodate the existence of distinct speed factors. (Theerror covariance between the letter comparison and patterncomparison measures was not significantly different fromzero and hence was omitted from the model.) This modelprovided an adequate fit to the data (i.e., X'(df : 51, N :1 7 3 ) : 1 1 4 . 8 6 ; R M R : . 0 5 3 ; G F f : . 9 1 ; A G F I : . 8 6 ) ,and the correlation between the speed and memory factorswas .75 (SE : .05).

Analyses were also conducted to determine whether thesame measurement model fit the data from the studentsample. Significant reductions in 12 values were obtained inthe two-sample analysis when factor correlations, factorloadings, and error variances were allowed to differ across

€ Ln

Pl98 SALTHOUSE

samples, either alone or in combination. It can therefore beinfened that all aspects of the factor structure differ acrossthe two groups. In other words, the relations between fac-tors, the relations of variables to factors, and the residualvariances of the variables all differed in the best-fittingmodels for the student and adult data. Perhaps of greatestinterest, the conelation between the speed and memoryfactors was .38 (SE : .12) in the student data but .75 (SE =.05) in the adult data.

A structural model for the adult data was created byadding age to the model, with paths from age to speed andfroln age to memory. The fit of this model was adequate(i.e., X' (df : 6l, N : 173) : 146.56: RMR : .055; GFI: .89; AGFI : .83), and the structural coefficients for thepath model are portrayed in Figure 3.

Although the direct path between age and the memorylatent construct was -.23, the correlation between age andthe memory construct was -.64. Because the total age-related variance on the memory construct was .410 (i.e., -.64'�), whereas the amount of age-related variance was only.053 (i.e., -.23'�) when speed was considered, these resultssuggest that a large proportion (i.e., [.410-.053]/.410 x100 : 87.l%o) of the age-related variance on the memoryconstruct was shared with measures of speed.

The proportions of shared age-related variance were alsoexamined for each combination of speed and memory vari-ables. That is, for each variable the proportion of age-relatedvariance shared with every other variable was estimated bydetermining the R'� for age, determining the increment in R2for age after control of the other variable, and then subtract-

Figure 3. Coefficients for the structural model illustrating relationsamong age, speed, and memory constructs. DCopy : Digit Copying,LetCom = Letter Comparison, PatCom = Pattem Comparison, RTL =

Reaction Time-lrtters, RTS : Reaction Time-Symbols, MMV : matrixmemory-verbal, MMS : matrix memory-spatial, EMV = elementmemory-verbal, EMS : element memory-spatial, KTV : keeping track-verbal, and KTS : keeping track-spatial.

ing the latter from the former, and dividing the difference bythe R'�for age. The resulting values, which are presented inTable 4, reflect the proportion of age-related variance in thetarget (column) variable that is shared with the controlled(row) variable. The matrix is not symmetric because theamount of age-related variance, and consequently the pro-portion of age-related variance that is shared, is not necessar-ily equivalent for the two variables in each pair.

The entries in Table 4have considerable variability, witha range from . 16, for the proportion of reaction time-symbolage-related variance shared with the element memory-verbal measure, to .99, for the proportion of reaction time-symbol age-related variance shared with the reaction time-verbal measure. An analysis of variance revealed that themeans for the four quadrants in Table 4 (i.e., top left speed-speed : .6671, top right, memory-speed : .584; bottomfeft, speed-memory : .452; and bottom right, memory-memory : .696) were significantly different, F(3,128) :ll .09, MS" : .036. Bonferroni contrasts indicated that themean for the age-related variance in the speed variablesshared with the memory variables (i.e., .452) was signifi-cantly lower than the other values, but that there were nodifferences between the means for the memory age-relatedvariance shared with other memory variables (i.e., .696) andfor the memory age-related variance shared with speedvariables (i.e., .584). At least based on these results, there-fore, it appears that speed is a more important influence onthe age-related variance in the memory measures than viceversa. In fact, the lack ofa significant difference between thememory-memory and memory-speed values suggests that,on the average, there is nearly as much overlap of age-related variance between any particular memory variableand any particular speed variable as between two memoryvariables.

An analysis of variance was also conducted on the datafrom the memory-memory pairs to determine whether theproportion of shared age-related variance varied accordingto the typ€ of information in the task. The means for theproportions of age-related variance were: for the spatialmemory measures, .677 shared with other spatial memorymeasures, and .696 shared with verbal memory measures;and for the verbal measures, .738 shared with other verbalmemory measures and .680 shared with spatial memorymeasures. An analysis of variance revealed that there was nosignificant difference among these conditions (i.e., F(3,26): .09, MS. : .005).

Estimates of the proportions of shared age-related variancecan be converted into a type of correlation coefficient, whichhas been termed the quasi-partial correlation (Salthouse,1994b). The quasi-partial correlation reflects the averageproportion of age-related variance that is shared between twovariables. It is computed by determining the square root of theproduct of the two proportions for a given pair of variablesand then taking the square root ofthat value. For example, theproportions for the MMV and MMS variables in Table 4 were.93 and .74, which correspond to a quasi-partial correlation of.91. The mean of the 66 quasi-partial correlations resultingfrom these computations was .75.

Two exploratory factor analyses were next conducted onthe matrix of quasi-partial correlations with both one-factor

,\

I

1I

I

VERBAL AND SPATIAL INFORMATION

Table 4. Proportions of Shared Age-Related Variance

Criterion Variable

Pl 99

ControlledVariable

t2l 1

I

I Boxes2 DCopy3 LetCom4 PatCom5 RTL6 RTS7 MMV8 MMS9 EMV

IO EMSI I KTV12 KTS

X.99.82.99.'79. 7 1.6'7.75.27.45.58. 6 1

.75X-82.95.59.54.54.u.26.26.54.45

.43

.70X.84.58.63.56.65. 3 1

.39

.25

.46

.67

.67X.47. 5 1.52.70.26. 3 1.46.3'7

.41

.49

.58

.62X.99.65.76.22- 2 3

.47

.40

.35

.43

. 6 1

.64

.99X

. f 9

.69

. 1 6

. 1 9

.49

.37

.30

.40

.50

. 6 1

.59

.55X.93.50.29.59.46

. J J

.42

.60

.50

.47

.74X.29. J Z

.5'7

. 4 1

- )+

.56

.'77

.84

.5'7

.44

.99

.97X.58.90.74

.70

.69

.76

.99

.'74

.68

.93

.94

.72X.90-92

A 1

.64

.56

.83

.67

. t l

.87

.99

.58

.45x. 7 1

. 5 1

.60

.79

.65

. @

. 8 1

.93

. 5 1

.54

.'t9X

Notes. DCopy : Digit Copying; Letcom : Letter Comparison; Patcom : Pattern Comparison; RTL = reaction time-letters; RTS = reaction trme-symbols; MMV : matrix memory-verbal; MMS : matrix memory-spatial; EMV = element memory-verbal; EMS : element memory-spatial; KTV :keeping track-verbal; and KTS : keeping track-spatial.

and two-factor solutions. The results with both solutions aresummarized in Table 5, where it can be seen that 77Vo of thevariance is accounted for by a single factor solution and thata second factor accounts for an additional 1.5Vo of thevariance. The correlation between the two factors was .69,and inspection of the rightmost two columns reveals thatwhen two factors are specified, the pattern of loadings for thetwo factors corresponds to the speed and memory variables.

Two important points should be noted about the data inTable 5. The first is that the communalities are all quite high,with an average of 777o of the age-related variance in thesevariables accounted for by a single factor and nearly 85Zoaccounted for by two factors. Because there is little residualvariance in the variables that is available to be associatedwith other factors, any more specific factors that might existcould presumably account for relatively small proportions ofthe age-related variance in these variables.

The second point to note from Table 5 is that the .69correlation between the two factors indicates that even when asecond factor is extracted, it is highly related to the firstfactor. This finding is consistent with the high correlation(.75) between the speed and memory factors in the measure-ment model and with the large path coefficient (.59) betweenthese constructs in the structural model portrayed in Figure 3.

DrscussroxBefore discussing the implications of the current results,

limitations of the present study should be mentioned. Onelimitation is that despite the intentions, the measures se-lected to involve different types of information may not havebeen completely parallel or equivalent in all respects exceptfor type of information. For example, the keeping tracli-spatial measure required two-dimensional transformations,whereas the addition and subtraction operations in the keep-ing track-verbal task were in a single dimension. Differ-ences in the processing requirements may therefore haveinfluenced the age relations in the verbal and spatial versionsof each task. Furthermore, although performance in bothtasks was assessed in terms of errors, the sensitivity of thedistance-from-the-target measure in the spatial task may not

Table 5. Community Estimates and Loadings From ExploratoryFactor Analyses With I and 2 Factors

Communality Loadings

Variable l-Factor 2-Factors

BoxesDigit CopyingLetter ComparisonPattern ComparisonReaction Time-lrttersReaction Time-SymbolsMMVMMSEMVEMSKTVKTS

Eigenvalue

Cumulative proportion of variance

Correlation between factors

.823 .906 .669

.862 .92'1 .680

.831 .910 .671

.853 .901 .7' t3

.836 .909 .700

.829 .909 .662

.852 .76t .904

.832 .774 .884

.881 .591 .935

.801 .679 .891

.9G04 .768 .936

.841 .116 .9100.907

.845

.69

Nores. MMV : matrix memory-verbal; MMS : matrix memory-spatial; EMV = element memory-verbal; EMS : element memory_spatial; KTV : keeping track-verbal; and KTS : keeping track-spatial.

have been equivalent to the percentage error measure in theverbal-symbolic task. A second possible limitation of thecurrent study is that the influence of type of informationbeing processed may have been low relative to other in-' fluences, such as the type, or amount, of required process-ing. To the extent that this was the case, the systematicvariance attributable to type of information may have beenobscured by the variance associated with other factors.These and other as yet unknown characteristics could havecontributed to the weak differentiation of verbal and spatialtasks in the present study. Nevertheless, it is important toemphasize that: (a) the memory measures in this study wereall quite reliable; (b) the average scores were generally in themiddle of the measurement range; and (c) the two memoryconstructs were moderately differentiated in the data fromthe student sample. These features, together with the plausi-

. '741

.7'7 |

.746

. 8 3 1

.'773

.'738

.812

.807

.675

. '7 t9

. utt,

.'772

9.237

.'770

1l

P200 SALTHOUSE

bility of the verbal-spatial distinction in the tasks portrayedin Figure l, suggest that the cunent study provides a reason-able, albeit not definitive, test of the hypothesis of differen-tial aging of verbal and spatial information processing.

The first major conclusion from the results of this study isthat there is little evidence for selectivity or specificity ofage-related effects on verbal and spatial memory tasks.Among the findings leading to this inference are the mixedresults from the comparison of individual pairs of measures.Specifically, the pattern with respect to which member of thepair exhibited the larger degree of age-sensitivity was incon-sistent across contrasts oforiginal scores, of scores in youngadult standard deviation units and of proportions of uniqueage-related variance. The factor analysis results also fail tosupport the hypothesis of differential age-related influenceson tasks involving visual-spatial compared to verbal-symbolic information. That is, the .98 correlation betweenthe verbal and spatial factors in the confirmatory factoranalysis indicates that the measures are not organized intodistinct factors in the data from the adult sample. The similarvalues of shared age-related variance for combinations ofmemory measures involving the same (i.e., means of .677and .738) and different (i.e. , means of .646 and .680) typesof information in Table 4 are also inconsistent with thehypothesis of selective or differential age-related influences.

The second major conclusion from the present study isthat a large proportion of the age-related variance in thememory measures is shared with that in the speed measures.This conclusion derives from several sources of evidenceobtained from different types of analytical procedures. Forexample, one relevant result is the .75 correlation betweenthe speed and memory factors in the measurement model fitto the data from the adult sample. Not only is this value highin absolute terms but it is also significantly greater than thecorrelation (.38) obtained in the model fit to the student data.The larger correlation in the age-heterogeneous sample sug-gests that the speed and memory constructs may be becom-ing more similar, or less differentiated, with increased age.

A second type ofrelevant evidence derives from analysesof the proportions of age-related variance shared acrosscombinations of memory and speed variables. An average ofover 597o of the age-related variance in the six memoryvariables and the six speed variables was shared with theother variables (Table 4), and a single factor accounted for77 Vo of the age-related variance in the I 2 variables (Table 5).Finally, the structural model summarized in Figure 3 led toestimates that 87. l7o of the age-related variance in thememory factor was shared with the speed factor.

The preceding results all seem to converge on the conclu-sion that there is considerable commonality in the age-related influences on variables representing memory func-tioning and variables reflecting speed of processing.Although the direction of the causal linkage is ambiguous onthe basis of the results summarized above, two additionalresults suggest that speed may be the more fundamentalconstruct. First, when the structural model was altered toreverse the direction of the path between speed and memory,it led to estimates that 72.8Vo of the age-related variance inthe speed construct was shared with the memory construct,which is less than the 87. l7o of ase-related variance in the

memory construct that was shared with the speed construct.And second, analyses of the proportions of shared age-related variance in pairs of individual variables revealed thatthe average amount of age-related variance in the memoryvariables shared with speed variables (i.e., .584) was signifi-cantly greater than the average amount of age-related vari-ance in the speed variables that was shared with memoryvariables (i.e., .452).

The present results are therefore consistent with previousspeculations (e.g., Salthouse, 1992) that the speed measuresreflect how quickly many different types of processingoperations can be executed. Furthermore, as processingspeed slows down with increased age, it may become acentral factor influencing the age-related effects in a varietyofdifferent cognitive tasks, including those involving differ-ent types of information.

In summary, the results of this study do not support thehypothesis that age-related effects are selectively or differ-entially greater for memory tasks involving spatial informa-tion compared to those involving verbal information. In-stead, the results suggest that there may be considerablecommonalities in the age-related influences across differenttypes of memory measures, as well as across measuresreflecting speed ofprocessing. Future research should there-fore not only continue to seek evidence for differential orselective age-related influences but in addition should inves-tigate reasons for the common age-related influences thatalso appear to exist.

ACKNowLEDCMENTS

This research was supported by NIA grant R37 AG-06826. I would liketo acknowledge the valuable assistance of M. Bridges, J. Crawford, A.May, R. Murray, N. Riecke, and B. Smith in recruiting and testing researchparticipants.

Address correspondence to Dr. Timothy A. Salthouse, School of Psy-chology, Georgia lnstitute ofTechnology, Atlanta, GA 30332-0170.

RspennNcEs

Chapman, L. J., & Chapman, J. P. ( 1973). Problems in the measurement ofcognitive deficit. P syc ho lo g ical B ul le t in, 70, 380-385.

Earles, J. L., & Salthouse, T. A. (1995). Interrelations of age, health,and speed. Journal of Gerontology: Psychological Sciences, 508,P33-P41.

Goldstein, G., & Shelly, C. (1981). Does the right hemisphere age morerapidly than the left'! J ournal of C linical N europsychology, 3, 65-78.

Jdreskog, K. G., & Stirbom, D. ( I 993). LI SREL 8 : U se r' s refe rence guide.Chicago: Scientifi c Software lnternational.

Kinsboume, M. (1974). Cognitive deficit and the aging brain: A behavioralanalysis. International Journal of Aging and Human Development, 5,4 l - 49 .

Klisz, D. (1978). Neuropsychological evaluation in older persons. ln M.Storandt, I. Siegler, & M. Elias (Eds.),Tfte clinical psychology ofaging, (pp.7l -95). New York: Plenum.

Salthouse, T. A. (1991). Theoretical perspectives on cognitive aginq.Hillsdale, NJ: Lawrence Erlbaum.

Salthouse, T. A. (1992). Mechanisms of age-cognition relations in adult-ftood. Hillsdale, NJ: Lawrence Erlbaum.

Salthouse, T. A. (1993a). Speed mediation of adult age differences incognition. Deve lopmental P syc holo gy, 29,'722-738.

Salthouse, T. A. (1993b). Speed and knowledge as determinants of adultage differences in verbal tasks. Journal ofGerontology: PsychologicalSciences.48. P29-P36.

Salthouse, T. A. ( I 994a). The nature of the influence of speed on adult agedifferences in cognition. Developme ntal P sycholo gy, 20, 240-259.

Salthouse, T. A. ( 1994b). How many causes are there of aging-related decre-ments in cognitive functioning? Developmental Review, 14, 413-43'7.

VERBAL AND SPATIAL INFORMATION P20l

Salthouse, T. A. , Babcock, R. L. , &Shaw, R. J. (1991). Ef fectsof adul tage.on structural and operational capacities of working memory. psy-chologt and Aging, 6, ll8-121.

Salthouse, T. A., & Coon, V. E. (1994). Interpretation of differentialdeficits: The case of aging and mental arithmetic. "/o urnal of Experimen-tcl Psychology: Learning, Memory and Cognition, 20, ll72_llg2.

Schear, J. M., & Nebes, R. D. (1990). Memory for verbal and spatialinformation as a function of age. Experimental Aging Researih, 6,2 '7 t -282.

Shelton, M. D., Parsons, O. A., & Leber, W. R. (19g2). Verbal andvisuospatial performance and aging: A neuropsychological approach.Journal of Gerontology, 37 , 336-341 .

Tubi, N., & Calev, A. (1989). Verbal and visuosparial recall by youngerand older subjects: Use of matched tasks. psycftalogy and Aging,-4,493-495.

ReceivedJune2l ,1994Accepted September 26, 1994