Copyriqhl 1992 byThe Gerontologicol Society ofAmerica l ournal of Geronto lo gy : P SYCH O LOGI C AL SCI ENC LS 1 9 9 2 .V o l . 4 7 , N o . 3 , P l 2 l 1 2 8 WhatDo Adult Age Differencesin the Digit SymbolSubstitutionTestReflect? Timothy A. Salthouse School of Psychology, Georgia Institute ofTechnology Resultsfrom three studiesare reported in which adults between18 and 84 years of age performed various versions of the Digit Symhol Substitution Test. The frst study revealed that the age-related declines in digit-symbol pedormance were largely independent of both the amount of education the participants had received and their self-reported health status, and h'ere characterfued by a gradual shift in the entire distribution of scores with little age-related increase in variance. The age relations were greatly attenuated after statistical control of a composite meosure of perceptual comparison speed,however, implying considerable commonalities betweenperceptual comparison speedand what the Digit Symbol Substitution Test measures. Two further studies indicated that young and old adults appeared to use similar strategies to perform the task, and were nearly equivalent in the proportions of time devoted to writing the responsesand searching the code table. THE Digit Symbol SubstitutionTest consistsof a code r tabledisplayingpairsof digitsandsymbols,androwsof double boxes with a digit in the top box and nothing in the bottom box. The task for the examineeis to use the code with eachdigit, and tableto determinethe symbol associated to write as many symbols as possible in the empty boxes below eachdigit. Although the test seemsquite simple, it may be importantfor the understandingof adult age differencesin cognitionbecausescoreson it are highly correlated age.In of intelligence andchronological bothwith measures fact, the relationswith intellectualmeasuresare so strong that Wechsler has included it as one of the subtestsin his adult intelligencescales.Correlationsof digit-symbol score with full-scale intelligence score reported in the WAIS (Wechsler,1955)and WAIS-R (Wechsler,l98l) manuals range from .51 to .74, with relatively little systematic variation in the magnitude of the correlationsacross the rangefrom 18 to 14 yearsof age. The relationsbetweenageand digit-symbolscoreare also large,as analysesof the WAIS (Binen & Monison, 1961) andWAIS-R (Kaufman,Reynolds,& Mclean, 1989)standardizationdatarevealedcorrelationsbetweenageand digitsymbol scoreof - .46 and - .54, respectively.Similar age trendshavealsobeenreportedwith a symbol-digitversionof the task in which digits are written insteadof symbols. To illustrate, correlationsbetweenage and symbol-digit score have been reportedto be - .6'7 for a sampleof 172 adults (Emmerson,Dustman,Shearer,& Turner, 1989), - .67 for a sampleof 125 adults(Gilmore,Royer, & Gruhn, 1983), and - .17 for a sampleof 386 adults (Royer, Gilmore, & G r u h n ,l 9 8 l ) . In light of thesestrongrelationswith both intelligenceand adult age, it seemslikely that the Digit Symbol Substitution Test is measuringprocessesimportant to the relation between age and cognition. A potentially fruitful research agendamight thereforeinvolve focusing on the causesand of the processesresponsiblefor age-related consequences variationsin digirsymbol performance. Three hypotheseshave been proposedin recent years to accountfor age-relateddifferencesin digit-symbol performance. What might be termed the peripheralmotor speed hypothesisattributesmost of the age differencesto an agerelatedreductionin the speedof making manualmovements, such as writing symbols.This interpretationis bolsteredby the fact that writing speedhas been found to decline with increasedage (e.g., Binen & Botwinick, l95l), but it is weakenedby its inability to explain the associationbetween digit-symbolscoreand measuresof higher order cognition. Severalsetsof empiricalfindingsreportedin the last l5 years with the peripheralmotor speedhypothearealsoinconsistent sis. For example, Erber (1986), Erber, Botwinick, and Storandt(1981),Salthouse(1988), and Storandt(1976)all measuredthe speed with which individuals could merely copy symbols,in additionto the speedof substitutingsymbols for digits in the standardtest. The ratiosof the copying measuresto the standardmeasureswere nearly identical for young andold adultsin eachstudy [i.e., .44 for young adults and .42 for older adultsin Erber (1986); .48 for young adults and .47 for older adultsin Erber et al. (1981); .45 for young adultsand .46 for olderadultsin Salthouse( I 988);and .50 for young adults and .53 for older adults in Storandt(1976)1. These results are therefore consistentin indicating that both young and old adultsdevotedapproximatelythe sameproportion of their total performancetime to writing the symbols. Another set of results suggestingthat writing speed is probablya minor factorcontributingto the agedifferencesin digit-symbolperformancederivesfrom studieswith a modified Yes/No version of the digit-symbol task. Both Salthouse (1978) in a paper-and-pencilversion, and Salthouse,Kausler, and Saults (1988) in a computeradministered version, found substantial age differences when the researchparticipantwas simply required to indicate whether the displayed digirsymbol pair was correct according to the code table. Age correlationsin the two Salthouse et al. (1988)studies,in which the decisionswere communicatedby a keypresson a computerkeyboard,were Pt2l Pt22 SALTHOUSE .54 and .56 - values quite similar in absolutemagnitude (but opposite in sign becausethey representtime per responserather than number of responsesin a fixed time) to thosereportedin the studiescited earlier. A secondhypothesisproposedto accountfor the negative relations betweenadult age and digit-symbol performance invokesmemory limitationsas the critical factor. That is, it has been proposedthat older adults may be lessefficient at learningor rememberingthe associationbetweenthe digif symbol pairsthan young adults,and consequentlythey must engagein.more time-consumingsearchesof the code table during the performanceof the task. This interpretationhas the advantageof relying on a mechanism(memory limitations) that can be plausibly relatedto other cognitive functions and is consistentwith the well-documentedfindingsof reduced efficiency of paired-associatelearning with increasedage [e.g., Salthouse et al. (1988)recentlyreported correlationsrangingfrom -. 15 to -.42 betweenage and measuresof paired-associateperformancel. However, a numberof resultssuggestthat this interpretation is unlikely to provide a completeexplanationof the negativerelation between age and digit-symbol performance. Among the relevant studiesis one by Erber et al. (198l) in which substantialage differenceswere still evident even after all the young andold researchparticipantshad learnedthe digitsymbol pairs to a criterion of perfect recall. There are also severalreportsthat agedifferenceseitherremainconstant,or evenincrease,acrossl0 (Erber, 1976),20(Grant,Storandt, & B o t w i n i c k ,1 9 7 8 ) ,a n d e v e n 1 0 0( B e r e s& B a r o n , l 9 8 l ) repetitionsof the task. The additionalpracticeshould have allowedmany of the digit-symbolpairsto be committedto memory, and yet in noneof thesestudieswas therea significant reductionof the magnitudeof the age differenceswith greaterpractice.Finally, Salthouse(1978)reportedthat the relative age differencesremained constantwhen memory loadwasminimizedby reducingthe numberof digit-symbol pairs from 9 to 6 to 3 to l. If older adultsdevoteda larger proportion of their performancetime to inspectionof the code table than young adults, then reducing the number of entriesin the code table should have benefitedolder adults more than young adults. The fact that this did not happen suggeststhat the two groups devoted approximately the sameproportionof time to inspectionand searchof the code table. The third hypothesisregardingthe agerelationson digitsymbol performanceis that they are a reflectionof an agerelated slowing of many cognitive operations(Salthouse, 1985).That is, ratherthanrepresentinga specificage-related deficit in one particularprocess,the agedifferencesin digir symbolperformancemay be a consequence of a ratherglobal slowing of many perceptual, motor, and cognitive processes.Unlike the other hypotheses,this hypothesiscurrently has little directly relevantevidence,and insteadhas receivedconsiderationin largepart becauseof the perceived limitations of the other hypotheses. Each of the precedinghypothesesis reexaminedin this report. Different methodologiesare used to investigatethe peripheral-motorand limited-memory interpretationsthan those used in the past, and the relation of digit-symbol performanceto performanceon othersimplespeededtasksis examinedas a meansof investigatingthe hypothesisbased on a relativelygeneralage-relatedslowing. Study I The rationalefor the first study was that if agedifferences in digit-symbolscoresprimarily reflectan age-relatedslowing of many processingoperations,then a substantialproportion of the age-relatedvariance in digit-symbol performance should be eliminated by removing the variance associatedwith other speededmeasures.The other measures of speedexamined in this study involved comparisonsof strings of letters or patternsof lines. Both tasks are very similarto thoseassumedto measureperceptualspeed(e.g., Number Comparison, Identical Pictures) in that simple (same/different)decisionswere to be made with respectto whetherthe two membersof the pair were identical. The availabilityof datafrom a relativelylarge (N : 9 l0) sampleof adultsfrom a wide rangeof agesalso allowed the exact natureof the age trendsto be investigated.For example, two possibilitiesare that with increasedage either (a) the entire distribution shifts toward lower scoresor (b) an increasingpercentageof people perform at relatively low levels,whereasthe othersmaintainhigh levelsof functioning. Negativerelationsbetweenage and performancecould be producedin each case, but the implicationswould be quite different. A shift in the entire distributionwould suggestthat most people are affected by the age-related processes,whereasa bimodal patternwould imply that only a portion ofthe relevantpopulationwas affectedby the agerelatedinfluences. MErHoo Subjects.- Researchparticipantswere recruited from newspaperadvertisementsrequestingvolunteersto participatein researchprojectsconcernedwith memory and cognition. A total of 910 adultsbetween18 and 84 yearsof age participatedin one of four separateprojects with between 220 and235 adultsin eachproject. (No analysesof the digitsymbol datawere describedin the reportsof theseprojects.) The meanageof the 9 10 adults(547owomen)with complete datarelevantto the currentpurposeswas 48.8 years(SD : 17.3).The meanyearsof educationwas 15.4 (SD : 2.5), and the meanhealthrating on a 5-point self-assessment scale (l : excellent,5 : poor)was2.l (SD : 1.2).Correlations of thesevariableswith chronologicalage were -.01 for educationand .13 (p < .01) for self-raredhealth. Procedure. - The WAIS-R Digit Symbol Substitution Test was administeredalong with other testsof perceptual comparison speed, memory span, working memory, and various types of cognition. Measurescommon acrossthe four projects, besidesthe Digit Symbol SubstiturionTest, were: letter comparison,pattern comparison,computation span, and listening span.The letter comparisonand pattern comparisontestsconsistedof pagescontainingpairsof 3, 6, or 9 letters,or patternscomposedof 3 , 6, or 9 line segments . One-half of the pairs were identical, and one-half were different (created by changing one of the letters or line DIGIT SYMBOL segmentsfrom a correct pair). The task for the participant was to classify each pair as SAME or DIFFERENT, by writing an S or a D on a line betweenthe two membersof the pair, as rapidly as possible.Trials with 3, 6, or 9 elements were separatelytimed, and the scoresaveragedto provide a single measurefor each type of comparison(letter or pattern). Becausethe time limits were 20 secondsper page in the first projectand 30 secondsper pagein the otherprojects, the scores from the participantsin the first project were multiplied by 1.5 to makethem comparablewith thosefrom participantsin the otherprojects. The computationspan and listening span tasks were designedto assessworking memoryby requiringparticipantsto rememberinformationwhile alsocanying out specifiedprocessing.In the computationspantask, arithmeticproblems were presentedauditorily, and the task was to answer the arithmeticproblem while also rememberingthe last digit in each problem. Short sentenceswere auditorily presentedin the listeningspantask, with participantsinstructedto answer a questionaboutthe sentencewhile alsorememberingthe last word in eachsentence.The numberof arithmeticproblemsor sentencespresentedon a trial increasedfrom one to seven, with threetrialsat eachsequence length.An individual'sspan was determinedby the highestnumberof digits or wordsthat could be rememberedon at leasttwo of the threetrials for a sequencelength,given that he or she was also conect in the answersto the relevantarithmeticand sentencecomprehension questions.This latterrequirementensuredthat the scores represented both storageand processing. REsulrs ANDDlscusstoN Digit-symbolscoresin this samplecorrelated-.54 with chronologicalage.This value is quite closeto thoseof the studiescited earlier,and the regressionslopeof - .47 items per year is also similar to the slope of - .43 reportedby Emmersonet al. (1989) for the Symbol-Digittest. These resultssuggestthat the presentsampleis probablyrepresentative of other samplesof adultswho participatein research projectsof this type. A multiple regressionanalysis with age and gender as variablesrevealedthat both main effectsandtheir interaction w e r e s i g n i f i c a n(ti . e . , a l l F ' s > 8 . 4 , 7 l < . 0 1 ) . S e p a r a t e analysesof the age relationsfor men and women indicated that the regressionequationswerel5.3l - .40 (years)with an r'of .25 for men,and87.06- .54 (years)with an r' of .35 for women. The interaction is therefore attributableto a higher initial level, but greater age-relateddecline, for women than for men. The distributionof scoresby decadeis illustratedin Figure l It is apparentin thesedatathat the negativeage relations are associatedwith a shift in the entire distribution toward lower scoreswith increasedage. There is little indicationof a tendencytoward bimodality with advancingage, and even the varianceappearsto remain relatively constantfrom the decadeof the 20s through the decadeof the 70s. Standard deviations for the six distributionsillustrated in Figure I w e r e 1 2 . 1 9 ,1 3 . 6 0 ,I 1 . 2 8 , 1 5 . 0 4 ,1 0 . 9 6 ,a n d 1 2 . 9 7 ,f r o m the 20s throughthe 70s, respectively. The two perceptualcomparisonspeedmeasureshad correlations with one anotherof .78, and the correlationbe- P123 tweenthe two working memory measureswas .59. Composite scoresfor each construct were created by averaging the z-scoresfor the letter comparisonand pattern comparison measuresfor perceptualcomparisonspeed, and averaging the z-scoresfor the computation span and listening span measuresfor working memory. Correlationsof thesecompositemeasureswith agewere - .64 for perceptualcomparison speedand - .49 for working memory, and those with digirsymbol score were .74 for perceptual comparison speedand .56 for working memory. A seriesof hierarchicalmultiple regressionanalyseswere conductedto examine the magnitudeof the relations between age and digit-symbol performancebefore, and after, statisticalcontrolofother variables.Resultsofthese analyses are summarizedin Table l. (The possibility that the variablesmight operateas moderatorsof the age relations was examinedby meansof cross-productinteractionterms enteredafter the main effects in the regressionequation. None of theseinteractionswas significant.) It can be seen in Table I that there was relatively little attenuationof the agerelationsaftercdntrollingthe healthand education variables. That is, 28.1o/oof the digit-symbol variancewas relatedto agebeforecontrollingthesevariables, and,27.3o/owas still relatedto age after both variableswere controlled. In contrastto the situation with the health and educationvariables,therewas substantialreductionofthe age relationsafter controlling the perceptualcomparisonspeed variable. The 0.97o of the variance related to age after removingthe varianceassociated with perceptualcomparison speedwas still statistically signi{icant,but it is clearlyquite small. It is also noteworthythat therewas moderateattenuation of the age-relatedvarianceby control of the working memoryvariable,but it was smallerthanthat associated with perceptualcomparisonspeed,and therewas little it could add to the further attenuationof the age relations above that attributableto perceptualcomparisonspeed. Three major conclusionsseemwarrantedfrom the results of this study. The first is that the large negativerelations betweenageanddigit-symbolperformancearecharacterized 95 85 5 I o O A < o " U) o 5 5 ; 4 5 (t) '6) 35 i5 25 15 5 I I I I 20s I I I I I I I I I I I I I I I I I I I I I I 50s I I I I I I I I T I I I T lr Age Decade Figure l. Distribution of WAIS-R Digit Symbol Substitutionscoresby decade.Between 129 and 169 adultsare reDresented in eachdecade. Pt24 SALTHOUSE Table 1. Age-RelatedVariancein Digit-Symbol Performance Before and After StatisticalControl of Other Variables,Study l (N : 910) Cumulative R2 Increment inR2 F for increment Age .287 .28'7 3 6 9l.3 x Health Age .025 .295 .025 .270 3t.6'7+ 348.42* Education Age .o43 .329 .043 .286 57.01* 388.20x Health Education Age .025 .060 .333 .025 .035 .273 33 . 4 9 x 41.96* 368.79x Health Education Perceptualcomparison speed Age .025 .060 .025 .035 50.64x 72.52* characteristicof the digit-symboltask that is contributingto the large negativerelations betweenage and digit-symbol performanceis also presentin the perceptualcomparison tasks. Becausethe perceptualcomparisontasks are superficially quite distinct from the digit-symboltask, this finding provides convergentevidence for the hypothesisthat age differencesin the digit-symboltestare a reflectionof an agerelatedslowingof many processingoperations.The next two studieswere designedto provide discriminantevidencein the form of resultsinconsistentwith intemretationsbasedon factorsrelatedto writing speedor r.roiy limitations. Study 2 Although the resultsof Study I appearconvincing in indicatingthat many of the age-relateddifferencesin digitsymbol performanceare associatedwith the sameprocesses .550 .490 1001.62+ responsiblefor agedifferencesin simpleperceptualcompar.559 .009 I 9.20* ison tasks,the analysesare limited by the coarseness of the Health .02s .025 38.24* measure of number of correct substitutions. A primary Education .060 .035 54.75* motivation for this study was the assumptionthat more Working memory . 3l 9 .259 399.I 5* detailed examinationof the processesinvolved in perforAge .4t6 .097 I5 0 . 6 7 * manceof the digit-symboltaskmight be possibleby impleHealth .025 .025 52.62* mentingthe task on a computer.That is, becausethe comEducation .060 .035 7 5 . 3 *5 puter can preservecompleteinformationabout responsesto Perceptualcomparison individual digit-symbolpairs, the data can be analyzedto speed 550 .490 1040.74* determinethe relationsbetweenmedian responsetime and Working memory 572 .o22 46.91* the serial position of the targetdigit in the code table. This Age 516 .004 9.22* information can, in turn, be used as an indication of the * p< . o l . searchstrategyusedby the individual, or his or her reliance on memory of the digit-symbol pairs while performing the task. As an example,if young adultshave less need than by a shift in the entire distribution, and do not appearto older adults to refer to the code table to retrieve the digitreflecta tendencyfor someolder adultsto havedramatically symbol pairing, then their serialpositionfunctionsmight be lower scores while others continue to perform like the expectedto be flatter than thoseof older adults. majority of young adults.Although thereare somerelatively Becausedataare to be collectedfrom both the paper-andhigh-performing older adults, there are also a number of pencil and computerversionsof the digit-symbol task, it is low-performingyoung adults,and the age-relatedreduction also of interestto examine the relations between the two in the meanlevel of performancewas not associatedwith an performancemeasures.Both the correlationandthe ratiosof increasein the varianceof the distribution. the time per responsecan be investigated.The correlation The secondconclusionis that the age relations in this indicatesthe extentto which the two measuresreflect comsample are largely independentof self-reportedhealth status mon processes,andthe ratio is informativeaboutthe relative and amountof education.This doesnot meanthat the factors time required to write symbols (in the paper-and-pencil of health and educationhave no effect on the age trendsin version)as opposedto merely pressingone of two keys (in digitsymbol performancebecausethe rangeof variation in the computerversion).If the peripheral-motorhypothesisis the health and education measures was rather restricted. correct,thenolder adultsshouldhavemuch smallerratiosof However,asthe valuesin this samplewereat the high end of scoresin the computerversion to scoresin the paper-andthe scales,it can be infened that the substantialnegative pencil version than young adults becausea greaterproporrelations betweenage and digit-symbol performanceare not tion of their responsetime is devotedto writing the symbols. an artifact of age differencesin health or education,because thoserelationsareevidentevenamonghighly educatedadults METHoD who report themselvesto be in good to excellent health. The third major conclusionfrom the resultsof this studyis Subjects.- Datawere availablefrom a total of 135young that there is considerablecommonality in the age-sensitive adults and 80 older adults. Young adults were college aspectsofthe digit-symboltest and the age-sensitive aspects studentswho receivedextracredit in a psychologycoursefor of the perceptualcomparisontasks.That is, statisticalcontheir participation.Older adultswererecruitedfrom newspatrol of the composite measureof perceptualcomparison per advertisements andreferralsfrom otherparticipants.The speedresultedin a 96.l%oattenuation(i.e., from 27.3 to age range of the young adults was l1 to 28 years, with a 0.9Vo)of the age-relatedvariancein Digit Symbol Substitumean of 19.7, and that for the older adults was 51 to 80 tion Test performance.Theseresultssuggestthat whatever years, with a mean of 64.5. Forty-six percentof the young Pt25 DIGIT SYMBOL adults were women, as were 55Vaof the older adults. The averageyearsofeducation were 13.6 (SD : 1.3) for young adultsand 15.7 (SD : 2.2) for older adults,and the mean self-assessed healthratings(on a scalefrom I : excellentto : 5 poor) were 1.5 (SD : 0.7) for young adultsand I .8 (SD : 0.9) for olderadults. Procedure. - The WAIS-R Digit Symbol Substitution Test and a computer-controlledversion of this test were administered,in this order, to the individuals before they participatedin one of severalexperimentsconcernedwith working memory.The computerversionof the testconsisted of a displayof the standarddigit-symbolcodetableat the top of the screen,and a display of a single digit-symbol pair centeredin the middle of the screen. The code table remained constant across trials but the digirsymbol pair changedfrom trial to trial. On one-halfof thetrials, the digitsymbol pair was correct, in that it conespondedto the associationrepresentedin the code table, and on one-halfof the trials it was incorrect.The task for the participantwas to classifythe pairs as CORRECT or INCORRECT as rapidly "/" as possible by pressingthe key on the keyboard for "Z" key CORRECT, and by pressingthe on the keyboard for INCORRECT. A practice sequenceof 20 trials was presentedfollowed by the experimentalsequenceof 90 trials (consistingof a randomarrangementof 5 CORRECT and 5 INCORRECTtrialsfor eachof the 9 disits). RESULTS ANDDISCUSSIoN The age differences in performance on the paper-andpencil digit-symboltask in this samplewere similar to those in other studiesas the age correlationwas -.61, with a regressionslope of - .41 items per year. Median time per responsein thecomputerversionhada correlationwith ageof .75, and a correlationof - .73 with the scoreon the standard Digit Symbol SubstitutionTest. Accuracy in the computer versionwas very similarin the two groups(i.e.,94.9Vofor young adults and 95.97ofor older adults),and thus was not analyzedfurther. Neither health,education,nor genderwas responsiblefor significantmediation(attenuation) or moderation (interaction)ofthe agerelations,andthusthesevariables were also ignoredin subsequent analyses. Two multiple regressionanalyseswere conductedto estimate the amount of age-relatedvariance in the standard paper-and-pencil scores.The initial analysis,with ageas the only predictor,yieldedan R2 for age of .316, F(l,2lO) : 126.30,p < .01. In the secondanalysis,themediantime per responsein the computer version of the digit-symbol task was enteredas a predictorbeforeage.This analysisrevealed with an incrementin R' of .009, that agewas only associated F(1,209) : 3.90, p < .05. As in the previous study, therefore,it appearsthat most ofthe age-relatedvariancein digirsymbol performanceis sharedwith a differenttask, and in this particularcaseit is one that does not even sharethe samekind of response. Estimatesof the relative amount of time devotedto processesother than writing the symbols were derived by test into convertingscoreson the standardpaper-and-pencil time per item, and thendividing thesevaluesinto the median time per responsein the computerversionof the test. These ratios averaged.83 for young adults and .94 for older adults (z : 4.76, p < .01), suggestingthat the young adults devotedapproximatelylTVoof their averageresponsetime to writing, whereasonly 67o of the total responsetime of older adultswas attributableto processesrelatedto writing. However, interpretationof this result is complicatedby a confoundingof task versionwith order of presentation.That is, becausethe computerversionwas alwaysperformedafter the paper-and-pencilversion, the differencein ratios could reflectgreaterlearningfrom the first to the secondteston the part of young adultsratherthan a greaterpercentageof time devotedto writing. Unfortunately,thesetwo factors could not be distinguishedwithin the presentdesign.It is neverthelessimportantto notethat thereis no indicationin thesedata that the requirementto write the symbolsexertedmore of an influenceon older adults than on young adults, and in this respectthe results are consistentwith those of the studies cited earlier (Erber, 1986; Erber et al., l98l; Salthouse, 1988; Storandt, 1976)in which young and old adults were found to have similar ratios of symbol-copying to digir symbol substitutionperformance. Medianresponsetimes in the computerversionof the test were analyzedaccordingto targetdigit (or serialposition in the code table)and trial type (positiveor negative).These data are illustratedin Figure 2, and results of the Age x . TargetPosition x Trial type analysisof variance(ANOVA) are summarizedin Table2. It is evident in Figure 2 that the serial position functions for responseson positive trials were similar for young and old adults.In both groups,the patternis for responsetimesto be somewhatfaster than averagefor digits at the extreme positions(i.e., l and9) andfor the digit 6. The advantage of the digits at the end positions is probably attributableto a strategyofsearchingthe codetablein an outside-inmanner, and responsesto the digit 6 are likely fast becausethe associatedsymbol (0) is distinctiveand easily remembered. The most important point regarding the serial position effects,however,is that they appearvery similar in the two 2000 6 tzso a ; 1s0o E i= 1250 o o 5 1000 Old - Negative t. ' )''-""i' "' z\ / r r : 1 \ ' Youno - Posilive + \ ' r r ! Old - Positive Young -l.le€ative r l_t r t r ct E zso (r 6 5UU o o > 250 0 r ' t r 2 l 3 l 4 r 5 r 6 l 7 A 9 Digit Figure 2. Median responsetime as a functionof targetserialpositionand trial type for the computer digit-symbol task in Study 2. Each data point representsthe median of an averageof 5 responsesfor 135 young adults or 80 older adults. SAL,THOUSE Pt26 Table2. ANOVA Resultsfrom Studies2 and3 Study3 Study2 Digit-Symbol Age (young vs. old) F df MSe DigirSymbol Digit-Digit 27.36* l,78 2295 26.26* l,78 35.53x 8,1688 78 12.'70* 8,624 88 | -94 8,624 l8 2.38 8,1688 78 0.82 8,624 88 t.t2 8,624 l8 79.27* 1,78 148 101.22* 1,78 35 15.56* l,2ll 107 0.08 1,78 148 1.69 1,78 35 ti.67* 8,1688 8l 2.24 8,624 90 1.90 8.624 l6 I.83 8,624 90 0.34 8,624 l6 206.03x l,2tl l 738 o-t / Position(of targetdigit in code table) df MSe Age x position F df MSe Trial type (positivevs. negative) 289.90* F l,2ll df l0'7 MSe Age x trial type df MSe Position x trial type F df MSe Age x position x trial type F df MSe 2.63* 8 ,16 8 8 8l * p< . o l . groups, and the Age x Position interaction was not significant. were generallyslower than positive Negativeresponses responses,andthe absolutedifferencebetweenthe two types oftrials was largerfor older adultsthan for young adults,as reflectedby the significantAge x Trial type interaction.It shouldbe noted, however,that the trial type differencewas very similar in the two groups when expressedin proportional terms. That is, ratios of times for negativetrials to times for positivetrials averaged1. l3 for young adultsand L 14 for olderadults(z : - .55). The major findings of this study were that the computercontrolled version of the Digit Symbol SubstitutionTest yielded measuresexhibiting a similar magnitudeof relation to chronologicalage as the standardversion, and also provided informationrelevantto the strategiesusedto perform the task. The lack of age differencesin the serial position pattemsand the constantproportionof negativeto positive responsetimes in the two groupssuggestthat young and old adultsperformedthe task in a similar manner. Study 3 The serialpositioneffectsin Study2 wereinterpretedas a reflectionof the strategyusedby the researchparticipantto searchthe code table for the correct digifsymbol pairing. However, becausethe order of the digits in the code table it is possiblethat the was fixed (i.e., in a I to 9 sequence), serial position effects actually representdifferencesin responsetimesto specificdigits. This alternativeinterpretation can be investigatedby examining the serial position functions in a version of the task in which no inspectionof the code table is required. If the serial position effects are attributableto variationsassociatedwith specificdigits, and not to the position of the digit within the code table, then thoseeffectsshouldstill be evident in this modified version of the task. In contrast,flat serial position functions would be expectedin this task if they originate becauseof the mannerin which the code table is searched. Two computer-controlledtasks were thereforeadministeredin this study- the digit-symboltask of Study 2 and a new digit-digittask.The digit-digittaskwas identicalto the digit-symbol task, except that no symbols were presented and, thus, there was no need to refer to the code table to evaluatethe correctnessof the pair. The availability of performancemeasuresfrom similar tasks diff'ering in the need to rely on the code table also provides an opportunity to examine the role of memory That is, the magniprocesses in digit-symbolperfbrmance. tude of the correlationbetweenthe digit-symbol and digitdigit measuresshould indicatethe extent to which they reflect common processes,and the ratio of digit-digit to digit-symbol performanceshould be informative about the relativetime neededto refer to the code table. METHoD Subjects.-A totalof l3l adultsbetween2l and80years of age (mean age : 46.5 years,SD : 15.6)contributed data to this study. All researchparticipantswere males recruitedfrom lettersand other contactsto groupspresumed usingspatialvisuto havemoderateto extensiveexperience alizationabilities.The averagenumberofyearsof education health was 16.2(SD : 2. l), and the averageself-assessed r a t i n gw a s 1 . 4( S D : 0 . 6 ) . Procedure. - The two tasks reported here were performed immediatelyafter participatingin a study concerned with influencesof age and experienceon spatial visualization abilities. The digit-symbol task was identical to the computer-controlledtask describedin Study 2. The digitdigit task had a similar format but no symbols were displayed.That is, the codetableconsisted ofrepetitionsofthe digit from the top box in the bottom box, and the probe displaysconsistedof two digits that were eitherthe same(on 50Voof the trials) or different (on 507o of the trials). All participants performed the digit-symbol task before the digit-digit task. ANDDISCUSSION RESULTS Median times in the digit-symbol and digirdigit tasks averagedl.5l and 0.67 seconds,respectively.Because accuracywas high (i.e.,94.8Voin the digirsymbol task and 97 .67oin the digirdigit task), it was not consideredfurther. The two time measurescorrelated.61 with eachother. and DIGIT SYMBOL had ageconelationsof .53 for the digitsymbol measureand .50 for the digit-digit measure.Neither healthnor education was significantlycorrelatedwith either digirsymbol time or digirdigit time and, thus, thesevariableswere isnored in subsequentanalyses. Hierarchical regressionanalysesrevealed that the age_ related variance in the digit-symbol measurewas reduc"ed from 27.l to 6.4Eoafter controlling the varianceassociated with the digit-digit measure.This result is similar to that of StudyI in indicatingthat thereis considerableoverlapin the age-relatedvariance in measuresof digit-symbol perfor_ manceand in measuresof performancein simpleperieptual comparisontasks. The ratio of mediantime in the digit-digit task to that in the digit-symboltaskaveraged.43, indicatingthat processes concernedwith determiningwhetherthe pairing of digit and symbol was correctrequiredan averageof about Sln Li.e., (1.0 - .43) x l00l of the total responsetime. The correla_ tion of this ratio with agewas only .007. indicatingthat the proportional time devoted to evaluating the digii-symbol associationremainedquite constantfrom about age 20 to age80. The sample was divided into three groups, and perfor_ manceof the 38 young adults(age2l to 34, mean age : 28.2 years)and42 olderadulrs(age56 ro 80, mean : 65.3 years)contrastedin a mannersimilar to Study 2. The serial positionfunctionsfor positiveand negativetrials in the digit_ symbol and digit-digit tasksare illustratedin Figure 3, ind the resultsof the ANOVA are summarizedin Table 2. Two results from Figure 3 and Table 2 are especially . important to note. The first is that the serial position func_ tions for the digit-symboltask closely resemblethose of Study 2. The major differenceseemsto be that the times of the young adultsin the currentstudyare slowerthanthoseof theyoung adultsin Study2, perhapsbecausethe averageage of the young adultsin this samplewas almost9 yearsgreater than that of the young adultsin the earlier sample. The secondinterestingaspectof the resultsin Figure 3 is OId DS NEG 2000 o o @ 1750 I 3 / I 1s00 .t -,, "t a i= o 1250 a . l rt -' I ord DS Pos -t.. OIdDD NEG ., r ' I c t r t a t 1 . t"-- --r, I t_ - ' ./ \ otd DD Pos \ . . Youno DS NEG : l t--- I 1000 YounoDS POS +Younq - aDD NEG o YounoDD POS a c I a c(U 6 500 o 250 l 1 t _ 2 L 3 t 4 t 5 L 6 L 7 8 9 Digit Figure 3. Median responsetime as a function of target serial position and trial type-for the computer digit-symbol (DS) and computer digit-digit (DD) tasks in Study 3. Each data point representsthe median of an averageof 5 responsesfor 38 young adults or 42 older adults. NEC, negatiue;pOS, Positive. Pl27 that there was no evidenceof a serial position effect in the responsetimes in the digit-digit task. As mentionedprevi_ ously, this finding is consistentwith the interpretatio; that the serial position effects in the digirsymbol task reflect processes associatedwith searchofthe codetableandarenot attributableto factorsspecificto particulardigits. General Discussion The resultsof thesestudieslead to two maior conclusions concerningthe relationbetweenage and performanceon the Digit Symbol SubstitutionTest. The firsi conclusion,based on the re-sults of Study l, is that the age trendsare produced by a gradualshift ofthe entiredistributionofscores.This is a somewhatsurprisingfinding, becauseit is occasionallysug_ gestedthat negativerelationsbetweenage and performanJe on various cognitive tests are a consequenceof a small number of people experiencingrather pionounced impair_ ments with increasedage, with many people continuing to perform at the level of young adults (Albert, lggg; Lich_ man, 1986;Schaie,1988).The parternsevidentin Figure I areclearly incompatiblewith this interpretation,and instead are more consistentwith a view that the age_related decline in digirsymbol performanceis normativeor typical of most individuals.Of course,longitudinalinformationis neededto reach definitive conclusionsabout patterns of individual aging, but the discoverythat the age-relateddeclinesin this testarecharacterizedby a gradualrepositioningofthe entire distribution,with little or no increasein interindividual variability, indicatesthat the negativerelationsare not sim_ ply attributableto the presenceof a few very low scoreson the part of a small numberof older adults. The secondmajor conclusionis that the agedifferencesin digirsymbol performanceprimarily seemtJreflect a slower rateof processinginformation,and not deficitsin memoryor in the efficiency of specific processes.Two sets of results lead to this conclusion.One set consistsof the findinssin Studies I and 3 that a large proportion of the age-reLted variancein digit-symbol performanceis sharedwith mea_ sures of simple perceptualcomparison speed. Of related interestare the high conelations betweenthe digirsymbol andperceptualcomparisonspeedmeasuresin Study t (i.e., r : betweenthe paper-and-penciland computerdigit 7ql, symbolmeasures in Study2 (i.e., r = - .73), andbetween the comprrterdigirsymbol and computer digit-digit mea_ suresin Study 3 (i.e., r : .61). [In order to place these values in proper context, it should be noted that Wechsler (1981) reportedrhat the test-retestreliability coefficienrfor the digit-symbolscoreis . 82.I Becausethe cbmmonelement in the paper-and-pencildigit-symbol substitution, letter comparison, pattern comparison, computer digit-symbol, and computer digit-digit tests seemsto be the speedwith which elementaryoperationscan be executed. it appears reasonableto infer that a major factor contributing to ttre age sensitivity of digit-symbol performance is the speed of carrying out simple processingoperations. The other setof resultssupportingthe conclusionthat age differencesin digirsymbol performanceare largely attribul_ able to reductions in the speed of processingare those concernedwith the influenceof searchor decisionprocesses, P128 SALTHOUSE memory processes,and factors related to writing speedor manualdexterity.The resultsof Studies2 and3 , andthoseof the other studiescited earlier, are remarkablyconsistentin makethe samerelativecontriindicatingthattheseprocesses bution to the performanceof young andold adults.Estimates of the magnitudeof the effectsof manipulationsdesignedto investigatethesefactorsare sometimesgreateramongolder adults in absoluteterms, but the relative contribution has generally been found to remain constantacross adults of differentages.This is preciselythe patternone would expect if increasedageis associatedwith an approximatelyproporoperations tional slowingof many elementaryprocessing It was mentionedin the introductionthat the Digit Symbol Substitution Test may have special significancebecause scoreson this test are stronglyrelatedboth to chronological age and to a variety of other measuresof cognition and intelligence.If, as the results of these studies seem to indicate,it reallyis the casethatthe agerelationson this test reflect a slowing in the rate of performingbasicoperations. then a worthy goal for future researchis to determine whethera similar mechanismis responsiblefor the relations betweenmeasuresof cognitivefunctioningand both ageand digirsymbol performance. AcKNOwLtDGt\4ENTS This study was supportedby National Instituteon Aging Grant AG06826. t<l Dr. Timothy A. Salthouse.School of PsyAddresscorrespondence c h o l o g y ,G e o r g i al n s t i t u t eo f T e c h n o l o g y ,A t l a n t a ,C A 3 0 3 3 2 - 0 1 7 0 . REt,ERENCF,S A l b e r t . M . S . ( 1 9 8 8 ) .C o g n i t i v el u n c t i o n .I n M . S . A l b e r t & M . B . M o s s (Eds.\, Geriatric neuropst'choloSy(pp. 33-53). New York: Guillirrd Press. ny B e r e s ,C . A . , & B a r o n ,A . ( 1 9 8 1 ) .l m p r o v e dd i g i t s y m b o ls u b s t i t u t i o b crlderwomen as a result o1'extendedpractice.Journul of Gercnktlog,t', 36.59t-597. B i n e n , J . 8 . , & B o t w i n i c k ,J . ( 1 9 5 1 ) .T h e r e l a t i o no f w r i t i n g s p e e dt o a g e 15, ancl to the senile psychoses.Journul ol Consulting Pst'chologt' 243-249. B i r r e n ,J . 8 . , & M o r r i s o n ,D . F . ( 1 9 6 1 ) .A n a l y s i so f t h e W A I S s u b t e s t isn relationto age and education.Journal ofGerontoktgy, 16,363-369. E m m e r s o n .R . Y . , D u s t m a n ,R . E . , S h e a r e r ,D . 8 . , & T u r n e r , C . W . (1989). P3 latencyand symbol digit performancecorrelationsin aging. E r p e r i m e n t aAl g i n g R e s e o r c h /, 5 , l 5 l - 1 5 9 . Erber, J. T. (1976). Age differencesin leaming and memory on a digitsymbol substilutiontask. ExperimentalAging Research,2, 45-53. Erber, J. T. (1986). Age-relatedeffectson spatialcontiguity and interference on coding performance. Journal ol Gerontologl', 4 I , 641-644. , . ( l 9 8 l ) . T h e i m p a c to f m e m o r y E r b e r ,J . T . , B o t w i n i c k ,J . , & S t o r a n d tM on age diff'erencesin digit symbol performance.Journal ol Gerontologt, 36,586-590. G i l m o r e ,G . C . , R o y e r , F . L . . & G r u h n , J . J . ( 1 9 8 3 ) .A g e d i f f e r e n c e si n symbol-digit substitutiontask perfbrmance.Journal tt CLinicnl Psvc h o l o g t , 3 9 ,l 1 4 - 1 2 4 . Grant, E. A. , Storandt,M. , & Botwinick, J. ( I 978). Incentiveand practice of the elderly. Journal of Gerontologr-, in the psychomotorperformance 3-i.413-41.5. K a u f m a n , A . S . , R e y n o l d s ,C . R . . & M c l - e a n , J . E . ( 1 9 8 9 ) . A g e a n d WAIS-R intelligencein a nationalsampleof adultsin the 20- to 74-year analysiswith educationallevel controlled. age range:A cross-sectional I nteI ligence, l -1,235-253 y n d i n t e l l e c t u al la c t o r si n L a c h m a n ,M . E , .( 1 9 8 6 ) .T h e r o l e o f p e r s o n a l i t a intellectuafaging.EducutktnaI G ero nto Io gt', I 2, 339-344. R o y e r ,F . L . , G i l m o r e ,G . C . , & G r u h n ,J . J . ( 1 9 8 1 ) .N o n n a t i v ed a t af o r 37, the syrnboldigit substitutiontask.Jourrul d Cliniul Psyt'holo11y, 60t]-614. S a l t h o u s eT, . A . ( 1 9 7 U ) .T h e r o l e o f m e m o r y i n t h e a g e d e c l i n ei n d i g i t symbol substitutionperlirrmance.J ournaI ol Gen t nn k t gt', -13, 232-238. Salthouse,T. A. (1985). A theory of t'ognitived8in8. Amsterdam:NorthHolland. S a l t h o u s eT, . A . ( l g l t u ) . T h e r o l e o 1 ' p r o c e s s i n rge s o u r c e \t n c o g n i t i v e a g i n g .I n M . L . H o w e & C . J . B r a i n e r d( E d s . ) ,C o g n i t l l ed e v e l o p m c n t in atlulthootl (pp. lti5-239). New York: Springer-Verlag. nl ' S a l r h o u s eT, . A . , K a u s l c r ,D . H . , & S a u l t s ,J . S . ( l 9 1 3 l t )l.n v e s t i g a t i o o s t u d e n ts t a t u s .b a c k g r o u n dv a r i a b l e s .a n d t h e t e a s i b i l i t yo l ' s t a n d a r d . s , - c h o l t t g t ' u nAdg i n g ,- 1 , 2 9 - 3 7 . t a s k si n c o g n i t i v ea g i n gr e s e a r c hP S c h a i e ,K . W . ( l 9 t t u ) . A g e i s m i n p s y c h o l o g i c arle s e a r c hA. m t r i t u n P s l t cfutlopist.4-1, 179 ltt3. Storandl,M. (1976). Speedand coding el'l'ectsin relationto age and ability l e ve l . D c veI o pm( nt oI Ps l ch ()I ()9 , " , I 2 , l ' 77- 1 1L . Wechsler, D. (1955). Manuul lbr the Wechslt'rAdult Intelligent'eScala. New York: PsychologicalCorporation. Wechsfer,D. (l9ul). Munuul lor thc WachslarAdult IntelligenL'eScaleRevisetl.New York: PsychokrgicalCorporation. ReteivedSeptamber26, 1990 AcceptedJune 6, l99l