Memory amp Cognition2005 33 (5) 833-839

The present study represents an attempt to reconcilefindings from two classic learning paradigms seriallearning and multitrial free recall In serial learning par-ticipants must recall a list in its presented order and thisorder is preserved from trial to trial In multitrial free re-call participants may recall list items in any order andthe presentation order is randomized from trial to trialOver the course of multiple studyndashtest trials partici-pantsrsquo overall recall level increases in both tasks thusdemonstrating transfer from earlier to later trials

Historically learning across trials has been a topic ofgreat interest in both free and serial recall paradigms (egHull et al 1940 Johnson 1991 L B Ward 1937) Con-siderable current attention however has been given in-stead to understanding the mechanisms underlying single-trial variants of free and serial recall (eg AndersonBothell Lebiere amp Matessa 1998 Botvinick amp Plaut2004 Brown Preece amp Hulme 2000 Cowan Badde-ley Elliott amp Norris 2003 Farrell amp Lewandowsky2002 Howard amp Kahana 2002 Nairne Neath Serra ampByun 1997 for exceptions see Becker amp Lim 2003and Raaijmakers 2003) Moreover although the vastmajority of studies of free and serial recall have focusedon understanding performance within each paradigm in-dividually our aim is to expose the relation betweenthese paradigms This aim derives from the observation ofsimilar trends across these two tasks which in aggregatesuggest that free and serial recall share common associa-tive mechanisms (Hulme Stuart Brown amp Morin 2003Kahana amp Caplan 2002 G Ward Woodward Stevensamp Stinson 2003)

How can one quantitatively measure and account forthe differences between serial and free recall Do the twoconditions have enough in common for comparisonConceptually serial and free recall tasks vary only in thatserial recall requires a participant to learn order infor-mation (Drewnowski amp Murdock 1980) Multitrial ver-sions of these two tasks also vary in that free recall listsare traditionally presented in different orders with eachsuccessive trial whereas serial recall lists remain in con-stant order over all trials

Waugh (1961) set out to determine whether the differ-ences between performance in multitrial free recall andserial learning could be attributed to recall instructionsor to differences in the order of presentation of the studylist across trials She compared serial recall with two dif-ferent free recall conditions Either the 48-word list to bestudied was held in constant order across six trials as inserial recall (free recall with constant presentation orderFRndashconstant) or it was shuffled after each of the six tri-als as in traditional free recall (free recall with variedpresentation order FRndashvaried) For each condition Waughcalculated learning curves which show how probabilityof recall increases across trials Whereas learning curvesin the two free recall conditions did not differ serial re-call performance started out at a lower level but over tri-als overlook that of the two free recall conditions Waughhypothesized that the difference in rate of learning be-tween free and serial recall resulted from the strategiesthat were employed due to the different recall instruc-tions On the basis of the similarity of the two free recallconditions Waugh further concluded that ldquorepeated con-tiguity between the items to be learned is in this instancean unimportant variablerdquo (p 499)

Given that associative processes play an important rolein episodic memory (Kahana 1996) it is surprising thatWaugh (1961) failed to detect an effect of repeated con-tiguity in her study Indeed Jung and Skeebo (1967)who replicated Waughrsquos free recall conditions found thatFRndashconstant leads to significantly faster learning thandoes FRndashvaried as would be expected if repeated conti-

833 Copyright 2005 Psychonomic Society Inc

The first two authors contributed equally to this research and theorder of these two authors was chosen randomly The authors acknowl-edge support from National Institutes of Health Grant MH55687 Wethank Geoff Ward Marc Howard Arne Ekstrom and Per Sederberg forhelpful comments on the manuscript Correspondence concerning thisarticle should be addressed to M J Kahana Department of PsychologyUniversity of Pennsylvania Suite 302C 3401 Walnut Street Philadel-phia PA 19107 (e-mail kahanasasupennedu)

A comparative analysis of serial and free recall

KRYSTAL A KLEIN KELLY M ADDIS and MICHAEL J KAHANABrandeis University Waltham Massachusetts

Multitrial free and serial recall tasks differ both in recall instruction and in presentation order acrosstrials Waugh (1961) compared these paradigms with an intermediate condition free recall with con-stant presentation order She concluded that differences between free and serial recall were due onlyto recall instructions and not to presentation order The present study reevaluated the relation be-tween free and serial recall using Waughrsquos three conditions By examining recall transitions and the or-ganization of information retained across trials we conclude that presentation order is an importantfactor causing participants to exhibit the same temporal associations in serial recall and in free recallwith constant presentation order

834 KLEIN ADDIS AND KAHANA

guity plays an important role in free recall Jung andSkeeborsquos study had a number of limitations includingthe fact that their participants were children who studiedjust one list each in a between-subjects design never-theless the results are discordant with Waughrsquos conclu-sions If Waugh was premature in concluding that inputorder across trials was unimportant further analysis ofthe FRndashconstant condition in comparison with standardmultitrial free and serial recall could provide valuableclues about the differences and similarities between freeand serial recall

In attempting to reevaluate the relation between freeand serial recall it is important to carefully analyze theaspects of performance in each paradigm that suggestcommon associative mechanisms An important newtool in the analysis of list memory experiments involvesthe measurement of response order during recall In bothfree and serial recall after recalling any given item par-ticipants tend to recall another item from a nearby posi-tion in the study list Moreover these transitions tend tobe to later list items rather than to earlier list items Thisoccurs both in free recall (Kahana 1996 G Ward et al2003) and in the pattern of errors made in probed recallfollowing serial learning (Kahana amp Caplan 2002 Raskinamp Cook 1937) As in Waugh (1961) in the present studymultitrial serial recall is compared with FRndashconstant andFRndashvaried conditions Our aim was to foster a more de-tailed and quantitative understanding of the relation be-tween the free and the serial learning paradigms com-paring them not only on classic analyses such as learningand serial position curves but also on measures of tem-poral associative tendencies

METHOD

ParticipantsTwelve native English speakers between the ages of 18 and 30

years were recruited to participate in a 10-session memory andlearning study for payment

DesignThe participants were tested in each of three conditions1 Free recall with varied presentation order (FRndashvaried condi-

tion) List words appeared in different random orders on each trialand the participants were instructed to recall the list in any orderwithout repeating words

2 Free recall with constant presentation order (FRndashconstant con-dition) List words appeared in the same order on each trial and theparticipants were instructed to recall the list in any order withoutrepeating words

3 Serial recall with constant presentation order (SR condition)List words appeared in the same order on each trial and the partici-pants were instructed to recall the list in order without backtracking

Experimental lists for each participant contained 19 words ran-domly selected from a large pool of two-syllable (mean word lengthof 65 letters) high-frequency (mean KucerandashFrancis written fre-quency of 319) nouns sampled from the MRC database (Wilson1988) Words were not repeated across lists for a given participantbut the same word pool was used to generate each participantrsquos listsSession 1 was a practice session consisting of one list from each ofthe experimental conditions described above Our reason for in-cluding a full hour-long practice session was to avoid differences in

learning-to-learn effects across the three conditions (Dallett 1963)Each subsequent session contained a single condition commencingwith one practice list followed by seven test lists The participantswere always informed about which condition they were going to re-ceive The nine test sessions were divided into 3 three-sessionblocks each block contained one session from each condition Theorder of the three conditions was the same across all three blocks(eg FRndashV FRndashC SR FRndashV FRndashC SR FRndashV FRndashC SR or SRFRndashV FRndashC SR FRndashV FRndashC SR FRndashV FRndashC) All six possiblepermutations of the three conditions were used and were counter-balanced across participants By excluding data from the practicesession (and practice lists from the beginning of each test session)we ensured that the participants were performing at a steady-statelevel and we were thus able to analyze the data from all test lists si-multaneously without worry of an effect of session number or otherlearning-to-learn effects

ProcedureThe study was carried out in soundproof testing rooms An ex-

perimenter was present during the entire practice session and forthe practice lists in all the subsequent sessions List presentationand recording of responses were controlled by computer List itemswere presented auditorily every 15 sec (each word in the pool wasrecorded by a female speaker with clear diction) Following the pre-sentation of the list the beginning of the vocal recall period wassignaled by a row of asterisks appearing on the screen a 300-msectone and the visual instructions to ldquoPlease begin recalling the listrdquoAfter recalling all remembered words the participant terminatedthe recall period by pressing the space bar on the keyboard Spokenresponses were recorded to the computer for later scoring The par-ticipants continued this studyndashtest routine for a total of five trialsbefore advancing to the next list Each session lasted approximately45 min and the participants were allowed to take short breaks be-tween lists The participants took part in no more than 1 session perday and completed all 10 sessions in no more than 2 weeks

RESULTS

Because serial recall requires the participant to recallin the order of list presentation whereas free recall doesnot two methods of scoring were considered when thedata from the present experiment were analyzed In itemscoring a recalled item was scored as correct if it hadbeen presented on the current list without regard toorder In relative order scoring an item was scored ascorrect if it was recalled directly following an item froman earlier study position (Drewnowski amp Murdock 1980)The first recall was always scored as correct in the rela-tive order scoring method The decision to use relativeorder scoring rather than strict positional scoring wasbased on the use of oral recall of supraspan length listsIn our experience instructions to indicate skipped itemsin recall are both awkward and unreliable especiallyafter several omissions of items at earlier serial posi-tions Strict positional scoring would therefore under-estimate the recall probabilities for later serial positionsRelative order scoring has been shown to be less sensi-tive than positional scoring to methodological issues ofthis type (Addis amp Kahana 2004) In addition becauseour main goal was to compare overall learning of listitems across the three recall conditions item scoringprovides a fair comparison of recall performance re-gardless of recall instructions

SERIAL AND FREE RECALL 835

Figure 1 shows the proportion of the list that was cor-rectly recalled using the item scoring method as a func-tion of trial number for each of the recall conditions aswell as the proportion of correct-orderndashscored recall forserial recall Because very few words were recalled outof order in the SR condition only item curves will bediscussed for this and subsequent analyses

A 3 (condition) 5 (trial) repeated measures analysisof variance (ANOVA) was conducted As was expectedthe proportion of the list correctly recalled increasedover trials in each of the three conditions as confirmedby a main effect of trial [F(444) 2500 MSe 0006p 001] In addition the learning curves were nega-tively accelerated This was confirmed by a significantquadratic term [F(111) 2326 MSe 0002 p 001] The main effect of condition [F(222) 144MSe 0003 p 001] was statistically significant re-flecting the finding that performance in the FRndashvariedcondition was significantly lower overall than perfor-mance in either the FRndashconstant [t (11) 490 p 001] or the SR [t(11) 51 p 001] condition TheSR and FRndashconstant conditions did not differ signifi-cantly overall [t(11) 07 ns] Furthermore the inter-action of condition and trial [F(888) 220 MSe 0001 p 001] was statistically significant Consistentwith this significant interaction we found that feweritems in Trial 1 were recalled in the SR condition than ineither FR condition During later trials the mean SRcurve overtakes both FR curves and the FRndashconstantcurve overtakes the FRndashvaried curve

Figure 2 plots serial position curves for lists in eachrecall condition for Trials 1ndash5 Of particular interest is

the comparison between Trials 1 and 2 because thisshows the change in recall for the first trial in which thepresentation of items differed between FR conditionsmdashthat is order was randomized in FRndashvaried but not in FRndashconstant To compare the changes in recall probabilitiesin the three recall conditions across trials a 3 (condi-tion) 3 (serial position bin) 5 (trial) repeated mea-sures ANOVA was computed where serial position binsconsisted of the primacy region (Serial Positions 1ndash4)the middle region (Serial Positions 5ndash15) and the re-cency region (Serial Positions 16ndash19) The participantsincreased their recall over trials as confirmed by a sig-nificant main effect of trial [F(444) 2985 MSe 0012 p 001] In addition main effects of condition[F(222) 220 MSe 0006 p 001] and serial po-sition bin [F(222) 313 MSe 0027 p 001] werefound A significant interaction of condition and serialposition bin [F(44) 246 MSe 0008 p 001] in-dicates that the level of recall in the primacy and recencyregions differed across conditions

Figure 2A shows that in Trial 1 both the FRndashconstantand the FRndashvaried conditions exhibit a large recency effectand a moderate primacy effect replicating the standardfree recall serial position curve (Murdock 1962) The SRcondition in contrast shows a moderate primacy effectand a small recency effect replicating the standard serialposition curve for serial recall (Drewnowski amp Murdock1980 L B Ward 1937) Pairwise t tests Bonferroni-corrected for multiple comparisons show that the primacyregion for FRndashconstant does not differ significantly fromthat for FRndashvaried [t(11) 06 ns] or SR [t(11) 28ns] The SR condition however does show significantly

Figure 1 Learning curves The four plotted curves show item scoring for freerecall with varied presentation order (FRndashvaried [item]) free recall with con-stant presentation order (FRndashconstant [item]) and serial recall (SR [item])conditions and relative order scoring for serial recall (SR [order])

836 KLEIN ADDIS AND KAHANA

lower recall in the recency region than does either FRndashvaried [t(11) 70 p 001] or FRndashconstant [t(11) 62 p 001] whereas FRndashconstant and FRndashvaried do notdiffer [t(11) 11 ns] Thus it seems that the partici-pants approached first-trial free recall in the same mannerfor both FRndashvaried and FRndashconstant regardless of theirknowledge that word order in later trials would be variedor constant respectively

A significant interaction of condition serial positionbin and trial [F(16176) 133 MSe 0003 p 001]indicates that the changes in the primacy and recency re-gions differ across conditions over the course of learn-ing Additional studyndashtest trials serve to increase theoverall level of the curves in all three recall conditionsbut the similarity between FRndashvaried and FRndashconstantattenuates Figure 2B shows that as early as Trial 2 theFRndashconstant curve exhibits a level of recall in the pri-macy region more similar to that seen in the SR curvethan to that of the FRndashvaried curve Bonferroni-correctedpairwise t tests confirm that the Trial 2 primacy regionsof the FRndashconstant and FRndashvaried conditions are signif-icantly different [t(11) 51 p 001] The FR condi-tions diverge further as trials proceed with the FRndashconstant condition looking more like the SR conditionand less like the FRndashvaried condition

Conditional Response ProbabilityAlthough learning curve and serial position curve

analyses are useful in examining which items partici-pants recall they provide very little insight into how re-call proceeds Because it is reasonable to hypothesize

that the specific trends in output order for a given list-learning paradigm could provide telling clues about theworkings of memory mechanisms newer methods ofanalysis have been developed One such analysis is theconditional response probability as a function of lag(lagndashCRP Kahana 1996) The lagndashCRP measures theprobability with which item j is recalled after recall ofitem i conditionalized on the availability of item j Theseprobabilities are plotted as a function of the lag k wherek j i1 Take an example list absence hollow pupilriver darling railway and an example recall trial riverdarling absence The transition from river to darlingis of lag k 1 but lags of k 3 (absence) k 2(hollow) k 1 (pupil) and k 2 (railway) werealso possible The transition from darling to absence isof lag k 4 but lags of k 3 (hollow) k 2(pupil) and k 1 (railway) were also possible Thusthe lagndashCRP for this trial would have the values P (k 4) 11 10 P (k 3) 02 0 P (k 2) 02 0 P (k 1) 01 0 P (k 1) 12 5and P (k 2) 01 0 By plotting the probabilitiesthat participants would make different kinds of temporaltransitions during recall Kahana found that transitionsoccur more frequently to items at nearby study positionsthan to items at distant study positions (termed associa-tive contiguity) and that there is a forward asymmetryduring recall

Figures 3Andash3C show the lagndashCRP functions for eachrecall condition over the course of learning Since powerfunctions have been found to describe lagndashCRP curves(Kahana Howard Zaromb amp Wingfield 2002) functions

Figure 2 Serial position curves (A) First-trial serial position curves for each of the three condi-tions free recall with varied presentation order (FRndashvaried) free recall with constant presentationorder (FRndashconstant) and serial recall (SR) (B) Second-trial serial position curves (C) Third-trialserial position curves (D) Fourth-trial serial position curves (E) Fifth-trial serial position curvesIn all cases serial position curves are based on item scoring (see the text for details)

SERIAL AND FREE RECALL 837

of the form CRP(lag) a | lag |b were fit separately forpositive and negative lags to each participantrsquos data in eachrecall condition The magnitude of the exponent b de-scribes the degree to which the participants exhibit asso-ciative contiguity in recall transitions Across-participantsaverage exponents over Trials 1ndash5 for each of the three re-call conditions appear in Figures 3Dndash3F A 3 (condi-tion) 5 (trial) 2 (transition direction) repeated mea-sures ANOVA was computed on these exponents Asignificant main effect of transition direction [F(111) 16516 MSe 028 p 001] illustrates the forwardasymmetry observed overall In addition a significantmain effect of recall condition [F(222) 5082 MSe 148 p 001] is moderated by a significant interactionof transition direction and recall condition [F(222) 1972 MSe 056 p 001] which indicates that thedegree of asymmetry differs across the three recall con-ditions Further analysis reveals that this interaction isthe result of significant simple effects of transition di-rection for the FRndashconstant [F(111) 1856 MSe 103 p 005] and the SR [F(111) 5534 MSe 069 p 001] conditions but not for the FRndashvariedcondition [F(111) 220 ns]

The changes in the lagndashCRP over trials seen in Fig-ures 3Andash3C for all three recall conditions are confirmedby a significant main effect of trial [F(444) 1601MSe 027 p 001] Significant interactions of trialand transition direction [F(444) 453 MSe 014p 005] and trial and recall condition [F(888) 3273MSe 027 p 001] as well as a marginally signifi-cant three-way interaction of trial transition direction

and recall condition [F(789) 215 MSe 022 p 053] show that the changes in asymmetry over trials dif-fer across the three recall conditions Analysis of simpleeffects reveals that in Trial 1 (panels A1 B1 and C1 ofFigure 3) the three conditions do not differ from one an-other [F(222) 193 ns] and all three conditions showa significant forward asymmetry [F(111) 1889 MSe 033 p 005] as is characteristic in free recall (Howardamp Kahana 1999 Kahana 1996 Kahana et al 2002)Over trials however the three recall conditions showvery different patterns of asymmetry as is illustrated byFigures 3Dndash3F As indicated by the significant interac-tions of trial and transition direction as well as of trialand recall condition the SR condition shows an increas-ing forward asymmetry over trials the FRndashconstant con-dition shows a consistent level of forward asymmetryand the FRndashvaried condition shows a decreasing forwardasymmetry In addition the nonzero level of the averageexponents in Figures 3E and 3F shows that the FRndashconstant and SR conditions show associative contiguityacross all trials whereas Figure 3D shows that associa-tive contiguity disappears over trials in the FRndashvariedcondition By Trial 5 virtually no associative contiguityis evident in the FRndashvaried condition (Figure 3 panel C5)In contrast the exponents in the FRndashconstant conditionremained at a constant level across trials (Figure 3E)and the SR condition forward exponent continued to in-crease over trials (Figure 3F) Although the degree offorward asymmetry differs between the FRndashconstant andthe SR conditions the transitions in the FRndashconstantcondition resemble transitions in the SR condition con-

1864201

8

6

4

201

8

6

4

20

ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5Lag

Con

ditio

nal R

espo

nse

Pro

babi

lity

Trial Number

Exp

onen

t

6

4

2

06

4

2

06

4

2

01 2 3 4 5

FRndashvaried

FRndashconstant

SR

FowardBackward

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

D

E

F

Figure 3 Associative contiguity effects The left panels (AndashC) show the conditional response probability as a functionof lag (lagndashCRP) across each of the five studyndashtest trials (columns 1ndash5) and for each of the three conditions A1ndashA5 freerecall with varied presentation order (FRndashvaried) B1ndashB5 free recall with constant presentation order (FRndashconstant)C1ndashC5 serial recall (SR) To further quantify the changes in contiguity effects across trials and conditions we fit thepower function CRP(lag) a | lag |b to individual participant lagndashCRP functions The rightmost panels (DndashF) show themean power function exponents b as a function of trial number for each of the three conditions D FRndashvaried EFRndashconstant F SR Error bars represent 95 confidence intervals

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

834 KLEIN ADDIS AND KAHANA

guity plays an important role in free recall Jung andSkeeborsquos study had a number of limitations includingthe fact that their participants were children who studiedjust one list each in a between-subjects design never-theless the results are discordant with Waughrsquos conclu-sions If Waugh was premature in concluding that inputorder across trials was unimportant further analysis ofthe FRndashconstant condition in comparison with standardmultitrial free and serial recall could provide valuableclues about the differences and similarities between freeand serial recall

In attempting to reevaluate the relation between freeand serial recall it is important to carefully analyze theaspects of performance in each paradigm that suggestcommon associative mechanisms An important newtool in the analysis of list memory experiments involvesthe measurement of response order during recall In bothfree and serial recall after recalling any given item par-ticipants tend to recall another item from a nearby posi-tion in the study list Moreover these transitions tend tobe to later list items rather than to earlier list items Thisoccurs both in free recall (Kahana 1996 G Ward et al2003) and in the pattern of errors made in probed recallfollowing serial learning (Kahana amp Caplan 2002 Raskinamp Cook 1937) As in Waugh (1961) in the present studymultitrial serial recall is compared with FRndashconstant andFRndashvaried conditions Our aim was to foster a more de-tailed and quantitative understanding of the relation be-tween the free and the serial learning paradigms com-paring them not only on classic analyses such as learningand serial position curves but also on measures of tem-poral associative tendencies

METHOD

ParticipantsTwelve native English speakers between the ages of 18 and 30

years were recruited to participate in a 10-session memory andlearning study for payment

DesignThe participants were tested in each of three conditions1 Free recall with varied presentation order (FRndashvaried condi-

tion) List words appeared in different random orders on each trialand the participants were instructed to recall the list in any orderwithout repeating words

2 Free recall with constant presentation order (FRndashconstant con-dition) List words appeared in the same order on each trial and theparticipants were instructed to recall the list in any order withoutrepeating words

3 Serial recall with constant presentation order (SR condition)List words appeared in the same order on each trial and the partici-pants were instructed to recall the list in order without backtracking

Experimental lists for each participant contained 19 words ran-domly selected from a large pool of two-syllable (mean word lengthof 65 letters) high-frequency (mean KucerandashFrancis written fre-quency of 319) nouns sampled from the MRC database (Wilson1988) Words were not repeated across lists for a given participantbut the same word pool was used to generate each participantrsquos listsSession 1 was a practice session consisting of one list from each ofthe experimental conditions described above Our reason for in-cluding a full hour-long practice session was to avoid differences in

learning-to-learn effects across the three conditions (Dallett 1963)Each subsequent session contained a single condition commencingwith one practice list followed by seven test lists The participantswere always informed about which condition they were going to re-ceive The nine test sessions were divided into 3 three-sessionblocks each block contained one session from each condition Theorder of the three conditions was the same across all three blocks(eg FRndashV FRndashC SR FRndashV FRndashC SR FRndashV FRndashC SR or SRFRndashV FRndashC SR FRndashV FRndashC SR FRndashV FRndashC) All six possiblepermutations of the three conditions were used and were counter-balanced across participants By excluding data from the practicesession (and practice lists from the beginning of each test session)we ensured that the participants were performing at a steady-statelevel and we were thus able to analyze the data from all test lists si-multaneously without worry of an effect of session number or otherlearning-to-learn effects

ProcedureThe study was carried out in soundproof testing rooms An ex-

perimenter was present during the entire practice session and forthe practice lists in all the subsequent sessions List presentationand recording of responses were controlled by computer List itemswere presented auditorily every 15 sec (each word in the pool wasrecorded by a female speaker with clear diction) Following the pre-sentation of the list the beginning of the vocal recall period wassignaled by a row of asterisks appearing on the screen a 300-msectone and the visual instructions to ldquoPlease begin recalling the listrdquoAfter recalling all remembered words the participant terminatedthe recall period by pressing the space bar on the keyboard Spokenresponses were recorded to the computer for later scoring The par-ticipants continued this studyndashtest routine for a total of five trialsbefore advancing to the next list Each session lasted approximately45 min and the participants were allowed to take short breaks be-tween lists The participants took part in no more than 1 session perday and completed all 10 sessions in no more than 2 weeks

RESULTS

Because serial recall requires the participant to recallin the order of list presentation whereas free recall doesnot two methods of scoring were considered when thedata from the present experiment were analyzed In itemscoring a recalled item was scored as correct if it hadbeen presented on the current list without regard toorder In relative order scoring an item was scored ascorrect if it was recalled directly following an item froman earlier study position (Drewnowski amp Murdock 1980)The first recall was always scored as correct in the rela-tive order scoring method The decision to use relativeorder scoring rather than strict positional scoring wasbased on the use of oral recall of supraspan length listsIn our experience instructions to indicate skipped itemsin recall are both awkward and unreliable especiallyafter several omissions of items at earlier serial posi-tions Strict positional scoring would therefore under-estimate the recall probabilities for later serial positionsRelative order scoring has been shown to be less sensi-tive than positional scoring to methodological issues ofthis type (Addis amp Kahana 2004) In addition becauseour main goal was to compare overall learning of listitems across the three recall conditions item scoringprovides a fair comparison of recall performance re-gardless of recall instructions

SERIAL AND FREE RECALL 835

Figure 1 shows the proportion of the list that was cor-rectly recalled using the item scoring method as a func-tion of trial number for each of the recall conditions aswell as the proportion of correct-orderndashscored recall forserial recall Because very few words were recalled outof order in the SR condition only item curves will bediscussed for this and subsequent analyses

A 3 (condition) 5 (trial) repeated measures analysisof variance (ANOVA) was conducted As was expectedthe proportion of the list correctly recalled increasedover trials in each of the three conditions as confirmedby a main effect of trial [F(444) 2500 MSe 0006p 001] In addition the learning curves were nega-tively accelerated This was confirmed by a significantquadratic term [F(111) 2326 MSe 0002 p 001] The main effect of condition [F(222) 144MSe 0003 p 001] was statistically significant re-flecting the finding that performance in the FRndashvariedcondition was significantly lower overall than perfor-mance in either the FRndashconstant [t (11) 490 p 001] or the SR [t(11) 51 p 001] condition TheSR and FRndashconstant conditions did not differ signifi-cantly overall [t(11) 07 ns] Furthermore the inter-action of condition and trial [F(888) 220 MSe 0001 p 001] was statistically significant Consistentwith this significant interaction we found that feweritems in Trial 1 were recalled in the SR condition than ineither FR condition During later trials the mean SRcurve overtakes both FR curves and the FRndashconstantcurve overtakes the FRndashvaried curve

Figure 2 plots serial position curves for lists in eachrecall condition for Trials 1ndash5 Of particular interest is

the comparison between Trials 1 and 2 because thisshows the change in recall for the first trial in which thepresentation of items differed between FR conditionsmdashthat is order was randomized in FRndashvaried but not in FRndashconstant To compare the changes in recall probabilitiesin the three recall conditions across trials a 3 (condi-tion) 3 (serial position bin) 5 (trial) repeated mea-sures ANOVA was computed where serial position binsconsisted of the primacy region (Serial Positions 1ndash4)the middle region (Serial Positions 5ndash15) and the re-cency region (Serial Positions 16ndash19) The participantsincreased their recall over trials as confirmed by a sig-nificant main effect of trial [F(444) 2985 MSe 0012 p 001] In addition main effects of condition[F(222) 220 MSe 0006 p 001] and serial po-sition bin [F(222) 313 MSe 0027 p 001] werefound A significant interaction of condition and serialposition bin [F(44) 246 MSe 0008 p 001] in-dicates that the level of recall in the primacy and recencyregions differed across conditions

Figure 2A shows that in Trial 1 both the FRndashconstantand the FRndashvaried conditions exhibit a large recency effectand a moderate primacy effect replicating the standardfree recall serial position curve (Murdock 1962) The SRcondition in contrast shows a moderate primacy effectand a small recency effect replicating the standard serialposition curve for serial recall (Drewnowski amp Murdock1980 L B Ward 1937) Pairwise t tests Bonferroni-corrected for multiple comparisons show that the primacyregion for FRndashconstant does not differ significantly fromthat for FRndashvaried [t(11) 06 ns] or SR [t(11) 28ns] The SR condition however does show significantly

Figure 1 Learning curves The four plotted curves show item scoring for freerecall with varied presentation order (FRndashvaried [item]) free recall with con-stant presentation order (FRndashconstant [item]) and serial recall (SR [item])conditions and relative order scoring for serial recall (SR [order])

836 KLEIN ADDIS AND KAHANA

lower recall in the recency region than does either FRndashvaried [t(11) 70 p 001] or FRndashconstant [t(11) 62 p 001] whereas FRndashconstant and FRndashvaried do notdiffer [t(11) 11 ns] Thus it seems that the partici-pants approached first-trial free recall in the same mannerfor both FRndashvaried and FRndashconstant regardless of theirknowledge that word order in later trials would be variedor constant respectively

A significant interaction of condition serial positionbin and trial [F(16176) 133 MSe 0003 p 001]indicates that the changes in the primacy and recency re-gions differ across conditions over the course of learn-ing Additional studyndashtest trials serve to increase theoverall level of the curves in all three recall conditionsbut the similarity between FRndashvaried and FRndashconstantattenuates Figure 2B shows that as early as Trial 2 theFRndashconstant curve exhibits a level of recall in the pri-macy region more similar to that seen in the SR curvethan to that of the FRndashvaried curve Bonferroni-correctedpairwise t tests confirm that the Trial 2 primacy regionsof the FRndashconstant and FRndashvaried conditions are signif-icantly different [t(11) 51 p 001] The FR condi-tions diverge further as trials proceed with the FRndashconstant condition looking more like the SR conditionand less like the FRndashvaried condition

Conditional Response ProbabilityAlthough learning curve and serial position curve

analyses are useful in examining which items partici-pants recall they provide very little insight into how re-call proceeds Because it is reasonable to hypothesize

that the specific trends in output order for a given list-learning paradigm could provide telling clues about theworkings of memory mechanisms newer methods ofanalysis have been developed One such analysis is theconditional response probability as a function of lag(lagndashCRP Kahana 1996) The lagndashCRP measures theprobability with which item j is recalled after recall ofitem i conditionalized on the availability of item j Theseprobabilities are plotted as a function of the lag k wherek j i1 Take an example list absence hollow pupilriver darling railway and an example recall trial riverdarling absence The transition from river to darlingis of lag k 1 but lags of k 3 (absence) k 2(hollow) k 1 (pupil) and k 2 (railway) werealso possible The transition from darling to absence isof lag k 4 but lags of k 3 (hollow) k 2(pupil) and k 1 (railway) were also possible Thusthe lagndashCRP for this trial would have the values P (k 4) 11 10 P (k 3) 02 0 P (k 2) 02 0 P (k 1) 01 0 P (k 1) 12 5and P (k 2) 01 0 By plotting the probabilitiesthat participants would make different kinds of temporaltransitions during recall Kahana found that transitionsoccur more frequently to items at nearby study positionsthan to items at distant study positions (termed associa-tive contiguity) and that there is a forward asymmetryduring recall

Figures 3Andash3C show the lagndashCRP functions for eachrecall condition over the course of learning Since powerfunctions have been found to describe lagndashCRP curves(Kahana Howard Zaromb amp Wingfield 2002) functions

Figure 2 Serial position curves (A) First-trial serial position curves for each of the three condi-tions free recall with varied presentation order (FRndashvaried) free recall with constant presentationorder (FRndashconstant) and serial recall (SR) (B) Second-trial serial position curves (C) Third-trialserial position curves (D) Fourth-trial serial position curves (E) Fifth-trial serial position curvesIn all cases serial position curves are based on item scoring (see the text for details)

SERIAL AND FREE RECALL 837

of the form CRP(lag) a | lag |b were fit separately forpositive and negative lags to each participantrsquos data in eachrecall condition The magnitude of the exponent b de-scribes the degree to which the participants exhibit asso-ciative contiguity in recall transitions Across-participantsaverage exponents over Trials 1ndash5 for each of the three re-call conditions appear in Figures 3Dndash3F A 3 (condi-tion) 5 (trial) 2 (transition direction) repeated mea-sures ANOVA was computed on these exponents Asignificant main effect of transition direction [F(111) 16516 MSe 028 p 001] illustrates the forwardasymmetry observed overall In addition a significantmain effect of recall condition [F(222) 5082 MSe 148 p 001] is moderated by a significant interactionof transition direction and recall condition [F(222) 1972 MSe 056 p 001] which indicates that thedegree of asymmetry differs across the three recall con-ditions Further analysis reveals that this interaction isthe result of significant simple effects of transition di-rection for the FRndashconstant [F(111) 1856 MSe 103 p 005] and the SR [F(111) 5534 MSe 069 p 001] conditions but not for the FRndashvariedcondition [F(111) 220 ns]

The changes in the lagndashCRP over trials seen in Fig-ures 3Andash3C for all three recall conditions are confirmedby a significant main effect of trial [F(444) 1601MSe 027 p 001] Significant interactions of trialand transition direction [F(444) 453 MSe 014p 005] and trial and recall condition [F(888) 3273MSe 027 p 001] as well as a marginally signifi-cant three-way interaction of trial transition direction

and recall condition [F(789) 215 MSe 022 p 053] show that the changes in asymmetry over trials dif-fer across the three recall conditions Analysis of simpleeffects reveals that in Trial 1 (panels A1 B1 and C1 ofFigure 3) the three conditions do not differ from one an-other [F(222) 193 ns] and all three conditions showa significant forward asymmetry [F(111) 1889 MSe 033 p 005] as is characteristic in free recall (Howardamp Kahana 1999 Kahana 1996 Kahana et al 2002)Over trials however the three recall conditions showvery different patterns of asymmetry as is illustrated byFigures 3Dndash3F As indicated by the significant interac-tions of trial and transition direction as well as of trialand recall condition the SR condition shows an increas-ing forward asymmetry over trials the FRndashconstant con-dition shows a consistent level of forward asymmetryand the FRndashvaried condition shows a decreasing forwardasymmetry In addition the nonzero level of the averageexponents in Figures 3E and 3F shows that the FRndashconstant and SR conditions show associative contiguityacross all trials whereas Figure 3D shows that associa-tive contiguity disappears over trials in the FRndashvariedcondition By Trial 5 virtually no associative contiguityis evident in the FRndashvaried condition (Figure 3 panel C5)In contrast the exponents in the FRndashconstant conditionremained at a constant level across trials (Figure 3E)and the SR condition forward exponent continued to in-crease over trials (Figure 3F) Although the degree offorward asymmetry differs between the FRndashconstant andthe SR conditions the transitions in the FRndashconstantcondition resemble transitions in the SR condition con-

1864201

8

6

4

201

8

6

4

20

ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5Lag

Con

ditio

nal R

espo

nse

Pro

babi

lity

Trial Number

Exp

onen

t

6

4

2

06

4

2

06

4

2

01 2 3 4 5

FRndashvaried

FRndashconstant

SR

FowardBackward

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

D

E

F

Figure 3 Associative contiguity effects The left panels (AndashC) show the conditional response probability as a functionof lag (lagndashCRP) across each of the five studyndashtest trials (columns 1ndash5) and for each of the three conditions A1ndashA5 freerecall with varied presentation order (FRndashvaried) B1ndashB5 free recall with constant presentation order (FRndashconstant)C1ndashC5 serial recall (SR) To further quantify the changes in contiguity effects across trials and conditions we fit thepower function CRP(lag) a | lag |b to individual participant lagndashCRP functions The rightmost panels (DndashF) show themean power function exponents b as a function of trial number for each of the three conditions D FRndashvaried EFRndashconstant F SR Error bars represent 95 confidence intervals

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

SERIAL AND FREE RECALL 835

Figure 1 shows the proportion of the list that was cor-rectly recalled using the item scoring method as a func-tion of trial number for each of the recall conditions aswell as the proportion of correct-orderndashscored recall forserial recall Because very few words were recalled outof order in the SR condition only item curves will bediscussed for this and subsequent analyses

A 3 (condition) 5 (trial) repeated measures analysisof variance (ANOVA) was conducted As was expectedthe proportion of the list correctly recalled increasedover trials in each of the three conditions as confirmedby a main effect of trial [F(444) 2500 MSe 0006p 001] In addition the learning curves were nega-tively accelerated This was confirmed by a significantquadratic term [F(111) 2326 MSe 0002 p 001] The main effect of condition [F(222) 144MSe 0003 p 001] was statistically significant re-flecting the finding that performance in the FRndashvariedcondition was significantly lower overall than perfor-mance in either the FRndashconstant [t (11) 490 p 001] or the SR [t(11) 51 p 001] condition TheSR and FRndashconstant conditions did not differ signifi-cantly overall [t(11) 07 ns] Furthermore the inter-action of condition and trial [F(888) 220 MSe 0001 p 001] was statistically significant Consistentwith this significant interaction we found that feweritems in Trial 1 were recalled in the SR condition than ineither FR condition During later trials the mean SRcurve overtakes both FR curves and the FRndashconstantcurve overtakes the FRndashvaried curve

Figure 2 plots serial position curves for lists in eachrecall condition for Trials 1ndash5 Of particular interest is

the comparison between Trials 1 and 2 because thisshows the change in recall for the first trial in which thepresentation of items differed between FR conditionsmdashthat is order was randomized in FRndashvaried but not in FRndashconstant To compare the changes in recall probabilitiesin the three recall conditions across trials a 3 (condi-tion) 3 (serial position bin) 5 (trial) repeated mea-sures ANOVA was computed where serial position binsconsisted of the primacy region (Serial Positions 1ndash4)the middle region (Serial Positions 5ndash15) and the re-cency region (Serial Positions 16ndash19) The participantsincreased their recall over trials as confirmed by a sig-nificant main effect of trial [F(444) 2985 MSe 0012 p 001] In addition main effects of condition[F(222) 220 MSe 0006 p 001] and serial po-sition bin [F(222) 313 MSe 0027 p 001] werefound A significant interaction of condition and serialposition bin [F(44) 246 MSe 0008 p 001] in-dicates that the level of recall in the primacy and recencyregions differed across conditions

Figure 2A shows that in Trial 1 both the FRndashconstantand the FRndashvaried conditions exhibit a large recency effectand a moderate primacy effect replicating the standardfree recall serial position curve (Murdock 1962) The SRcondition in contrast shows a moderate primacy effectand a small recency effect replicating the standard serialposition curve for serial recall (Drewnowski amp Murdock1980 L B Ward 1937) Pairwise t tests Bonferroni-corrected for multiple comparisons show that the primacyregion for FRndashconstant does not differ significantly fromthat for FRndashvaried [t(11) 06 ns] or SR [t(11) 28ns] The SR condition however does show significantly

Figure 1 Learning curves The four plotted curves show item scoring for freerecall with varied presentation order (FRndashvaried [item]) free recall with con-stant presentation order (FRndashconstant [item]) and serial recall (SR [item])conditions and relative order scoring for serial recall (SR [order])

836 KLEIN ADDIS AND KAHANA

lower recall in the recency region than does either FRndashvaried [t(11) 70 p 001] or FRndashconstant [t(11) 62 p 001] whereas FRndashconstant and FRndashvaried do notdiffer [t(11) 11 ns] Thus it seems that the partici-pants approached first-trial free recall in the same mannerfor both FRndashvaried and FRndashconstant regardless of theirknowledge that word order in later trials would be variedor constant respectively

A significant interaction of condition serial positionbin and trial [F(16176) 133 MSe 0003 p 001]indicates that the changes in the primacy and recency re-gions differ across conditions over the course of learn-ing Additional studyndashtest trials serve to increase theoverall level of the curves in all three recall conditionsbut the similarity between FRndashvaried and FRndashconstantattenuates Figure 2B shows that as early as Trial 2 theFRndashconstant curve exhibits a level of recall in the pri-macy region more similar to that seen in the SR curvethan to that of the FRndashvaried curve Bonferroni-correctedpairwise t tests confirm that the Trial 2 primacy regionsof the FRndashconstant and FRndashvaried conditions are signif-icantly different [t(11) 51 p 001] The FR condi-tions diverge further as trials proceed with the FRndashconstant condition looking more like the SR conditionand less like the FRndashvaried condition

Conditional Response ProbabilityAlthough learning curve and serial position curve

analyses are useful in examining which items partici-pants recall they provide very little insight into how re-call proceeds Because it is reasonable to hypothesize

that the specific trends in output order for a given list-learning paradigm could provide telling clues about theworkings of memory mechanisms newer methods ofanalysis have been developed One such analysis is theconditional response probability as a function of lag(lagndashCRP Kahana 1996) The lagndashCRP measures theprobability with which item j is recalled after recall ofitem i conditionalized on the availability of item j Theseprobabilities are plotted as a function of the lag k wherek j i1 Take an example list absence hollow pupilriver darling railway and an example recall trial riverdarling absence The transition from river to darlingis of lag k 1 but lags of k 3 (absence) k 2(hollow) k 1 (pupil) and k 2 (railway) werealso possible The transition from darling to absence isof lag k 4 but lags of k 3 (hollow) k 2(pupil) and k 1 (railway) were also possible Thusthe lagndashCRP for this trial would have the values P (k 4) 11 10 P (k 3) 02 0 P (k 2) 02 0 P (k 1) 01 0 P (k 1) 12 5and P (k 2) 01 0 By plotting the probabilitiesthat participants would make different kinds of temporaltransitions during recall Kahana found that transitionsoccur more frequently to items at nearby study positionsthan to items at distant study positions (termed associa-tive contiguity) and that there is a forward asymmetryduring recall

Figures 3Andash3C show the lagndashCRP functions for eachrecall condition over the course of learning Since powerfunctions have been found to describe lagndashCRP curves(Kahana Howard Zaromb amp Wingfield 2002) functions

Figure 2 Serial position curves (A) First-trial serial position curves for each of the three condi-tions free recall with varied presentation order (FRndashvaried) free recall with constant presentationorder (FRndashconstant) and serial recall (SR) (B) Second-trial serial position curves (C) Third-trialserial position curves (D) Fourth-trial serial position curves (E) Fifth-trial serial position curvesIn all cases serial position curves are based on item scoring (see the text for details)

SERIAL AND FREE RECALL 837

of the form CRP(lag) a | lag |b were fit separately forpositive and negative lags to each participantrsquos data in eachrecall condition The magnitude of the exponent b de-scribes the degree to which the participants exhibit asso-ciative contiguity in recall transitions Across-participantsaverage exponents over Trials 1ndash5 for each of the three re-call conditions appear in Figures 3Dndash3F A 3 (condi-tion) 5 (trial) 2 (transition direction) repeated mea-sures ANOVA was computed on these exponents Asignificant main effect of transition direction [F(111) 16516 MSe 028 p 001] illustrates the forwardasymmetry observed overall In addition a significantmain effect of recall condition [F(222) 5082 MSe 148 p 001] is moderated by a significant interactionof transition direction and recall condition [F(222) 1972 MSe 056 p 001] which indicates that thedegree of asymmetry differs across the three recall con-ditions Further analysis reveals that this interaction isthe result of significant simple effects of transition di-rection for the FRndashconstant [F(111) 1856 MSe 103 p 005] and the SR [F(111) 5534 MSe 069 p 001] conditions but not for the FRndashvariedcondition [F(111) 220 ns]

The changes in the lagndashCRP over trials seen in Fig-ures 3Andash3C for all three recall conditions are confirmedby a significant main effect of trial [F(444) 1601MSe 027 p 001] Significant interactions of trialand transition direction [F(444) 453 MSe 014p 005] and trial and recall condition [F(888) 3273MSe 027 p 001] as well as a marginally signifi-cant three-way interaction of trial transition direction

and recall condition [F(789) 215 MSe 022 p 053] show that the changes in asymmetry over trials dif-fer across the three recall conditions Analysis of simpleeffects reveals that in Trial 1 (panels A1 B1 and C1 ofFigure 3) the three conditions do not differ from one an-other [F(222) 193 ns] and all three conditions showa significant forward asymmetry [F(111) 1889 MSe 033 p 005] as is characteristic in free recall (Howardamp Kahana 1999 Kahana 1996 Kahana et al 2002)Over trials however the three recall conditions showvery different patterns of asymmetry as is illustrated byFigures 3Dndash3F As indicated by the significant interac-tions of trial and transition direction as well as of trialand recall condition the SR condition shows an increas-ing forward asymmetry over trials the FRndashconstant con-dition shows a consistent level of forward asymmetryand the FRndashvaried condition shows a decreasing forwardasymmetry In addition the nonzero level of the averageexponents in Figures 3E and 3F shows that the FRndashconstant and SR conditions show associative contiguityacross all trials whereas Figure 3D shows that associa-tive contiguity disappears over trials in the FRndashvariedcondition By Trial 5 virtually no associative contiguityis evident in the FRndashvaried condition (Figure 3 panel C5)In contrast the exponents in the FRndashconstant conditionremained at a constant level across trials (Figure 3E)and the SR condition forward exponent continued to in-crease over trials (Figure 3F) Although the degree offorward asymmetry differs between the FRndashconstant andthe SR conditions the transitions in the FRndashconstantcondition resemble transitions in the SR condition con-

1864201

8

6

4

201

8

6

4

20

ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5Lag

Con

ditio

nal R

espo

nse

Pro

babi

lity

Trial Number

Exp

onen

t

6

4

2

06

4

2

06

4

2

01 2 3 4 5

FRndashvaried

FRndashconstant

SR

FowardBackward

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

D

E

F

Figure 3 Associative contiguity effects The left panels (AndashC) show the conditional response probability as a functionof lag (lagndashCRP) across each of the five studyndashtest trials (columns 1ndash5) and for each of the three conditions A1ndashA5 freerecall with varied presentation order (FRndashvaried) B1ndashB5 free recall with constant presentation order (FRndashconstant)C1ndashC5 serial recall (SR) To further quantify the changes in contiguity effects across trials and conditions we fit thepower function CRP(lag) a | lag |b to individual participant lagndashCRP functions The rightmost panels (DndashF) show themean power function exponents b as a function of trial number for each of the three conditions D FRndashvaried EFRndashconstant F SR Error bars represent 95 confidence intervals

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

836 KLEIN ADDIS AND KAHANA

lower recall in the recency region than does either FRndashvaried [t(11) 70 p 001] or FRndashconstant [t(11) 62 p 001] whereas FRndashconstant and FRndashvaried do notdiffer [t(11) 11 ns] Thus it seems that the partici-pants approached first-trial free recall in the same mannerfor both FRndashvaried and FRndashconstant regardless of theirknowledge that word order in later trials would be variedor constant respectively

A significant interaction of condition serial positionbin and trial [F(16176) 133 MSe 0003 p 001]indicates that the changes in the primacy and recency re-gions differ across conditions over the course of learn-ing Additional studyndashtest trials serve to increase theoverall level of the curves in all three recall conditionsbut the similarity between FRndashvaried and FRndashconstantattenuates Figure 2B shows that as early as Trial 2 theFRndashconstant curve exhibits a level of recall in the pri-macy region more similar to that seen in the SR curvethan to that of the FRndashvaried curve Bonferroni-correctedpairwise t tests confirm that the Trial 2 primacy regionsof the FRndashconstant and FRndashvaried conditions are signif-icantly different [t(11) 51 p 001] The FR condi-tions diverge further as trials proceed with the FRndashconstant condition looking more like the SR conditionand less like the FRndashvaried condition

Conditional Response ProbabilityAlthough learning curve and serial position curve

analyses are useful in examining which items partici-pants recall they provide very little insight into how re-call proceeds Because it is reasonable to hypothesize

that the specific trends in output order for a given list-learning paradigm could provide telling clues about theworkings of memory mechanisms newer methods ofanalysis have been developed One such analysis is theconditional response probability as a function of lag(lagndashCRP Kahana 1996) The lagndashCRP measures theprobability with which item j is recalled after recall ofitem i conditionalized on the availability of item j Theseprobabilities are plotted as a function of the lag k wherek j i1 Take an example list absence hollow pupilriver darling railway and an example recall trial riverdarling absence The transition from river to darlingis of lag k 1 but lags of k 3 (absence) k 2(hollow) k 1 (pupil) and k 2 (railway) werealso possible The transition from darling to absence isof lag k 4 but lags of k 3 (hollow) k 2(pupil) and k 1 (railway) were also possible Thusthe lagndashCRP for this trial would have the values P (k 4) 11 10 P (k 3) 02 0 P (k 2) 02 0 P (k 1) 01 0 P (k 1) 12 5and P (k 2) 01 0 By plotting the probabilitiesthat participants would make different kinds of temporaltransitions during recall Kahana found that transitionsoccur more frequently to items at nearby study positionsthan to items at distant study positions (termed associa-tive contiguity) and that there is a forward asymmetryduring recall

Figures 3Andash3C show the lagndashCRP functions for eachrecall condition over the course of learning Since powerfunctions have been found to describe lagndashCRP curves(Kahana Howard Zaromb amp Wingfield 2002) functions

Figure 2 Serial position curves (A) First-trial serial position curves for each of the three condi-tions free recall with varied presentation order (FRndashvaried) free recall with constant presentationorder (FRndashconstant) and serial recall (SR) (B) Second-trial serial position curves (C) Third-trialserial position curves (D) Fourth-trial serial position curves (E) Fifth-trial serial position curvesIn all cases serial position curves are based on item scoring (see the text for details)

SERIAL AND FREE RECALL 837

of the form CRP(lag) a | lag |b were fit separately forpositive and negative lags to each participantrsquos data in eachrecall condition The magnitude of the exponent b de-scribes the degree to which the participants exhibit asso-ciative contiguity in recall transitions Across-participantsaverage exponents over Trials 1ndash5 for each of the three re-call conditions appear in Figures 3Dndash3F A 3 (condi-tion) 5 (trial) 2 (transition direction) repeated mea-sures ANOVA was computed on these exponents Asignificant main effect of transition direction [F(111) 16516 MSe 028 p 001] illustrates the forwardasymmetry observed overall In addition a significantmain effect of recall condition [F(222) 5082 MSe 148 p 001] is moderated by a significant interactionof transition direction and recall condition [F(222) 1972 MSe 056 p 001] which indicates that thedegree of asymmetry differs across the three recall con-ditions Further analysis reveals that this interaction isthe result of significant simple effects of transition di-rection for the FRndashconstant [F(111) 1856 MSe 103 p 005] and the SR [F(111) 5534 MSe 069 p 001] conditions but not for the FRndashvariedcondition [F(111) 220 ns]

The changes in the lagndashCRP over trials seen in Fig-ures 3Andash3C for all three recall conditions are confirmedby a significant main effect of trial [F(444) 1601MSe 027 p 001] Significant interactions of trialand transition direction [F(444) 453 MSe 014p 005] and trial and recall condition [F(888) 3273MSe 027 p 001] as well as a marginally signifi-cant three-way interaction of trial transition direction

and recall condition [F(789) 215 MSe 022 p 053] show that the changes in asymmetry over trials dif-fer across the three recall conditions Analysis of simpleeffects reveals that in Trial 1 (panels A1 B1 and C1 ofFigure 3) the three conditions do not differ from one an-other [F(222) 193 ns] and all three conditions showa significant forward asymmetry [F(111) 1889 MSe 033 p 005] as is characteristic in free recall (Howardamp Kahana 1999 Kahana 1996 Kahana et al 2002)Over trials however the three recall conditions showvery different patterns of asymmetry as is illustrated byFigures 3Dndash3F As indicated by the significant interac-tions of trial and transition direction as well as of trialand recall condition the SR condition shows an increas-ing forward asymmetry over trials the FRndashconstant con-dition shows a consistent level of forward asymmetryand the FRndashvaried condition shows a decreasing forwardasymmetry In addition the nonzero level of the averageexponents in Figures 3E and 3F shows that the FRndashconstant and SR conditions show associative contiguityacross all trials whereas Figure 3D shows that associa-tive contiguity disappears over trials in the FRndashvariedcondition By Trial 5 virtually no associative contiguityis evident in the FRndashvaried condition (Figure 3 panel C5)In contrast the exponents in the FRndashconstant conditionremained at a constant level across trials (Figure 3E)and the SR condition forward exponent continued to in-crease over trials (Figure 3F) Although the degree offorward asymmetry differs between the FRndashconstant andthe SR conditions the transitions in the FRndashconstantcondition resemble transitions in the SR condition con-

1864201

8

6

4

201

8

6

4

20

ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5Lag

Con

ditio

nal R

espo

nse

Pro

babi

lity

Trial Number

Exp

onen

t

6

4

2

06

4

2

06

4

2

01 2 3 4 5

FRndashvaried

FRndashconstant

SR

FowardBackward

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

D

E

F

Figure 3 Associative contiguity effects The left panels (AndashC) show the conditional response probability as a functionof lag (lagndashCRP) across each of the five studyndashtest trials (columns 1ndash5) and for each of the three conditions A1ndashA5 freerecall with varied presentation order (FRndashvaried) B1ndashB5 free recall with constant presentation order (FRndashconstant)C1ndashC5 serial recall (SR) To further quantify the changes in contiguity effects across trials and conditions we fit thepower function CRP(lag) a | lag |b to individual participant lagndashCRP functions The rightmost panels (DndashF) show themean power function exponents b as a function of trial number for each of the three conditions D FRndashvaried EFRndashconstant F SR Error bars represent 95 confidence intervals

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

SERIAL AND FREE RECALL 837

of the form CRP(lag) a | lag |b were fit separately forpositive and negative lags to each participantrsquos data in eachrecall condition The magnitude of the exponent b de-scribes the degree to which the participants exhibit asso-ciative contiguity in recall transitions Across-participantsaverage exponents over Trials 1ndash5 for each of the three re-call conditions appear in Figures 3Dndash3F A 3 (condi-tion) 5 (trial) 2 (transition direction) repeated mea-sures ANOVA was computed on these exponents Asignificant main effect of transition direction [F(111) 16516 MSe 028 p 001] illustrates the forwardasymmetry observed overall In addition a significantmain effect of recall condition [F(222) 5082 MSe 148 p 001] is moderated by a significant interactionof transition direction and recall condition [F(222) 1972 MSe 056 p 001] which indicates that thedegree of asymmetry differs across the three recall con-ditions Further analysis reveals that this interaction isthe result of significant simple effects of transition di-rection for the FRndashconstant [F(111) 1856 MSe 103 p 005] and the SR [F(111) 5534 MSe 069 p 001] conditions but not for the FRndashvariedcondition [F(111) 220 ns]

The changes in the lagndashCRP over trials seen in Fig-ures 3Andash3C for all three recall conditions are confirmedby a significant main effect of trial [F(444) 1601MSe 027 p 001] Significant interactions of trialand transition direction [F(444) 453 MSe 014p 005] and trial and recall condition [F(888) 3273MSe 027 p 001] as well as a marginally signifi-cant three-way interaction of trial transition direction

and recall condition [F(789) 215 MSe 022 p 053] show that the changes in asymmetry over trials dif-fer across the three recall conditions Analysis of simpleeffects reveals that in Trial 1 (panels A1 B1 and C1 ofFigure 3) the three conditions do not differ from one an-other [F(222) 193 ns] and all three conditions showa significant forward asymmetry [F(111) 1889 MSe 033 p 005] as is characteristic in free recall (Howardamp Kahana 1999 Kahana 1996 Kahana et al 2002)Over trials however the three recall conditions showvery different patterns of asymmetry as is illustrated byFigures 3Dndash3F As indicated by the significant interac-tions of trial and transition direction as well as of trialand recall condition the SR condition shows an increas-ing forward asymmetry over trials the FRndashconstant con-dition shows a consistent level of forward asymmetryand the FRndashvaried condition shows a decreasing forwardasymmetry In addition the nonzero level of the averageexponents in Figures 3E and 3F shows that the FRndashconstant and SR conditions show associative contiguityacross all trials whereas Figure 3D shows that associa-tive contiguity disappears over trials in the FRndashvariedcondition By Trial 5 virtually no associative contiguityis evident in the FRndashvaried condition (Figure 3 panel C5)In contrast the exponents in the FRndashconstant conditionremained at a constant level across trials (Figure 3E)and the SR condition forward exponent continued to in-crease over trials (Figure 3F) Although the degree offorward asymmetry differs between the FRndashconstant andthe SR conditions the transitions in the FRndashconstantcondition resemble transitions in the SR condition con-

1864201

8

6

4

201

8

6

4

20

ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5ndash5 ndash3 ndash1 1 3 5 ndash5 ndash3 ndash1 1 3 5Lag

Con

ditio

nal R

espo

nse

Pro

babi

lity

Trial Number

Exp

onen

t

6

4

2

06

4

2

06

4

2

01 2 3 4 5

FRndashvaried

FRndashconstant

SR

FowardBackward

A1 A2 A3 A4 A5

B1 B2 B3 B4 B5

C1 C2 C3 C4 C5

D

E

F

Figure 3 Associative contiguity effects The left panels (AndashC) show the conditional response probability as a functionof lag (lagndashCRP) across each of the five studyndashtest trials (columns 1ndash5) and for each of the three conditions A1ndashA5 freerecall with varied presentation order (FRndashvaried) B1ndashB5 free recall with constant presentation order (FRndashconstant)C1ndashC5 serial recall (SR) To further quantify the changes in contiguity effects across trials and conditions we fit thepower function CRP(lag) a | lag |b to individual participant lagndashCRP functions The rightmost panels (DndashF) show themean power function exponents b as a function of trial number for each of the three conditions D FRndashvaried EFRndashconstant F SR Error bars represent 95 confidence intervals

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

838 KLEIN ADDIS AND KAHANA

siderably more than they resemble transitions in the FRndashvaried condition which shows no associative contiguityand no forward bias

DISCUSSION

We considered the relation between two classic learn-ing paradigms serial learning and multitrial free recallIn attempting to settle the question of whether the dif-ferences between these paradigms are due to recall in-structions (free vs serial) or item presentation orderacross trials (varied vs constant) we compared threeconditions free recall with varied presentation orderfree recall with constant presentation order and serial re-call Specifically the repeated contiguity between listitems facilitates learning in the FRndashconstant conditionbeyond what is seen in traditional multitrial free recallWe find that temporal contiguity of study words acrosstrials allows participants to more easily organize listitems than when temporal contiguity is not present lead-ing to superior learning Our results are consistent withthe view that temporally defined interitem associationshelp to guide retrieval in all three recall conditions Thereason for the observed differences between the FRndashconstant and the FRndashvaried conditions is that in the FRndashconstant condition the associations learned on the previ-ous presentations of the list reinforced those on the cur-rent presentation whereas in the FRndashvaried conditionthey competed In the SR condition by requiring partic-ipants to recall the list in the order of study one sees ad-ditional reinforcement of the same temporally defined as-sociations during output Thus FRndashvaried FRndashconstantand SR form a continuum representing the degree towhich contiguity at study and test plays a role in the as-sociative structures formed in memory that ultimatelyguide recall

On their first learning trial the participantsrsquo perfor-mance was comparable to results from single-trial ex-periments in immediate free and serial recall Both freerecall conditions exhibited large recency effects andmore moderate primacy effects whereas in the SR con-dition primacy was much larger than recency The over-all level of recall was higher in the free recall conditionsthan in serial recall and analysis of item-to-item transi-tions revealed significant temporal contiguity effects inall conditions

Considering that the participants were always awareof which condition they were being tested in they couldhave easily adopted different Trial 1 strategies betweenthe two FR conditions However it was not until later tri-als when they could take advantage of the trial-to-trialtemporal contiguity of list items provided in the FRndashconstant condition that performance in the FRndashconstantcondition began to diverge from performance in the FRndashvaried condition Beginning in Trial 2 and continuingthrough Trial 5 FRndashconstant performance graduallytook on similarities to performance in SR In later trialsthe shape of the FRndashconstant serial position curve strongly

resembles the SR curve Each exhibits nearly symmetri-cal primacy and recency effects whereas the FRndashvariedserial position curve shows only recency

Temporal contiguity effects characterized by the lagndashCRP functions shown in Figure 3 decreased dramati-cally over trials in the FRndashvaried condition remainedconstant over trials in the FRndashconstant condition and in-creased asymmetrically in the SR condition By Trial 5temporal contiguity had virtually no effect on recall inthe FRndashvaried condition whereas in the SR conditionthe participants made transitions almost exclusively tothe next item in the presentation order In the FRndashconstantcondition the participants used the constant temporal in-formation to their advantage They made transitions tonearby study positions from negative lags as well asfrom positive lags yielding a ldquolooserdquo serial organizationduring recall

These results follow from a straightforward associa-tive account of list learning In multitrial learning para-digms participants have every reason to utilize associa-tive information encoded on all previous trials duringboth study and recall When the presentation order isconstant across trials the current trial serves to reinforceprevious trials When the list order is shuffled on eachtrial as in the FRndashvaried condition item-to-item transi-tions reflect contributions of previous presentation or-ders and output orders in addition to that of the currentpresentation order Thus the resultant lagndashCRP functionsare flat when calculated using only the current presenta-tion order

Several factors may help to explain why Waugh (1961)did not detect a significant difference between learningcurves in the FRndashvaried and the FRndashconstant conditionsFirst her data set was very small Nine participants eachlearned 6 lists in each condition The present study in-cluded 12 participants each of whom learned 21 exper-imental and 4 practice lists in each recall condition An-other drawback of Waughrsquos study was its lack of controlin the SR condition Although the participants were toldto recall ldquoearly words before late onesrdquo because thestudy used written recall this instruction may not havealways been followed by the participants making it dif-ficult to properly analyze errors in recall The presentstudyrsquos use of recorded vocal recall prohibited the par-ticipants from returning to earlier serial positions laterin the recall period Finally Waugh drew her conclusionsentirely on the basis of learning curves which provide anincomplete picture of the recall process By analyzingrecall transitions the present study directly comparedtemporal contiguity effects in the three recall conditions

The defining characteristic of the serial learning methodis the requirement that participants recall list items intheir presentation order This is not so with free recallthe other principal method for studying list learningacross multiple studyndashtest trials Our results suggest acommon associative basis to both free and serial recallwith temporal contiguity effects being reinforced acrossmultiple trials in serial recall but competing across trials

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=

SERIAL AND FREE RECALL 839

in free recall When presentation order is preserved acrosstrials during free recall (as in our FRndashconstant condi-tion) temporal contiguity is reinforced during the en-coding phase but the opportunity to recall items out oforder often introduces competing associations This con-dition thus reveals the distinct contributions of presenta-tion order and recall instructions on the retrieval process

REFERENCES

Addis K M amp Kahana M J (2004) Decomposing serial learningWhat is missing from the learning curve Psychonomic Bulletin ampReview 11 118-124

Anderson J R Bothell D Lebiere C amp Matessa M (1998)An integrated theory of list memory Journal of Memory amp Language38 341-380

Becker S amp Lim J (2003) A computational model of prefrontal con-trol in free recall Strategic memory use in the California verballearning task Journal of Cognitive Neuroscience 15 821-832

Botvinick M amp Plaut D C (2004) Doing without schema hierar-chies A recurrent connectionist approach to routine sequential actionand its pathologies Psychological Review 111 395-425

Brown G D A Preece T amp Hulme C (2000) Oscillator-basedmemory for serial order Psychological Review 107 127-181

Cowan N Baddeley A D Elliott E M amp Norris J (2003) Listcomposition and the word length effect in immediate recall A com-parison of localist and globalist assumptions Psychonomic Bulletinamp Review 10 74-79

Dallett K M (1963) Practice effects in free and ordered recall Jour-nal of Experimental Psychology 66 65-71

Drewnowski A amp Murdock B B (1980) The role of auditory fea-tures in memory span for words Journal of Experimental Psychol-ogy Human Learning amp Memory 6 319-332

Farrell S amp Lewandowsky S (2002) An endogenous distributedmodel of ordering in serial recall Psychonomic Bulletin amp Review 959-79

Howard M W amp Kahana M J (1999) Contextual variability andserial position effects in free recall Journal of Experimental Psy-chology Learning Memory amp Cognition 25 923-941

Howard M W amp Kahana M J (2002) A distributed representationof temporal context Journal of Mathematical Psychology 46 269-299

Hull C L Hovland C I Ross R T Hall M Perkins D T ampFitch F B (1940) Mathematico-deductive theory of rote learningA study in scientific methodology New Haven CT Yale UniversityPress

Hulme C Stuart G Brown G D A amp Morin C (2003) High-and low-frequency words are recalled equally well in alternating listsEvidence for associative effects in serial recall Journal of Memory ampLanguage 49 500-518

Johnson G J (1991) A distinctiveness model of serial learning Psy-chological Review 98 204-217

Jung J amp Skeebo S (1967) Multitrial free recall as a function ofconstant versus varied input orders and list length Canadian Journalof Psychology 21 329-336

Kahana M J (1996) Associative retrieval processes in free recallMemory amp Cognition 24 103-109

Kahana M J amp Caplan J B (2002) Associative asymmetry inprobed recall of serial lists Memory amp Cognition 30 841-849

Kahana M J Howard M W Zaromb F amp Wingfield A (2002)Age dissociates recency and lag recency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 28530-540

Murdock B B (1962) The serial position effect of free recall Jour-nal of Experimental Psychology 64 482-488

Nairne J S Neath I Serra M amp Byun E (1997) Positional dis-tinctiveness and ratio rule in free recall Journal of Memory amp Lan-guage 37 155-166

Raaijmakers J G W (2003) Spacing and repetition effects in humanmemory Application of the SAM model Cognitive Science 27 431-452

Raskin E amp Cook S W (1937) The strength and direction of asso-ciations formed in the learning of nonsense syllables Journal of Ex-perimental Psychology 20 381-395

Ward G Woodward G Stevens A amp Stinson C (2003) Usingovert rehearsals to explain word frequency effects in free recall Jour-nal of Experimental Psychology Learning Memory amp Cognition 29186-210

Ward L B (1937) Reminiscence and rote learning PsychologicalMonographs 49 64

Waugh N C (1961) Free versus serial recall Journal of Experimen-tal Psychology 62 496-502

Wilson M (1988) MRC Psycholinguistic Database Machine-usabledictionary version 200 Behavior Research Methods Instruments ampComputers 20 6-10

NOTE

1 The lagndashCRP estimates the probability of a transition to a given lagfor successively recalled list items conditional on which list items areavailable for recall Consider the recall of item i followed by the recallof item j The lag of the transition is calculated as k j i A tally ofsuch transitions accumulated over output positions measures the ob-served pattern of recall transitions as a function of lag However theavailability of a transition of a given lag depends on the serial positionof the current recalled item i Transitions cannot extend beyond theboundaries of the list Therefore if the current recalled item is the lastlist item a transition of lag k 1 would not be possible To correctfor the difference in availabilities of various lags we conditionalize thistally by dividing the number of observed transitions n(k) by the num-ber of possible transitions d(k) In addition because repetitions in re-call are extremely rare transitions to list items that have already beenrecalled on the current trial are not counted in either the numerator orthe denominator The lagndashCRP is then calculated across lists for eachparticipant as

Unlike Kahana (1996) we collapse across all output positions We pres-ent lagndashCRPs that represent the across-participants average values

(Manuscript received February 10 2004revision accepted for publication August 17 2004)

CRP( )( )( )

kn kd k

=