Primacy and Recency in the Continuous Distractor Paradigm

Journal of Experimental Psychology:Human Learning and Memory1977, Vol. 3, No. 5, 560-571

Primacy and Recency in the Continuous Distractor Paradigm

Steven E. Poltrock and Colin M. MacLeodUniversity of Washington

Three experiments investigated serial position effects in immediate and finalfree recall. Each word in a 10-item list was both preceded and followed bya 15-sec period of distraction activity. In Experiment 1, half of the lists wereimmediately followed by either a recall test or a recognition test; the remain-ing lists were not tested, but were followed by a different distraction activity.After presentation of all lists, a final recall or recognition test was given.Primacy was observed only in immediate free recall and in all final testsfollowing immediate free recall, demonstrating that primacy develops from afree-recall storage strategy. No recency was observed in Experiment 1. InExperiment 2, every list was followed by an immediate free-recall test, witha final free-recall test after the last list. The primacy results of Experiment 1were replicated. Furthermore, the appearance of recency in Expriment 2 sug-gests that recency results from a retrieval strategy that failed to develop inExperiment 1 because some lists were not tested immediately. To eliminatean artifact account, Experiment 3 used an experimenter-paced distractor taskand replicated the findings of Experiment 2, which used a subject-paced dis-tractar task. Contrary to previous claims, the pattern of results in the con-tinuous distractor paradigm is seen as completely consistent with the accountoffered by multistore models of serial position effects in standard free recall.

A crucial test of any model of humanmemory is its ability to account for theserial position curve obtained in single-trialfree recall. Multistore models have success-fully accounted for serial position effectsthrough the operation of short-term andlong-term stores (Atkinson & Stuff rin,1968; Waugh & Norman, 1965). Briefly,multistore models explain the familiarbowed curve as follows: Superior retentionof the first few list items (the primacyeffect) results from differential storage. The

Both authors contributed equally, with the flipof a coin determining the order of authorship.This research was partially supported by Na-tional Institute of Mental Health Grant MH21795 to Earl Hunt. We are grateful for thehelpful comments of Veronica Dark, Marcy Lans-man, and our two reviewers.

Steven E. Poltrock is now at the University oiDenver.

Requests for reprints should be sent to ColinM. MacLeod, Department of Psychology, NI-2S,University of Washington, Seattle, Washington98195.

probability of entering an item in long-termstore is directly related to the amount ofprocessing given to that item, and theamount of processing per item decreases toan asymptote as the number of presenteditems increases (cf. Rundus, 1971). Thus,midlist items are poorly recalled in com-parison to primacy items. Superior retentionof terminal list items (the recency effect) isattributed to readout of the last few itemsfrom short-term store. The absence of re-cency when a filled delay separates studyand test (Glanzer & Cunitz, 1966; Postman& Phillips, 1965) supports this account.

Until recently, the multistore explanationhas fared exceptionally well in handling thelarge serial position literature. However, amodification of the free-recall procedure thatwe call the continuous distractor paradigmappears to pose a problem for this account.In fact, from their investigations using thisparadigm, Bjork and Whitten (1974) haveconcluded that "the customary two-processtheoretical account of immediate free recallis certainly incomplete, if not wrong" (p.

560

SERIAL POSITION IN CONTINUOUS DISTRACTION 561

189). This conclusion notwithstanding, in-consistencies in previous findings describedbelow suggest that the processes involved inthe continuous distractor paradigm remaininadequately understood. In the present ar-ticle, we seek to resolve these inconsistenciesand to account for serial position effects inthe continuous distractor paradigm throughsubject strategies that are consistent withtwo-process theories.

The continuous distractor paradigm wasinvented by Whitten and Bjork (Note 1) toexamine certain within-list effects freedfrom the confounding of serial position ef-fects. The only difference between the con-tinuous distractor paradigm and the stan-dard free-recall paradigm is in the mannerin which lists are studied. Basically, theparadigm involves preceding and followingeach item in the list by 12-20 sec of dis-tractor activity like that typically used inthe Brown-Peterson paradigm. This proce-dure has the effect of embedding each itemin an otherwise uninterrupted distractortask, thereby isolating one item from thenext. The isolation is further accentuatedby instructions encouraging the subject torehearse only the current item. Each ofseveral lists is followed by an immediatefree-recall test on only that list. As with thestandard free-recall paradigm, after all ofthe lists have been studied and tested, anunexpected final test over all of the itemsmay be administered.

The aim of the continuous distractor pro-cedure was to prevent operation of the pro-cesses proposed by multistore models to ac-count for serial position effects. Consider theserial position function for immediate freerecall. If subjects obey the instructions torehearse only the current item, then eachitem should receive the same amount ofprocessing and therefore have an equalchance of entering long-term store. Thus,the differential storage responsible for pri-macy in free recall should be eliminated;there should be no effect of serial positionat the beginning of the list. Furthermore,at the end of the list, the distractor activ-ity separating list items should eliminate

the opportunity for recall of items fromshort-term store; no effect of serial positionshould be found at the end of the list. Inshort, the serial position curve should beessentially flat in the continuous distractorparadigm. Instead, in three separate studiesusing this paradigm (Bjork & Whitten,1974; Tzeng, 1973; Whitten & Bjork, Note1) the familiar bowed serial position curvehas emerged completely intact!

The predictions regarding serial positionin final free recall are identical to the pre-dictions for immediate free recall. The onlyadditional stipulation required is that thenegative recency observed in the usual free-recall paradigm (Craik, 1970) should notappear, because negative recency relies onthe short-term readout strategy preventedby continuous distraction. Therefore, theserial position function for the final testshould also be flat. Here, the results ob-tained by previous investigators are incon-sistent. Whereas Bjork and Whitten (1974)obtained virtually flat serial position curvesin final free recall, Tzeng (1973) foundbowed curves identical in shape to thoseobtained in immediate recall (but depressedin terms of proportion correct). However,this inconsistency may have resulted fromprocedural differences. Bjork and Whittenpresented eight lists with 10 word pairs ineach, so final recall required retrieval of160 words. Tzeng argued that the proce-dure used by Bjork and Whitten led to ex-cessive output interference in final recallsuch that the resulting floor effect maskedserial position effects. Consequently, Tzeng'sresults were obtained by requiring final re-call of only 40 words. Because his studywas designed to remedy this interferenceproblem, Tzeng's final recall results areprobably more indicative of the processesunderlying the continuous distractor phe-nomenon.

How should this pattern of results beinterpreted? Consider recency first. Positiverecency in immediate recall appears to beinconsistent with the explanations givenabove for recency in the standard free-recallparadigm. On the basis of finding recency

562 STEVEN E. POLTROCK AND COLIN M. MAcLEOD

in both immediate and final recall, Tzeng(1973) concluded that recency results fromthe operation of a "long-term retrievalprocess." Although Tzeng fails to specifythis retrieval process, there is a more criti-cal problem with his conclusion. A storageprocess could also account for Tzeng's re-sults. In fact, there are no means for sepa-rating the effects of storage and retrievalprocesses in Tzeng's experiment.

Bjork and Whitten (1974) also con-cluded that recency results from a retrievalprocess, but for the opposite reason. Find-ing no recency in final free recall in Experi-ment 2, they reasoned that recency couldnot be due to differential storage but mustresult from a retrieval process. This con-clusion relies on the questionable findingthat recency is absent in final recall, a find-ing requiring acceptance of the null hypoth-esis. Given that output interference mayhave masked serial position effects, Bjorkand Whitten also failed to present resultsthat separate the effects of storage and re-trieval processes.

Now consider primacy. While Tzengfound primacy in both immediate and finalrecall, he offered no explanation for hisfinding. Bjork and Whitten, on the otherhand, found primacy only in immediate re-call and therefore attributed primacy to theoperation of retrieval processes. Again, theirconclusion is based on the questionable find-ing of no serial position effect in final recall.If Tzeng's final recall results are acceptedinstead, then primacy may result from eitherstorage or retrieval processes.

In summary, neither Tzeng (1973) norBjork and Whitten (1974) obtained resultsindicating whether storage or retrieval pro-cesses are responsible for the observed serialposition effects. In this article we reportthree experiments designed to reveal theprocesses underlying primacy and recencyin the continuous distractor paradigm. Ourfirst experiment provides an independentassessment of the contributions of storageand retrieval processes by using recall and •recognition tests and by initially testing onlyhalf the lists. The second experiment focuses

on the recency effect as a possible outgrowthof a developing output strategy. In the thirdexperiment, the possibility that subject-paced distractor tasks have produced arti-factual primacy and recency is tested.Finally, in the general discussion section,we present an overall account of serial posi-tion effects in the continuous distractorparadigm in terms of processing strategiesoperating in free recall.

Experiment 1

Previous research with the continuousdistractor paradigm has almost exclusivelyused the free-recall test procedure, whichdepends on both storage and retrieval pro-cesses. To the extent that recognition testsminimize (Kintsch, 1968; Murdock, 1968)or eliminate (Anderson & Bower, 1972)the retrieval component, the use of a recog-nition test permits assessment of the con-tribution of primarily storage processes toserial position effects. If a serial positioneffect is observed both in free recall and inrecognition, then storage processes are im-plicated. However, when an effect is foundonly in free recall and not in recognition,then retrieval processes are more likely tobe the source. In Experiment 3 of theirstudy, Bjork and Whitten (1974) did in-clude immediate recognition tests. Theirfinding of recency in immediate recall butnot in immediate recognition providesstrong inductive support for the hypothesisthat recency in free recall is due to a re-trieval strategy.

Comparisons of performance on immedi-ate and final tests can also help to reveal theprocesses responsible for serial position ef-fects. For instance, a long-term storageprocess should affect both immediate andfinal performance, while a short-term re-trieval process should affect only immediateperformance. One problem with compari-sons of immediate and final test perform-ance is that performance on the final testmay be affected by performance on the im-mediate test (cf. Dark & Loftus, 1976; Mo-digliani, 1976). Consequently, to permit un-confounded comparisons of immediate and


final tests, only half of the lists in Experi-ment 1 were initially tested. Final perform-ance on the nontested lists should be un-contaminated, and a comparison of finalperformance for tested and nontested listsshould indicate the extent to which immedi-ate testing contaminated final performance,

In Experiment 1, immediate and finalrecall and recognition tests were combinedfactorially between subjects. Each subjectreceived the same type of immediate test onthree of six lists and no immediate test onthe three remaining lists. Thus, there werefour combinations of testing: (a) immediatefree recall-final free recall, (to) immediatefree recall-final recognition, (c) immediaterecognition-final free recall, and (d) imme-diate recognition-final recognition.

Examination of the serial position func-tions obtained in these four types of test(i.e., immediate and final recall and recog-nition) should reveal the processes respon-sible for serial position effects. If a serialposition effect, either primacy or recency,results from a long-term retrieval processas Tzeng (1973) argued, then on both im-mediate and final tests that effect should bemaximized in free recall and minimized inrecognition because retrieval processes playa lesser role in recognition. If a serialposition effect is due to a transient retrievalprocess as Bjork and Whitten (1974) ar-gued, then the effect should be observedonly in immediate free recall. If differentialstorage operates while a list is being studiedin the continuous distractor paradigm, anyresulting serial position effects should beobserved on all tests, since a storage processwill influence both recall and 'recognitionperformance. Finally, consider the possibil-ity that a serial position effect could resultfrom a storage process used during studyfor a particular type of test, recall or recog-nition (cf. Tversky, 1973). If a test-specificstorage strategy were responsible for a serialposition effect, then the effect should beobserved in only that type of immediatetest, Furthermore, the effect should be ob-served on both types of final test, but onlyfor those subjects who studied for that typeof immediate test.

MethodSubjects. The subjects were 60 students in

introductory psychology at the University ofWashington, who participated for extra coursecredit. For purposes of testing, subgroups of 1 to4 subjects were assigned randomly to one of thefour experimental conditions, with the constraintthat 15 subjects were assigned to each condition.

Stimuli and apparatus, A pool of ISO wordswas selected from the Paivio, Yuille, and Madigan(1968) norms; all were common A and AAwords from four to seven letters long, Choice ofstudy word lists, recognition test lists, and dis-tractor materials was under the control of a pro-gram implemented on a NOVA 820 computer.The program randomly selected six unique 10-word lists for each subject from the word pool.Distractor words for immediate and final recog-nition were randomly selected from the remaining90 words, The recognition tests were YES-NOand used all list items plus an equal number ofdistractors, presented in random order.

Stimuli were presented to subjects in individ-ual soundproof booths on Tektronix 602 displayscopes under computer control. Recall tests werewritten; recognition test responses were collectedby the computer,

Procedure, The four experimental conditionsconsisted of all factorial combinations of imme-diate and final test type (recall or recognition, inboth instances). Of the six lists studied by eachsubject, three were tested immediately and threewere not tested immediately. Tested lists wereselected randomly with the requirement that notmore than two consecutive lists could be tested ornontested.

Each of the 10 words in a list was presentedfor l.S sec. Before the first word and after everyword in the list, a three-digit number was pre-sented that stayed on the screen for IS sec, Theinstructions were to memorize each word as itappeared and to count backward by threes asrapidly as possible from the number, beginningas soon as the number appeared. The subject wastold that several lists would be presented in thisway but that only certain randomly selected listswould be tested. No mention was made of thefinal teat. Furthermore, no instructions regardingrehearsal were given, since Brodie and Prytulak(197S) have shown that these instructions are notas effective as had been previously assumed.

On nontested lists, subjects spent 1 rain shadow-ing digits presented at the rate of two per second.To indicate nontested lists, the computer dis-played the words shadow digits. Subjects in thetwo immediate-recall conditions were given free-recall instructions and were told that the wordrecall following a list was a signal that duringthe next minute they were to write down as manywords as they could remember from the currentlist, Subjects in the two immediate recognitionconditions were given YES-NO recognition in-

564 STEVEN E. POLTROCK AND COLIN M. MACLEOD

1.00

,90

.80

.70

.60

,90

.40

.30

.20 RECALL-

1 2 3 4 B 8 7 8 8 1 0

SERIAL POSITION

Figure 1. Proportion correct in immediate free re-call and immediate recognition as a function ofinput serial position in Experiment 1.

structions and were told that the word recogni-tion would signal a tested list. Subjects indicatedtheir responses by key presses; a 3-sec limit wasimposed on response time to permit the test tobe completed in 1 min. No feedback was pro-vided following tests, and the recall protocolswere removed after each test,

A phony debriefing, lasting approximately 2min, followed the last immediate test (or digit-shadowing) period, Following this, instructionswere given for the final test, half of the subjectsswitching to the other type of test and half con-tinuing with the same type of test. Final recog-nition tests were computer-paced at 5 sec/word;final recall tests were subject-paced.

Results and Discussion1

Immediate tests. Figure 1 presents theserial position curves for proportion correctin immediate recall and recognition, eachcurve based on the performance of 30 sub-jects. The results are straightforward: Asexpected, recognition performance was con-siderably superior to recall performance,as indicated by the highly significant maineffect of test type, F(l, 58) = 137.25, MSe

= 1.46. Most important, there was also ahighly significant effect of serial position,F(9, 522) = 3.68, MS* = .60, and a signifi-cant Test Type X Serial Position interac-tion, F(9, 522) = 3.09. Together, the serialposition effects reflected the different pat-terns of serial position in immediate recall

and recognition. Immediate recognition per-formance increased slightly over serial posi-tion, while immediate free recall demon-strated a strong primacy effect and no re-cency effect.

The presence of primacy in free recall butnot in recognition replicates Bjork andWhitten's (1974) findings and requiresrejection of the hypothesis that primacyresults from a differential storage processthat necessarily occurs during study in thecontinuous distractor paradigm. However,these data are consistent with two of thehypotheses advanced earlier, that primacyresults from some type of retrieval strategy(whether long-term or short-term) and thatprimacy results from a storage strategy usedspecifically during study for a free-recalltest. Discrimination between these hypoth-eses must await examination of performanceon the final tests.

The surprising aspect of the immediatetest results is the complete absence of re-cency in immediate free recall. Perhaps re-cency depends on the development of a free-recall strategy, as Wing and Thomson(1965) found in the standard free-recallprocedure. Two procedural differences be-tween our paradigm and that of earlier in-vestigators may have prevented that strat-egy from developing. First, the presence ofthe nontested lists may have been disruptive.Second, the omission of item-by-item re-hearsal instructions may have been critical(cf. Brodie & Prytulak, 1975). We examinethese possibilities in Experiment 2.

Final tests. Figure 2 presents the serialposition curves for final recall and recogni-tion as a function of the type of immediatetest. The results for tested and nontestedlists are presented in the upper and lowerpanels of Figure 2, respectively. A 2 X 2 X2 X 10 analysis of variance was conductedon the number of correct responses in thefinal tests. The two between-subjects factorswere type of immediate test and type of final

J-A11 significant statistics are reported for atleast p < .05; all nonsignificant statistks representa p > .10.


test (recall or recognition, in both in-stances). The two within-subjects factorswere serial position (10-word lists) andwhether a given list was tested or nontestedinitially,

The main effects of immediate and finaltest type were both significant. Obviously,final recognition performance was far betterthan final recall performance, F(l, 56) =577.25, MSt = 1.43. Also, final performancewas generally better following study for animmediate recall test rather than for an im-mediate recognition test, F(l, 56) =4.93.However, most of the advantage of priorrecall learning is in final recall, not in finalrecognition, as indicated by the significantinteraction of immediate and final test type,F(l, 56) =6.54.

Our greatest interest in final performanceis in the effects of serial position. However,we first point out that, as expected fromother studies of the effects of test trials onretention (Birnbaum & Eichner, 1971; Dar-ley & Murdock, 1971; Hogan & Kintsch,1971), previously tested items were betterretained than previously nontested items,F(l, 56) = 46.79, MS6 = .73. This findingis qualified by two significant interactions,the Immediate Test Type X Tested-Non-tested interaction, F(l, 56) = 18.79, andthe Immediate Test Type X Final TestType X Tested-Nontested interaction, F(l,56) = 14.90. The interpretations of theseinteractions are necessarily post hoc and arenot relevant to the issues considered in thisarticle. The important finding is that all theinteractions of the tested-nontested variablewith serial position were nonsignificant (allFs^l.24), indicating that the immediatetests did not influence the shape of the serialposition functions obtained in final tests.8

Thus, results are reported below for bothtested and nontested lists.

Finally, we turn to the effects of serialposition. The main effect of serial positionwas highly significant, F(9, 504) = 5.64,MSe = .47, as was the Immediate TestType X Serial Position interaction, F(9,504) = 3.75. The only serial position effectsevident in Figure 2 are primacy effects oc-

0 102 3 4 8 6 7SERIAL POSITION

Figure 2. Proportion correct in final recognition(FRN) and final free recall (FFR) as a functionof input serial position and type of immediatetest, recognition (IRN) or recall (IFR) in Ex-periment 1. (The top panel represents tested lists;the bottom panel represents nontested lists.)

curring in every condition involving imme-diate free-recall tests. The absence of anyother interactions with serial position (allFs ^ 1.24) indicates that primacy did notdepend on the type of final test nor onwhether lists were originally tested. To theextent that recognition is a retrieval-mini-mizing process, a primacy effect in finalrecognition cannot reflect a retrieval proc-ess. Therefore, contrary to Bjork and Whit-ten (1974), primacy must reflect differential

2 After immediate recall testing, both types offinal test show a slight "negative recency" fornontested lists. Although nonsignificant, we thankone of our reviewers for noting this trend, whichis consistent with findings in standard free recall(cf. Craik, 1970).


FINAL

1 2 3 4 5 8 7 8 9 1 0

SERIAL POSITIONFigure 3. Proportion correct in immediate andfinal free recall as a function of serial position inExperiment 2.

storage resulting from a strategy used dur-ing study for free-recall tests (cf. Tversky,1973). This result is consistent with thenotion proposed by Loftus (1971) that stor-age differences account at least in part fordifferences in recall and recognition per-formance.

Experiment 2

The most surprising finding in Experi-ment 1 was the absence of recency, a resultconflicting with those of Bjork and Whitten(1974) and Tzeng (1973). To examinethis finding further, Experiment 2 was areplication of Tzeng's study, but withoutitem-by-item rehearsal instructions. Thepurpose of Experiment 2 was to assure thatrecency is a replicable finding in the con-tinuous distractor paradigm when all listsare tested.

Brodie and Prytulak (1975) have shownthat the instruction to rehearse each itemequally leads to increased recency in stan-dard free recall. Perhaps, then, this instruc-tion accounts for the recency observed byTzeng (1973) and Bjork and Whitten(1974) in the continuous distractor para-digm. If so, recency should still be absentin Experiment 2, since no rehearsal instruc-tions were given. On the other hand, if

recency stems from a developing retrievalstrategy that was impeded by nontested listsin Experiment 1, then recency should ap-pear in Experiment 2 despite the absenceof rehearsal instructions. Furthermore, adeveloping retrieval strategy should lead toan increase in recency over lists in imme-diate recall, as demonstrated in standardfree recall by Wing and Thomson (1965).

Method

The subjects were 22 students from the samepool as in Experiment 1 who had not participatedin that experiment. The apparatus and stimulus-word pool were unchanged from Experiment 1.The procedure was very similar to that of Ex-periment 1, with the following three changes: (a)Each subject studied only four 10-word lists, (b)every 10-word list was tested immediately, and(c) all tests, immediate and final, were free-recalltests. To maintain comparability with Experiment1, the four immediate tests each lasted 1 min,while the final test was subject-paced.

Results and Discussion

The serial position curves for immediateand final recall are presented in Figure 3.A 2 X 10 analysis of variance was con-ducted that involved two within-subjectsfactors, time of test (immediate vs. final)and serial position (10 items). As expected,immediate recall was better than final re-call, F(l, 21) =64.09, MSe = A2. Moreimportant, the main effect of serial positionwas highly significant, F(9, 189) = 5.78,MSe= 1,24, as was the Time of Test XSerial Position interaction, F(9, 189) =2.68, MS6 = .27.

The interpretation of these results isstraightforward. The presence of strong pri-macy in both immediate and final recallreplicates the findings of Experiment 1,as well as those of Tzeng (1973), andstrengthens the argument that primacy isthe result of a storage phenomenon. Moreimportant, the presence of recency in imme-diate free recall also replicates Tzeng, butstands in sharp contrast to the absence ofrecency in Experiment 1. Thus, the recencyeffect in immediate recall is replicable in thecontinuous distractor paradigm even with-out item-by-item rehearsal instructions.


However, one necessary condition for re-cency to appear is a relatively continuoussequence of immediate recall tests, Appar-

• ently, recency results from a strategy thatdevelops after practice at recall, a notion towhich we will return shortly.

Other aspects of the results presented inFigure 3 are quite different from those ob-tained by Tzeng. Compared with Tzeng's(1973) results for immediate recall, thepresent results demonstrate much less re-cency and more primacy. Our results infinal recall are also quite different fromTzeng's; he found a positive recency effect,whereas Figure 3 reveals no recency in finalrecall. This lack of recency in final recallsupports the argument that recency in im-mediate recall resulted from a transient re-trieval process, at least in our experiment.On the other hand, Tzeng's finding of re-cency in final recall could have resultedfrom a differential storage process or per-haps from a long-term retrieval process, ashe argued. However, we must point out thatthe meaning of the concept "long-term re-trieval process" is badly in need of explica-tion. Our experiment differs in procedurefrom Tzeng's experiment primarily in theabsence of rehearsal instructions, Item-by-item rehearsal probably decreases the likeli-hood of implementing a short-term readoutstrategy, thereby improving storage of ter-minal list items. This interpretation is con-sonant with the results of Brodie and Pry-tulak (1975) for standard free recall. Fur-thermore, the fact that terminal items arerecalled correctly in the immediate test moreoften than items at other positions providesan additional benefit in storage. This betterstorage should in turn result in the greaterrecency observed by Tzeng on final free-recall tests.

Comparison of immediate and final testperformance suggests that a retrieval pro-cess is responsible for recency in immediaterecall, whereas comparisons with Experi-ment 1 suggest that recency depends onfree-recall experience. If recency resultsfrom a developing retrieval strategy, thenrecency should increase over lists. To evalu-

ate this prediction, an analysis of variancewas conducted on immediate-recall perform-ance with list (1-4) as a factor. The maineffect of list was nonsignificant (P < 1),but the effect of list on recency was con-firmed by the significant List X Serial Posi-tion interaction, F(27, 567) = 2.12, MS* =.21, and the significant serial position effect,F(9, 189) =5.74, MS1. = .21. Inspectionof individual list serial position functionsindicates that recency increases steadily overlists, whereas primacy decreases only be-tween Lists 1 and 2, This result supportsthe prediction regarding increasing recencyover lists.

Goodwin (1976) has found very similarchanges in primacy and recency over listsin standard free recall. He attributes thedecrease in primacy to the buildup of pro-active interference and the increase in re-cency to an early-output retrieval strategythat develops over lists. Thus, Goodwinmaintains that the gains in recency andlosses in primacy represent coincident inde-pendent processes rather than aspects of asingle process. Unfortunately, considerablevariance in output order in the present ex-periment prevented observation of any con-sistent relationship between recency andoutput order, but in other respects Good-win's interpretation is consistent with ourfindings, The relationship between outputorder and increasing recency is consideredfurther in Experiment 3.

Experiment 3

Both Experiments 1 and 2 used a subject-paced backward-counting task as the con-tinuous distractor activity during list acqui-sition, as Tzeng (1973) had done previously.Such tasks permit the subjects to allocateprocessing effort (cf. Kahneman, 1973) dif-ferentially to the list-learning and distractortasks at their discretion. Conceivably, then,the serial position effects observed in thecontinuous distractor paradigm could resultfrom less attention begin given to the dis-tractor task and more to the items them-selves at the beginning and end of a list.


1.00

SERIAL POSITION

Figure 4. Proportion correct in immediate andfinal free recall as a function of serial positionin Experiment 3.

An informative procedural change wouldbe the substitution of an experimenter-paceddistractor task for the subject-paced task.8

Here, the processing demands of the dis-tractor task would be constant across serialpositions. Any serial position effects ob-tained could not subsequently be ascribed toshifting resource-allocation strategies withinthe distractor segments (e.g., Norman &Bobrow, 1975), but must be due to ongoinglist-learning strategies. Consequently, Ex-periment 3 used a rapid, experimenter-paceddistractor task (digit shadowing at the rateof 3/sec) in 15-sec segments to examine thepossibility that previous attempts at contin-uous distraction may not have been entirelysuccessful in eliminating procedural arti-facts.

MethodThe subjects were 22 students from the same

pool as Experiments 1 and 2 who had not par-ticipated in the earlier experiments. The appara-tus and stimulus-word pool were identical to thosein Experiments 1 and 2. Additionally, the proce-dures of Experiments 2 and 3 were very similar,with only the following two changes: (a) Eachsubject studied six 10-word lists (as in Experi-ment 1), and (b) a number-shadowing task re-placed the backward-counting task. In the num-ber-shadowing task, subjects were required torepeat aloud single digits as they appeared; threerandomly selected digits were presented eachsecond.

Results and Discussion

The serial position functions for immedi-ate and final recall in Experiment 3 arepresented in Figure 4, A 2 X 10 analysis ofvariance was conducted with time of test(immediate vs. final) and serial position(10-item lists) as within-subjects factors.The analysis revealed the-same pattern ofresults as obtained in Experiment 2. Again,immediate recall was better than final recall,F(l, 21) = 10276, MS1. = 2.11. Of moreimportance, the main effect of serial positionwas highly significant, F(9, 189) = 8.85,M5"e=1.89, as was the Time of Test XSerial Position interaction, F(9, 189) =4.32, MSe = .62.

The serial position curves for Experiment3 (Figure 4) are very similar to those forExperiment 2 (Figure 3), although per-formance is somewhat better overall in Ex-periment 3. Because the same pattern ofresults emerges in both experiments, it isclear that serial position effects in the con-tinuous distractor paradigm are not artifactsof a changing allocation strategy. Processingeffort across serial position on the distractortask is roughly constant when the task isexperimenter-paced, yet the same serialposition pattern emerges for the list itemsas when distraction is subject-paced. Speci-fically, primacy appears in both immediateand final recall, with recency only in imme-diate recall, Again, the evidence is in accordwith the hypothesis advanced earlier thatprimacy is a storage phenomenon and re-cency is a retrieval phenomenon.

In Experiment 2, examination of changesin serial position over lists indicated thatprimacy decreased and recency increased.Although the same pattern of changes inserial position occurred in Experiment 3,an analysis of variance indicated that theList X Serial Position interaction was non-

"Bjork and Whitten (1974) did use an experi-menter-paced distractor task, consisting of simplearithmetic problems presented at a 2-sec rate.However, other procedural differences (e.g., theiruse of word doubles) complicate comparisons be-tween studies,


significant, F(45, 945) = 1.22, MS* = .21.The better overall performance in Experi-ment 3 than in Experiment 2 permits eval-uation of the early-output account of de-veloping recency. Analyses of output orderindicate that terminal list items were indeedrecalled progressively earlier over lists inExperiment 3. The average output per-centiles (see Bjork & Whitten, 1974) ofthe last 3 items in Lists 1 through 6, re-spectively, were 79, 71, 60, 60, 61, and 57.Taken together, then, the results of Experi-ments 2 and 3 provide evidence that re-cency in the continuous distractor paradigminvolves a retrieval strategy of the sort pro-posed by Goodwin (1976) in his study ofstandard free recall.

General Discussion

The three experiments reported in thisarticle replicate and extend the finding ofserial position effects in the continuous dis-tractor paradigm (Bjork.& Whitten, 1974;Tzeng, 1973). As mentioned earlier, thegoal of this paradigm was to eliminate thesources of serial position effects in standardfree recall (Whitten & Bjork, Note 1).However, despite the marked proceduraldifferences in the two paradigms, the patternof results is virtually identical. Two con-clusions are possible, given this outcome.First, assuming the distraction procedurewas successful, multistore models cannotaccount for the unanticipated serial positioneffects. In accord with this conclusion,Tzeng (1973) and Bjork and Whitten(1974) proposed that their observed pri-macy and recency were the result of un-specified long-term retrieval processes. Thesecond conclusion provides a more par-simonious interpretation. Assuming that thedistraction procedure (including the re-hearsal instructions) was not successful,multistore models account for the serial po-sition effects in both paradigms withoutmodification. Unfortunately, the critical as-sumption that the procedure failed is notdirectly testable within the continuous dis-tractor paradigm. We therefore examine the

evidence from other studies of distractionactivity and rehearsal instructions.

There are several reasons why, a priori,the continuous idistractor paradigm mightbe expected to fail. First, consider the sepa-rate effects of rehearsal instructions and ofdistraction on storage processes in long-term store. Fischler, Rundus, and Atkinson(1970) have shown that primacy in stan-dard recall is reduced but not eliminatedwhen subjects are told to rehearse only thecurrent item. In replicating this finding,Brodie and Prytulak (1975) observed thatsubjects covertly use a differential rehearsalstrategy even while overtly rehearsing onlythe current item. Thus, item-by-item re-hearsal instructions do not eliminate pri-macy resulting from differential storage.Furthermore, the effectiveness of distrac-tion in preventing rehearsal is questionable;Reitman (1974) has shown that subjects dorehearse during distraction, despite instruc-tions to avoid rehearsal. Taken together,these studies suggest that a differential stor-age strategy could be adopted in the con-tinuous distractor paradigm despite the on-going distraction.

Now consider the effects of rehearsal in-structions and distraction on retrieval ofrecency items from short-term store. Cer-tainly, longer periods of distraction resultin reduced recall from short-term store(Peterson & Peterson, 1959). Furthermore,a single terminal distractor period elimi-nates recency completely in standard freerecall (Glanzer & Cunitz, 1966; Postman &Phillips, 1965). But Murdock (1961) andMelton (1963) demonstrated that forgettingfrom short-term store under distraction con-ditions decreases as the number of items inthat store decreases. Thus, in the continu-ous distractor paradigm, retrieval fromshort-term store may not be prevented iffew items are held there. In fact, instruc-tions to rehearse only the current itemshould encourage such a strategy.

These considerations suggest that the dis-traction procedure was not successful, andmultistore models can account for serialposition effects in the continuous distrac-


tion paradigm. Furthermore, several aspectsof the results of Experiments 1-3 suggestthat the same processes are responsible forserial position effects in the continuous dis-tractor paradigm as in standard free recall.Experiment 1 demonstrated that primacyresults from a storage strategy operatingduring istudy for a free-recall test. The ab-sence of recency in Experiment 1 and inthe final recall of Experiments 2 and 3 sug-gests that recency results from a short-termretrieval process. Recency tended to increaseover lists in both Experiments 2 and 3, andin Experiment 3 the increase in recencywas accompanied by earlier output of theterminal list items, supporting the hypothe-sis that a developing retrieval strategy isresponsible for recency. Similarly, in stan-dard free recall, recency is found in immedi-ate recall but not in final recall (Craik,1970), recency increases over lists (Wing &Thomson, 1965), and recency appears to re-sult from earlier output of the last items inthe list (Goodwin, 1976).

The only difference in form between theserial position functions obtained in stan-dard free recall and in Experiments 2 and3 is in the magnitude of the recency effect.In comparison to Murdock's (1962) resultsusing the same list length in standard freerecall, the recency effect in the continuousdistractor paradigm extends over feweritems and is considerably attenuated. Thisfinding is consistent with the notion sug-gested earlier that distraction activity re-duces the number of items held in short-term store. In addition, the slight negativerecency in the final recall of Experiments1 and 3 is also consistent with the accountof recency in terms of readout from short-term store (cf. Craik, 1970).

In several respects, the present results aredifferent from those of previous investiga-tors. In the introduction, we noted Tzeng's(1973) suggestion that excessive output in-terference could account for the discrepancybetween his results and the results of Bjorkand Whitten (1974), so we will not con-sider the latter further. The important pointis that rnultistore models do provide a rea-

sonable explanation of differences betweenthe results obtained in our Experiments 2and 3 and those of Tzeng (1973). Tzeng(like Bjork and Whitten) instructed sub-jects to rehearse only the current list item,whereas the present study omitted rehearsalinstructions. Although the precise effect ofrehearsal instructions is somewhat unclear,early list items should be rehearsed lessoften, reducing the primacy effect, and ter-minal list items should be rehearsed moreoften, increasing the probability that theseitems will enter long-term store. Consistentwith these hypotheses, Tzeng found a muchlarger recency effect and a smaller primacyeffect than we obtained in Experiments 2and 3. The results of Brodie and Prytulak(1975) suggest that instructions to rehearseeach item equally have the same effect instandard free recall; recency is increasedand primacy is decreased. Furthermore,Tzeng found positive recency in final recallas well as in immediate recall, suggestingthat rehearsal instructions affect storagestrategies, increasing storage of terminal listitems. While we know of no studies thathave addressed this issue, we predict thatsimilar rehearsal instructions would lead topositive recency in final recall in the stan-dard free-recall paradigm.

The initial intention of the present re-search was to examine the processes under-lying serial position effects in the continu-ous distractor paradigm. At the outset, itappeared that the familiar rnultistore ex-planation used in standard free recall couldnot be invoked due to the major proceduraldifferences. It now appears, however, thatthe addition of rehearsal instructions andcontinuous distraction was not very effec-tive. Rather, these changes simply alteredthe magnitude of serial position effects ina way consistent with the multistore ac-count. That the storage and retrieval strate-gies underlying primacy and recency per-sisted indicates that these strategies arequite robust. In so demonstrating, the con-tinuous distractor paradigm has served auseful purpose. However, the paradigm hasfailed in its original aim (cf. Bjork & Whit-


ten, 1974) of eliminating unwanted serialposition effects. In addition, we have shownthat the observed serial position effects donot conflict with the multistore explanation.Therefore we suggest that future researchexamine the active roles of rehearsal andshort-term memory processes in the contin-uous distractor paradigm and not assumethat the paradigm eliminates these processes.

Reference Note1. Whitten, W. B., & Bjork, R. A. Test events as

learning trials: The importance of being im-perfect. Paper presented at the MidwesternMathematical Psychology Meetings, Blooming-ton, Indiana, April 1972.

ReferencesAnderson, J. R., & Bower, G. H. Recognition and

retrieval processes in free recall. PsychologicalReview, 1972, 79, 97-123.

Atkinson, R. C., & Shiffrin, R. M. Human mem-ory: A proposed system and its control pro-cesses. In K. W. Spence & J. T. Spence (Eds.),The psychology of learning and motivation: Ad-vances in research and theory (Vol. 2). NewYork: Academic Press, 1968.

Birnbaum, I. M., & Eichner, J. T. Study versustest trials and .long-term retention in free-recalllearning. Journal of Verbal Learning and Ver-bal Behavior, 1971, 10, 516-521.

Bjork, R. A., & Whitten, W. B. Recency-sensi-tive retrieval processes in long-term free recall.Cognitive Psychology, 1974, 6, 173-189.

Brodie, D. A., & Prytulak, L. S. Free recallcurves: Nothing but rehearsing some items moreor recalling them sooner? Journal of VerbalLearning and Verbal Behavior, 1975, 14, 549-563.

Craik, F. I. M. The fate of primary memoryitems in free recall. Journal of Verbal Learningand Verbal Behavior, 1970, 9, 143-148.

Dark, V. J., & Loftus, G. R. The role of rehearsalin long-term memory performance. Journal ofVerbal Learning and Verbal Behavior, 1976, 15,479-490.

Darley, C. F., & Murdock, B. B., Jr. Effects ofprior free recall testing on final recall andrecognition. Journal of Experimental Psychol-ogy, 1971, 91, 66-73.

Fischler, I., Rundus, D., & Atkinson, R. C. Effectsof overt rehearsal procedures on free recall.Psychonomic Science, 1970, 19, 249-250.

Glanzer, M., & Cunitz, A. R. Two storage mecha-nisms in free recall. Journal of Verbal Learningand Verbal Behavior, 1966, 5, 351-360.

Goodwin, C. J. Changes in primacy and recencywith practice in single-trial free recall. Journalof Verbal Learning and Verbal Behavior, 1976,15, 119-132.

Hogan, R. M., & Kintsch, W. Differential effectsof study and test trials on long-term recogni-tion and recall. Journal of Verbal Learning andVerbal Behavior, 1971, 10, 562-567.

Kahneman, D. Attention and effort. EnglewoodCliffs, N. J.: Prentice-Hall, 1973.

Kintsch, W. Recognition and free recall of or-ganized lists. Journal of Experimental Psychol-ogy, 1968, 78, 481-487.

Loftus, G. R, Comparisons of recognition and re-call in a continuous memory task. Journal ofExperimental Psychology, 1971, 91, 220-226.

Melton, A. W. Implications of short-term memoryfor a general theory of memory. Journal ofVerbal Learning and Verbal Behavior, 1963, 2,1-21.

Modigliani, V. Effects on a later recall by delay-ing initial recall. Journal of Experimental Psy-chology: Human Learning and Memory, 1976,2, 609-622.

Murdock, B. B., Jr. The retention of individualitems. Journal of Experimental Psychology,1961, 62, 618-625.

Murdock, B. B., Jr. The serial position effect infree recall. Journal of Experimental Psychology,1962, 64, 482-488.

Murdock, B. B., Jr. Modality effects in short-term memory: Storage or retrieval? Journal ofExperimental Psychology, 1968, 77, 79-86.

Norman, D. A., & Bobrow, D. G. On data-limitedand resource-limited processes. Cognitive Psy-chology, 1975, 7, 44-64.

Paivio, A., Yuille, J. C., & Madigan, S. A. Con-creteness, imagery, and meaningfulness valuesof 925 nouns. Journal of Experimental Psychol-ogy Monograph, 1968, 76(1, Pt. 2).

Peterson, L. R., & Peterson, M. J. Short-termretention of individual items. Journal of Ex-perimental Psychology, 1959, 58, 193-198.

Postman, L., & Phillips, L. W. Short-term tem-poral changes in free recall. Quarterly Journalof Experimental Psychology, 1965, 17, 132-138.

Reitman, J. S. Without surreptitious rehearsal,information in short-term memory decays. Jour-nal of Verbal Learning and Verbal Behavior,1974, 13, 365-377.

Rundus, D. Analysis of rehearsal processes infree recall. Journal of Experimental Psychology,1971, S9,63-77.

Tversky, B. Encoding processes in recognition andrecall. Cognitive Psychology, 1973, 5, 275-287.

Tzeng, O. J. L. Positive recency effect in adelayed free recall. Journal of Verbal Learn-ing and Verbal Behavior, 1973, 12, 436-439.

Waugh, N. C., & Norman, D. A. Primary mem-ory. Psychological Review, 1965, 72, 89-104.

Wing, J. F., & Thomson, B. P. Primacy-recencyeffects in free recall. Proceedings of the 73rdannual convention of the American Psychologi-cal Association. Washington, D.C.: AmericanPsychological Association, 1965.

Received January 26, 1977 •

Documents

Primacy and Recency in the Continuous Distractor Paradigm