CLUSTERING AND ORGANIZATION IN FREE RECALL...between organization and memory. This pur-pose can best be represented by a quote from Bousfield's (1953) original paper The theoretical

Psychological Bulletin1969, Vol. 72, No. 5, 353-374

CLUSTERING AND ORGANIZATION IN FREE RECALL 1

THOMAS J. SHUELL *

State University of New York at Buffalo

Research on clustering and subjective organization (SO) in free recall is re-viewed and evaluated. Various indexes developed to measure clustering andSO are evaluated, and two intercorrelation matrices among clustering mea-sures and the number of words recalled are presented. The existence of alarge negative bias in the correlation between the ratio of repetition (RR)measure and recall is demonstrated. Various theoretical issues which have de-veloped from the study of organization in free recall are presented and dis-cussed.

Psychologists have long been interested inthe way organization affects memory. Overthe years there have been several, rather dif-ferent, approaches to the problem. Probablythe most productive to date has been thestudy of clustering and subjective organiza-tion in free recall since an attempt has beenmade to quantify the organizational process.

In free recall the subject is presented alist of words to learn, and he is told to recallthe items in the order in which he thinks ofthem. Interestingly, certain regularities appearin the order in which the items are recalled.For example, there is a tendency for itemswhich are somehow related to one another tobe recalled together even though these itemswere not contiguous during presentation. Thisdiscrepancy between the order in which theitems were presented and the order in whichthey were recalled is presumed to representa tendency on the part of the subject to or-ganize his recall on the basis of various sec-ond-order habits, that is, preexperimental as-sociations or conceptual relationships. Thistendency for related items to be recalled to-gether has been termed clustering.

Clustering was apparently first observed byBousfield and Sedgewick (1944) while they

1 An earlier draft of this paper was written duringthe author's tenure as a predoctoral National In-stitutes of Health trainee at the Institute of HumanLearning, University of California, Berkeley. The In-stitute is supported by grants from the NationalScience Foundation and the National Institutes ofHealth.

2 Requests for reprints should be addressed toThomas J. Shuell, Educational Psychology Depart-ment, State University of New York, Buffalo, NewYork 14214.

were studying sequential characteristics of as-sociative responses. When they asked subjectsto list items in specified categories, they no-ticed that the subjects tended to respond withsequences of related items. For example, ifasked to list a series of animals, the subjectsmight first write down several felines, thenseveral canines, etc. Later, Bousfield (1953)developed a technique for quantifying cluster-ing. During the ensuing IS years, a greatdeal of research has been done on clusteringin free recall. The main purpose of this re-search has been to explicate the relationshipbetween organization and memory. This pur-pose can best be represented by a quote fromBousfield's (1953) original paper

The theoretical significance of this undertaking de-rived in part from the assumption that clustering isa consequence of organization in thinking and recall.If clustering can be quantified, we are provided witha means of obtaining additional information on thenature of organization as it operates in the highermental processes [p. 229].

The purpose of the current paper is to reviewand evaluate this research, to determine howwell its original purpose has been fulfilled, andto raise some questions as to what directionfuture research in this area might take.

Three paradigms have been developed forthe study of organization in free recall. Thesediffer primarily in the experimental situationused for inducing clustering. Two of theseparadigms are similar in that the basis oforganization is determined by the experi-menter. The first of these is called categoricalclustering. So far, the bulk of the researchin the area, including all of the early studies

353

354 THOMAS J. SHUELL

by Bousfield, has used this paradigm. In thissituation the stimulus list is comprised ofwords representing two or more mutually ex-clusive conceptual categories. The other par-adigm is called associative clustering and re-fers to the situation in which the stimulus listis comprised of associatively related words(as determined from associative norms) whichare not members of the same conceptualcategory.

Subjective organization differs from theother two paradigms in that the basis oforganization is not predetermined by the ex-perimenter. Rather, the stimulus list is com-prised of so-called unrelated words, that is, arandom sample of words in which the experi-menter has made no attempt to include wordswhich are categorically or associatively re-lated. In fact, he frequently tries to mini-mize or eliminate such relationships amongthe words. Thus, the subject is more or lessfree to organize the words in any way thathe wishes. Organization is determined by theextent to which the subject recalls the wordsin the same order on two successive trials.This differs from categorical and associativeclustering which compare the order of recallon any given trial with the organization, asdefined by the experimenter, which is presentin the stimulus list. Therefore, a multitrialexperiment is required in order to measuresubjective organization while categorical andassociative clustering can be measured aftera single trial.

It seems most probable that all three para-digms are dealing with the same basic psy-chological processes. The main differences arethe experimental procedures which are em-ployed and the extent to which some type oforganization or relatedness is apparent to thesubject. Any paradigm will, of course, in-fluence the types of questions which are askedby investigators using that paradigm. Thepresent situation is no exception, althoughthese influences are not always made explicit.A case in point, which will be discussed inmore detail later, is the single-trial versusmultitrial nature of the various paradigms(cf. Gofer, 1967; Tulving, 1968).

The present paper is divided into threemajor sections. The first section deals withthe various techniques which have been de-

veloped for measuring organization in freerecall. The ability to quantify organization isa major reason for the interest in clusteringand subjective organization. The second sec-tion discusses some of the major findings ofthe various research studies on organizationin free recall. The final section considers themajor theoretical interpretations of the rela-tionship between organization and memorywhich have resulted from the study of cluster-ing and subjective organization.

THE MEASUREMENT OF ORGANIZATION

A variety of measures have been developedto measure organization in free recall. It ispossible to classify these measures in severaldifferent ways. For purposes of this paper,the major distinction is made in terms ofwhether they were designed to measure or-ganization as determined by the experimenteror organization as determined by the subject.Then, within each of these two classes thevarious measures are discussed in terms of theassumptions underlying the method.

Predetermined Basis of Organization

The procedure typically followed in thestudy of categorical clustering can be ex-emplified by Bousfield's original experiment(1953). A list of 60 nouns containing IS ex-amples from each of four mutually exclusivecategories (i.e., animals, names, professions,and vegetables) was constructed. These wordswere then read to the subjects in a randomorder. Following the presentation of the list,the subjects were given 10 minutes to writedown as many of the words as they couldremember. The order in which the words arerecalled is used to determine the amount oforganization present.

All of the indexes used to measure cate-gorical clustering, with the exception of thefirst one discussed, are based on the numberof repetitions which occur during recall. Arepetition is said to occur any time two itemsfrom the same category are recalled one afterthe other. The number of repetitions for asequence of items from a given category isalways the number of items in the sequenceminus one, since the first item in the se-quence is not a repetition. Summing over allof the sequences in the protocol for one sub-

CLUSTERING IN FREE RECALL 355

ject gives the total number of repetitions forthat subject.8

Cluster size. If subjects organize the wordsto be remembered on the basis of the cate-gories present in the stimulus list, then onereasonable way to measure the amount ofthis organization would be to count the num-ber of words appearing in clusters representingone of the categories. If there is little or noorganization, then most of the words wouldprobably appear by themselves. If, however,organization does occur, then there should bea relatively large number of words appearingin clusters. Bousfield (1953) tabulated thenumber of times single (unclustered) wordsand clusters of varying numbers of words oc-curred in the subjects' recalls. He made asimilar tabulation for a parallel, artificial ex-periment in which 100 sequences, matched inlength with those of the actual subjects, weredrawn without replacement from a box con-taining IS capsules each of four differentcolors. While the results of the artificial ex-periment exhibited a greater occurrence ofsingle words and clusters of two words thanwere produced by the real subjects, the sub-jects produced many more clusters of largersizes than were obtained from the artificialexperiment. While this tabulation appears tobe a valid measure of clustering, it seldom hasbeen used in subsequent papers. Probably themain reason it has not been used more fre-quently is that it does not provide a single,summary measure.

Items equally available. If one is willing toassume that all of the words presented forlearning are equally available for recall, thenit is possible to apply a probability model todetermine if the amount of clustering ob-served exceeds a chance level. Using the nor-mal approximation to the binomial, Bousfield(1953) developed what he called the indexof repetition ( I R ) . The IR is based on theprobability of a given word being followed by

3 In computing the various measures of clustering,categorical intrusions (i.e., words recalled whichwere not presented to the subject but which belongto one of the categories represented in the stimuluslist) have generally been treated the same as correctitems. The effect which this procedure has on themeasures is probably minimal, but it has never beencompared with the procedure in which categoricalintrusions are excluded.

another word in the same category if theorder of recall occurs on a random basis. Theindex can be represented by the followingformula:

IR =cr — n

V«(c - 1)[1]

in which r is the number of repetitions ob-served, n is the total number of sequences ina given protocal and is always the total num-ber of words recalled minus one, and c is thenumber of categories in the stimulus list.4

The IR is normally distributed with a meanof zero and a standard deviation of one, andit is independent of the number of words re-called. The extent to which the obtained meanIR deviates from the expected value of zeroreflects the extent to which clustering oc-curred during recall. However, the model isone of sampling with replacement, and thevalidity of this model may be seriously ques-tioned particularly when short lists are used.The IR has not been used a great deal insubsequent papers.

The measure which has been used mostfrequently in studies of clustering is the ratioof repetition (RR—Bousfield, 1953). The RRis a simple ratio of the number of obtainedrepetitions to the number of repetitions pos-sible for the words recalled. This can berepresented by

RR =N - 1 [2]

where N is the total number of words re-called.8

Tables of chance values of RR for varioustypes of lists have been constructed by Cohen,Sakoda, and Bousfield (1954). They used amodel of sampling without replacement froma pool comprising all the items in a given

4 The assumption made here, of course, is that allof the items, and therefore all of the categories, areequally available for recall. If one wished to assumethat unless at least one item from a given category isrecalled, the category is not available for recall andshould not be counted in computing the probabilityof obtaining a repetition, then c would be the numberof categories represented by the words recalled.

5 The formula presented in the original 1953 paperwas r/N. However, since r/(N — \) is the formulawhich has been used most frequently in subsequentpapers, it is the one presented.

356 THOMAS J. SHVELL

list. These tables were used in many subse-quent papers to determine the amount ofclustering that could be expected by chancefor a given categorized list.

The main advantage of RR over IR is theease with which the measure can be com-puted. However, the RR does not take intoaccount the transition from one category toanother which occurs even with perfect clus-tering. Because of this, it is not independentof the number of categories represented bythe words recalled. For example, the onlysituation in which RR can equal one is whenonly a single category is recalled.

Robinson (1966) developed the item-clus-tering index (1CI) in order to allow for thispossibility of perfect clustering in spite ofthe gaps in repetitions resulting from cate-gorical transitions. The formula which hepresents is

ICI =c(Wc - 1) [3]

in which r is the total number of repetitionsoccurring in recall, c is the number of cate-gories represented by the words recalled, andthe We is the number of items per category inthe stimulus list. Chance values were notcomputed. The ICI, however, does not takeinto consideration the possibility of perfectclustering occurring for the words actually re-called; that is, the only time ICI can equalone is when all of the words in a certain num-ber of categories are recalled. Thus, the as-sumption is made that all of the items withina given category are available for recall giventhat at least one word in that category isrecalled.

The methods which have been developedfor measuring associative clustering (Jenkins,Mink, & Russell, 19S8; Jenkins & Russell,19S2) also make the assumption that all ofthe items are equally available for recall.Clustering was said to occur when the twowords in a given stimulus-response pair oc-curred together in recall. Jenkins and Russell(19S2) used the occurrence of arbitrary pairs,that is, a stimulus word followed not by itsown response but by another specified andrandomly selected response from the list asa chance base line against which the oc-currence of both forward and backward as-

sociations was compared. Jenkins, Mink, andRussell (1958) developed an alternative pro-cedure in which the opportunity to cluster istaken into consideration. This method is usedfor comparing groups receiving different ex-perimental manipulations and does not use achance base line. Two studies (Jenkins, Mink,& Russell, 19S8; Wicklund, Palermo, & Jen-kins, 196S) have reported very similar con-clusions when both of the methods are usedfor analyzing the same data.

All of the measures discussed in this sec-tion are based on the assumption that all ofthe items presented for learning are equallyavailable for recall. However, this is not avalid assumption. For example, there is sub-stantial evidence that a serial position effectis obtained in free recall under a wide varietyof conditions (Murdock, 1962; Shuell, 1967;Tulving & Patterson, 1968). If all of theitems are equally available such an effectshould not be obtained. Therefore, the useful-ness of these measures is limited since theyare based on a faulty assumption.

Limited availability of items. Since it doesnot appear reasonable to assume that theitems are equally available for recall, thequestion arises as to how one might proceedin developing a more valid model for themeasurement of clustering. One such at-tempt has been made by Bousfield (Bousfield& Bousfield, 1966; Bousfield & Puff, 1964).The difference between the obtained and ex-pected number of repetitions is used as themeasure of clustering; the larger this differ-ence the greater the clustering or organization.A distinction is made between item properties,that is, what items are recalled, and orderproperties, that is, what is the sequentialordering of those items which are recalled.In determining the number of repetitions thatcan be expected on the basis of chance, oneis concerned with order properties rather thanitem properties. Therefore, the following as-sumptions are made: (a) When a subjectbegins his recall a certain number of itemsare not available and therefore can have noeffect on the incidence of clustering, and (b)at every stage of recall all of the words re-maining to be recalled are equally availableand are chosen without replacement. The ex-pected number of repetitions calculated on

CLUSTERING IN PREE RECALL 357

the basis of this model is

^ m(m - 1)[4]

where m is the number of words recalled ina category k, and N is the total number ofwords recalled. Summing over all of thecategories in a particular list, the expectednumber of repetitions in the list as a wholecan be represented by

*zK(r) = -i [5]

This deviation is, of course, computed foreach subject.

Bousfield and Bousfield (1966) also presenta formula for calculating the expected numberof times an item in one category, for example,Category i will be followed by an item fromanother specified category, for example, Cate-gory ;. This formula is

p(r.\ — ——'- ffi"|t^Vij) iy L°J

for i =7^=;'.One shortcoming of the deviation measure

is that no allowance is made for the maximumamount of clustering which can be obtained

with the words recalled. This limitation be-comes particularly important when there arefairly large discrepancies in the number ofwords recalled by subjects, for example, whenlists of different length are used. Perfectclustering could be obtained in a short list,yet the value of the deviation measure couldbe less than that obtained in a longer list inwhich clustering is not perfect. In such situa-tions it would be relatively simple to convertthe deviation measure into a proportion simi-lar to the one developed by Pagan (1968)for measuring subjective organization.

Comparison of the various measures. Thevarious measures were empirically comparedin a study by the present author (Shuell,1967). Specifically, 336 college students re-ceived four alternate study-recall trials on a35-item list in which there were five wordsrepresenting each of seven conceptual cate-gories. The words were selected from the 10most frequent responses to the category namesas reported in the norms of Cohen, Bousfield,and Whitmarsh (19S7). Intrusions were dis-regarded in counting repetitions. The indexesused were the actual number of repetitionsobtained, the observed minus the expected(Equation 5) number of repetitions (O — E),the IR (Equation 1), the RR (Equation 2),and the 1C I (Equation 3). The intercorrela-

TABLE 1

INTERCORRELATIONS OP VARIOUS CLUSTERING MEASURES

Measure O-E" RR IR ,VR ICI W/C

Trial 1

Words recalledWords per category (W/C)Item clustering index (ICI)Number of repetitions (NR)Index of repetition (IR)Ratio of repetition (RR)

.76

.70

.91

.99

.99

.87

.44

.63

.93

.87

.89

.73

.65

.89

.98

.79 .61 .75

.78 .84

.94

Trial 4

Words recalledWords per categoryICI measureNumber of repetitionsIndex of repetitionRatio of repetition

.89

.85

.951.00 \.99.89

.62

.74

.94

.89

.92

.85

.79

.93

.99

.90 .78 .85

.88 .92

.96

• O —E is the observed minus the expected number of repetitions.


tion matrices for these measures, the num-ber of words recalled, and the number ofwords recalled per category are presented forboth Trial 1 and Trial 4 in Table 1. It canbe seen that the correlations among the clus-tering measures are all quite high. Since allof the measures are primarily based on thenumber of repetitions obtained and differ onlyin the way corrections for various factors areapplied, this finding is not very surprising.However, there are substantial differences inthe degree to which the various measures cor-relate with the number of words recalled. Forexample, the correlation coefficient betweenthe traditional RR measure and the numberof words recalled is substantially lower thanthe coefficients for the other measures on bothtrials. The values of the correlation betweenRR and recall are in the range typically re-ported in other studies.

There are at least two possible explana-tions for these differences. First, it has alreadybeen noted that the assumption of equalavailability which underlies the RR measureis not a reasonable assumption. However, thesame assumption is made for the IR measure,and the correlation between IR and recallappears to be much larger than the one be-tween RR and recall. Second, there is a sta-tistical artifact in the correlation coefficientbetween RR and recall. All of the measuresmake some adjustment for differential recall.There is, of course, a direct relationship be-tween the total number of words recalled andthe opportunities for repetitions to occur. TheRR makes this correction by calculating aratio of the number of repetitions obtained tothe total number of repetitions possible giventhe number of words recalled (N — l ) . R Ineffect, an attempt is made to provide a baseline for comparison by partialing out the num-ber of words recalled. In such a case, RRcould not be expected to correlate highly withrecall since the effect of the latter variablehas already been removed. In fact, it can beshown that the expected value of the cor-relation coefficient between two random vari-ables X and Y/X, assuming that X and Y

alt has already been pointed out that this is abiased measure since no allowance is made for transi-tions from one category to another, but this is notthe main concern here.

are independent, is not zero but some nega-tive value which is dependent upon the param-eters of X and YJ This expectation can beexpressed approximately by

£(',<„/.>) « _^£f_ [7]

where C, the coefficient of variation, is equalto the standard deviation divided by themean. Using the present data to estimate thevalue of Cj, and Cv, the expected value ofthe correlation coefficient between RR andrecall is -.43 for Trial 1 and -.48 for Trial4. Thus, the obtained values of +.44 and + .62are substantial in view of the large negativebias inherent in the relationship.

It appears most likely that this statisticalartifact is the main reason for obtaining thelower values for the RR measure. Thus, whilethe RR measure is obviously able to detectclustering, it is clear that at least under cer-tain conditions, use of this measure will re-sult in somewhat different conclusions thanwill be reached from use of the other mea-sures. However, there may be situations inwhich it is desirable to use a proportion mea-sure, for example, when comparing lists ofdifferent lengths. In these situations, it maybe best to convert the deviation measure intoa proportion. It is interesting to note that atleast for the conditions used in this studythe various measures, with the exception ofRR, appear to be roughly equivalent. Themeasure used will, of course, depend in parton the purpose of the investigator and theassumptions he is willing to make. Neverthe-less, the present author considers the devia-tion measure, or some form of it, to be thebest measure currently available since themodel on which it is based seems to be themost valid.

Subject-Determined Basis of Organization

Probably the biggest drawback in using pre-determined basis of organization is the im-possibility of determining the extent to whichthe subjects are in fact using the organizationpresent in the stimulus list. It is possiblethat bases of organization other than those

7 Y/X is taken as a reasonably close approximationto Y/(X-\).


specified by the experimenter are being usedby the subjects to organize their recall. Also,to the extent that other bases of organizationare being used, clustering, or the use of pre-determined bases of organization, underesti-mates the amount of organization being usedby the subjects. All of the measures discussedso far have been based exclusively on pre-determined aspects of organization. Thus, itis possible that many interesting facets of theorganizational process have been overlooked.

The study of subjective organization in freerecall is an attempt to investigate those basesof organization which the subjects are actuallyusing. No assumption is made concerning theorganization present in the stimulus list.Rather, organization is measured by the sub-ject's tendency to recall words in the sameorder on successive trials. The measurementof subjective organization was first developedby Tulving (1962b). Since then Bousfield andhis associates (Bousfield & Bousfield, 1966;Bousfield, Puff, & Cowan, 1964) have de-veloped alternative methods for measuringsubjective organization.

While the measures of clustering are basedon repetitions, the measures of subjective or-ganization are based on intertrial repetitions(ITR). In determining the number of ITRswhich occurred during recall, a matrix is con-structed for each subject. All of the words inthe stimulus list are represented along boththe rows and the columns of the matrix. Therows represent the wth word recalled while thecolumns represent the ( w + l ) t h word re-called. The matrix can represent any numberof trials from two to the total number oftrials given. There is one slight difference be-tween the matrix used by Tulving and the oneused by Bousfield. The matrix used by Tul-ving has extra positions for no word immedi-ately preceding the first word and for no wordimmediately following the last word recalled.The matrix used by Bousfield, however, doesnot have these additional blank positions. Thisis done since repetitions cannot be formedfor these two situations. The matrix isused for making a tabulation of the frequencywith which given pairs occur adjacent to oneanother on two immediately successive trials.An ITR is said to occur whenever a given cellis checked twice in succession as a consequence

of recording the data for a given pair onTrial » and Trial n + 1.

Items equally available. Bousfield, Puff, andCowan (1964) have developed a deviationmeasure of subjective organization which issimilar in many respects to the deviationmeasure used in the study of clustering. Themodel used in calculating the expected valueis one in which two successive random draw-ings are made from the same pool of W itemsin which W is the total number of words pre-sented for learning. Each drawing is madewithout replacement, and at each stage ofboth drawings it is assumed that the remain-ing items are equally available. This ex-pected value can be represented by the form-ula

E(ITR) = W(W - 1) [9]

in which h is the number of words recalled onTrial n, and k is the number of words re-called on Trial n + 1. The difference betweenthe number of ITRs obtained and the ex-pected value is then taken as a measure ofsubjective organization.

Unfortunately, this measure is based on afaulty assumption. It has already been pointedout that in free recall items are not equallyavailable. It is not entirely clear how impor-tant the violation of this assumption is. Nev-ertheless, one should keep it in mind whenusing the measure.

Limited availability of items. Tulving's(1962b) original measure was derived frominformation theory, and he quite appropriatelytermed it the SO measure. It is a ratio measurein which the organization obtained is expressedas a fraction of the maximum organizationpossible. The rows and the columns of thematrix described above are summed, and thefollowing formula is used:

so =^2 Hij log Hij

ii ---2_. ni log ni

[8]

In the formula % represents the numericalvalue of the cell in the tth row and the ;thcolumn, and nt represents the marginal totalof the ith row. The value of SO can vary fromzero to one, although it can assume the valueof one only when the same words are recalled


on all of the trials represented by the matrix.When new words are recalled or old wordsomitted on some of the trials, the maximumvalue of 50 is some value less than one.

While this formula considers only se-quences occurring in one direction, it can beeasily modified to consider sequences occurringin either direction, that is, Word A followedby Word B as well as Word B followed byWord A, by summing the corresponding cellsacross the diagonal. The matrix can easily bemodified to provide an estimate of higherorder organization, for example, by letting thecolumns represent the (« + 2)th word ratherthan the (n + l)th word. The method canalso be extended to determine higher orderdependencies, but the computations might beprohibitive.

Bousfield and Bousfield (1966) have re-cently revised the model used in calculatingthe expected number of ITRs. The model isthe same one used in the measurement of pre-determined organization. It will be remem-bered that a distinction is made between re-sponse recall and the sequential character-istics of the words recalled. They argue thatthe calculation of the chance value should bebased on the assumption "that the given re-call sequence is a random sample from amongall possible orderings of the recalled items [p.939]." The new formula for calculating theexpected value is

E(ITR) =c(c -

hk [10]

where c is the number of items common tothe two recalls. It can be readily seen thatthe value of E(ITR) is always less than one.

One limitation of the deviation measurenoted previously was that the maximumamount of clustering possible is not taken intoaccount. Fagan (1968) has recently shownhow the deviation measure can be convertedinto a ratio measure, and this convertedmeasure can be used when it is desirable touse a proportion measure.

Comparison of the measures. Puff and Hy-son (1967) have compared the deviationmeasure, using the limited-availability model,with the 50 measure. Over 20 trials of free-recall learning they obtained virtually identi-cal results with the two measures. Correlation

coefficients between the two measures werecalculated for each of the 19 trial pairs. Theyreport that the lowest of these correlationswas .90. They also calculated the correlationbetween the mean values (across subjects) ofthe two measures over the 19 trial pairs. Thevalue of this coefficient was .94. The correla-tion between the observed number of ITRsand the value of 50 was .92 across the 19trial pairs. Unfortunately, they do not reportcorrelations of the measures with the numberof words recalled. Therefore, it is unclearwhether or not the biases apparent for theratio measure of clustering (RR) are alsopresent in the ratio measures of subjectiveorganization, for example, the 50 measure.

Independent measures oj organization. Itwould be desirable, of course, to obtain in-dependent estimates of organization and re-call. All of the measures discussed aboveestimate both variables from the same set ofdata, namely, the words which the subject re-called. A promising technique has recentlybeen developed by Seibel8 at the PennsylvaniaState University which permits such inde-pendent estimates. With this method, a sub-ject is given a matrix, that is, a blank sheetof paper with, for example, 10 rows and 10columns, prior to the presentation of thestimulus list. The subject is instructed to writeeach word, as it appears, wherever he wantsto in the matrix. After the last word has beenpresented and recorded, the matrix is re-moved, and the subject free recalls as manywords as he can on another sheet of paper. Atthe beginning of the next trial, the recallsheet is removed, and the subject is given anew matrix. While this method allows thesubject additional, unsystematic study time,it has the substantial advantage of providingan estimate of organization independent ofthe recall data. While any of the conventionalmeasures of organization can be used, thedata in the matrix also permit one to measurethe consistency of organization independentof sequence and the degree to which eachsubject utilizes his own subjective organiza-tion in his recall tests.

A related procedure has been used byHandler (1967; Handler & Pearlstone, 1966),

8 R. Seibel, personal communication, January 1968.


and it has the advantage of reducing theamount of unsystematic study time inherentin Seibel's procedure. Mandler has subjectssort lists of words into stacks so that onlythe top word of each stack is showing. WhileMandler asks his subjects to recall the wordsonly after they have reached a predeterminedsorting criterion, it would be a simple matterto ask the subjects to recall the words aftereach sort. The present author has satisfac-torily used this technique in an unpublishedstudy.

Both of these techniques have one majoradvantage over the other measures of sub-jective organization. Both the 50 and thedeviation measures are based on pairwise con-tingencies, and in some instances these con-tingencies may provide only a crude indexof the organization actually present. For ex-ample, it is possible that a unit of fourwords will be recalled together on everytrial, but if the words are recalled in a dif-ferent order on successive trials, both mea-sures will underestimate the organization pres-ent. The two measures discussed in this sec-tion, however, permit the investigator to lookat larger size subjective units.

Units oj Analysis

The response unit selected for measuringthe amount of material recalled can fre-quently influence the conclusions drawn fromthe data. The unit typically used in verballearning studies is some experimentally de-nned single item such as a word, trigram, etc.While the use of such a unit is somewhatarbitrary, there is at least some face validityfor selecting a word as the basic unit ofmeasurement. However, a distinction can bemade between such experimentally denned ornominal unit and the functional unit actuallyused by the subject. Tulving (1968) refersto the former as E-units and the latter asS-units. In any situation in which the sub-jects group the words together into sometype of higher order units, for example, infree recall, there will be a discrepancy be-tween E-units and S-units. While there area number of situations in which this dis-crepancy is of little interest (cf. Tulving,1968), there are many others in which im-portant theoretical considerations depend on

the unit selected. This is particularly truewhen one is concerned with the amount ofmaterial recalled or with the capacity ofmemory. For example, Tulving and Patkau(1962) were concerned with the effects ofcontextual constraint and word frequency onfree recall. When the number of words re-called was used as the dependent variable,there were significant effects due to wordfrequency and to the interaction betweenfrequency and order of approximation to En-glish (contextual constraint). However, whenthe number of word sequences correspondingto uninterrupted sequences of words fromthe stimulus list ("adopted chunks") wereused as the dependent variable, neither fre-quency nor the interaction was significant.Similar types of findings have been obtainedby Cohen (1963b) and McNulty (1966).

Deese (1968), on the other hand, has sug-gested that units more basic than a wordshould be considered. Basically, he is sug-gesting a distinctive feature analysis in whichthe intersection of a set of distinctive sourcefeatures enables the subject to retrieve orproduce the single item, for example, the word.The unit which an investigator uses will de-pend upon both his theoretical predilectionand the type of question he is asking.

THE STUDY OF ORGANIZATION INFREE RECALL

In this section of the paper, considerationwill be given to the variety of experimentalstudies on the organizational process. Tulving(1968) has distinguished between two typesof organization in free recall. The first ofthese is referred to as primary organization.This type of organization is defined as theconsistent discrepancies between input andoutput orders which are independent of thesubject's prior familiarity with the inputitems. The serial position effect (e.g., Mur-dock, 1962) and the tendency for subjects torecall the terminal items first (Postman &Keppel, 1968; Shuell & Keppel, 1968) areexamples of primary organization. The othertype of organization is referred to as sec-ondary organization. This type of organiza-tion is dependent upon the subject's prioracquaintance with the items in the stimuluslist. Clustering on the basis of meaning would


be an example of secondary organization.Most of the studies to date have been con-cerned with this latter type of organization.

Variables Influencing Organization

Much of the research on organization infree recall has been concerned with the de-termination of the variables and conditionswhich influence the amount of clustering ob-tained. Recall performance has also been fre-quently used as a dependent variable. A largevariety of independent variables have beenused, and some of the more important onesare considered below.

Number 0} categories. A number of studieshave investigated the effect of using varyingnumbers of categories for organizing thewords in the stimulus list. The main purposeof these studies has been to determine ifthere is an optimum number of categories.Much of the more recent work has beenconcerned with the limits of memory and theprocess of chunking as suggested by Miller(1956). While the analogy between categoriz-ing and chunking or receding as discussed byMiller (1956) is not exact, especially whennonexaustive categories are used, it has beenfrequently assumed that the same process isbeing studied (cf. Mandler, 1967). Unfor-tunately, all of the studies investigating num-ber of categories have held list length con-stant while varying the number of cate-gories used. As a result, number of categorieshas been confounded with the number ofwords per category since the number of wordsper category must decrease as the number ofcategories is increased. Consequently, the re-sults of these studies are somewhat equivocal.

The effect of varying numbers of categoriesappears to be dependent, at least in part, onthe length of the list and on whether or notrecall is cued (Dallett, 1964; Tulving &Pearlstone, 1966). In general, the relationshipbetween recall and number of categories ap-pears to be a direct one when cued recall isused and an inverse or curvilinear one whennoncued recall is used. While it is not possibleto presently separate the effects of number ofcategories and number of words per category,Earhard's (1967a) data indicate that at leastfor cued recall the use of categorized lists is

effective only when the number of words percategory is fewer than six or seven items.

With regard to measures of organization,however, the data are somewhat equivocal.Bousfield and Cohen (1956a) and Mandler(1967) found a direct relationship betweenclustering and number of categories used. Dal-lett (1964), however, found no consistentrelationship for 12-item lists and an inverserelationship for 24-item lists. It is quite pos-sible that these differences result from differ-ent measures of organization being used in thevarious studies. The Bousfield and Cohen andthe Mandler studies used the RR measurewhile Dallett used the deviation measure(0 — E). If each category is viewed as a sub-list, then when the number of categories is in-creased, the length of each sublist is cor-respondingly decreased. The RR measuremakes some provision for the maximumamount of clustering which can be obtained,although it has already been noted that thisadjustment is somewhat limited. The devia-tion measure, however, does not consider themaximum amount of clustering possible. Whenonly a small number of words per categoryis used, clustering or organization may be ator near maximum; however, when categorysize is increased, that is, number of cate-gories decreased, the value of the deviationmeasure may increase merely because of theincrease in the maximum number of repeti-tions which can occur.

Cohen and Bousfield (1956) used a two-level list in which each of the four maincategories in a 40-item list could be dividedinto two subgroups (e.g., animals: felines andcanines). The results were analyzed on thebasis of both a four- and eight-category listand compared with the results of earlier ex-periments in which a four- or eight-category,single-level list had been used. For the eight-category analysis of the dual-level list theresults were very similar to the results of theearlier experiment using an eight-category,single-level list. For the four-category analy-sis, however, recall was higher for the dual-level list, although RR was about the same inthe two experiments.

Several studies have been concerned withthe relationship between the number of cate-gories used in a classification task and free-


recall performance. In her 19S2 dissertation,Mathews (1954) had subjects classify namesof 24 famous people into either two, three,or six categories such as poet, scientist, etc.Ten minutes after completing the classifica-tion task the subjects were given 5 minutes torecall as many names as they could. The meannumber of words recalled was directly re-lated to the number of categories used, al-though she mentions in the discussion thatanother group receiving a one-category list re-called just as many words as the six-categorygroup. Later, she showed (Helson & Cover,1956) that recall was facilitated if subjectsused specific, rather than general, categoriesfor classifying the names; each of the spe-cific categories was a subcategory of one ofthe general ones.

Mandler (1967; Mandler & Pearlstone,1966) has used a similar procedure with un-related words. The subjects were asked tosort from 52 to 100 "unrelated" words intovarious classifications of their choosing;typically the subjects were asked to use fromtwo to seven categories. The same words weresorted on successive trials until the subjectachieved two identical sorts. Then, he wasasked to free recall as many of words as hecould remember. In general, both the meannumber of words recalled and the amount ofclustering (defined in terms of the subject-defined categories) were directly related tothe number of categories used. Unfortunately,however, these data should probably be in-terpreted with some caution since only about55% to 60% of the subjects provided usefuldata.

Exhaustive versus nonexhaustive categories.Exhaustive categories are those whose ex-amples exhaust, or nearly exhaust, all of theitems generally subsumed under the categorylabel, for example, north, east, south, andwest. Nonexhaustive categories, on the otherhand, are those whose examples only partiallyexhaust the members included in the category,for example, dog, lion, horse. Cohen (1963a,1963b) has shown that the mean number ofcategories recalled (i.e., represented by thewords recalled) is the same for the two typesof categories; however, significantly morewords are recalled for the exhaustive than forthe nonexhaustive categories. Interitem as-

sociate strength was significantly related towithin-category recall (i.e., recall of otherwords in a category given that at least one ofthe words is recalled) but was not related tocategory recall.

Blocked versus random presentation.Blocked presentation refers to the experi-mental situation in which all members of acategory are presented contiguously in thestimulus list, for example, all of the examplesof one category are presented before those ofanother category are presented. The order ofcategories and the order of items within eachcategory, however, can still be varied whenmore than one trial is used. All of the earlystudies of clustering used random presenta-tion in that the words in the stimulus listwere presented in a random order, and some-times restrictions were placed on the numberof examples of a given category that could ap-pear together.

Blocked presentation is frequently con-sidered to be more effective than randompresentation for helping the subject perceivethe categorized nature of the list. This isthought to be particularly true for lists com-prised of low-frequency associates to the cate-gory name and for categories with only afew items. However, the results of variousstudies by Cofer and his associates (Cofer,1967; Cofer, Bruce, & Reicher, 1966) raisesome doubts as to the validity of this in-terpretation. These studies indicate thatblocked presentation augments clustering toat least equivalent degrees in lists of highand low taxonomic frequency; there is someindication that this effect may be greater forlists of high taxonomic frequency. Also,blocked presentation facilitated recall only inthe high-frequency list.

Dallett (1964) obtained both superior re-call and superior clustering with blocked pre-sentation, but the difference between blockedand random presentation interacted with thenumber of categories in the list such thatthe largest difference between the two methodswas obtained for the condition in which therewere three words per category. Puff (1966)used lists containing 10 words from each ofthree categories. The lists were constructedso that there were either 0, 9, 18, or 27 cate-gory repetitions in the order of presentation


which the subject received; the first of theseconditions is the typical random presentationwhile the latter is the typical blocked methodof presentation. With one exception, both re-call and clustering were directly related tothe number of repetitions in the stimulus list.

Thus, blocked presentation appears to fa-cilitate both clustering and recall. However,the facilitation for clustering may be partlydue to the fact that all members of certaincategories appear in the most favorable posi-tions, that is, the first and last serial positionswith the terminal items tending to be recalledfirst (Postman & Keppel, 1968; Shuell &Keppel, 1968). In an unpublished study,Cohen9 compared blocked and random pre-sentation of a 70-item list comprised of 20categories of three or four items each. Whilethe mean number of words recalled was equiv-alent for the two methods of presentation,there was a significant tendency for theblocked presentation to increase the numberof words recalled per category while decreas-ing the mean number of categories representedin recall. Thus, it is possible that at leastunder certain conditions, blocked presentationmay facilitate the coding or organization ofthe predefined categories while decreasing thelikelihood that stable intercategory associa-tions will be developed.

Changes in organization and recall as ajunction of time. Changes in performanceover time can occur either in the presence orabsence of practice. In the presence of prac-tice the stimulus list is presented for a seriesof trials. Recall or test trials may be inter-spersed between the study trials, or theremay be a single recall trial after a givennumber of presentations. Both modes of pre-sentation have been used in the study ofclustering. Bousfield and Cohen (1953) pre-sented a four-category, 40-item list for eitherone, two, three, four, or five presentationsprior to a single 10-minute recall period. Bothmean recall and mean IR were directly re-lated to the number of presentations. Themean number of categorical intrusions, how-ever, was inversely related to the number ofpresentations.

Several studies (Bousfield, Berkowitz, &9 B. H. Cohen, personal communication, September

1968.

Whitmarsh, 1959; Marshall, 1967b; Robin-son, 1966; Shuell, 1968) employing the al-ternate study-recall procedure have shown thatclustering (several different measures wereused in the various studies), mean recall, andthe mean number of categories recalled (asdetermined by the words recalled) increaseprogressively as a function of trials. Onceagain, there is a tendency for categorical in-trusions to decrease over trials indicating thatlist differentiation is directly related to thenumber of trials.

Several studies (e.g., Bousfield, Puff, &Cowan, 1964; Tulving, 1962b) have shownthat subjective organization increases as afunction of trials. While the amount of organi-zation obtained on the first few trials is at achance level, the difference between the ob-tained and expected organization increasesprogressively over trials. The chance level forthe SO measure (Tulving, 1962b) was cal-culated from data obtained from statisticalsubjects. For the ITR measure, of course,Bousfield, Puff, and Cowan (1964) used theirmodel to compute the expected value. Tulving,McNulty, and Ozier (1965) obtained a posi-tive relationship between SO and the numberof lists which the subjects had learned. May-hew (1967) had subjects learn two lists oneach of 3 consecutive days and found a sig-nificant increase in SO both within and be-tween experimental sessions. This increase inSO proceeded at a faster rate if the subjectswere told to organize the words during recallthan if they were given standard FR in-structions.

The question can be raised as to whetherthe single-trial or the multitrial procedure isthe better method for studying organization infree recall. Cofer (1967) has argued that thesingle-trial situation is more representativeof free recall as it appears in daily life. Onthe other hand, Tulving (1968) argues infavor of the multitrial procedure. It must berealized, of course, that somewhat differenttypes of questions are being asked in thetwo situations. The single-trial procedure usu-ally deals with well-defined categories andprovides a useful paradigm for studying theeffects of a variety of independent variables.The multitrial procedure, especially when it isused to investigate subjective organization, is


more concerned with the development ofstable organization in the absence of anywell-defined relationships. Both situations canprovide us with useful information on or-ganization and memory, although it is im-portant to keep in mind the type of questionwhich is being asked.

In the absence of practice, a retentionsituation exists. Brand and Woods (1958)compared the retention of a categorized list at1-, 2-, and 3-week intervals. In one condi-tion, a single group was tested at each of theretention intervals. In the other condition, in-dependent groups of subjects were tested atjust one of the three intervals. For the inde-pendent groups there was a tendency for bothRR and recall to decrease as a function oftime. For the group receiving repeated test-ing, there was an initial drop at the firstretention interval, but there was no furtherdecrease after that. It should probably benoted that subjects who did not demonstrateclustering above a chance level on an initialrecall test were excluded.

Using "unrelated" words, Handler (1967)obtained a negatively accelerated retentionfunction over a 15-week period. The asymp-tote, approached about the fourth week, wasbetween 20 and 30%. The RR measure (asdetermined from the S-defined categories) de-creased from about .54 at one-half week toabout .24 after 14 weeks; the value of .24still exceeds the chance value of .16.

In a series of studies, Cofer and his as-sociates (Cofer, 1967; Cofer et al., 1966;Gonzales & Cofer, 1959) have investigatedchanges in clustering and recall from an im-mediate-recall test to a second recall test 5minutes later. In general, there was an in-crease in clustering and a decrease in recall.The clustering obtained on the second test issignificantly greater than the clustering ob-tained in a control group which waited anequivalent amount of time but did not havethe interpolated recall test (Cofer et al.,1966).

Several studies (Gonzalez & Cofer, 1959;Holroyd & Holroyd, 1961; Shuell, 1968;Winograd, 1968) have used categorized listsfor studying retroactive inhibition (RI). Ingeneral, RI was obtained in all of the studies.Retroactive inhibition has also been demon-

strated in free recall with unrelated words(Postman & Keppel, 1967; Shuell & Kep-pel, 1967; Tulving & Thornton, 1959). Per-haps the most interesting finding, however, isthat significantly fewer words per category arerecalled when the two lists contain membersrepresenting the same categories than whendifferent categories are represented by thewords in the two lists (Shuell, 1968; Wino-grad, 1968). At the present time, it is notclear whether this differential recall is theresult of increasing the number of wordsper category or intracategory interference.Category recall, on the other hand, was sig-nificantly poorer when different categorieswere represented in the two lists.

Associated t-elatedness. The role of associa-tive relationships in free recall is a complexand hotly debated one. A demonstration ex-periment by Jenkins and Russell (1952)showed that associative relationships, as wellas categorical relationships, can influenceclustering.10 Twenty-four stimulus-responsepairs of words from the Kent-Rosanoff wordlist were presented in a random order withthe restriction that a response word couldnot follow its stimulus. Clustering was saidto occur when the two words in a given pairoccurred together in recall. The occurrence ofarbitrary pairs, that is, the stimulus word fol-lowed not by its own response but by another,specified and randomly selected response fromthe list, was used as a chance base lineagainst which the occurrence of both forwardand backward associations was compared.There was a decided tendency to recall thepairs of words together in the stimulus-re-sponse order. The reverse associations, orrecall of the pairs in the response-stimulusorder, occurred significantly more frequentlythan the arbitrary pairings but significantlyless frequently than the forward sequences.

The occurrence of associative clustering inboth the forward and reverse direction hasbeen shown to be a positive function of the

10 While the publication date on this paper pre-dates Bousfield's 1953 paper, the authors include afootnote referring to a paper which Bousfield pre-sented at the American Psychological Associationmeeting in 1951. This 1951 paper apparently reportsthe results later presented in the frequently cited 1953paper.


associative strength between the two wordsin the pair. With adult subjects (Jenkins etal., 1958), clustering was above the chancelevel in the pairs of low associative strength.However, with children (Wicklund et al.,1965), clustering occurred at a chance levelin the pairs of low associative strength. In ad-dition, Wicklund, et al. examined the degreeof forward and reverse clustering for each in-dividual pair and concluded that the tendencyto cluster is strongly related to the presenceof bidirectional associative strength.

Associative clustering also occurs when thetwo words of a pair are not directly relatedbut are related to a third word not appearingin a stimulus list, that is, A elicits B, Belicits C, but A does not elicit C; A and Care the words presented (Cramer, 1965;Shapiro, & Palermo, 1967). Several studies(Bousfield, Whitmarsh, & Berkowitz, 1960;Marshall, 1967a, 1967b) have shown that as-sociative clustering can be predicted by theextent to which two words elicit common as-sociates, and the results of the study by Bous-field, Steward, and Cowan (1964) seem toindicate that clustering in a categorized listcan be predicted better with an index of as-sociative overlap, that is, the extent of whichthe words elicit common responses, than bymeans of an index of interitem associations,that is the extent to which the items inthe list elicit one another. In general, thereis a tendency for associative clustering to in-crease as a function of study-tests trials(Marshall, 1967a, 1967b). However, this re-lationship appears to be partly dependentupon the sex of the subject, the direction ofassociation, and the normative strength of theassociative pairs (Rosenberg, 1966). Withmediated clustering, there appears to be anincrease in clustering only when blocked pre-sentation is used (Shapiro & Palermo, 1967).

In a series of frequently cited studies, Deese(1959a, 1959b, 1960, 1961) demonstrated apositive relationship between interitem as-sociative strength (HAS) and free-recall per-formance. This relationship existed for bothintrusions and the probability of a word beingcorrectly recalled. The same relationship hasalso been found with children (Hess & Simon,1964; Simon & Hess, 1965). Also, the proba-bility that a given word will be recalled is

directly related to the number of other wordsin the list which elicit the given word as aresponse (Bodin, Crapsi, Deak, Morday, &Rust, 1965; Rothkopf & Coke, 1961).

Marshall and Cofer (1963) have consideredvarious indexes of associative relatedness.They note that although these various in-dexes are able to predict performance in avariety of situations, there is presently nosatisfactory theoretical interpretation of theway in which these associative relations func-tion. Deese (1968) has recently argued thatassociations are only manifestations or indexesof underlying patterns of relation and there-fore cannot cause organization as such. Never-theless, it is clear that preexperimental rela-tions influence performance on a free-recalltask. A better understanding of the way inwhich these relationships operate must awaitfuture research.

Bases of Organization

Factors which mediate clustering. Most ofthe studies investigating clustering have usedstimulus lists comprised of concrete nounswhich are clearly members of some abstractclass, for example, animals, countries, vege-tables, etc. Cohen et al. (1957) collected cul-tural norms for use in clustering studies. Fourhundred undergraduates at the University ofConnecticut responded to 43 different categorynames with the first four specific associatesthey could think of for each class of items.These responses were then ranked on thebasis of their frequency of occurrence to eachof the category names. Several studies (Bous-field, Cohen, & Whitmarsh, 1958; Bousfield,Steward, & Cowan, 1964; Cofer et al., 1966;Holroyd & Holroyd, 1961) have demonstratedboth superior recall and superior clustering inlists comprised of high-frequency responsesfrom these norms than for lists comprised oflow-frequency responses. Unfortunately, mostof these studies have confounded normativefrequency with frequency of usage as deter-mined from the Thorndike-Lorge (1944)count. However, one study (Bousfield, Stew-ard, & Cowan, 1964) indicates that clustering,but not recall, is apparently unaffected by afairly wide variation in Thorndike-Lorge fre-quency as long as the members of the cate-gories are equated on normative frequency.


A variety of attempts have been made todetermine if clustering could be obtained onother dimensions. Studies which have variedresponse dominance (Bousfield & Puff, 1964),structural characteristics of geometric designs(Bousfield et al., 1959), and structural char-acteristics of words (Pellet, 19S7) have metwith limited success. Attempts to obtain clus-tering on the basis of form class (Cofer &Bruce, 1965; Gonzales & Cofer, 19S9), syn-onyms (Cofer, 1959), and the semantic differ-ential (Cowan, 1964) have generally been un-successful. Bousfield and Cohen (1956b) dem-onstrated that for both males and femalesthere is a greater recall and greater clusteringof words representing interests of their ownsex than of words representing interests ofthe opposite sex.

Another way of determining various fac-tors which can mediate clustering is to lookat the various clusters of words which con-sistently occur in studies of subjective or-ganization. Mandler (1967) and Tulving(1968) have presented some interesting ex-amples of these types of clusters. The basisof organization in some of these clusters isquite obvious, while in others it is not at allobvious. Probably a whole variety of factorsdetermine the basis of organization which thesubject will use in studies of this nature.When an obvious form of organization isprovided, for example, studies of clustering,the subject will usually use the organizationprovided. When no obvious form of organiza-tion is provided, for example, studies of sub-jective organization, the subject will find moresubtle forms of organization to use. Whilesubjective organization is primarily basedupon intrasubject organization, there appearsto be substantial commonality among sub-jects in the way the recalls are organized(Earhard, 1967b; Tulving, 1962b, 1965). Thisindicates that the subjects are using relation-ships present in the stimulus list even thoughthe experimenter has assumed the words are"unrelated." While the nature of this or-ganization is somewhat subtle, the subjectsare nevertheless able to locate and make useof it in similar ways.

The role of context. Various contextualfactors present during the presentation of thestimulus list can also influence the amount

and type of organization which occurs. Gon-zalez and Cofer (1959) did a series of ex-periments on the effect of context on cluster-ing. Their technique was to present pairs ofwords rather than individual words, althoughthe subjects were usually asked to recall onlyone member of the pair. Five types of effectsare discussed.

1. Specificity effect. Forty unrelated ad-jectives were selected to modify 40 nouns rep-resenting four different categories. Clusteringof the nouns was reduced to the chance level.

2. Mediational effect. When a list of un-categorized nouns, which did not cluster whenpresented alone, was modified by adjectivesrepresenting discreet categories, which didcluster when presented alone, the recall of thenouns alone revealed significant clusteringwhen scored on the basis of the adjectiveswhich modified them during presentation. Thereverse was also found to be true; uncate-gorized adjectives would cluster on the basisof the nouns which they modified. Similar ef-fects were obtained using either normal wordorder, adjective-noun, or reverse word order,noun-adjective.

3. Mediated facilitation effects. When clus-tering adjectives and clustering nouns werepresented so that the adjective categories andthe noun categories were congruent, clusteringand recall of the nouns alone were facilitated.Similar effects were not. obtained with adverband verb combinations or with four wordcombinations, for example, adjective-noun-verb-adverb.

4. Mediated conflict effects. When cluster-ing adjectives and clustering nouns werepresented so that the adjective categories andnoun categories were not congruent, that is,adjectives from a given category modifiednouns from all of the categories, clusteringand recall were impaired even though for thespecific pairs the adjective appropriately modi-fied the noun.

5. Inappropriate modification effects. Whenthe adjective modifying a given noun is inap-propriate, for example, leafy dog, clusteringand recall were impaired, and the effect wasmore pronounced when the adjective-nounpair was recalled than when the noun onlywas recalled. Later, association data werecollected for these various situations (Cofer,


1960). Except for the mediated facilitationcondition, the results of the various condi-tions could be fairly well accounted for onthe basis of changes in the associative rela-tionships involved. In a more recent study,Cofer (1968) has shown that clustering canbe disrupted by presenting nouns in the con-text of a sentence.

Thus, it is clear that context can influenceorganization. Whether this effect results froma modification of the associative relationships(Cofer, 1960) or from the fact that contextdetermines the dimensions or distinctive fea-tures( Deese, 1968) which the subject selectsas the bases of his organization is not entirelyclear at present. It is quite possible thatthese two interpretations are in actuality re-ferring to the same basic process.

Organization and Recall

Generally speaking, there is a positive cor-relation between measures of organization andthe number of words recalled. However, thereis evidence (cf. Cofer, 1967, 1968; Cofe-r etal., 1966; Dallett, 1964) that at least undercertain conditions measures of organizationand recall can vary independently. In addi-tion, two studies (Laurence, 1966; Handler &Stephens, 1967) have indicated that at leastfor subjective organization different resultsare obtained with children than are obtainedwith adult subjects. Laurence (1966) ob-tained a direct relationship between age andrecall, but the values of SO were virtually thesame for all of the age groups, although thevalues obtained by college students were sig-nificantly higher than those for the children.While these results are contrary to the re-sults of two studies on clustering (Bousfield,Esterson, & Whitmarsh, 1968; Rossi, 1964)which found a direct relationship betweenclustering and age, they do raise an importantquestion. The extent to which this variationmay result from limitations of the measuresor from other factors is not entirely clear atthe present time. Also, it should be realizedthat the amount recalled will depend on theunit of analysis which is used.

A number of writers (e.g., Handler, 1967;Tulving, 1968) have suggested that recallis dependent upon organization. Probablythe best empirical evidence in support of such

a position is the study on alphabetic organi-zation (Tulving, 1962a). However, the situa-tions discussed above in which organizationand recall can vary independently raise somequestions as to the validity of this interpreta-tion. The explication of the relationship be-tween organization and recall is still in needof further investigation.

While only one study (Earhard, 1967a)has made a direct comparison between cate-gorized and unrelated lists, several studies(Dallett, 1964; Earhard, 1967a; Hathews,1954) have used a condition in which all ofthe words represented the same category,and one study (Tulving & Pearlstone, 1966)used a condition in which each word in thelist represented a different category. In gen-eral, there has been superior recall in thecategorized lists, although there is some in-dication (Dallett, 1964; Earhard, 1967a;Hathews, 1954) that this facilitation maydepend on whether recall is cued or noncuedand on the number of words per category inthe stimulus list.

Earhard, (1967a) included two free-recall(noncued) conditions in her study. One ofthese conditions consisted of words which allbegan with the same letter, and the otherconsisted of words which all began with differ-ent letters. Since alphabetic organization isnot effective unless the subjects are told aboutthe nature of the list (Tulving, 1962a), thelater condition is equivalent to a list of un-related words. There is virtually no differencebetween the two conditions over the 20 learn-ing trials, although this finding cannot begeneralized to other studies of categoricalclustering since the subjects were not toldabout the nature of the list. The mean num-ber of words recalled per trial in the cued-recall conditions (the subjects in these con-ditions were instructed as to the nature ofthe list) was superior to that in the free-recall conditions for 1, 2, 3, and 4 words percategory, equivalent for 6 and 8 words percategory, and inferior for 12 words per cate-gory.

THEORETICAL ISSUES

Various attempts have been made to de-velop a theoretical explanation for the cluster-ing phenomenon. None of the present inter-


pretations is completely satisfactory, that is,none can explain all of the relevant data. Someof the major theoretical positions and issueswhich have developed from the study of or-ganization in free recall will be discussed inthis section of the paper.

Perception of the Superordinate

The early Bousfield papers (e.g., Bousfield,19S3; Bousfield & Cohen, 1953) interpretedclustering in terms of (a) the habit strengthof an item which derived from the reinforce-ment it received before and during the ex-periment and (b) a relatedness incrementwhich was taken as a hypothetical incrementadded to habit strength. It was hypothesizedthat when a word is recalled an increment isadded to other words which are related to it.The strength of these two factors, subject tooscillation, determines the probability ofwhether the next item recalled will be fromthe same or from a different category and,therefore, determines the amount of cluster-ing obtained. The finding that clusteringchanges as a function of the stage of recall(Bousfield, 1953; Bousfield & Cohen, 1953,1955; Cohen & Bousfield, 1956) was ex-plained by making the assumption that thelatent period is an index of habit strength,although as Tulving (1968) has pointed out,these changes may be partly the result ofsuch factors as the tendency to recall terminalitems first.

The relatedness increment was assumed tobe related to Hebb's (1949) conception ofthe development of superordinant perceptions.According to this conception, the repeatedperception of a set of related items resultsin the formation of a new superordinant sys-tem. Eventually the activation of a singleperception unit will suffice to arouse thesuperordinant system which, in turn, willfacilitate its subordinant units. Phrased interms of the experimental situation of theBousfield studies, the perception or recall ofa single word will tend to activate the super-ordinant systems which correspond to thecategory represented by the word. Once thissuperordinant system is activated, it will tendto facilitate the perception and recall of otherwords belonging to the same category.

In more recent papers (e.g., Bousfield &

Bousfield, 1966; Bousfield & Puff, 1965;Bousfield, Steward, & Cowan, 1964), Bous-field has tended to interpret clustering morein terms of associative factors than in termsof the superordinate mechanism suggestedearlier. Recently, Deese (1968) has sug-gested that while this supraordinate-sub-ordinate structure is too elementary to de-scribe adequately the complexities or therelationships involved, it is probably closer tothe truth than the alternative interpretationspresently available.

Coding Interpretation

Somewhat related to the foregoing positionis the coding interpretation of clustering.This position is based primarily on Miller's(1956) concept of chunking. The basic no-tion is that the subject recodes the words inthe stimulus list into the respective categoriesand then stores the category label (or somerepresentation of it) in memory. When itcomes time to recall the words he recalls thecategory as a whole. This interpretation issomewhat supported by the finding that if atleast one word in a category is recalled, theaverage number of words recalled per cate-gory is remarkably consistent over a fairlywide range of conditions. (Cohen, 1966; Tulv-ing & Pearlstone, 1966). Whether recall isactually a two-level search process in whichthe subject first searches for, or tries to recall,a category, for example, its name, and thentries to recall the words within that categoryas some have suggested (Cohen, 1966; Shep-herd, 1966) is something on which little dataare presently available. The best evidence tosupport such an interpretation comes froma study by Segal (1969). In this study acategorized list of words was used which in-cluded the category names. There was adistinct tendency for the category name tobe recalled prior to the specific instances ofthat category.

One of the ways in which the coding in-terpretation differs from the superordinateinterpretation is that the latter is made interms of the habit strength of individualitems. This habit strength can be augmented,apparently via some sort of generalizationmechanism, by the recall or perception of re-lated items. Thus, this augmentation can


occur either at the time of presentation orat the time of recall. The coding interpreta-tion, on the other hand, assumes that therelated words are encoded at the time ofpresentation into their respective categories,or chunks, and that during recall thesechunks are recalled as a unit.

It is worthwhile to note one limitation tothe notion that the category name as such isused as the coding cue. If instances of a non-exhaustive category, for example animals, areused in the stimulus list, it is rather difficultto explain why the subject recalls only thosemembers of the category presented. If theinstances were coded only as animals, therewould be no way for the subject to differ-entiate between those animals which werepresented and those which were not. Thus, itseems unlikely that in the strict sense thecategory name is used as the coding cue,although it is certainly possible that this maybe one facet of the cue used by the subject.This is perhaps one explanation why ex-haustive categories are better recalled thannonexhaustive categories.

Associative Interpretation

The role of associative relatedness in freerecall was discussed in an earlier section. Asmight be expected, the question has beenraised as to whether or not clustering can beexplained solely in terms of associative re-lationships, that is, is it necessary to postulatean additional concept such as superordinationor coding? Several studies (Bousfield & Puff,1965; Field, 1969; Marshall, 1967b) havecontrasted the associative and the codinginterpretations. In general, the conclusionhas been that both associational and codingprocesses are involved in most free-recallsituations (cf. Gofer, 1965, 1966). Marshall(1967b) has shown that when pairs of wordsare matched for association relatedness,those pairs which are categorized cluster toa greater extent than noncategorized pairs atlow and middle levels of associative related-ness but not at high levels.

It has been suggested (e.g., Deese, 1961)that subjects reconstruct the list at the timeof recall. Cofer (1967) has found that whensubjects are asked to generate additionalitems to match list length they apparently

make use of both list-name and list-memberassociations. The fact that subjects do notmatch list length under typical free-recall pro-cedures has been taken as evidence that sub-jects edit their recall (Cofer, 1967). Theadditional finding that differential perform-ance as a function of associative strength isnot obtained with a recognition test (Cofer,1967; Field, 1969) suggests that perhapsassociative relationships influence performanceby means of varying the likelihood that agiven word is generated by the subject. Thus,it is possible that during presentation thesubject learns certain characteristics aboutthe list in addition to some of the specificwords and then uses this information duringrecall to generate words and to make decisionsas to whether or not a particular word wason the list.

Deese (1968) has recently argued that as-sociations cannot cause organization, althoughPostman (1968) has suggested that the termassociation can be used in at least six differentways. It is possible, of course, that bothassociative and categorical relationships canbe explained in terms of some common under-lying process.

Subjective Organization

Subjective organization does not actuallyoffer a separate theoretical interpretation ofclustering, although a number of importanttheoretical contributions have resulted fromthe study of subjective organization. It wassuggested earlier in this paper that the samebehavioral processes are being studied in allthree paradigms. The main difference is inthe way the problem is approached, althoughthis difference may prove to be very importantto the future understanding of the processesinvolved. Perhaps the most important con-tribution which the study of subjectiveorganization has provided is the demonstrationthat there are discrepancies between the learn-ing situation as defined by the experimenterand the learning situation as perceived by thesubject.

Another issue which has been raised isthe distinction between availability and ac-cessibility (Tulving & Pearlstone, 1966). Ac-cording to this interpretation, items can beavailable for recall, that is, in storage, but


unless sufficient retrieval cues are present atthe time of recall the items will not be re-called. Several studies (Dong & Kintsch,1968; Tulving & Osier, 1968; Tulving &Pearlstone, 1966) have demonstrated that re-call can be increased if subjects are providedwith relevant cues at the time of recall. How-ever, Slamecka (1968), in a series of ex-periments, was unable to obtain facilitationby providing the subjects with varying num-bers of items from the stimulus list as re-trieval cues. Tulving (1967) has suggesteda limited-capacity retrieval system in whichthe main limitation in memory is not in thestorage of items but is determined by thenumber of accessible memory units. Thecontents or nature of the units is deemed tobe of little importance.

The occurrence of intertrial repetitionsabove a chance level, as discussed in thesection on measurement, provides evidence forthe existence of higher order units in free re-call. However, Tulving (1966; Tulving &Osier, 1967) obtained additional verificationin a series of transfer studies. When the sec-ond list contains items from the first list,negative transfer is obtained relative to acontrol group which receives completely dif-ferent items in the two lists. This effect isobtained both when subjects are transferredfrom a shorter list to a longer list (Tulving,1966) and when they are transferred from alonger list to a shorter list (Tulving & Osier,1967). Similarly, differential transfer has beenobtained with categorized lists when the wordsin successive lists represent either the same ordifferent categories (Roberts & Smith, 1966;Shuell, 1968). Both studies obtained superiorrecall and clustering when different categoriesare represented in the successive lists.

Throughout the present paper an attempthas been made to evaluate the research onorganization in free recall. The quantificationof organization has been achieved, and muchhas been learned about the relationship be-tween organization and memory. However,many questions remain unanswered, and someof the directions which future research maytake have been suggested at various pointsthroughout the paper. A number of writers(e.g., Handler, 1967; Tulving, 1968) havesuggested that recall is dependent upon organ-

ization. While this suggestion appears reason-able, it must be admitted that the evidencewhich actually supports such a position is notoverwhelming at present. Perhaps the use ofmeasures of organization which are independ-ent of recall will permit us to obtain addi-tional data relevant to this important ques-tion.

It should be remembered that clustering isbasically an output phenomena from whichwe infer some sort of organizational processon the part of the subject. Since organizationis present in the stimulus list, and this istrue for subjective organization as well as forthe more traditional clustering studies, theinference is basically concerned with the wayin which the subject uses this organization.

Finally, it should be noted that the locusof the organizational process has not beendetermined. The organization which is ob-served during recall could occur at the time ofinput, that is, presentation of the stimuluslist, or at the time of retrieval, that is, whenthe subject is recalling the words. Slamecka(1968) has suggested that the items are storedindependently along with a general represen-tation of the list structure. At the time ofrecall this general representation forms aretrieval plan which then guides the subject'ssearch for the specific items. However, thereis increasing evidence (Anisfeld & Knapp,1968; Rohwer, Shuell, & Levin, 1967; Tulving& Osier, 1968) that encoding or organiza-tion must occur at the time of presentation inorder to be effective.

REFERENCES

ANISFELD, M., & KNAPP, M. Association, synonymity,and directionality in false recognition. Journalof Experimental Psychology, 1968, 77, 171-179.

BODIN, A. M., CRAPSI, L. A., DEAK, M. R., MORDAY,T. R., & RUST, L. D. Prediction of free recall

from word-association measures: A replication.Journal of Experimental Psychology, 1965, 69,103-105.

BOUSFIELD, A. K., & BOUSFIELD, W. A. Measurementof clustering and of sequential constancies in re-peated free recall. Psychological Reports. 1966,19, 935-942.

BOUSFIKLD, W. A. The occurrence of clustering inthe recall of randomly arranged associates. Journalof General Psychology, 1953, 49, 229-240.

BOUSFIELD, W. A., BERKOWITZ, H., & WHITMARSH,G. A. Associative clustering in the recall of


minimally meaningful geometric designs. CanadianJournal of Psychology, 1959, 13, 281-287.

BOUSFIELD, W. A., & COHEN, B. H. The effects ofreinforcement on the occurrence of clustering in therecall of randomly arranged associates. Journal ofPsychology, 1953, 36, 67-81.

BOUSFIELD, W. A., & COHEN, B. H. The occurrenceof clustering in the recall of randomly arrangedwords of different frequencies-of-usage. Journalof General Psychology, 1955, 52, 83-95.

BOUSFIELD, W. A., & COHEN, B. H. Clustering inrecall as a function of the number of word-categories in stimulus-word lists. Journal of Gen-eral Psychology, 1956, 54, 95-106. (a)

BOTJSFIELD, W. A., & COHEN, B. H. Masculinity-femininity in the free recall of a categorized stim-ulus word list. Perceptual and Motor Skills, 1956,6, 159-165. (b)

BOUSFIELD, W. A., COHEN, B. H., & WHITMARSH,G. A. Associative clustering in the recall of wordsof different taxonomic frequencies of occurrence.Psychological Reports, 1958, 4, 39-44.

BOUSFIELD, W. A., ESTERSON, J., & WHITMARSH,G. A. A study of developmental changes in con-ceptual and perceptual associative clustering.Journal of Genetic Psychology, 1958, 92, 95-102.

BOUSPIELD, W. A., & PUFF, C. R. Clustering as afunction of response dominance. Journal of Experi-mental Psychology, 1964, 67, 76-79.

BOUSFIELD, W. A., & PUFF, C. R. Determinants ofthe clustering of taxonomically and associativelyrelated word pairs. Journal of General Psy-chology, 1965, 73, 211-221.

BOUSFIELD, W. A., PUFF, C. R., & COWAN, T. M.The development of constancies in sequentialorganization during repeated free recall. Journalof Verbal learning and Verbal Behavior, 1964,3, 489-495.

BOUSFIELD, W. A., & SEDOEWICK, C. H. W. Ananalysis of sequences of restricted associative re-sponses. Journal of General Psychology, 1944, 30,149-16S.

BOUSFIELD, W. A., STEWARD, J. R., & COWAN, T. M.The use of free associational norms for the pre-diction of clustering. Journal of General Psy-chology, 1964, 70, 205-214.

BOUSFIELD, W. A., WHITMARSH, G. A., & BERKO-WITZ, H. Partial response identities in associativeclustering. Journal of General Psychology, 1960,63, 233-238.

BRAND, H., & WOODS, P. J. The organization of theretention of verbal material. Journal of GeneralPsychology, 1958, 58, 55-68.

COFER, C. N. A study of clustering in free recallbased on synonyms. Journal of General Psy-chology, 1959, 60, 3-10.

COFER, C. N. An experimental analysis of the roleof context in verbal behavior. Transactions of theNew York Academy of Sciences, 1960, 22, 341-347.

COFER, C. N. On some factors in the organizationalcharacteristics of free recall. American Psycholo-gist, 1965,20, 261-272.

COPER, C. N. Some evidence for coding processesderived from clustering in free recall. Journal ofVerbal Learning and Verbal Behavior, 1966, 5,188-192.

COFER, C. N. Does conceptual organization in-fluence the amount retained in immediate freerecall. In B. Kleinmutz (Ed.), Concepts and thestructure of memory. New York: Wiley, 1967.

COFER, C. N. Free recall of nouns after presentationin sentences. Journal of Experimental Psychology,1968, 78, 145-152.

COFER, C. N., & BRUCE, D. R. Form-class as thebasis for clustering in the recall of nonassociatedwords. Journal of Verbal Learning and VerbalBehavior, 1965, 4, 386-389.

COFER, C. N., BRUCE, D. R., & REICHER, G. M.Clustering in free recall as a function of certainmethodological variations. Journal of Experi-mental Psychology, 1966, 71, 858-866.

COHEN, B. H. An investigation of receding in freerecall, Journal of Experimental Psychology, 1963,65, 368-376. (a)

COHEN, B. H. Recall of categorized word lists.Journal of Experimental Psychology, 1963, 66,227-234. (b)

COHEN, B. H. Some-or-none characteristics of codingbehavior. Journal of Verbal Learning and VerbalBehavior, 1966, 5, 182-187.

COHEN, B. H., & BOUSFIELD, W. A. The effects of adual-level stimulus-word list on the occurrence ofclustering in recall. Journal of General Psychology,1956, 55, 51-58.

COHEN, B. H., BOUSFIELD, W. A., & WHITMARSH,G. A. Cultural norms for verbal items in 43 cate-gories. (ONR Tech. Rep. No. 22, Contract Nonr-631(00)) Washington, D. C.: United StatesGovernment Printing Office, 1957.

COHEN, B. H., SAKODA, J. M., & BOUSFIELD, W. A.The statistical analysis of the incidence of cluster-ing in the recall of randomly arranged associates.(ONR Tech. Rep. No. 10, Contract Nonr-631(00)Washington, D. C. : United States GovernmentPrinting Office, 1954.

COWAN, T. M. The semantic differential as a basisfor obtaining clustering. Unpublished doctoral dis-sertation, University of Connecticut, 1964. (Dis-sertation Abstracts, 1965, 25, 4824-4825)

CRAMER, P. Mediated clustering and importationwith implicit verbal chains. Psychonomic Science,1965, 2, 165-166.

DALLETT, K. M. Number of categories and categoryinformation in free recall. Journal of ExperimentalPsychology, 1964, 66, 1-12.

DEESE, J. Influence of inter-item associative strengthupon immediate free recall. Psychological Re-ports, 1959, 5, 305-312. (a)

DEESE, J. On the prediction of occurrence of par-ticular verbal intrusions in immediate recall.Journal of Experimental Psychology, 1959, 58,17-22. (b)

DEESE, J. Frequency of usage and number of wordsin free recall: The role of association. PsychologicalReports, 1960, 7, 337-344.

CLUSTERING Iff FREE RECALL 373

DEESE, J. From the isolated verbal unit to con-nected discourse. In C. N. Cofer (Ed.), Verballearning and verbal behavior. New York: Mc-Graw-Hill, 1961.

DEESE, J. Association and memory. In T. R. Dixon& D. L, Horton (Eds.), Verbal behavior andgeneral behavior theory. Englewood Cliffs, N. J.:Prentice-Hall, 1968.

DONG, T. & KINTSCH, W. Subjective retrieval cuesin free recall. Journal of Verbal Learning andVerbal Behavior, 1968, 7, 813-816.

EARHARD, M. Cued recall and free recall as a func-tion of the number of items per cue. Journal ofVerbal Learning and Verbal Behavior, 1967, 6,2S7-26S. (a)

EARHARD, M. Subjective organization and list organi-zation as determinants of free-recall and serial-recall memorization. Journal of Verbal Learningand Verbal Behavior, 1967, 6, 501-507. (b)

FAOAN, J. F. Ill Measuring verbal recall II: TheITR score expressed as a ratio. PsychonomicScience, 1968, 11, 205.

FIELD, W. H. Coding and association in the recalland recognition of categorized words. Unpublisheddoctoral dissertation, State University of NewYork at Buffalo, 1969.

GONZALES, R. C., St COFER, C. N. Exploratorystudies of verbal context by means of clustering infree recall. Journal of Genetic Psychology, 1959,95, 293-320.

HEBB, D. O. The organization of behavior. NewYork: Wiley, 1949.

HELSON, R. M., & COVER, A. Specificity-generalityof classificatory categories as a variable in recall.Perceptual and Motor Skills, 1956, 6, 233-236.

HESS, J. L., & SIMON, S. Extra-list intrusions inimmediate free recall as a function of associativestrength in children. Psychological Reports, 1964,14, 92.

HOLROYD, R. G., & HOLROYD, J. C. Associativeclustering in a retroaction paradigm. Journal ofGeneral Psychology, 1961, 64, 101-104.

JENKINS, J. J., MINK, W. D., & RUSSELL, W. A.Associative clustering as a function of verbalassociation strength. Psychological Reports, 1958,4, 127-136.

JENKINS, J. J., & RUSSELL, W. A. Associative cluster-ing during recall. Journal of Abnormal and SocialPsychology, 1952, 47, 818-821.

LAURENCE, M. W. Age differences in performance andsubjective organization in the free-recall learningof pictorial material. Canadian Journal of Psy-chology, 1966, 20, 388-399.

MANDLER, G. Organization and memory. In K. W.Spence & J. T. Spence (Eds.), The psychology oflearning and motivation. Vol. 1. New York:Academic Press, 1967.

MANDLER, G., & PEARLSTONE, Z. Free and constrainedconcept learning and subsequent recall. Journal ofVerbal Learning and Verbal Behavior, 1966, 5,126-131.

MANDLER, G., & STEPHENS, D. The development offree and constrained conceptualization and sub-

sequent verbal memory. Journal of ExperimentalChild Psychology, 1967, 5, 86-93.

MARSHALL, G. R. Effect of total association andconceptual cohesiveness among words on recall,clustering, and recognition association. Psycholog-ical Reports, 1967, 20, 29-44. (a)

MARSHALL, G. R. Stimulus characteristics contributingto organization in free recall. Journal of VerbalLearning and Verbal Behavior, 1967, 6, 364-374.(b)

MARSHALL, G. R., & COFER, C. N. Associative in-dices as measures of word relatedness: A summaryand comparison of ten methods. Journal of VerbalLearning and Verbal Behavior, 1963, 1, 408-421.

MATHEWS, R. Recall as a function of number ofclassificatory categories. Journal of ExperimentalPsychology, 1954, 47, 241-247.

MAYHEW, A. J. Interlist changes in subjective organ-ization during free-recall learning. Journal ofExperimental Psychology, 1967, 74, 425-^30.

MCNULTY, J. A. The measurement of "adoptedchunks" in free-recall learning. Psychonomic Sci-ence, 1966, 4, 71-72.

MILLER, G. A. The magical number seven, plus orminus two: Some limits on our capacity toprocess information. Psychological Review, 1956,63, 81-97.

MURDOCK, B. B., JR. The serial position effect of freerecall. Journal of Experimental Psychology, 1962,64, 482-488.

PELLET, B. The use of structural characteristics inthe clustering of verbal associates. Unpublisheddissertation, University of Connecticut, 1957. (Dis-sertation Absracts, 1957, 17, 2680).

POSTMAN, L. Association and performance in theanalysis of verbal learning. In T. R. Dixon &D. L. Horton (Eds.), Verbal behavior and gen-eral behavior theory. Englewood Cliffs, N. J.:Prentice-Hall, 1968.

POSTMAN, L., & KEPPEL, G. Retroactive inhibition infree recall. Journal of Experimental Psychology,1967, 74, 203-211.

POSTMAN, L. & KEPPEL, G. Conditions determiningthe priority of new items in free recall. Journalof Verbal Learning and Verbal Behavior, 1968,7, 260-263.

PUFF, C. R. Clustering as a function of the sequentialorganization of stimulus word lists. Journal ofVerbal Learning and Verbal Behavior, 1966, 5,503-506.

PUFF, C. R., & HYSON, S. P. An empirical comparisonof two measures of intertrial organization in freerecall. Psychonomic Science, 1967, 9, 329-330.

ROBERTS, W. A., & SMITH, I. The effects of categ-orization and categorical repetition on successiverecalls of word lists. Psychonomic Science, 1966,S, 225-226.

ROBINSON, J. A. Category clustering in free recall.Journal of Psychology, 1966, 62, 279-285.

ROHWER, W. D., JR., SHUELL, T. J., & LEVIN, J. R.Context effects in the initial storage and retrievalof noun pairs. Journal of Verbal Learning andVerbal Behavior, 1967, 6, 796-801.


ROSENBERG, S. Associative clustering and repeatedtrials. Journal of General Psychology, 1966, 74,89-96.

Rossi, E. Development of classificatory behavior.Child Development, 1964, 35, 137-142.

ROTHKOPF, E. Z., & COKE, E. U. The prediction offree recall from word association measures.Journal of Experimental Psychology, 1961, 62,433-438.

SEGAL, E. M. Hierarchical structure in free recall.Journal of Experimental Psychology, 1969, 80,59-63.

SHAPIRO, S. L., & PALERMO, D. S. Mediated cluster-ing in free recall. Journal of Experimental Psy-chology, 1967, 75, 36S-371.

SHEPARD, R. N. Learning and recall as organizationand search. Journal of Verbal Learning and VerbalBehavior, 1966, 5, 201-204.

SHUELL, T. J. Retroactive inhibition in free-recalllearning of categorized lists. Unpublished doctoraldissertation, University of California, Berkeley,1967.

SHUELL, T. J. Retroactive inhibition in free-recalllearning of categorized lists. Journal of VerbalLearning and Verbal Behavior, 1968, 7, 797-805.

SHUELL, T. J., & KEPPEL, G. Retroactive inhibitionas a function of learning method. Journal of Ex-perimental Psychology, 1967, 75, 457-463.

SHUELL, T. J., & KEPPF.L, G. Item priority in freerecall. Journal of Verbal Learning and VerbalBehavior, 1968, 7, 969-971.

SIMON, S., & HESS, J. L. Supplementary report:Influence of interitem associative strength uponimmediate free recall in children. PsychologicalReports, 1965, 16, 451-455.

SLAMECKA, N. J. An examination of trace storage infree recall. Journal of Experimental Psychology,1968, 76, 504-513.

THORNDIKE, E. L., & LOROE, I. The teachers wordbook of 30,000 words. New York: Columbia Uni-versity, Teachers College, 1944.

TULVINO, E. The effect of alphabetical subjectiveorganization on memorizing unrelated words.Canadian Journal of Psychology, 1962, 16, 185-191. (a)

TULVING, E. Subjective organization in free recallof "unrelated" words. Psychological Review,1962, 69, 344-354. (b)

TULVING, E. The effect of order of presentation onlearning of "unrelated" words. Psychonomic Sci-ence, 1965, 3, 337-338.

TULVINO, E. Subjective organization and effects ofrepetition in multitrial free-recall learning. Journalof Verbal Learning and Verbal Behavior, 1966,5, 193-197.

TULVING, E. The effects of presentation and recallof material in free-recall learning. Journal ofVerbal Learning and Verbal Behavior, 1967, 6,175-184.

TULVING, E. Theoretical issues in free recall. In T. R.Dixon & D. L. Horton (Eds.), Verbal behaviorand general behavior theory. Englewood Cliffs,N. J.: Prentice-Hall, 1968.

TULVING, E., McNuLiy, J. A., & OZIER, M. Vividnessof words and learning to learn in free-recall learn-ing. Canadian Journal of Psychology, 1965, 19,242-252.

TULVING, E., & OSLER, S. Transfer effects in whole/part free-recall learning. Canadian Journal of Psy-chology, 1967, 21, 253-262.

TULVING, E., & OSLER, S. Effectiveness of retrievalcues in memory for words. Journal of Experi-mental Psychology, 1968, 77, 593-601.

TULVING, E., & PATKAU, J. E. Concurrent effects ofcontextual constraint and word frequency on im-mediate recall and learning of verbal material.Canadian Journal of Psychology, 1962, 16, 83-95.

TULVING, E., & PATTERSON, R. D. Functional unitsand retrieval processes in free recall. Journalof Experimental Psychology, 1968, 77, 239-248.

TULVING, E., & PEARLSTONE, Z. Availability versusaccessibility of information in memory for words.Journal of Verbal Learning and Verbal Behavior,1966, 5, 381-391.

TULVING, E., & THORTON, G. B. Interaction betweenproaction and retroaction in short-term retention.Canadian Journal of Psychology, 1959, 13, 255-265.

WICKLUND, D. A., PALERMO, D. S., & JENKINS, J. J.Associative clustering in the recall of children asa function of verbal association strength. Journalof Experimental Child Psychology, 1965, 2, 58-66.

WINOGRAD, E. List differentiation, recall, and categorysimilarity. Journal of Experimental Psychology,1968, 78, 510-515.

(Received January 21, 1969)

Documents

CLUSTERING AND ORGANIZATION IN FREE RECALL...between organization and memory. This pur-pose can best be represented by a quote from Bousfield's (1953) original paper The theoretical