self-reference and th encodine g of personal information · in the processin ogf people-related...

Journal of Personality and Social Psychology1977, Vol. 35, No. 9, 677-688

Self-Reference and the Encoding of Personal Information

T. B. Rogers, N. A. Kuiper, and W. S. KirkerUniversity of Calgary, Canada

The degree to which the self is implicated in processing personal informationwas investigated. Subjects rated adjectives on four tasks designed to forcevarying kinds of encoding: structural, phonemic, semantic, and self-reference.In two experiments, incidental recall of the rated words indicated that adjec-tives rated under the self-reference task were recalled the best. These resultsindicate that self-reference is a rich and powerful encoding process. As anaspect of the human information-processing system, the self appears to func-tion as a superordinate schema that is deeply involved in the processing, inter-pretation, and memory of personal information.

Present research and theory in personalityappear to be placing more and more empha-sis on how a person has organized his or herpsychological world. Starting with Kelly's(1955) formulation of personal constructs,we see a gradual emergence of a number ofavenues of inquiry that use this as their focalpoint. In person perception, the concept oflay personality theory stresses that the ob-server's analytic network of expected traitcovariations is an integral part of how heprocesses (and generates) interpersonal data(Hastorf, Schneider, & Polefka, 1970). Bernand Allen (1974), in their embellishment ofAllport's (1937) idiographic position, arguethat an important determinant of predictiveutility of trait measurement is the mannerin which the respondent has organized his orher view of the trait being measured. Theseauthors see the overlap between the respond-ent's and the experimenter's concept of thetrait as a necessary prerequisite of predic-tion. Attribution theory (Jones et al., 1971)is another example of this increased accenton personal organization. Here the emphasisis on how the subject explains past behavior

This research was supported by a grant from theCanada Council. We would like to thank the fol-lowing persons for their useful ideas and commentson earlier drafts: F. I. M. Craik, E. J. Rowe, P. J.Rogers, H. Lytton, J. Clark, J. Ells, C. G. Costello,and especially one anonymous reviewer.

Requests for reprints should be sent to T. B.Rogers, Department of Psychology, The Universityof Calgary, Calgary, Alberta, Canada, T2N 1N4.

and how these explanations are organized inan attributional network. The commonthread in all of these contemporary researchareas is the notion that the cognitions of aperson, particularly their manner of organ-ization, should be an integral part of ourattempts to explain personality and behavior.

Of concern in the present article is theconstruct of self and how it is implicated inthe organization of personal data. Our gen-eral position is that the self is an extremelyactive and powerful agent in the organizationof the person's world. More specifically, thepresent research was designed to determineif self-reference serves a meaningful functionin the processing of certain kinds of informa-tion. That is, we attempted to determine therelative strength of self-reference as an agentin the processing of people-related informa-tion.

The self is defined as an abstract repre-sentation of past experience with personaldata. Phenomenologically, it is a kind of vagueidea about who the person thinks he or sheis. It probably develops to help the personkeep track of the vast amounts of self-rele-vant information encountered over a lifetime.The self, then, represents the abstracted es-sence of a person's perception of him or her-self. A more formal definition of self is toview it as a list of terms or features that havebeen derived from a lifetime of experiencewith personal data. More than likely a por-tion of the list consists of general terms—not unlike traits—that represent the ab-

677

678 T. ROGERS, N. KUIPER, AND W. KIRKER

stracted essentials of a person's view of self.In addition to these general terms, there arealso some more specific entries in the self.These relate to less salient and more situa-tion-specific aspects of self-perception aswell as to specific behaviors. This definitionis supported by Jones, Sensening, and Haley(1974). They simply asked subjects to de-scribe their "most significant characteristics."The most frequent entries in the obtainedresponse protocols were positively wordedterms such as sensitive, intelligent, andfriendly. These appear to be the generalterms in the self and appear to resembletraits. Jones et al. (1974, p. 38) also presenta single response protocol. Of interest hereis the tendency for conditionals (e.g., suchsituational hedges as sometimes) to emergelater in the protocol. Furthermore, as theprotocol develops, the entries tend to relateto more specific situations than did the earlierterms (e.g., "have a hang-up about authorityfigures"). These latter entries appear to bethe specific terms in the self.

One of the main functions of the self isto help the individual process personal data.When a person encounters a situation involv-ing personal information, this structure isactivated and becomes part of the availableinformation-processing system. For example,when students encounter a list of character-istics of a psychopathological state (e.g., inan introductory psychology lecture), theytend to interpret (and attempt to remember)these by referring them to their own views ofself. Such a strategy could lead to the "medi-cal student syndrome," where students beginto see themselves in the varying states de-scribed by the lecturer. In extreme cases,some students can be convinced they areraving lunatics—despite repeated warningsof the instructor. Other examples of this self-reference phenomenon can occur in situationsinvolving personal feedback, expressive be-haviors, and the processing of informationabout other people.

The central aspect of self-reference is thatthe self acts as a background or settingagainst which incoming data are interpretedor coded. This process involves an interaction

between the previous experience of the in-dividual (in the form of the abstract struc-ture of self) and the incoming materials. Theinvolvement of the self in the interpretationof new stimuli imparts a degree of richnessand fullness to the input because of the avail-ability of the immense amounts of previousexperience embodied in the self.

The interaction between new input andprevious experience, postulated to be centralto self-reference, has been modeled in thecognitive literature under the concept ofschema or prototype (Bartlett, 1932; Posner& Keele, 1968). For example, subjects showna series of dot patterns tend to abstract aprototypical visual pattern and use it as astandard in a memory task (Posner & Keele,1970). In personality, several recent papershave suggested that personal data are pro-cessed using schemata or prototypes. Markus(1977), using a series of measurements, cate-gorized subjects by whether dependence waspart of their self. Schematics, or those withdependence as part of their self, were thosewho rated themselves as extreme on severaldependence items, as well as indicating thatthey viewed dependence as important. As-chematics were midrange on dependence andlow on importance ratings, representing sub-jects who did not have dependence in theirgeneral concept of self. On the basis of im-pressive convergent evidence, Markus (1977)found that schematics and aschematicsshowed differences in how they processed per-sonal data. For example, schematics weremore resistant to incorrect personal informa-tion than were the aschematics. These datasuggest that the traits, such as those foundin the self, serve an important function inprocessing certain kinds of information.

Cantor and Mischel (1977) tested theproposition that traits function as prototypesusing a recognition memory task. Subjectswere shown a series of statements that rep-resented an introvert. When faced with atask requiring recognition of these statementsfrom among some new introvert statements,subjects tended to misidentify some of thenew items as having been original state-ments. This suggests that the concept of in-

SELF-REFERENCE AND ENCODING 679

trovert mediated subjects' processing of theinitial set of statements. This mediation waspostulated to be in the form of a prototype,which represents an abstraction of the con-cept of introvert from the initial items. Theobserved memory bias for new, yet conceptu-ally related items reflected the involvementof this abstraction.

The Markus (1977) and Cantor and Mis-chel (1977) data indicate that traits are im-plicated in information-processing functions.They appear to be involved in the organiza-tion, storage, and retrieval of personality-related information. Our view of self placesthese traitlike schemata or prototypes as gen-eral terms in the feature list making up theself.

Rogers, Rogers, and Kuiper (Note 1) ex-plored the manner in which this set ofschemata is involved in processing personaldata. They hypothesized that the self func-tions like a grand or superordinate schema.If the self is a schema, it should be possibleto observe the kinds of memory biases docu-mented by Cantor and Mischel (1977). Inone study (Rogers et al., Note 1, Experiment2), subjects filled out self-ratings on 84 ad-jectives. Two and a half months later, thesesame subjects participated in a recognitionmemory study involving these same adjec-tives. They first saw a randomly selected setof 42 of the words, and then had to recognizethese from among the total set of 84. If theself was involved as a schema, subjectsshould tend to falsely recognize new itemsthat were rated as self-descriptive (i.e., Pos-ner & Keele, 1970). Each subject's recogni-tion protocol was divided into high, neutral,and low self-descriptive categories on thebasis of their self-ratings. Performance forthe 42 items initially shown in the recogni-tion study was not affected by degree ofself-reference. However, performance on thenew or distractor items (correct rejects) be-came poorer as degree of self-reference in-creased. In other words, more false alarmsoccurred as the adjectives became more self-descriptive. This clearly confirms the predic-tion derived from viewing the self as aschema.

To review, self-reference can be seen as aprocess involving the schema of self. Thisprocess involves the interaction between pre-vious experience with personal data and newstimulus input. When self-reference is in-volved, it should provide a useful device forencoding or interpreting incoming informa-tion by virtue of accessing the extensive pastexperience abstracted in the self. Contactwith the reservoir of history embodied in theself should provide considerable embellish-ment and richness to an incoming stimulus.Rogers (in press) explored this possibilityusing recognition memory for personalityitems. Subjects instructed to "read the item,decide if it describes you, and use this tohelp your memory" performed significantlybetter than subjects receiving either no ordifferent (i.e., imagery) instructions. Thesedata, in combination with some older re-search (e.g., Cartwright, 1956), indicate thatexplicit instructions to use the self in amemory task increase performance, whichsupports the claim that self-reference servesto enrich input.

This enriching aspect of self-reference isthe focal point of the present article. Ourmajor concern is how powerful self-referenceis as an encoding device. More specifically,self-reference is compared to several otherencoding processes in an effort to determinethe relative degree of richness and embellish-ment that self-reference imparts to the en-coding of adjectives.

The experimental manipulation used in thepresent context is an incidental recall para-digm, in which subjects make different kindsof ratings on a set of words. For example, asubject rates whether a given word means thesame as a target word. This would be asemantic rating, as the subject must extractthe meaning of the word to perform the task.The same subject rates whether another wordis written in big letters. This is a structuralcoding task, since all the subject has to dois inspect the structure of the stimulus itemrather than extract the meaning of the word.Other words are rated on a phonemic task,which involves deciding whether a wordrhymes with a target. The self-reference rat-


ing, which subjects perform on some of thewords, involves the respondent's decidingwhether the word describes him or her. Whenthe rating task is completed, each subject hasrated one fourth of the words on each of thefour rating tasks. These four tasks arethought to vary in depth, or semantic rich-ness, from the structural task as the mostshallow to the semantic and/or self-ratingtask as the deepest. The test of codingstrength comes when the subjects are givena surprise recall task at the conclusion ofthe ratings. According to Craik and Lock-hart (1972), words that have been deeplycoded during the rating task should be re-called better than words with shallow coding.This manipulation permits us to determinethe relative deepness of self-reference as acoding device.

Craik and Tulving (1975) have done aseries of studies using this methodology.They have restricted their efforts to thestructural, phonemic, and semantic types oftasks. Their results indicate that recall (orrecognition) is best for semantic tasks andpoorest for structural ratings, with phonemicin the middle. These data are interpreted assupport for the position that the strength ofthe memory trace is "a positive function of'depth' of processing, where depth refers togreater degrees of semantic involvement"(Craik & Tulving, 1975, p. 268). Presuma-bly the rating tasks (structural, phonemic,etc.) force the subject to code the word to aspecific level, and the incidental recall is afunction of the depth of these tasks. Thesekinds of results have emerged quite consist-ently in the cognitive literature (e.g., D'Agos-tino, O'Neill, & Paivio, 1977; Klein & Saltz,1976; Schulman, 1974; Walsh & Jenkins,1973).

Of particular concern in the present studyis the comparison between incidental recallfor words rated under the semantic and self-reference tasks. Both of these tasks involvesemantic encoding, but there is an importantdifference between them. The self-referencetask forces the subject to use the self inthe rating task, whereas the semantic taskdoes not. The self-reference/semantic com-

parison permits assessment of the degree towhich the self aids in producing a strongertrace, in contrast to usual semantic encoding.If the self is an active agent in the encodingof personal data, we predicted that the self-reference rating would produce good inci-dental recall in this depth-of-processing para-digm. If incidental recall of the self-refer-ence words is superior to that for semanticwords, the hypothesis that the self servesan active and powerful role in processingpersonal data would be supported.

The present article offers two experimentsthat examine this proposition. The first studyinvolves a close replication of Craik andTulving's (1975) initial experiments, withthe self-reference task included. The secondexperiment replicates and extends the firststudy by using a different technique anddifferent semantic rating task.

Experiment 1

This experiment is intended to determinethe relative position of self-reference inCraik's (Craik & Lockart, 1972) depth hier-archy. The procedural details have beenchosen to closely approximate Craik and Tul-ving's (1975) initial experiments in an effortto maximize the degree of comparability ofthe present results.

Method

The study has two main parts. First, subjectsrated 40 adjectives on one of four tasks. This in-volved presenting a cue question, followed by 1of the 40 adjectives. Subjects answered yes or noto the cue question as it applied to the adjective.The cue questions, along with the manipulationsfor each task, are presented in Table 1. After com-pleting the ratings, subjects attempted to recall theadjectives in the second part of the study.

Materials. The main items for this study were40 adjectives that were deemed appropriate for aself-description task. They were chosen to representa broad spectrum of possible characteristics andwere selected from all of the trait descriptions foundin Jackson's (1967) Personality Research Form AManual. Thirty-eight of the adjectives, selected tobe familiar to the subject population, came fromthis source. Two other adjectives (shy and out-going) were added to make up the total of 40.

We used Roget's Thesaurus to construct a furtherset of 40 synonyms for the semantic tasks. The


Table 1Examples of the Rating Tasks

Task Cue question Manipulation

Structural

Phonemic

Semantic

Self-reference

Big letters?

Rhymes with x x x x ?

Means same as YYYY?

Describes you ?

The adjective was either presented in thesame size type as the question or twiceas large.

x x x x was a word that either rhymed ordid not rhyme with the adjective.

YYYY was either a synonym or unrelatedword to the presented adjective.

Subjects simply responded yes or no toindicate the self-reference quality of thepresented adjective.

final synonyms chosen represented consensus amongthe three authors.

The phonemic task dictated a second supple-mentary list of 40 words that rhymed with themain adjective set. The authors generated a set ofpossible rhyming words, and consensus among our-selves was the final criterion for selection. Most(90%) of these words were adjectives.

A third supplementary list of nonsynonym, non-rhyming words was also required, so that one halfof the cue questions could result in a no rating.Kirby and Gardner's (1972) set of adjectives wasconsulted to derive this list. Again, author con-sensus regarding the nonrhyming and nonsynonymquality of the adjectives dictated the final list.

A set of eight further adjectives and supplemen-tary words was generated to provide buffer itemsof four ratings each at the beginning and end ofthe list. These items, which were constant acrosslists, were not included in the data analysis. Thiswas intended to minimize the effects of primacy andrecency in the incidental recall task.

Four lists of adjectives were constructed, suchthat 10 adjectives in each list were rated undereach cue question, and over the four lists, eachadjective was rated under each cue question.

To guard against the possibility that wo-ratedwords are recalled differently than yes-rated words,each of the four lists was reversed to generate eightlists in total. For example, if in a given list, underthe structural task, a word appeared in small letters(generating a no response), the reversed list wouldhave the word presented in big letters (generatinga yes response). The one exception to this counter-balancing was the self-reference task. Here it wasimpossible to have experimental control over yesand no responses, since the person's view of selfwould dictate his or her response.

In all lists, order of the cue questions was ran-domly assigned in blocks of eight trials, such thateach combination of cue question and expectedresponse was represented once every eight trials.

Procedure. Subjects were tested individually. Ini-

tial instructions did not indicate that recall wasexpected. All stimuli were displayed on a televisionmonitor driven by a PDP8/1 computer, which alsorecorded the ratings and rating times. Including thefour buffer items at the end and beginning of thelist, there were 48 rating trials. Each of these con-sisted of (a) a 3-sec presentation of the cue ques-tion, (b) a 500-msec blank interval, (c) presentationof the target adjective, which was terminated bythe subject's response, indicated on a two-buttonresponse panel placed comfortably in front of thesubject, and (d) a 2-sec intertrial interval beforethe next cue question was presented. After the rat-ing task, the subject was given a piece of paperand was asked to recall, in any order, the adjectiveshe or she had rated. Three minutes were allowed forrecall.

Subjects. Volunteers from the introductory psy-chology subject pool served as subjects. There were32 subjects (16 female and 16 male) with an averageage of 20.2 years. Each was paid $1.50 for par-ticipating. Subjects were randomly assigned to theeight list conditions, yielding 4 subjects per order. .

Results and Discussion

For each subject, the number of adjectivesrecalled as a function of rating task (struc-tural, phonemic, etc.) and observed yes or norating was calculated. The means of thesefigures are presented in the top panel ofTable 2. A 4 (rating tasks) X 2 (yes/no rat-ing) two-way analysis of variance revealeda significant main effect of rating task, F(3,93) = 29.01, p < .001. Newman-Keuls testsindicated meaningful differences (p < .05 orbetter) in the recall for all points in thismain effect except for the structural-pho-nemic comparison. The main effect of ratingwas also significant, F(l, 31) = 4.22, p<


.05, indicating superior recall for words givena yes rating. The Rating Task X Rating in-teraction was also significant, F(3, 93) =3.47, p < .05. Post hoc tests of this inter-action revealed a meaningful yes/no differ-ence for the self-reference rating, £(31) =2.62, p < .05.

The overall pattern of these results is simi-lar to that typically found in the literature(e.g., Experiments 1, 2, and 3 in Craik &Tulving, 1975). The main effect of ratingtask is used to suggest that the depth towhich items are processed during the ratingtask determines the strength of the memorytrace. As recall is a function of trace strength,the present results support this position.

Of central interest is the finding that theself-reference task develops a stronger tracethan the semantic task, as shown by the sig-nificant recall differences between these twoconditions. This result clearly supports theidea that self-reference functions as a pow-erful coding device. In the case of self-refer-ence ratings, the subject uses his or her con-cept of self to respond to the adjective. Theself-ratings involve comparison of the in-coming adjective with the terms and sche-

mata that are part of the self (see Rogers,1974). This comparison culminates in thesubject's yes/no response, which leads to astrong and specific encoding of the rateditem. During the recall phase of the study,items with this detailed and specific encodingare easily retrieved, producing good recallperformance.

In the semantic rating task it is unneces-sary for an elaborate structure such as theself to be involved. Rather, the subject ac-cesses his associative memory (e.g., Estes,1976) for the target adjective and makes hissynonymity judgment from this. The result-ing trace is not as specific or detailed as thatinvolved with self-reference. Clearly, the ac-cess of associative memory produces a moredetailed trace than either the structural orphonemic tasks. However, when compared toself-reference, the trace derived from a syno-nymity judgment is relatively weak. Thisdifference in the specificity of the self-refer-ence and semantic tasks seems to be themajor reason for the inferior incidental re-call of the semantically rated words.

The time required to make the ratings istypically used to monitor the effectiveness

Table 2Recall, Rating Time, and Adjusted Recall as a Function of Rating Task and Rating forExperiment 1

Rating

yesno

Total

yesno

M

yesno

M

Structural

.28

.06

.34

1,2671,474

1,371

.05a

.01

.03

Rating

Phonemic

Mean recall

.34

.34

.68

task

Semantic

.65

.68

1.33

Mean rating time (msec)

2,1772,104

2,141

Mean adjusted recall

.08

.06

.07

2,2553,006

2,631

.14

.12

.13

Self-reference

1.781.06

2.84

3,1942,689

2,941

.30

.29

.30

Total

3.052.14

5.19

M

2,2232,318

2,271

M

.14

.13

.13

a This figure represents the mean (over 32 subjects) of the number of recalled ^5-rated structural itemsdivided by the number of yes ratings made on the structural task.


of the experimental manipulations in thisparadigm. Further, these reaction time (RT)data provide convergent evidence for self-reference as a useful encoding task. The RTsfrom the present study were sorted separatelyfor each subject into yes/no by rating-taskcategories. The means are presented in themiddle panel of Table 2. Only the main effectof rating task was significant in this analysis,F(3, 93) = 10.35, p < .001. A clear lineartrend in RTs is evident in these data, indi-cating maximal RT for the self-referencerating task.1 This analysis replicates Craikand Tulving (1975) and is clearly compat-ible with the recall data presented above,supporting the involvement of the self as acoding device.

The finding that yes-rated words are re-called better than «0-rated words occurs inother studies and has a number of interestingimplications. Craik explains these data byarguing that in the case of yes-ra,ted words,the "encoding questions or context forms anintegrated unit with the target word" (Craik& Tulving, 1975, p. 291). Presumably thisintegrated unit forms a stronger trace thanless integrated ones (no-rated words),thereby augmenting recall. The interactionobserved in the present data indicates thatthis yes/no difference occurred only for theself-reference case, which suggests that itemsviewed as self-descriptive (yes-rated words)form a "more integrated unit" than donon-self-descriptive terms. These resultsstrengthen even more our view of self, as itappears that terms that match the subject'sself-view become more integrated than thosethat do not match. This finding is consonantwith both Markus's (1977) and Cantor andMischePs (1977) finding that personal dataare processed using schemalike structures.

There are several aspects of these datathat require examination before the previousconclusions are fully warranted. The yes/nodifference in recall for self-reference wordscould be due to a differential number of yesresponses as a function of rating task. Sinceexperimental control over the number of yesresponses was not possible for the self-refer-ence task, this is a distinct possibility. For

each subject, the number of yes responsesmade under each rating task (maximum =10) was calculated. The means were 5.00,4.34, 4.06, and 6.13 for the structural, pho-nemic, semantic and self-reference tasks, re-spectively. A simple analysis of variance onthese figures revealed a significant effect,F(3, 93) = 16.99, p < .001, indicating thatnumber of yes responses is related to ratingtask. The deviations from 50% yes responsesfor the phonemic and semantic tasks are dueto the difficulty of constructing exact rhymesand synonyms for the adjectives.

More important than the significant varia-tion in number of yes responses is the pos-sible effect this might have on the recall data.Since it is already known that yes-ratedwords are better recalled (e.g., Craik &Tulving, 1975), it is possible that self-refer-ence recall was superior because subjectsmade more yes responses in the self-referencetask. To assess this, the recall data weretransformed to a proportion score that ad-justs for differential numbers of yes re-sponses. Specifically, a particular subject's re-call of ;yej-rated words under a specific ratingtask was divided by the number of yes rat-ings the subject made while doing the task.This transformed score represents the pro-portion of recalled words the subject ratedas yes. Similarly, the wo-rated word recallunder a given rating task can be divided bythe number of no responses made on this rat-

1 It is possible that items with large RTs arebetter recalled, calling into question this interpreta-tion. If study time is the important factor, itseffects should be observable within each task aswell as across tasks. Thus, within a given ratingtask, the items with the longer study times shouldbe recalled better. To explore this, the 10 RTsunder each of the four rating tasks were subdividedseparately for each subject into a fast and a slowsubset (5 RTs in each). The recall for these sub-sets was analyzed in a 4 (rating tasks) X 2 (fastand slow study times) two-way analysis of variance.The study-time hypothesis predicts significant effectsfor the terms involving study time. The analysis in-dicated only the expected main effect of rating task,F(3, 93) =30.85, p < .001. This analysis weakensthe study-time interpretation and reinforces theinterpretation that the recall data are due to thequalitative nature of the various encoding tasks.


ing task to provide a score representing theproportion of recalled words rated as no.Note that this is a subject-specific correctionthat reflects recall corrected for differentialnumbers of yes and no ratings. The means ofthe adjusted recall scores are presented inthe bottom panel of Table 2. An analysis ofvariance of these data revealed only a maineffect of rating task, F(3, 93) =31.63, p <.001. The important recall difference betweensemantic and self-reference survived this ad-justment (p<.0l), but the yes/no differ-ence for the self-reference task did not. Thisanalysis reaffirms self-reference as a codingtool but questions the possibility that yes-rated items form a more integrated unit.

In summary, the data from Experiment 1provide evidence that self-reference is a pow-erful encoding device. The superior incidentalrecall of adjectives rated under the self-refer-ence task, in combination with the RT data,suggests that self-reference provides a richand powerful encoding. The involvement ofself in the rating task provides a good en-coding unit, which functions effectively as amemory cue.

Experiment 2

It is possible that the superiority of self-reference encoding documented in Experi-ment 1 is specific to synonymity ratings.Maybe other kinds of semantic tasks wouldproduce equally powerful results. Semantic

tasks can be thought of as a family of judg-ments, all of which involve the extraction(and possibly some elaboration) of the mean-ing of the target item. Such tasks as syno-nymity ratings, judgments of semantic speci-ficity, and deciding whether a word fits intoa sentence frame can be considered membersof this semantic family. Experiment 2 em-ployed meaningfulness ratings as the semantictask. Since recall is a function of meaning-fulness (see Noble, 1952; Paivio, Yuille, &Rogers, 1969), this encoding task should bevery beneficial for recall, particularly forwords given a yes rating. If self-referenceemerges as superior to meaningfulness rat-ings, evidence confirming the strength and re-liability of self-reference as an encoding de-vice will be provided.

A second purpose of this experiment is toexplore the robustness of the self-referencefindings. Experiment 1 was performed usingfairly tight experimental controls. The pres-ent experiment deviates from this by usinga group testing procedure. Craik and Tulving(1975) and Klein and Saltz (1976) haveused similar procedures and replicated thefindings from more rigorous paradigms, sug-gesting that the self-reference finding shouldstand up in this group procedure.

Method

The four rating tasks used for this experimentare outlined in Table 3. Subjects were given a

Table 3Rating Tasks and Mean Adjusted Recall for Experiment 2

Rating task

Structural

Phonemic

Semantic

Self-reference

M

Cue question

Long?

Rhythmic?

Meaningful?

Describes you ?

Definition

Rate whether you feel the wordis long or short.

Rate whether you feel the wordhas a rhythmic or lyrical sound.

Rate whether you feel the wordis meaningful to you.

Indicate whether the worddescribes you.

Mean

Yesrating

.21

.20

.23

.33

.24

adjusted

Norating

.18

.18

.15

.31

.21

recall

M

.20

.20

.19

.32

.23


rating- sheet which indicated which of the fourtasks they were to perform on a given word; thiswas indicated by the cue questions from Table 3.After the subjects had read the task cue to them-selves, an adjective was read aloud by the experi-menter, and the subjects made their yes or no re-sponses on the sheet. After the ratings, subjectsturned over their rating sheets and attempted torecall the adjectives.

Materials. The 40 adjectives used in Experiment1 made up the target items in this study. Fourdifferent task orders were generated, such that eachadjective was rated under each task consideredacross the four orders, and within each order onefourth of the words were rated under each task.Within each list the order of tasks was randomizedin blocks of four, such that each task was repre-sented once in every four trials.

Procedure. Subjects were run in one group. Afterinstructions, the experimenter read the item number,said the word task (which cued the subjects to readthe cue question), and then read out the adjective.After 40 such trials, subjects were given 3 minutesto recall, in any order, as many of the adjectives asthey could. Subjects were not expecting this free-recall task.

Subjects. Twenty-seven students in a fourth-year summer class served as subjects. The mean agewas 27.7 years.

Results and Discussion

For each subject, recall as a function ofrating task and rating was calculated. Thesefigures were converted to adjusted recallscores following the procedures for Experi-ment 1. The means of the adjusted recallscores are presented in the far right columnsof Table 3. Analysis of variance of the ad-justed recall scores produced a lone signifi-cant main effect of rating task, F(3, 78) =4.20, p < .01, and a meaningful semantic/self-reference recall difference (p < .05).

Experiment 2 demonstrates self-referencerecall superiority when a meaningful ratingtask is used. This kind of task has been pre-viously implicated in recall, resulting in aseemingly powerful semantic encoding task.However, the present results indicate thatself-reference still is the more useful encod-ing task in this paradigm.2

Taken in total, the results of these ex-periments indicate that self-reference inducessuperior incidental recall compared to a di-versity of strictly semantic rating tasks. Theimportant thing is that self-reference appears

to produce recall that is superior to any othertask ever used in the incidental recall para-digm. This by itself attests to the power ofself-reference.

General Discussion

As a test of encoding strength, the depth-of-processing paradigm forces the subject toprocess stimuli to a specific depth by havingthe subject rate the words on different tasks.During the rating task a memory trace of therated word is created. Tasks that are deep orsemantically rich produce strong traces,which in turn serve as useful cues in theincidental recall of the rated words. The rela-tive power of an encoding device is corre-lated with incidental recall in this paradigm.The present data indicate unequivocally thatwords rated under the self-reference taskshow superior recall. This indicates that self-reference represents a powerful and rich en-coding device. Clearly, self-reference pro-duces a rich encoding unit that can functioneffectively during information processing (seealso Markus, 1977; Rogers, in press; Rogerset al., Note 1).

The major difference between the semanticand self-reference encoding tasks lies in theinvolvement of self in the latter rating. Theself is a superordinate schema that containsan abstracted record of a person's past ex-perience with personal data. The richness ofself-reference encoding shown in the presentarticle is due to the access of this schema.The semantic rating task does not force in-volvement of a powerful schema, and hencefails to induce as powerful and rich an en-

2 The same pattern of results has been replicatedtwice for this group procedure using differentsemantic rating tasks. Typically this group proce-dure fails to replicate Craik and Tulving's (1975)findings for the structural, phonemic, and semantictasks (see Table 2). This is probably due to therating tasks used. For example, in the structuraltask some subjects may have rated whether theword was a "big" word (rather than long), whichwould be a semantic task. Regardless of this prob-lem, the important semantic/self-reference differenceclearly emerged in all studies using the group pro-cedure.


coding unit. The mere act of making a self-referent decision produces such powerful in-ternal reactions that the attending memorytrace is stronger than any observed in thepresent experimental situation.

In order for self-reference to be such auseful encoding process, the self must be auniform, well-structured concept. During therecall phase of the study, subjects probablyuse the self as a retrieval cue (e.g., Mosco-vitch & Craik, 1976). In order for this tobe functional, the self must be a consistentand uniform schema. This property of theself is also shown by Rogers et al. (Note 1),who were able to predict memory perform-ance with a measure of self taken 2\ monthsearlier. The present data support the con-tention that the self is a well-structured andpowerful schema. Presumably the self-refer-ent decision activates the superordinateschema of self as well as the salient sub-schemata. The strength of the trace devel-oped from this activation suggests that aconsistent and well-structured schema under-lies these decisions. This consistency pro-duces a rich and effective encoding unit,which accounts for the present data.

The present data permit some furtherstatements about the schema properties ofthe self. Considering the four experimentsreported here (including the two supplemen-tary studies in Footnote 2), it has been con-sistently found that yes-rated items arebetter recalled than »0-rated items in theself-reference task.3 These data support theschema view, since ;yes-rated items would fitmore easily into the schema, and therebyform a more integrated encoding unit (Schul-man, 1974). Such a pattern of results is com-patible with Markus (1977) and Cantor andMischel (1977) and further reinforces a viewof the self as a schema.

The data indicating that the self is aschema prompt consideration of how the vari-ous traits (i.e., subschemata) and specificelements (i.e., individual behaviors) are or-ganized within this structure. The terms ofthe self are organized in an hierarchicalfashion, with the most central traits repre-sented initially. At first blush, it seems rea-

sonable to think that this hierarchy relatesto extremity. That is, the initial trait in theself would be a person's most extreme trait,followed by the second most extreme, and soon, until meaningful traits for the personwere exhausted. This simple extremity or-ganization may hold for some persons, butanother aspect of the traits must also be con-sidered. Specifically, the salience of a traitfor a person also adds to the organization ofself. For example, a person who views himor herself in the midrange on "friendly,"may perceive this characteristic as excep-tionally important, and thereby have it in-cluded in the self. Markus (1977) includedthis consideration when she used importanceratings in her definition of schematics. Thisaspect of traits or constructs has also beendiscussed under the labels of centrality(Snygg & Combs, 1949) and salience (Jack-son, Note 2). The important thing to note isthat the traits involved in a person's self arenot necessarily the most extreme ones.Rather, they represent a mix of salience andextremity.

The inclusion of specific behaviors in theself derives from the work in cognition. Pos-ner and Keele (1968) postulate that a per-son stores both the prototype and some in-dication of how a given stimulus deviatesfrom this norm. This means that a schema,by virtue of its abstract property, must alsocontain specific data indicating aspects ofthe previous input that do not conform tothe abstracted structure.

In sum, the self contains a set of orderedfeatures. The ordering appears to be fromgeneral to specific, with the general terms(e.g., traits) ordered by a combination ofsalience and extremity. The general termscan serve as schemata when studied inde-pendently of a person's idiographic view ofself (e.g. Cantor & Mischel, 1977; Markus,1977).

The present data stress how the self can

^Statistical significance for this difference tendsto disappear in the adjusted recall analyses. How-ever, the consistent emergence of the effect acrossthis series of experiments (even in the adjusted re-call data) suggests a substantial effect.


become involved in the encoding of personaldata. Focusing on the organizational proper-ties of the self is by no means new (e.g.,Bertocci, 1945; Gergen, 1971). In fact, asearly as James (1890) the self has beenpostulated to be an active agent in the over-all human cognitive apparatus. The presentresearch adds to this tradition by providinga strong empirical test of a proposition deriv-ing from such a view of self. Our emphasis isupon the information-processing functions ofthe self, specifically relating to self-referenceas an encoding device. This represents a neo-mentalistic approach (Paivio, 1975) to theself. While behavioral evidence (i.e., memoryperformance) is the key element in this ap-proach, the focal concern is upon the inferredconstruct—in this case, the self.

It should be noted that there are certainclasses of information likely to receive self-reference encoding. Only after certain con-textual information indicates that the selfmay be a functional aid to processing willthe schema be activated. In the present con-text, we forced this process with the encod-ing task. In real life situations, it seemslikely that the self would be functional in anumber of situations involving feedback ofpersonal data, such as conversations, expres-sive behaviors, and attempts to assess per-sonal impact on others. The kinds of situa-tions that activate this schema or possibleindividual differences in the frequency andintensity of the involvement of self in dataprocessing may prove to be very useful ex-tensions of the present formulation.

Probably the main advantage of the pro-cess-oriented view of information processingunderlying our approach to self is the oppor-tunity to move toward less descriptive modelsof social behavior. If we understand theprocesses and mechanisms underlying theprocessing of personal information, we willhave a real opportunity to construct sub-stantive models based on hard experimentalfindings (see Sechrest, 1976). For example,the finding that the self induces certainbiases during the processing of personal data(Cantor & Mischel, 1977; Markus, 1977)can be related to the cross-situational con-

sistency issue. As noted by Bern and Allen(1974), our intuitions tell us that there arecross-situational consistencies in behavior,even though the research data do not tend toconfirm this. Our process approach wouldinterpret this as follows: (a) Personal dataare processed using the self (e.g., Experiment1). (b) The self induces people to view novelbut self-relevant data as having been previ-ously experienced (Rogers et al., Note 1).(c) Therefore new personal data will appearto conform to expectation (i.e., fit into thescheme), which could produce a perceptionof consistency. This interpretation focusesupon the organizational and biasing aspectsof the human information processor, whichprovides an alternate construction of theseimportant data. Although the utility of thisinterpretation rests on further empirical tests,the amenability of such a model to direct ex-perimental scrutiny argues in its favor.

In summary, the present article offers datato suggest that self-reference is a very potentencoding device. The pattern of results indi-cates that the use of self during the encodingof adjectives produces as elaborate and inte-grated a memory trace as has been foundusing the present experimental paradigm.These data suggest the self is an extremelyimportant aspect of the processing of per-sonal information. In the realm of humaninformation processing it is difficult to con-ceive of an encoding device that carries morepotential for the rich embellishment of stim-ulus input than does self-reference.

Reference Notes

1. Rogers, T. B., Rogers, P. J., & Kuiper, N. A.Recognition memory for personal adjectives:Some evidence for self-reference as an aspect ofmemory. Unpublished manuscript, University ofCalgary, 1977.

2. Jackson, D. N. A threshold model for stylisticresponding. Paper presented at the meeting ofthe American Psychological Association, SanFrancisco, September 1968.

References

Allport, G. W. Personality: A psychological inter-pretation. New York: Holt, 1937.


Bartlett, F. C. Remembering. Cambridge, England:Cambridge University Press, 1932.

Bern, D. J., & Allen, A. On predicting some of thepeople some of the time: The search for cross-situational consistencies in behavior. PsychologicalReview, 1974, 81, SO6-S2O.

Bertocci, P. A. The psychological self, the ego, andpersonality. Psychological Review, 1945, 52, 91-99.

Cantor, N., & Mischel, W. Traits as prototypes:Effects on recognition memory. Journal of Per-sonality and Social Psychology, 1977', 35, 38-48.

Cart wright, D. Self-consistency as a factor affectingimmediate recall. Journal of Abnormal and SocialPsychology, 1956, 52, 212-219.

Craik, F. I. M., & Lockart, R. S. Levels of process-ing: A framework for memory research. Journalof Verbal Learning and Verbal Behavior, 1972,11, 671-684.

Craik, F. I. M., & Tulving, E. Depth of processingand the retention of words in episodic memory.Journal of Experimental Psychology: General,1975, 104, 268-294.

D'Agostino, P. R., O'Neill, B. J., & Paivio, A.Memory for pictures and words as a function oflevels of processing: Depth or dual coding?Memory & Cognition, 1977, 5, 252-256.

Estes, W. K. Structural aspects of associative modelsfor memory. In C. N. Cofer (Ed.), The structureof human memory. San Francisco: Freeman, 1976.

Gergen, R. J. The concept of self. New York:Holt, Rinehart & Winston, 1971.

Hastorf, A. H., Schneider, D. J., & Polefka, J.Person perception. Don Mills, Ontario: Addison-Wesley, 1970.

Jackson, D. N. A manual for the Personality Re-search Form. Goshen, N.Y.: Research Psycholo-gists Press, 1967.

James, W. Principles of psychology. New York:Holt, 1890.

Jones, E. E., et al. (Eds.). Attribution: Perceivingthe causes of success and failure. New York:General Learning Press, 1971.

Jones, R. A., Sensenig, J., & Haley, J. V. Self-descriptions: Configurations of content and ordereffects. Journal of Personality and Social Psy-chology, 1974, 30, 36-45.

Kelly, G. A. Psychology of personal constructs.New York: Norton, 1955.

Kirby, D. M., & Gardner, R. C. Ethnic stereotypes:

Norms on 208 words typically used in their as-sessment. Canadian Journal of Psychology, 1972,26, 140-154.

Klein, K., & Saltz, E. Specifying the mechanismsin a levels-of-processing approach to memory.Journal of Experimental Psychology: HumanLearning and Memory, 1976, 2, 671-679.

Markus, H. Self-schemata and processing informa-tion about the self. Journal of Personality andSocial Psychology, 1977, 35, 63-78.

Moscovitch, M., & Craik, F. I. M. Depth ofprocessing, retrieval cues, and uniqueness of en-coding as a factor in recall. Journal of VerbalLearning and Verbal Behavior, 1976, 15, 447-458.

Noble, C. E. An analysis of meaning. PsychologicalReview, 1952, 59, 421-430.

Paivio, A. Neomentalism. Canadian Journal of Psy-chology, 1975, 29, 263-291.

Paivio, A., Yuille, J. C, & Rogers, T. B. Nounimagery and meaningfulness in free and serialrecall. Journal of Experimental Psychology, 1969,79, 509-514.

Posner, M. I., & Keele, S. W. On the genesis ofabstract ideas. Journal of Experimental Psychol-ogy, 1968, 77, 353-363.

Posner, M. I , & Keele, S. W. Retention of abstractideas. Journal of Experimental Psychology, 1970,83, 304-308.

Rogers, T. B. An analysis of two central stagesunderlying responding to personality items: Theself-referent decision and response selection. Jour-nal of Research in Personality, 1974, 8, 128-138.

Rogers, T. B. Self-reference in memory: Recogni-tion of personality items. Journal of Research inPersonality, in press.

Schulman, A. I. Memory for words recently classi-fied. Memory & Cognition, 1974, 2, 47-52.

Sechrest, L. Personality. In M. R. Rosenzweig &L. W. Porter (Eds.), Annual review of psychol-ogy. Palo Alto: Annual Reviews, 1976.

Snygg, D., & Combs, A. W. Individual behavior.New York: Harper, 1949.

Walsh, D. A., & Jenkins, J. J. Effects of orientingtasks on free recall in incidental learning: "Diffi-culty," "effort," and "process" explanations. Jour-nal of Verbal Learning and Verbal Behavior, 1973,12, 481-488.

Received January 24, 1977 •

self-reference and th encodine g of personal information · in the processin ogf people-related...

Documents