Reference: Bower, G. H., & Gilligan, S. G. (1979). Remembering information related to one's self. Journal of Research in Personality 13, 420-432.

Remembering Information Related to One's Self

Gordon H. Bower and Stephen G. Gilligan

Stanford University

How does memory for an incident vary depending on whether, and how, the person relates the information to himself? Trait adjectives are better remembered if they were judged in reference to oneself rather than judged for meaning or sound. Our first experiment found a similar mnemonic advantage of referring a described episode or object to some event from one's life. Pleasant events were remembered better than unpleasant ones. A second experiment found incidental memory for trait adjectives was equally enhanced by judging each directly in reference to one's self-concept or indirectly by retrieving an episode either from one's life or from one's mother's life. Contrariwise, memory was poorer when traits were judged in reference to a less familiar person. Thus, good memory depends on relating the inputs to a well-differentiated memory structure.

People often deal with new information by relating it to themselves. This habit surrounds us in daily life. If you describe an interesting incident to someone, his probable reply is to relate a similar incident about himself. If you describe an interesting psychological phenomenon to students, their frequent reaction is to check it out against their experiences. A familiar illustration of self-reference is the "medical student's syndrome" of students in abnormal psychology who believe that most psychopathic descriptions fit themselves. Since memory retrieval follows resemblance of the new information to stored information, the self reference process has a built-in "confirmation bias." When the professor describes a patient with an obsessional fantasy, the student remembers an occasion when he too experienced a compelling fantasy, and he concludes he too is "abnormal," completely ignoring in his decision the countless days of his life when he was not obsessed with that fantasy.

What is the function of this self-referencing process in memory? Besides promoting agreeable conversation and affiliation, self-referencing of new information doubtless serves other functions. From the perspective of learning theory, the self-referencing process is another instance of the general rule that new information is assimilated and learned by relating it to prior information in memory. The relation of new to old information may either be (I) subsumption of the new within an old generic schema, or (2) resemblance or contrast of the new in analogy to a specific memory of a specific incident.

How does referring an input to one's self or one's experiences affect its encoding and storage? A recent report by Rogers, Kuipers, and Kirker (1977) is particularly relevant. They investigated how subjects' incidental memory for trait adjectives (like honest. hardworking) varied with the orienting task subjects carried out. The judgment tasks required accessing different information about the visually presented words. Some words were judged for their surface appearance ("Is this word printed in all capital letters?"), some for their phonetic features ("Does it rhyme with fly?"), some for their semantic meaning ("Does it mean the same as assertive?"), and some for their accuracy as a self-description ("Does this adjective describe you?"). Subjects responded yes or no as each stimulus word was shown. They were later asked unexpectedly to recall the stimulus words. As found previously (e.g., Craik & Tulving, 1975), Rogers et al. (1977) found poorest recall for words judged on surface or phonetic features and better recall for words judged for semantic features. The novel finding was that the best remembered trait adjectives were those judged in relation to the self. Recall of those adjectives was two- to eightfold greater than recall of adjectives processed in the other manners.

Rogers et al. (1977) describe the "self-concept" as a "superordinate” schema that contains an abstracted record of a person's past experience with personal data." Elsewhere they write that the self-schema contains not only general terms like trait names the person believes characterize himself across situations, but also situation-specific variations in his behavior (e.g., "Although I'm normally extraverted, I don't behave that way at funeral services"). Rogers et al. believe that when the subject refers a trait adjective to himself, the process results in a "richer, deeper encoding" of that word. Unfortunately, such suggestive words do not constitute an explanation of the result (see Baddeley, 1978, for a critique of the "depth of processing" metaphor). It now appears that the actual advantage is that during the recall phase the subject can "use the self as a retrieval cue" (Rogers et al., p. 686). That is, the subject can implicitly generate a list of his own traits and check to see whether each one (or its opposite) was one of those mentioned in the context of the experimental list. If so, then he will overtly recall that adjective. An advantage of this retrieval procedure is that it generates a number of different cues to each of which one or more list items might be associated. No comparable retrieval scheme is available to aid recall of the words processed for their surface or phonetic features. For example, the fact that a word was in "all capital letters" gives no useful information about the word to be recalled.

The retrieval cuing idea also helps explain why items that fit the question (are answered Yes) are better recalled than items that do not (yield No answers); the Yes items will be in the set of retrieval cues generated from the self-concept during later recall, whereas the No items must be derived by another step (e.g., "I'm independent, which is the opposite of dependent, and dependent was presented in the list").

Rogers et al. (1977) describe the self in terms of relatively general traits and consider the advantage of self-referencing to result from a match of an input word to one of these general traits. But it seemed likely to us that a similar mnemonic advantage could be produced by relating an input to any prior self-information, even memories of specific episodes and incidents. We hypothesized that subjects' memory for a described incident would be enhanced if they were asked to try to locate a similar incident in their autobiographical memory. Accordingly, we repeated the Rogers et al. type of experiment except some subjects were asked to decide whether they had ever experienced an episode of a certain kind. We expect this to cause greater incidental memory for that input than would making a semantic or surface decision about it.

EXPERIMENT 1

The episodes used as learning materials in Experiment 1 were adjective noun phrases like a boring lecture, a faithful pet, a painful spanking, and a sunburned back. Three groups of eight subjects were assigned to three orienting decisions: a Self reference group decided whether each phrase referred to a specific personal experience or object they could remember; a Semantic group decided whether the phrase involved some social, interpersonal interaction; a Superficial group decided whether the words of the phrase jointly contained more than two occurrences of the letter e. After processing 48 such phrases, the subjects were unexpectedly given memory tests over the phrases. We expected the Self-reference group to recall best, and the Superficial group worst.

To test subsidiary hypotheses, features of the adjective-noun phrases were varied systematically. The noun in the phrase was either relatively concrete or abstract (based on the noun's rating in the concreteness norms of Paivio, Yuille, & Madigan, 1968), the incident or object referred to was either relatively pleasant or unpleasant, and was either relatively frequent or infrequent in most people's life-times. The pleasantness and frequency of the incidents were judged by six colleagues. Examples are: a cheerful mood (abstract, pleasant, high frequency), a broken bone (concrete, unpleasant, low frequency), a wedding ceremony (abstract, pleasant, low frequency), a comfortable chair (concrete, pleasant, high frequency). We believed that phrases with concrete nouns would be recalled better than those with abstract ones, that pleasant incidents would be recalled more than unpleasant ones, and that culturally common experiences would be located more easily in memory and hence recalled better than uncommon experiences.

Method

Subjects

Twenty-four Stanford University students participated in groups of four in order to fulfill their introductory psychology course requirement.

Design

The design was between subjects, with three different groups of eight subjects performing different orienting tasks. As noted, the task judgments were: Self-reference ("Describes you?"), Semantic ("A social interaction?"), and Superficial ("Two or more e's?"). The latter two criteria were chosen to yield approximately half Yes and No judgments to the 48 adjective-noun phrases.

The experiment also had four within-subject factors with repeated measures. Three were materials variables, i.e., the phrases differed in concreteness/abstractness (of noun). high/ low frequency (of occurrence of an experience in an individual's life), and pleasantness/ unpleasantness (of experience). The fourth factor was the Yes-No rating given to each phrase by the subject.

Materials

Stimuli. Fifty-four phrases of the form “(article) (modifier) (noun)" were projected as slides on a screen for 7 sec. each. The first and last three phrases served as primary and recency buffers and were never analyzed; thus, there were actually 48 target phrases. Order of presentation was randomized but constant for all subjects.

The 48 phrases were composed by crossing the variables of abstractness/concreteness, high/low frequency of experience, and pleasant/unpleasant experience, thus producing eight categories of six items each. That is, each phrase was rated as either concrete (e.g., an absorbing book) or abstract (e.g., an amazing discovery), was representative (as judged by six graduate students' ratings) of an experience that had occurred many limes (e.g., a boring lecture) or only a few times (e.g., a high school graduation) in an individual's experience, and was representative (again judged by graduate students' ratings) of either a pleasant or unpleasant experience.

Procedure

Subjects were run in groups of four, and were told the experimenter was investigating judgments of certain types of phrases. Word phrases were flashed on the screen one every 7 sec with the experimenter calling out the number of each slide as it was shown. Each subject judged each phrase by asking and answering a question (as determined by his condition), indicating his answer by circling (for that numbered phrase) the appropriate "yes" or "no" column on his rating sheet.

After showing the slides, the rating sheets were collected, and subjects were asked to solve some arithmetic addition problems given on a paper handed out to each subject. These problems served as an interpolated distractor task; after 10 min, these sheets were collected.

The subject then received three unexpected memory tests. The first test was for free recall; the subject was given 10 min to recall and write down on a blank sheet as many input phrases as he could. in any order. Following this, subjects completed the cued recall test, for which they had 8 min. The cued recall test consisted of the adjective of 24 of the original 48 phrases, with a blank space for the noun beside each. The subject was tow to use these adjectives as cues for retrieval of the noun of each phrase. The 24 recall cues consisted of 3 randomly selected phrases from each of the 8-item categories. After the cued recall test was collected, the recognition memory test was given for 8 min. The recognition test used the 24 phrases (of the original 48) not used in the cued recall test. The test included 3 classes of items: 8 Trues, which were identical to those presented; 12 Relateds, which were derived from the original phrases by replacing the adjective, the noun, or both by synonyms (e.g., "a stimulating talk" was replaced by "an interesting talk"); and 8 Falses, which were phrases unrelated to any shown during the learning phrase. Subjects were instructed on using a 7-point rating scale, with a 1 indicating that they were certain that the recognition phrase had appeared in the slide display, and a 7 indicating that they were certain it had not been shown before. After this test had been completed, subjects were debriefed and dismissed.

In scoring the free recall, credit was given for closely synonymous substitutions, and half credit was given for recalling one word from a phrase correctly. These scoring conventions do not affect the pattern of significances in the results.

Results

The recall results are shown in Table 1, with free recall in the top half and cued recall in the bottom half. Several conclusions are apparent. First, both indices show that subjects in the Self-reference condition remembered more than those in the Semantic condition, who in turn remembered more than those in the Superficial condition. The treatment effect was significant for unconditional free recall (F(2,21) = 57.2, p < .01) and for unconditional cued recall (F(2,21) = 9.5, p < .01). Self reference encoding enhances memory very substantially. Second, cued recall significantly exceeded free recall in every condition. Third, Yes items are better recalled than No items in the Self-reference condition but not in the others. An analysis of variance on the arc-sine transforms of the conditional probabilities in the top half of Table 1 revealed a significant effect due to treatments (p < .01), another due to Yes vs No (p < .01), and another due to the interaction of treatments with the Yes-No variable, (p < .05).

The three groups differed in the percentages of Yes decisions. These were 77, 55, and 42% for the Self-reference, Semantic, and Superficial groups respectively (p < .01). However, as Table 1 shows, the Self reference group outperformed the other two in free recall both for Yes items (t(22) = 4.57, p < .01) and for No items (t(22) = 2.76, p < .01). A similar difference occurred for cued recall for Yes items (t(22) = 4.92, p < .01), but not for No items, an unexplainable curiosity. The fact that the Self-reference group free recalled more than the other two groups on both Yes and No items implies that group differences in Yes responding cannot account for the differing levels of recall among the groups.

Next, we examined whether recall varied with the type of phrase. Pleasantness of a phrase increased its free recall significantly for the Self-reference and Semantic groups separately. For the Self-reference group, free recall of pleasant vs unpleasant phrases was 45 vs 33%, t(14) = 2.58, p < .02; for the Semantic group, it was 27 vs 18%, t(14) = 3.98, p < .01. However, there were no differences in cued recall for either group. The Superficial group did not differ in free or cued recall of pleasant and unpleasant phrases. So, to conclude here, those subjects who processed the meaning of the phrases recalled more pleasant than unpleasant ones.

In contrast to their pleasantness, frequency rating of the phrases had no consistent effect on free or cued recall. Free recall percentages for high vs low frequency phrases, were 39 vs 38% for the Self-reference group, 27 vs 18% for the Semantic group, and 8 vs 7% for the Superficial group. A recall difference failed to appear for the Self-reference group despite high-frequency phrases being accepted more often (84% Yes responses) than low-frequency phrases (70% Yes responses). Our frequency manipulation confounds the percentages of subjects having an experience with its distinctiveness for individuals. Therefore, our frequency variable probably has no simple relation to memory.

Similarly, the concreteness of the noun of the phrases did not influence free or cued recall. Free recall percentages for abstract vs concrete phrases were 41 vs 41%, 22 vs 24%, and 8 vs 7% for the Self-reference, Semantic, and Superficial groups, respectively. A problem with our manipulation was that the concreteness of the encoding of a phrase depended on a complex relation between the adjective and noun, and the vividness of the experience or object selected as its referent. For example, the phrase a mystical feeling is composed of abstract words, but for some students it revives a particularly vivid and distinctive incident of their spiritual life.

Following the recall tests, subjects took the recognition memory test, rating phrases on a 7-point scale, with 1 denoting certainty that that exact phrase was presented and 7 denoting certainty that it was not. An index of memory is the accuracy of discriminating old True statements from new False lures, estimated by the difference in their ratings. The average ratings for True vs False items was 1.1 vs 6.2 for the Self-reference group, 1.6 vs 5.8 for the Semantic group, and 3.0 vs 4.8 for the Superficial group. The Self-reference group showed the highest accuracy in recognition memory, and the Superficial group the lowest. The recognition data yielded conclusions substantially the same as those from recall, so will not be presented further.

To summarize, this experiment has shown superior memory from relating event descriptions to episodes from one's history. Incidental memory established by reference to autobiographic events significantly exceeded that due to a semantic analysis or a surface analysis of the phrase itself. We suppose that in Self-reference an event description (a dying dog) may arouse from memory an episode matching that description (e.g., "When I was ten my dog, Spunky, died in my arms after she'd been hit by a truck in the road near our house."). As the ideas, concepts, and images of that episode are aroused in active memory, they become associated to the "list context" as ideas thought of during the experiment. The same process operates to some extent with No phrases since these would call up perhaps a succession of partially matching episodes from memory. Later, when asked to recall, the subject retrieves from memory recently activated episodes of his life, and outputs phrases describing those that are associated to the list context. (This latter editing is logically required to prevent intrusions of all experiences that were thought of but which were not elicited by a presented phrase.) This seems a plausible reconstruction of what is going on in the Self-reference condition.

EXPERIMENT 2

Rogers et al. (1977) found superior recall when subjects decided whether particular traits were part of their abstract self-schema. Experiment 1 found superior recall when subjects decided whether particular episode descriptions resembled some episode they could remember from their life. Comparison of the two tasks raises several questions. First, which orienting task will produce better memory? In the following experiment, we presented trait adjectives and had some subjects refer them for a match to their abstract self-schema, whereas others searched for a specific autobiographic episode which illustrated this trait in their behavior. If retrieval of concrete images with the discrete episode improves the quality or "depth" of encoding, then the episodic-memory judgments might enhance recall over that for judgments of abstract traits.

We also varied the likableness of the trait names, half being likable (friendly, loyal) and half unlikable (grouchy, anxious). Assuming subjects judge themselves favorably, those judging the abstract traits should accept more likable than unlikable traits as self-descriptive. On the other hand, since most people can think of at least one incident when they did some unlikable deed, subjects instructed to search for an exemplifying episode should not differ so much in Yes saying to likable vs unlikable adjectives.

We claimed earlier that the Self-reference advantage exemplified the general rule that memory is improved by relating an input to any well-articulated information structure. As such, the effect should not depend on relating an input to one's self-concept or autobiographic memory. Any well-differentiated information structure should do as well. To test this, we had some subjects decide for each trait whether they knew of an episode involving their mother as the agent who exemplified that trait. It had to be a specific episode involving the subject's mother, one which he may have witnessed or which he knew had happened to her. Mother served as a referent (other than self) about whom the subject had many episodic memories for relating to the input materials. The question is whether adjectives related to specific incidents from one's mother's life will be recalled just as well as those related to incidents of one's own life.

Finally, we asked whether recall would be poorer if the subject had to relate the adjectives to a familiar person about whom he knew relatively little. For this purpose, and following an experiment by Lord (Note 1), we used Walter Cronkite, the television newscaster. College students have a relatively undifferentiated impression of Cronkite's personality—typical traits cited are warm, authoritative, and trustworthy. Therefore, most trait questions about Cronkite would be answered according to their resemblance to or consistency with the few traits the subject had already ascribed to Cronkite. In any event, the trait questions should not arouse specific episode memories of Cronkite, nor would they be placed within a highly differentiated conceptual structure. Accordingly, subjects judging whether the presented adjectives described Cronkite should recall them more poorly than subjects who refer the adjectives to themselves.

Method

Subjects

Forty Stanford University students participated in groups of two to four in order to fulfill their Introductory Psychology course requirements.

Design

The design was a 4 x 2 factorial. The first factor was between subjects (Orienting task) and the second was within subjects (pleasant/unpleasant adjectives). The four orienting-task groups of 10 subjects were: (1) the Self Reference Episodic group (with the question: "Can you access a personal experience in which you exemplified this trait?"); (2) the Self Reference Abstract group ("Does this adjective describe you?"); (3) the Mother Reference Episodic group ("Can you access an incident, either directly experienced by you or told to you, in which your mother exemplified this trait?"); and (4) Walter Cronkite Abstract group ("Does this adjective describe Walter Cronkite?"). An incidental learning procedure was used for all subjects.

An incidental factor in the design was the subject-selected variable of "yes/no." depending on how each subject answered each question.

Materials

Fifty-four trait adjectives were presented by a slide projector. They were selected from Anderson's (1968) likableness ratings, half "likable" (upper third of the norms), half "unlikable" (lower third). "Likable" examples are capable, friendly, and gracious; "unlikable" examples are prejudiced, lifeless, and disobedient.

Procedure

Subjects were tested in groups of two to four, and were told the experimenter was investigating judgments of adjectives. Words were projected on the wall, one every 7 sec, to be rated Yes or No according to whatever orienting task the subjects were given.

After presentation of all slides, the rating sheets were collected and subjects received a sheet of addition problems. Alter 10 min of this distraction test, the experimenter collected the arithmetic sheets, distributed the unexpected free recall test, and asked subjects to recall (in any order) as many of the list adjectives as they could. They had 10 min for free recall, after which they did the recognition test for 12 min. The recognition test was composed by randomly ordering the 48 originally presented adjectives with 48 new trait adjectives not shown earlier. These were selected from Anderson's norms, half "likable" and half "unlikable" Subjects used a rating scale of 1 to 4, with 1 indicating "sure this adjective was presented," 2 indicating "think it was presented but not sure," 3 indicating "think it was not presented, but not sure," and 4 indicating "sure it was not presented." Following the recognition test, subjects filled out a brief questionnaire asking them to note any thoughts, images, etc., they were aware of during the slide presentations. It also asked specifically whether they were aware of recalling personal episodic experiences to answer carne of the trait questions.

Results

Free Recall

Table 2 shows free recall percentages segregated by groups and type of adjective. There is a marginally significant difference in recall among treatment groups, F(3,36) = 2.40, p < .10. However, a preplanned comparison showed that the Cronkite group recalled significantly less than the other three groups combined, t(38) = 2.49, p < .05. Therefore, the evidence suggests that recall is poorer when the items are related to a poorly differentiated cognitive structure. The lack of difference among the top three groups also supports our hypothesis. Results of the Self-Reference Episodic and Abstract groups suggest that recall is equally enhanced whether the trait adjective is matched to an abstract self-schema or is used to search episodic memory for a relevant personal incident. Moreover, 9 of the 10 subjects in the Abstract Self-Reference group reported that they often answered a trait question by retrieving a specific autobiographic incident to support or refute the trait as self-descriptive.

The near equality of the Mother Episodic condition to the two Self-reference groups shows that enhanced recall does not depend on self-reference of the list items. The equality suggests that any well-differentiated cognitive structure can serve as a "hitching post" for evaluating and attaching to the items to be remembered. In this sense, the self is not a uniquely distinguished conceptual structure.

As Table 2 reveals, likable adjectives were recalled 1.5-fold more often than unlikable adjectives. The effect is statistically significant, F(1,36) = 31.1, p < .01, and it does not vary with the type of orienting task.

Yes Saying

The three reference characters were similarly likable. In the two Abstract judgment groups, 80% of likable adjectives were judged to apply compared to 25% of unlikable adjectives. In the two Episodic judgment groups, 75% of likable adjectives elicited memories of exemplifying episodes compared to 60% for unlikable adjectives. The difference in Yes saying to unlikable adjectives in the Abstract vs Episode conditions is understandable. Although one might admit to an incident illustrating an undesirable trait (in himself or his mother), he is loathe to concede that he or she generally acts so undesirably.

The probabilities of recall conditionalized upon an item being categorized as Yes or No were analyzed, but yielded no new information. As before, conditional recall probabilities differed across groups (with the Cronkite group the poorest), pleasant adjectives were recalled more than unpleasant ones, and Yes items were recalled only slightly more than No adjectives.

Finally, we examined recognition ratings of Old vs New test items. Accuracy was high for all groups. Indexing memory by the difference is mean ratings between Old vs New items (on a 1 to 4 scale), the scores were 1.96 for the Self-Reference Episodic group, 1.98 for the Self Reference Abstract group, 1.89 for the Mother Episodic group, and 1.47 for the Cronkite Abstract group. The first three groups do not differ, but the Cronkite group is significantly poorer than the other three combined, t(38) = 3.08 p < .01. Thus, recognition indicates the same conclusion as recall, that is, poor memory in the Cronkite group.

DISCUSSION

We have shown that memory for words is promoted by having the subject relate them to some person he knows a lot about, either himself or his mother. Thus, there is nothing special about the self-schema as a mnemonic peg: any well-differentiated person will do. The benefit to memory is also about the same whether an item is referred to an abstract personality profile or is referred to the subject's specific episodic memories. Perhaps a similar benefit would arise by relating items to a known fictional character from a novel or movie.

We can understand such results by casting them within current memory models like HAM (Anderson & Bower, 1973) or ACT (Anderson, 1976). Figure 1 shows a tiny fragment of a possible conceptual network surrounding a subject's self-concept. By convention in such graphs, circles or nodes stand for concepts and lines stand for directed associations that are labeled according to the logical relationship between the two concepts so connected. The unit of thought or conceptualization is the subject-predicate (S-P) construction, which is interpreted to mean that the denotation of the concept on the S-link is a subset (or member) of the denotation of the concept at the end of the P-link. (For details, see the ACT model of Anderson, 1976). The same network is used to record specific episodes as well as generic information. For example, the structure in Figure 1 records the fact that Self returned a lost wallet (at some time and location, not depicted here for simplicity) and helped an old woman cross the street. General semantic rules are indicated by links like those asserting that "returning a lost wallet" is an instance of an honest action, and "helping an old lady cross the street" is an instance of kindness.

Certain facts about self-concepts have a natural interpretation in such conceptual networks. For example, Markus (1977) found that some people do not "define themselves" with respect to certain traits-they do not locate themselves along that dimension, and it is neither salient nor important to them. This could correspond in the graph of Figure 1 to the absence of a link from Self to Honest (Markus would call this person Aschematic with respect to Honesty). But this person would define himself in terms of Kindness, if that link were strong.

When this (model) person is asked whether he exhibits a certain trait, he looks for a link-up between self and that trait. The answer comes directly and swiftly for Kind, but must be derived for Honest (see differences in latency for Markus' subjects). The latter derivation requires retrieving an episode and applying a classification rule to it, i.e., that returning lost wallets is an example of honest acts. In either event, suppose that whatever memory structure is active when the question is answered becomes associated to that experimental-list context, as represented by links in Figure 1.

This context association is used later for recall or recognition of items from the list. In particular, when the signal for recall is given, we presume that the person activates two entry points in his network, Self and List, and sends activation out along all pathways emanating from these nodes. Occurrence of an intersection of activation from the two sources identifies a list item somewhere along it, so that item can be overtly recalled. The recall advantage of the Self-reference condition over the Semantic or Superficial conditions of Experiment I stems from the fact that in the Self-reference condition the subject has a second known entry point (namely, Self) for starting his activation-search process, whereas the other conditions have only List as an entry point for searching. The same reasoning explains why Mother serves as a usable reference and attachment point for enhancing recall of list items. On the other hand, Cronkite is a relatively poor attachment point because too little is known about him. Suppose that a subject only has an undifferentiated profile of Cronkite as a "nice" person. When asked 48 trait questions about Cronkite, the subject could answer only by checking the trait's similarity in meaning to "nice." He would then set up associations of resemblance or contrast from Cronkite to each of the trait concepts in his memory and thence to List. Such a graph structure has two problems: First, the new Cronkite-to-trait associations must be learned from scratch, which is difficult; second, this overloads the number of new associations that must be retrieved out of one and the same concept, namely, Cronkite. But we know that such overload situations create competition, retroactive interference, and forgetting of associations at recall. For such reasons, item memory will be poorer when they are related to a less differentiated schema.

REFERENCES

Anderson, J. R. Language, memory, and thought. Hillsdale. New Jersey: Lawrence Erlbaum, 1976.

Anderson, J. R., & Bower, O. H. Human associative memory. Washington: V. H. Winston, 1973.

Anderson, N. H. likableness ratings of 555 personality-trait words. Journal of Personality and Social Psychology, 1968, 9, 272-279.

Baddeley, A. D. The trouble with levels: A reexamination of Craik & Lockhart's framework for memory research. Psychological Review, 1978, 85, 139-152.

Craik, F. I. M., & Tulving, E. Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 1975, 104, 266-291.

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

Paivio, A., Yuille, J. C., & Madigan, S. A. Concreteness, imagery, and meaningfulness values for 923 nouns. Journal of Experimental Psychology Monograph Supplement, 1968, 76, (1).

Rogers, T. 8., Kuipers. N. A., & Kirker W. S. Self-reference and the encoding of personal information. Journal of Personality and Social Psychology, 1977, 33, 677-688.

REFERENCE NOTE

I. Lord, C. G. Self-schemas and self-images as recall cues. Unpublished manuscript. Stanford University, 1978.