Adult age differences in impressio change processes

Copyright 1994 by the American Psychological Association, Inc. 0882-7974/94/S3.00

Psychology and Aging 1994, Vol. 9, No. 2, 237-250

Adult Age Differences in Impression Change Processes Thomas M. Hess and Sharon M. Pullen Three experiments examined the factors influencing impression change in young and older adults. In each study, Ss formed an impression of a fictitious target person and then read additional behavioral information that varied in its consistency with this initial impression. On the basis of previous work, older adults were expected to be less likely than younger adults to integrate new, inconsistent information in the schema-based memory representation, which would result in less impression change. No support for this prediction was found; instead, young and older adults varied in their weighting of different types of information (e.g., negative behaviors), which subsequently affected their impressions and memory for specific behavioral information. These results highlight the importance of considering the impact of age differences in implicit theories about behavior on social cognition.

cognitive processes that affect the integration of information within the memory representation (e.g., Hess & Tate, 1992). At the same time, we also assume that aging-related variations in relevant knowledge systems will affect the representation of social information by influencing the perceptions and interpretations of social events. Such variations may go beyond simple age differences in the amount of knowledge possessed. For example, as a result of their more extensive experience with people in social situations, older adults may have beliefs very different from those of younger adults about personality (e.g., its situational specificity or its plasticity), the specific indices of its underlying nature, and the causal factors associated with behavior (e.g., Blanchard-Fields, 1986; J. Erber, 1989; Heckhausen & Bakes, 1991). Such beliefs should have an impact on the interpretation of specific behavioral information (e.g., the degree to which specific behaviors indicate latent personality traits) and the extent to which it is used in making judgments about others. In the three experiments reported here we examined issues concerning potential age differences in both processing and knowledge factors.

We often make judgments about other people, either formally, such as when we evaluate someone's job performance, or informally, such as when we describe an acquaintance to a friend. Theories of social cognition emphasize that the manner in which behavioral information about specific people is represented in memory is an important determinant of such judgments (Fiske & Neuberg, 1990; Wyer & Srull, 1989). In general, the nature of these representations is affected both by information-processing operations associated with encoding processes and by knowledge-based interpretive processes. For example, the accuracy and specificity of the behavioral information encoded about a person, and its later accessibility, will be largely determined by the efficiency of basic processing operations, processing demands, and the availability of cognitive resources (e.g., Bargh & Thein, 1985; Srull & Wyer, 1983). Alternatively, processing goals and the motivation to attend to an event, the type of information encoded, and the organization of this information in memory will be influenced by the knowledge (e.g., expectancies and stereotypes) that an individual brings to the situation (e.g., R. Erber & Fiske, 1984; Gordon & Wyer, 1987). In the present set of studies we examined the impact of aging on the representation and use of social information within the context of the impression formation process. Our concern was to identify those factors associated with different patterns of impression change across age groups. Consistent with current theory in social psychology, the perspective we adopt conceptualizes age differences in social cognition in terms of variations in functions associated with both processing skills and knowledge systems (Hess, in press). Specifically, we assume that age differences in the representation of social event information in memory will be, in part, influenced by normative changes in basic

Experiment 1 Our initial study dealt with the consequences of hypothesized age differences in specific processing mechanisms on memorybased judgments regarding social events. Previous work that examined memory for scripted events (e.g., Hess, Donley, & Vandermaas, 1989; Hess & Tate, 1992) and for people (Hess & Tate, 1991) indicated that age differences in the intentional reproduction of schema-inconsistent actions are greater than those observed for schema-consistent actions. Interestingly, Hess and Tate (1992) found that this interaction between age and consistency was evident in a direct memory task, but not an indirect one. They argued that the disproportionate age difference observed in memory for inconsistent information in some direct assessments is not due to older adults' problems in comprehension of individual actions and their availability in memory, but rather to older adults' being less likely than young adults to integrate these actions within the event representation. This results in age differences in the subsequent accessibility of schema-inconsistent information when the task emphasizes the use of the schema as a retrieval mechanism, as in recall (Smith

Thomas M. Hess and Sharon M. Pullen, Department of Psychology, North Carolina State University. This research was supported by National Institute on Aging Grant AG05552 awarded to Thomas M. Hess. We would like to thank Christiene Epting, Sheryl Manthey, and Karen McGee for their assistance in data collection and scoring. Correspondence concerning this article should be addressed to Thomas M. Hess, Department of Psychology, North Carolina State University, Raleigh, North Carolina 27695-7801. Electronic mail may be sent to [email protected].

237

238

THOMAS M. HESS AND SHARON M. PULLEN

& Graesser, 1981). We assume that the relative memorability of inconsistent versus consistent information reflects the success of integration (and hence accessibility in memory) of inconsistent information. Our concern in this study was to examine the consequences of these age-related integration processes on judgments based on the event representation. If older adults are less likely than young adults to integrate schema-inconsistent information in memory, we would expect their memory representations of specific people to be less likely to change with the presentation of new, impressioninconsistent information. We would also predict little subsequent change in judgments relating to these people. This prediction is based on the assumption that judgments about people (e.g., impressions) are determined by information accessible in the memory representation. Though not designed to specifically test these ideas, Hess, Vandermaas, Donley, and Snyder's (1987) Experiment 2 provides some supportive evidence. They had young and older adults read a description of a target person that biased formation of a coherent impression conforming to a specific sex role stereotype. After presenting additional behavioral information, some of which was inconsistent with the initial target description, Hess et al. obtained ratings from their subjects on their impressions of the target. They found that the ratings of the older adults were more consistent with the initial target description than were those of the young adults, which suggests that the new information had less of an impact on the older adults' original memory representation. Unfortunately, such a conclusion is complicated by the facts that impression change was not systematically assessed and that there were no age differences in the nature of the behavioral information available in memory. In our study, to test age differences in impression change in a more systematic fashion, we made two important modifications in the design used by Hess et al. (1987). First, we collected impression ratings using the same scale both before and after the new behavioral information was presented so that impression change could be assessed. Second, we presented subjects with different types of additional behavioral information. In some cases, this new information was entirely consistent with the initial description and reinforced subjects' impressions. In other cases, the new information contained varying numbers of behaviors that were inconsistent with the initial description. On the basis of previous observations of age differences in schemaintegration processes, we predicted that younger adults would be more likely to integrate this new, inconsistent information into their schema-based person representation and, therefore, would exhibit more change in their impression ratings than would older adults after exposure to this information. That is, we expected to find a significant interaction between age group and change in ratings across assessments. In addition, because older adults are less consistent in integrating unexpected information, we predicted that they would require more inconsistent information than would younger adults to exhibit the same degree of impression change. Thus, an interaction among age, amount of new inconsistent information presented, and ratings change was also expected. Finally, we obtained a measure of working memory capacity (reading span) to see if age-related variations in capacity could account for the expected age effects in impression change.

Method Subjects. Forty-eight young adults (mean age = 20.9 years, range = 18-32 years) were recruited from introductory psychology classes at North Carolina State University (NCSU) and participated for optional class credits. Forty-eight community-dwelling older adults (mean age = 68.9 years, range = 54-80 years) were recruited through the NCSU Alumni Association and various local organizations and were paid $6 for participating. There were 24 men and 24 women in each age group. Using a 5-point scale (1 = excellent, 5 = poor), the young subjects rated themselves as more healthy (M = 1.6, SD = 0.6) than did the older subjects (M = 1.9, SD = 0.9), f(94) = 2.24, p < .03. The older adults had significantly higher levels of formal education and verbal skills than did the young adults (/» < .001). The mean years of education were 13.7 for the young adults (SD = 1.3) and 15.7 for the older adults (SD = 2.5). The young adults had a mean scaled score of 11.2 (SD - 2.6) on the Vocabulary subtest of the Wechsler Adult Intelligence Scale—Revised (WAIS-R; Wechsler, 1981), and the older adults had a mean score of 13.2 (SD = 1.8). Because of these variations in education and ability, we duplicated all formal data analyses while controlling for these factors. The results of these analyses of covariance (ANCOVA) are reported only when the outcome was affected. Materials. A short paragraph (139 words) was written that described a fictitious man whose characteristics conformed to the traditional male stereotype in our society. For example, he was depicted as a bank executive with a traditional marriage who likes competitive games, carpentry, and the outdoors. This description was typed on a single sheet of white paper. We created a set of 30 additional stimulus behaviors using materials and norms developed by Hess et al. (1987). They had obtained ratings from young and older men and women regarding the typicality (1 = very atypical, 6 = very typical) of individual behaviors for each sex. The 30 behaviors we chose had high agreement in these ratings across both sex and age groups (correlations ranged from .94 to .98). Ten of the 30 behaviors were masculine sex-typed (mean typicality ratings were 5.2 for men and 1.9 for women), 10 were feminine sextyped behaviors (mean typicality ratings were 2.3 for men and 5.5 for women), and 10 were gender neutral (mean typicality ratings were 4.8 for men and 4.7 for women). Using these 30 behaviors, we constructed three separate sets of 20 behaviors each, with each set varying in its consistency with the preface. The low-inconsistency set comprised 10 masculine and 10 neutral behaviors, the medium-inconsistency set comprised 10 masculine, 5 neutral, and 5 feminine behaviors, and the high-inconsistency set comprised 10 masculine and 10 feminine behaviors. Across all three sets, we controlled for the mean length of the individual behavioral descriptions within each gender category in terms of both words (5.5) and syllables (7.5). The behaviors were randomly arranged within each set with the stipulation that no more than two behaviors of one type could be presented in sequence. We also created a recognition memory test for each behavior set by randomly arranging the 20 target behaviors with 20 new behaviors of a similar kind. The trait-rating task consisted of the 20 masculine, 20 feminine, and 20 neutral traits in the Bern Sex-Role Inventory (BSRI; Bern, 1974). These traits were listed in random order on a single piece of standardsized paper, with the rating scale described at the top. To get some indication about knowledge of sex role stereotypes, we also used a scale adapted from the stereotype questionnaire developed by Broverman, Broverman, Clarkson, Rosenkrantz, and Vogel (1970). The scale consisted of 58 bipolar items, each of which described a particular trait. For 38 of the items, one pole of each item was characterized as typically masculine and the other as typically feminine, whereas the remaining 20 items were neutral with respect to gender. For the reading span task, 60 sentences of 13 to 16 words in length were drawn from novels. No obvious relation existed between any of the sentences, and no two ended in the same word. Each sentence was typed on a 20.3 cm X 12.7 cm white index card. Procedure. Subjects were tested individually and were told that the

239

IMPRESSION CHANGE AND AGING purpose of the study was to investigate how we form impressions about other people and how we use these impressions to make judgments. They were given 1 min to read the description (or preface) of the target individual and were asked to form an impression of what he was like. Immediately after reading the preface, they were given the list of 60 personality traits from the BSRI. Using a scale ranging from 1 (not at all descriptive) to 7 (very descriptive), subjects rated the degree to which they believed each trait was consistent with their impression of the target. After completing the trait-rating task, subjects were told that they would now learn more about the character by reading a list of behaviors that he had engaged in during a typical week of his life. These behaviors were presented one at a time on a computer screen. The subjects read through the list at their own pace, pressing the spacebar when they had understood the description, and reading times were recorded for each behavior. Equal numbers of men and women in each age group were randomly assigned to each of three information (inconsistency) conditions. The order of presentation of the behaviors within each set was counterbalanced across subjects. After presentation of the behaviors, subjects were asked to think about all of the information they had now received about the character and to once again rate the degree to which each trait on the BSRI described him. Subjects were then given a recognition test for the behaviors they had read on the computer. The appropriate recognition list was presented on the computer screen, one behavior at a time, and subjects indicated for each item whether it was one of the original set of 20 behaviors. Subjects were next given the sex role stereotype questionnaire and asked to rate on a 7-point scale the degree to which they believed each trait described the typical male in our society. We calculated a score based on the 38 sex-typed items from Broverman et al. (1970) that indicated the degree to which subjects agreed with the stereotyped gender association for each trait. Scores could range from 38 to 266, with higher scores indicating greater stereotype agreement. The young adults had significantly higher scores (M = 173,SZ>= 16) than the older adults (M

= 162, SD = 12). Note, however, that subjects in both age groups held relatively unstereotypic views in that the mean ratings for both young (4.6) and older (4.3) adults were just above the midpoint (4) on the scale, which indicated gender neutrality. Finally, the reading span task was given. The index cards were shuffled for each subject, and three groups of 2, 3, 4, 5, and 6 sentences were created. The groups were arranged by increasing size, with a blank card inserted between each group. After practice, subjects were given the deck of index cards and read each sentence aloud until they encountered a blank card. At that point, they attempted to recall the last word of each sentence in the just-read order. Reading span was determined by the highest set size at which the subject correctly recalled the final words in at least two of the three groups. An additional 0.5 was added to a subject's span if he or she had correct recall for one of the three groups at the next level. The younger adults had significantly higher spans (M = 3.0, SD = 0.8) than the older adults (M = 2.6, SD = 0.6), ) = 2.27,p2 values are reported for within-subjects effects. For each mixed-model ANOYA in this article, Huynh-Feldt « estimates were used to indicate possible violations of the multisample sphericity assumption. When these estimates were below .90 for specific repeated-measures effects, we used adjusted degrees of freedom for significance tests (Hertzog & Rovine, 1985).

Table 1 Experiment 1: Mean Trait Ratings Before and After Presentation of New Information Time 2

Time 1 Information condition

Masculine

Neutral

Feminine

Masculine

Neutral

Feminine

Young adults Low inconsistency M SD Medium inconsistency M SD High inconsistency M SD

6.0 0.4

4.2 0.4

3.5 0.7

5.9 0.5

4.6 0.6

4.5 0.4

5.8 0.4

4.0 0.3

3.6 0.7

5.3 0.7

4.1 0.3

4.4 0.5

5.8 0.8

4.0 0.4

3.7 0.7

5.4 0.8

4.4 0.4

4.9 0.6

4.4 0.8

Older adults Low inconsistency M SD Medium inconsistency M SD High inconsistency M SD

6.1 0.6

4.2

3.9 0.9

6.1 0.6

4.3

0.5

6.1 0.5

4.3 0.5

3.9 0.6

5.9 0.6

4.3 0.3

4.7

6.1 0.5

4.1

3.6 0.6

5.7 0.5

4.2 0.3

4.6 0.4

Note. Ratings could range from 1 to 7.

0.3

0.5

0.5

240

THOMAS M. HESS AND SHARON M. PULLEN

Significant effects were obtained for time, F(l, 90) = 73.41, MSC = 0.12, a)2 = .38; BSRI item, F(2, 180) = 631.56, MSC = 0.29, a2 = .81; and their interaction, F(2, 180) = 143.31, MSe = 0.10, a;2 = .33. These effects reflected the facts that ratings for masculine items were greater than those for feminine items but that the masculine ratings decreased when additional behavioral information was provided whereas the feminine ratings exhibited a significant increase. Ratings for neutral traits also increased over time, but not to the same degree as for feminine traits. As expected, time and item also interacted with age, F(2, 180) = 5.05, MSe = 0.10, o>2 = .02, reflecting the fact that the older adults exhibited less overall change than the younger adults regardless of condition. Contrasts involving each type of item indicated significant Age X Time interactions for feminine, F(l, 90) = 4.24, and neutral, F ( l , 90) = 6.24, items, with older adults exhibiting less change in ratings than younger adults in each case. The Age X Time interaction did not achieve significance, F( 1, 90) = 2.80, p < . 10, for masculine items, but its form was identical to that for the other two types of items. The Age X Time X Item interaction was not significant when the Vocabulary subtest score was used as a covariate, F(2, 178) = 1.61, p < .21, apparently reflecting a correlation between Vocabulary score and ratings change (Time 1 — Time 2). However, these correlations were significant only for the young adults and only for neutral (.35) and feminine (.35) traits; the corresponding correlations for the older adults were —.05 and .06. This disparity in correlations indicates a violation of the homogeneity-of-slopes assumption for ANCOVA, and thus it is unclear what should be made of this finding. The only other significant effects obtained in the main analysis were due to the interactions between condition and item, F(4, 180) = 2.68, M5e = 0.29, o>2 = .02, and among condition, item, and time, F(4, 180) = 4.17, MSC = 0.10, a2 = .02, and reflected the expected pattern of change associated with the type of additional information. Specifically, as more inconsistent (feminine) information was presented, changes in both masculine and feminine ratings across times of test increased: The Time X Condition interaction was significant for both types of items, Fs(2, 90) > 3.80. As would be expected, neutral ratings were unaffected by condition (F < 1). Contrary to our expectations, the impact of information condition on ratings did not vary across age groups, as indicated by the nonsignificant fourway interaction (F < 1.0). (The power of the present analysis to detect a moderate-sized effect [/= .25] was .78.) We had some concern that the small but significant variations across age groups in beliefs about sex role stereotypes might have influenced the amount of impression change observed. Therefore, we performed an ANCOVA on these data using the Broverman et al. (1970) scale score as a covariate. This did not change the pattern of results, however, in that the Age X Item X Time interaction was still significant, F(2, 178) = 6.49, MSC = 0.10, oj2 = .04. An additional ANCOVA with reading span as a covariate produced results identical to those of the original ANOVA, which suggests that age differences in working memory capacity did not mediate performance. Memory for behavior information. To obtain an indication of the information that was available for making judgments, we examined memory for the additional behaviors by using hit and false-alarm rates to calculate d' scores (see Table 2). (Perfect recognition was assigned a d' of 4.64.) Because subjects were

Table 2 Experiment 1: Mean Recognition (d')for New Behavioral Information Behavior type Information condition

Masculine

Neutral

Feminine

3.89 0.91

2.97 0.98

— —

3.37 0.96

3.42 1.13

3.96 0.81

3.58 0.78

— —

3.61 1.08

Young adults Low inconsistency M SD Medium inconsistency M SD High inconsistency M SD

Older adults Low inconsistency M SD Medium inconsistency M SD High inconsistency M SD

2.43 1.24

1.65 0.88

2.68 1.10

2.53 1.49

2.77 1.32

3.26 1.31 3.12 1.18

Note. Dashes indicate that no data is available for these cells.

tested only for the behavior categories to which they were actually exposed, two separate 2 X 2 X 2 (Age Group X Information Condition X Behavior Type) ANOVAs were performed, each comparing conditions that viewed the same types of behaviors. In the first analysis, we contrasted the low- and moderateinconsistency conditions and examined memory for masculine and neutral behaviors. Younger adults had better recognition memory than older adults, F(l, 60) = 38.09, MSf = 1.88, u2 = . 18, and masculine behaviors were recognized better than neutral ones, F( 1,60) = 11.83, MSe = 0.55, u2 = .08. An interaction between behavior and condition was also obtained, F( 1, 60) = 9.32, MSC = 0.55, o>2 = .06, because d' scores for neutral behaviors were significantly greater in the moderate-inconsistency condition than in the low-inconsistency condition, F( 1, 60) = 5.41. In the second analysis, memory for masculine and feminine behaviors was compared in the moderate- and high-inconsistency conditions. Once again, the young had better recognition memory than the old, F(l, 60) = 14.56, MSe = 1.80, w2 = .06. In addition, feminine behaviors were recognized more accurately than masculine behaviors, F(l, 60) = 8.69, MSC = 0.55, o>2 = .06. No other effects were significant. Interestingly, the Age X Item interaction was not significant in either analysis (Fs < 1.0), which indicated no age differences in the effects of consistency on memory performance. (The power to detect a moderate-sized interaction was .50.) ANCOVAs with reading span as a covariate did not alter the age effects in either analysis. Recognition performance was close to ceiling for the young adults in several conditions. Thus, we decided to analyze the data in a different fashion, examining the proportion of both targets and distractors given positive recognition responses. This gave us an idea of the extent to which each type of mfor-

241

IMPRESSION CHANGE AND AGING

mation—whether correctly recognized or not—was attributed to the behavior list. No age differences in performance were obtained when this was done, which further suggests little between-groups variation in the nature of the memory representation. Study times. We examined the time spent studying individual behaviors to see if there was any evidence that young and older adults allocated their attention differently during reading. Study times for entire behavior descriptions that were less than 500 ms were eliminated from consideration. To control for variations in the length of the behavior descriptions, we then divided each remaining time by the number of words in the corresponding description in order to obtain per-word study times. Times greater than 3 standard deviations above a subject's mean were eliminated from further consideration. We then used the remaining times to calculate mean per-word study times for each type of behavior for the first and second halves of the list. (This rather broad division was used because of the relatively small number of each type of behavior within lists, especially in the moderate-inconsistency condition.) Because subjects did not all read the same types of behaviors, two separate 2 X 2 X 2 X 2 (Age Group X Information Condition X Behavior Type X List Position) ANOVAs were performed once again. When the low- and moderate-inconsistency conditions were compared, significant effects were obtained for behavior, F( \, 60) = 4.17, MS; = 4,310, to2 = .03, and list position, F(l, 60) = 38.37, MS; = 17,881, a;2 = .23. Specifically, neutral items were read more quickly than masculine items (560 ms vs. 577 ms), and items from the first half of the list were read more slowly than items from the second half (621 ms vs. 517 ms). In comparing the moderate- and high-inconsistency conditions, we observed significant effects for age, F ( l , 60) = 7.54, MSe = 164,062, u2 = .07; behavior, F(l, 60) = 15.40, MSe = 19,848, a;2 = .10; and list position, F(l, 60) = 37.32, MSe = 18,754, o>2 = .22. Older adults had longer study times than young adults (664 ms vs. 525 ms), masculine items were read more quickly than feminine ones (560 ms vs. 630 ms), and first-half items were read more slowly than second-half items (647 ms vs. 543 ms). No other effects were significant in these two analyses, including interactions involving age (power to detect moderatesized effects = .51). Thus, there is no evidence of systematic differences in attention allocation across age groups.

Discussion The results of this study were not entirely consistent with our expectations. We predicted that older adults would exhibit less impression change than younger adults, and this was in fact observed. The lack of a significant four-way interaction for ratings, however, provided no support for our prediction that older adults would be less sensitive than young adults to the degree to which the additional behavioral information was inconsistent with the initial target description. In addition, there is no evidence that the age differences in impression change can be related to variations between groups in either attention allocation or the relative availability of consistent versus inconsistent information in memory. Age did not interact with any other variable in the analyses of reading times and recognition memory. Interestingly, these null results were obtained in spite of significant age differences in reading span scores, which are thought

to reflect working memory capacity, which in turn affects encoding efficiency. These results appear on the surface to be incompatible. On the one hand, young and older adults seem to be studying the information in the same manner and to have similar types of information available to them in memory. On the other hand, when making judgments, older adults appear to be less influenced by the new information, a finding similar to that of Hess et al. (1987). Several factors could be operating to create this seeming confusion. First, it should be recognized that similar patterns of study times across age groups do not necessarily imply the use of the same processing operations. Other research that used similar materials (e.g., Hess & Tate, 1992; Hess et al., 1989) showed null age effects for study times but systematic agerelated variations in the accessibility of information in memory. In addition, Hess and Tate (1991) showed that young and older adults do differ in the types of operations they use to process behavioral information and that these variations affect the accessibility of such information in memory. Thus, although the temporal characteristics associated with the stimuli are similar and reflect schema-based expectations, the underlying processing operations may vary with age. Recognition memory may also not be an accurate index of what information is accessible for making judgments about people, because recognition is not thought to necessarily require an active search on the part of the subject and thus may not require use of subject-generated links between items in memory for retrieval. Thus, the information that subjects access in making judgments may be very different from the information available to them for making recognition responses, and age differences in integration may emerge only when accessibility is emphasized (e.g., Hess & Tate, 1992). Alternatively, it is possible that the variations in impression change across age groups may reflect differences in belief systems that affect how information in memory is used in making decisions. Even though we controlled for age differences both in perceptions of the sex role relatedness of the stimulus behaviors and in stereotypical beliefs, young and older adults may still weight information differently in making decisions. Such differences may reflect variations in implicit theories about people and personality and the degree to which specific behaviors represent underlying traits.

Experiment 2 In our second experiment we used a variant of the procedure from Experiment 1 in an attempt to deal with these and other concerns. First, we used a cleaner design that allowed more direct comparisons between conditions for memory and study times. In Experiment 1, the fact that one condition contained no feminine behaviors and one contained no neutral behaviors precluded systematic comparisons across information conditions. In Experiment 2, only two information conditions were used, across which the ratio of consistent to inconsistent behaviors was simply reversed. As before, we were interested primarily in seeing if the first impressions of older adults would be more impervious to new and small amounts of inconsistent information than those of younger adults. If so, a significant interaction would be expected among age, amount of inconsistent information, and ratings changes over times of assessment.

242

THOMAS M. HESS AND SHARON M. PULLEN

Second, instead of using sex-role-related stimuli, we used behavioral descriptions that were evaluatively positive or negative. It has been proposed that the primary basis for impressions about others is evaluative, especially when people are given little prior information about specific trait categories (Srull & Wyer, 1989). The use of such descriptions provides a better tie-in with existing theory and research on impression formation and eliminates potential confounds that are due to developmental or cohort variations in sex role identity or knowledge (e.g., Hyde & Phillis, 1979; Windle & Sinnott, 1985). We also used two different target persons, each depicted in a different manner (i.e., one in positive terms, the other in negative terms). This allowed us to examine the generality of age effects across different stimulus conditions. If the previous results were due to older adults' being less likely in general to integrate inconsistent behavioral information regardless of its nature, then the pattern of performance across age groups should not vary with stimulus materials. If, however, age differences in beliefs about the nature of specific personality attributes were influencing the interpretation and subsequent representation of behavioral information, then variation in the pattern of performance across age groups along with the type of stimulus information would be expected (i.e., target type should interact with time and age in determining changes in ratings). For example, Heckhausen and Bakes (1991; Heckhausen, Dixon, & Baltes, 1989) found that, although high agreement exists between age groups regarding the desirability of specific personal attributes, older adults are less likely than younger adults to believe that negative personal attributes are controllable. If these implicit beliefs influence perceptions of behavior, then age differences in impression formation should result when the task requires subjects to evaluate behaviors that are linked with negative traits (e.g., older adults will infer negative behavioral information to be more fundamental to a person's personality). Such between-groups variations should be independent of specific age-related problems in integration. Finally, we assessed memory for behavioral information by using recall rather than recognition. Because recall typically requires more active search than recognition and is also more schema driven (Smith & Graesser, 1981), it should provide a better indication of the information that is immediately accessible to subjects, both for recall and for making judgments. This accessibility should reflect the effectiveness of initial encoding operations. Thus, if older adults are less effective than young adults at integrating unexpected information, a significant interaction should be observed between age and consistency. Method Subjects. Forty-eight young adults (mean age = 20.0 years, range = 18-31 years) and 48 older adults (mean age = 67.5 years, range = 5989 years) participated, with 24 men and 24 women in each age group. Subjects were recruited and compensated as in Experiment 1. None of these subjects had participated in any previous impression-formation study. The older adults had significantly more years of formal education (M = 15.6, SD = 2.5) than the young adults (M = 13.3, SD = 1.2), t(94) = 5.75, and their WAIS-R Vocabulary scaled scores (M = 14.2, SD = 2.1) were also greater than those of the young adults (M = 12.7, SD = 2.1), t(94) = 3.60. All analyses were redone while controlling for these two variables, but the results were not affected. The self-rated health of the young adults (M = 1.7, SD = 0.7) did not differ from that of the older adults (M = 1.8, SD = 0.8).

Materials. Two short prefaces, similar to the one used in Experiment 1, were developed and typed on standard white paper. One described a male actor in positive terms, the other a male actor in negative terms. We obtained the additional behavioral information by constructing a set of 60 behaviors and then having an independent sample of 24 younger and 8 older adults read each preface and rate the likelihood that the described actor would engage in each of the behaviors on a scale from 1 (very unlikely) to 7 (very likely). On the basis of these ratings, we selected 16 positive and 16 negative behaviors. Correlations between the young and older adults' mean evaluative ratings for these behaviors were .96 for the positive preface and .93 for the negative preface. For the positive preface, positive behaviors had a mean likelihood rating of 6.2 and negative behaviors a rating of 2.0. For the negative preface, the mean likelihood rating was 2.2 for the positive and 5.5 for the negative behaviors. The behavioral descriptions had a mean of 10.5 words and 14.3 syllables. Stimulus sets were constructed that consisted of either 16 positive and 8 negative behaviors or 8 positive and 16 negative behaviors. Note that the former set constituted the low-inconsistency condition for the positive preface and the high-inconsistency condition for the negative preface. The opposite was true for the second set. Eight different lists were constructed for each of these stimulus sets so that each of the 32 behaviors appeared equally often across lists with each of the other behaviors in each list quartile. Four neutral behaviors were included as buffers at both the beginning and the end of each list. The trait-rating task used a list consisting of 15 positive, 11 negative, and 12 neutral traits. The traits were obtained from Hess, Follett, and Perry (1992), who had 30 younger and 30 older adults provide evaluative ratings for 50 different traits on a scale from 1 (good) to 7 (bad). The between-ages correlation on the ratings for these 38 traits was .97. Mean ratings were 1.6 for the positive traits, 6.3 for the negative traits, and 3.8 for the neutral traits. The traits were listed in random order on a sheet of paper, with the rating scale specified at the top. Materials for the reading span assessment were identical to those in Experiment 1. We also obtained an additional measure of working memory capacity using a computation span task (Salthouse & Babcock, 1991), which was similar in format to the reading span task except that two-digit arithmetic problems were used instead of sentences. Three groups of 2, 3,4, 5,6, and 7 problems were constructed for oral presentation. Procedure. The general procedure for the impression formation task was identical to that used in Experiment 1. Within each age group, equal numbers of men and women were assigned to the two preface conditions. Immediately after reading the appropriate preface, subjects were given the list of 38 traits to make their impression ratings. Subjects were then given the same instructions as in Experiment 1, and the additional behavioral information was presented via microcomputer. Half of the subjects in each preface condition were randomly assigned to the low-inconsistency condition (16 preface-consistent and 8 prefaceinconsistent behaviors) and half to the high-inconsistency condition (8 consistent and 16 inconsistent behaviors). After reading through the list of behaviors, subjects were asked once again to rate the extent to which each of the 38 traits described the actor based on everything they now knew about him. After the second rating task was completed, subjects were given a minimum of 3 min to orally recall the behaviors they read on the screen. Recall was recorded on audiotape-. The test was concluded only when subjects had not recalled a new item for 30 s. The procedure for the reading span task was the same as before. The younger adults' scores (M = 3.2, SD = 1.0) were only marginally higher than those of the older adults (M = 2.9, SD = .7), /(94) = 1.93, p < .06. In the computation span task, the problems in each set were orally presented to the subject, who then had to select the correct answer of the three listed on a sheet of paper in front of him or her. After all the problems in a group had been presented, the subject recalled the second digit in each problem in the order presented. Computation span was calculated as the highest set size at which the subject had correct serial recall of the target digits for two of the three groupings, plus an addi-

243

IMPRESSION CHANGE AND AGING tional 0.5 for correct recall of one of three at the next level. This measure was only moderately correlated with reading span (r = .34), and the young adults had significantly higher computation spans (M = 3.3, SD = 1.5) than the older adults (M = 2.7, OT = 1 .0), 494) = 2.25.

Results Impression change. Mean ratings for consistent, neutral, and inconsistent traits were calculated at each time of test, and a 2 X 2 X 2 x 2 x 3 (Age Group X Preface X Information Condition X Time of Test X Trait Type) ANOVA was performed on these means. Significant effects were observed for time, F(\, 88) = 62.30, MS, = 0.25, u2 = .24; trait, F(2, 176) = 326.94, MS, = 0.53, a;2 = .69; and the Time X Trait, F(2,176) = 152.36, MS, = 0.31, o>2 = .34; Age X Trait, F(2, 176) = 5.06, MSC = 0.53, o>2 = .03; Preface X Trait, F(2, 176) = 70.81, MS, = 0.53, a;2 = .33; and Preface X Trait X Time, F(2, 176) = 40.43, MS, = 0.31, a;2 = . 12, interactions. The expected Age X Trait X Time interaction was not significant, F(2, 176) = 2.32, p > . 10. There was, however, a significant Age X Preface X Trait X Time interaction, F(2, 176) = 3.31, MS, = 0.31, u2 = .01 (see Table 3), which indicated that the expected interaction was moderated by the preface that subjects read. No other significant effects involving age were obtained. As can be seen, both age groups had similar patterns of performance, with preface-consistent trait ratings decreasing over time and preface-inconsistent ratings increasing. In addition, this effect varied as a function of preface, with subjects appearing to weight negative information more than positive information in their ratings. Specifically, when the target was initially portrayed in negative terms, new positive information had less of an impact on trait ratings than did new negative information presented about the positively portrayed target. Ratings for neutral traits increased slightly in all conditions but were unaffected by preface. The four-way interaction indicates that this weighting effect was more pronounced for the older adults than for the young adults. That is, the change in ratings from Time 1 to Time 2 was greater for the older adults than for the young adults

in the positive preface condition (—1.4 vs. —1.2 for consistent traits, 1.9 vs. 1.5 for inconsistent traits) and less than that for the young adults in the negative preface condition (0 vs. -0.4 for consistent traits, 0.7 vs. 0.9 for inconsistent traits). This variation in the impact of preface across age groups can be seen in the relatively greater strength of the Preface X Item X Time interaction for the older adults, F(2, 88) = 34.41, MSC = 0.30, o>2 = .19, than for the young adults, F(2, 88) = 10.03, MS, = 0.32, «)2 = .06. The Time X Item interactions were similar in strength for the two groups, Fs(2, 88) > 77, MS, = 0.51 (young) and0.41(old),o>2s = .35. The only other significant effects obtained in the main analysis involved the interaction of information condition with item, F(2, 176) = 10.93, MS, = 0.53, 2 = .07. These effects reflected the expected impact of information condition, with consistent and inconsistent trait ratings changing (Time 1 Time 2) more in the high-inconsistency condition (1.05 and — 1.64, respectively) than in the low-inconsistency condition (0.41 and -0.80). As in Experiment 1, the impact of information condition did not vary across age groups: The Age X Condition X Trait X Time interaction was not significant (F < 1), and the power of the present analysis to detect a moderate-sized effect was .85. Finally, the results of ANCOVAs that used either reading span or computation span as covariates did not alter the previously described pattern of results, which suggests once more that age differences in working memory capacity were not mediating performance. Memory for behavioral information. We examined the proportion of behaviors recalled by using a 2 X 2 X 2 X 2 (Age Group X Preface X Information Condition X Consistency) ANO\A. As can be seen in Table 4, older adults recalled less than younger adults, F(l, 88) = 30.33, MS, = 0.04, w2 = .16, recall was higher with the negative preface than with the positive one, F(l, 88) = 4.85, oi2 = .03, and inconsistent items were recalled better than consistent ones, F( 1,88) = 5.82, MS, = 0.02, iii2 = .02. A significant interaction between preface and consis-

Table 3 Experiment 2: Mean Trait Ratings Before and After Presentation of New Information Time 2

Time 1 Preface

Consistent

Neutral

Inconsistent

Consistent

Neutral

Inconsistent

Young adults Positive M SD Negative M SD

6.2 0.6

3.6 0.6

1.9 0.7

5.0 0.8

4.2 0.8

3.4 0.8

5.0 0.6

3.6 0.7

3.1 0.7

4.6 0.8

4.2 0.4

4.0 0.9

4.7

4.3 0.6

3.7

0.9 4.4

4.3

0.9

0.6

3.9 0.7

Older adults Positive M SD Negative M SD

6.1 0.6

3.9 0.4

1.8 0.6

4.4

4.0 0.6

3.2 0.7

0.8

1.0

244

THOMAS M. HESS AND SHARON M. PULLEN

Table 4 Experiment 2: Proportions of New Behaviors Recalled Positive preface Information condition

Consistent

Negative preface

Inconsistent

Consistent

Inconsistent

Young adults Low inconsistency

M SD

.35 .13

.49 .16

.41 .11

.48 .15

.29 .18

.43 .15

.59 .22

.42 .15

High inconsistency

M SD

Older adults Low inconsistency

M SD

.20 .13

.34 .22

.31 .10

.30 .16

.22 .18

.30 .16

.31 .18

.28 .15

High inconsistency

M SD

tency, F(l, 88) = 19.34, MSe = 0.02, 2 = .01, which reflected nonsystematic variations in this basic pattern of results across age groups and information conditions. The power to detect other age-related interactions ranged from .65 to .99.

Discussion The results of this experiment once again provide little evidence in support of our hypothesis that the first impressions of older adults would be more difficult to change than those of young adults. Both young and older adults changed their impression ratings in response to new behavioral information, but age did not interact with the amount of inconsistent information presented in determining ratings change. Instead, the results suggest that there are systematic age-related biases in the

245

IMPRESSION CHANGE AND AGING 750 700 650

Preface/Consistency: 0Q A A

Positive/Consistent Positive/Inconsistent Negative/Consistent Negative/Inconsistent

600 >. 550

a 500 450 •0

400 350 List Quartile

Figure 1. Mean study times from Experiment 2 by preface read, consistency of the stimulus behavior with the preface, and position of the behavior in the study list.

manner in which impressions change. In both age groups, subjects exhibited a negativity bias in their ratings, with more weight being given to negative information than positive information in determining impression change. Specifically, when an individual was initially depicted in negative terms, new positive information had a relatively small impact on the ratings associated with the original impression. In contrast, when the target was initially portrayed in a more positive light, negative information had a more substantial impact on subsequent ratings, with consistent ratings decreasing and inconsistent ratings increasing more than with the negative preface. This negativity effect has been observed elsewhere and has been hypothesized to reflect people's implicit theories of personality (Skowronski & Carlston, 1989). In terms of the concept of cue validity (Rosch, 1978), negative information appears to have a high degree of diagnosticity with respect to underlying personality traits and thus has high cue validity when used to assign people to categories. Specifically, negative behaviors are seen as being disproportionately performed by people who we view in negative terms, and thus negative behaviors can be reliably used to identify people with negative personality traits. In contrast, positive behaviors have a lower level of diagnosticity because, although they may be indicative of underlying personality traits, even "bad" people perform good behaviors every now and then. Thus, positive behaviors cannot be used as reliably as negative behaviors in making categorizations. Interestingly, the results of Experiment 2 indicate that the older adults weighted the negative information even more than did the younger adults, with the previously described variations in ratings across preface conditions being greater for the old than for the young. The recall data also support this interpretation. Older adults' memory for positive behaviors varied according to its consistency with the preface, with recall of these

behaviors being greater when they were inconsistent. In contrast, negative behaviors were recalled at the same high level regardless of their consistency. The typically observed superiority in recall of inconsistent information in on-line situations has been attributed to the more extensive processing accorded these items (e.g., Hastie, 1984; Srull & Wyer, 1989). Thus, the fact that negative behaviors were being recalled at the same level as inconsistent positive information suggests that they were being processed more extensively than would be expected solely on the basis of their consistency with the initial impression information. This pattern of performance was also observed for the young adults, but to a lesser degree. Given the absence of systematic age effects in study times, we hypothesized that the observed between-groups variations reflected age-related differences in implicit theories of behavior and personality. That is, relative to the ratings and recall of the young adults, those of the older adults suggest that they are operating under a belief system that disproportionately highlights the importance of negative information in judging people. Such an interpretation is consistent with Heckhausen and Baltes's (1991) observation that older adults have a stronger belief in the uncontrollability of negative traits, which suggests a concomitant belief that negative behaviors should reflect fundamental, unchanging aspects of personality. To pursue this explanation further, we conducted a third study.

Experiment 3 Our primary purpose in Experiment 3 was to replicate the age-related negativity bias observed in the impression and memory data of Experiment 2. To do this, we used the same design, but we eliminated the information condition manipulation and presented all subjects with 16 consistent and 8 inconsistent behaviors. We also attempted to assess people's perceptions of the diagnosticity of behaviors more directly by having them rate the degree to which different types of behaviors would affect their impression of a specific target person. If older adults do disproportionately weight negative information, we would expect them to indicate that they would be swayed more than young adults by negative information when judging a target depicted previously in positive terms. Likewise, we would also expect them to be swayed less by positive information about a person depicted originally in negative terms. Method Subjects. Subjects were recruited from the same sources and compensated in the same manner as in the previous experiments. The young group consisted of 16 men and 16 women with a mean age of 18.3 years (range = 17-20 years) and an average of 12.5 years (SD = 0.7) of formal education. There were also 16 men and 16 women in the older adult group. These individuals had a mean age of 66.9 years (range = 60-76 years) and an average of 14.5 years (SD = 2.3) of education. The difference between groups in education level was significant, t(62) = 4.67. There were, however, no between-groups differences (ps > .69) in scaled WAIS-R Vocabulary scores (young, M = 13.7, SD = 2.6; old, M = 13.9, •SD = 2.5) or in self-rated health (M = 1.8 for both groups). Given their low explanatory power in the previous experiments, no span measures were obtained. Once again, controlling for education and verbal ability had no impact on the results of the main data analyses. Materials and procedure. The impression formation and recall tasks were almost identical to those in the low-inconsistency condition

246

THOMAS M. HESS AND SHARON M. PULLEN

Table 5 Experiment 3: Mean Trait Ratings Before and After Presentation of New Information Time 1 Preface

Consistent

Neutral

Time 2 Inconsistent

Consistent

Neutral

Inconsistent

Young adults Positive M SD Negative M SD

6.2 0.5

3.4 0.6

1.6 0.5

5.5 0.8

4.1 0.8

2.5 0.7

4.9 0.6

3.9 0.7

3.3 1.0

4.5 0.6

4.2 0.5

3.9 0.6

Older adults Positive M SD Negative M SD

6.4 0.5

4.0 0.5

1.5 0.8

5.3 0.9

4.3 0.7

2.8 1.0

4.6 0.8

3.6 0.5

3.3 1.0

4.6 0.6

4.1 0.6

3.7 0.8

of Experiment 2. The only change was that the lengths of the behavior descriptions were reduced to a mean of 5.2 words and 7.1 syllables. We thought that this might decrease some of the variability associated with reading times and thereby make these measures more informative with respect to potential age differences in processing operations. After subjects had finished with the recall test, they were given the impressionchange rating task. Subjects were instructed to read the target description that they had not read previously (i.e., those in the positive preface condition read the negative preface, and vice versa) and to form an impression of that person. They were then presented with the 16 positive behaviors, 16 negative behaviors, and 8 neutral buffer behaviors that were used as stimuli, randomly listed on two sheets of paper. Two separate random orders were used. Subjects read each behavior and then rated the degree to which it would change their impression of the person about whom they had just read. They used a 7-point scale, with 1 indicating no change and 7 indicating great change.

Results and Discussion Impression change. We examined mean trait ratings using a 2 x 2 X 3 X 2 (Age Group X Preface X Item X Time of Test) mixed-model ANOVA. The only significant effects obtained were for time, F(l, 60) = 25.80, MSe = 0.18, u>2 = .16; item, F(2,120) = 244.75, MSe = 0.77, u2 = .72; and the Item X Time, F(2, 120) = 59.96, MSe = 0.27, o>2 = .23; Preface X Item, F(2, 120) = 72.35, MSe = 0.77, u2 = .43; Preface X Item X Time, F(2, 120) = 12.72, MSC = 0.27, w2 = .06; and Age X Preface X Item X Time, F(2, 120) = 3.42, MS, = 0.27, o>2 = .02, interactions. This last interaction was of primary interest and, as can be seen in Table 5, reflected the same pattern of performance observed in Experiment 2. That is, both age groups weighted negative information more than positive information in determining their ratings, but once again the variation in ratings change across prefaces was greater in the older adults than in the young adults. This is indicated by the fact that a significant Preface X Item X Time interaction was observed for the older adults, F(2, 60) = 14.65, MSt = 0.26, o>2 = .22, but not for the young adults, F(\, 60) = \.ll,p < .18,MSe = 0.28, 2 = .03, and consistency, F(2, 120) = 65.99, MSe = 1.50, to2 = .40. Change ratings were higher for the negative preface (M = 3.4) than for the positive preface (M = 2.8) and were greater for inconsistent behaviors (M = 4.6) than for consistent (M = 2.3) or neutral (M = 2.5) behaviors. Because of the exploratory nature of this specific task, we conducted specific planned contrasts to test two specific hypotheses even though the three-way interaction was not significant (F < 1). First, we predicted that older adults would weight negative information more than younger adults in the positive preface condition. This contrast was not significant (F < I). Second, we expected that older adults would weight positive information less than would younger adults in the negative preface condition. This prediction was supported, F(l, 30) = 4.47, with older adults indicating that they would be less likely than younger adults to change their opinion about a negatively perceived person if they were presented with new positive information. Thus, this examination of more direct judgments of behavioral information provides some preliminary support for our hypothesis regarding age differences in implicit personality theories. It is likely that other assessment procedures (e.g., greater separation of the impression formation and judgment tasks, more specific questions regarding beliefs, use of independent samples) might prove more sensitive to age effects.

General Discussion 1000 950

900

Preface/Consistency: o G A A

Positive/Consistent Positive/Inconsistent Negative/Consistent Negative/Inconsistent

850

800

750 700 650 600 2

3

List Quartile

Figure 2. Mean study times from Experiment 3 by preface read, consistency of the stimulus behavior with the preface, and position of the behavior in the study list.

In these three experiments we examined the impact of agingrelated variations in both basic processing operations and knowledge-based mechanisms on impression change. With respect to the former factor, the results of Experiments 1 and 2 offered little support for the hypothesis that older adults' first impressions would be more resistant to change than those of younger adults because of age-related variations in schemabased integration processes. Contrary to expectations, the main determinants of age differences in performance were related to the type of materials rather than the amount of inconsistent information presented. In addition, age differences in impression change were not mediated by measures of working memory capacity. Although these findings are not supportive of the hypothesis that age-related differences in basic cognitive skills will affect the representation of social events in memory, we are not willing at this point to reject this general thesis. Other research (e.g., Hess & Tate, 1991, 1992) has provided data that are consistent with such a view. It may be that variations in task requirements influence the extent to which such differences are observed. For example, Hess and Tate's (1991) subjects simultaneously formed an impression while studying the to-be-recalled behaviors. Such demands may limit older adults' capacity to effectively process individual behaviors and result in poorer memory, especially for unexpected (inconsistent) behaviors. In

248

THOMAS M. HESS AND SHARON M. PULLEN Table 7 Experiment 3: Impression Change Ratings for Positive, Negative, and Neutral Behaviors Positive preface Age group

Young M SD Old M SD

Negative preface

Positive

Negative

Neutral

Positive

Negative

Neutral

2.3 1.8

4.5 1.0

1.6 0.8

5.3 1.1

2.5 1.4

3.5 1.2

2.1

4.1 1.6

2.1

4.3 1.5

2.3 1.2

2.8 1.2

1.7

1.6

Note. Ratings are on a scale from 1 to 7, with 7 indicating great change.

the present research, and in other cases where age differences in consistency effects have not been observed (e.g., Hess et al., 1987), subjects had already formed an impression of the target prior to presentation of the to-be-remembered behaviors. The separation of tasks may disproportionately benefit memory operations in older adults, who tend to be less effective than young adults when their attention is divided (e.g., Kausler & Puckett, 1981). Indeed, research with young adults has shown that the recall advantage for inconsistent behaviors is eliminated when working memory capacity is occupied (e.g., Srull & Wyer, 1983). More research is obviously needed if we are to discover the reasons behind inconsistencies in findings across studies. The results of our research do, however, provide support for the notion that age differences in knowledge-based factors can influence cognitive performance. Although the effects are not especially strong, consistently observed performance patterns across studies suggest that young and older adults weight different types of information in making decisions and that these variations affect the types of judgments they make about people. These age-related informational biases also appear to affect the processing and subsequent accessibility of information in memory. In Experiments 2 and 3, the older adults' negativity bias was evidenced by an elimination of the typically observed consistency effect in memory; that is, there were disproportionately high levels of recall for negative information independent of its consistency with previously presented information. This suggests that the nature of negative information resulted in more extensive processing and greater elaboration within the memory representation. It is quite clear that these effects are not attributable to simple age differences in basic cognitive skills. If factors such as age-related difficulties in integration processes were operative in differentiating the performances of older and young adults, then similar patterns of performance would have been expected regardless of the stimulus materials. These were not observed. It is also clear that the effects are not due to age differences in perceptions of the negativity or masculinity of individual traits or behaviors because we controlled for these factors in the stimulus materials. Instead, a more reasonable explanation may be based in between-groups variations in implicit theories associated with the stability and controllability of traits and the causes of behavior. We assume that specific beliefs associated with implicit theories of personality operate to influence the" perceptions pf specific behaviors and the manner in which they are processed. There is evidence that beliefs about both people and the nature and

developmental trajectory of traits do vary with age (e.g., Heckhausen & Baltes, 1991; Heckhausen et al., 1989; Luszcz & Fitzgerald, 1986). Our results suggest that older adults have a somewhat stronger belief than do young adults in the basic stability of negative personality characteristics and, perhaps, masculinity. Thus, when presented with information that is indicative of either of these traits, older adults appear to attach more significance to it than younger adults and to weight such information more in determining their judgments about people. They also seem to be more willing than younger adults to discount behaviors that are inconsistent with these traits. We have fairly good evidence regarding the hypothesis that older adults weight negative information more than do younger adults. First, the pattern of performance on ratings and memory varied with age across preface conditions, which indicates that the bias in impression change in the older adults was not just one toward ensuring consistency. If this had been the case, the degree of impression change, and the form of the associated age effects, would have been identical across the two types of preface. Second, subjective ratings indicated that older adults were less likely than younger adults to change their negative impressions when given additional positive information. Although this evidence was not particularly strong, it does suggest a linkage between beliefs and on-line impression judgments. As noted before, the age-related variations in performance are also consistent with other research that has examined beliefs about abilities. Heckhausen and Baltes (1991) observed a positive correlation between beliefs concerning the desirability of traits and beliefs about their controllability. Specifically, they found that most adults believe that positive characteristics are more within their ability to change than are more negative traits; this apparently reflects an adaptive attributional device that gives the individual credit for positive things in his or her life and thereby allows for the possibility of positive change. Negative attributes, however, are seen as beyond the individual's control, a view that serves to protect feelings of self-worth. An interesting qualification to this pattern of performance, also observed by Heckhausen and Baltes, was that the correlation between desirability beliefs and controllability beliefs was significantly stronger for middle-aged and older adults than it was for young adults. Heckhausen and Baltes argued that this difference may also be adaptive in that young adults' viewing both positive and negative things as controllable might provide a more optimistic view of the future regarding paths of development.

249

IMPRESSION CHANGE AND AGING

In suggesting that positive behaviors are within one's control whereas negative behaviors reflect more basic, uncontrollable characteristics, this pattern of beliefs is generally consistent with the negativity effect observed here and elsewhere. Our finding that older adults give more weight to negative information in changing their impressions than do young adults suggests a somewhat stronger belief among older adults in the more basic, uncontrollable nature of negative information, a result also consistent with the belief patterns observed by Heckhausen and Baltes (1991). We have less evidence for age differences in beliefs regarding the stability or controllability of sex-role-related traits because a feminine preface condition was not included in Experiment 1. However, findings from Hess et al. (1987, Experiment 2) are also consistent with the notion that older adults view genderrelated traits as relatively stable. Specifically, in spite of no between-groups variations in the accessibility of inconsistent behavioral information in memory, older adults were observed to maintain more stereotyped impressions of a target than were young adults after presentation of this inconsistent information. This suggests that subjects in the two age groups were weighting the information in memory differently when making impression judgments. In conclusion, we have demonstrated the importance of considering knowledge-based factors in understanding age differences in cognitive performance on an impression-formation task. In particular, this research suggests that age-related variations in belief systems regarding personality traits will influence the processing of behavioral information, the accessibility of such information in memory, and the types of decisions that are made about this information. The results argue for the importance of examining alternative mechanisms and multiple determinants of performance in the study of cognition in adulthood. Indeed, by examining the results of the three experiments reported here, one can see not only the importance of considering knowledge-based factors in studying cognitive skills but also how ignoring such factors might lead to faulty conclusions about age differences. For example, if only the negative preface conditions had been considered in Experiments 2 and 3, we might have come to the conclusion that older adults are less likely to remember inconsistent information (i.e., they exhibited no consistency effect in recall) and are less likely to change their impressions when presented with such information. Only by systematically manipulating the knowledge base accessed were we able to achieve our present understanding of the data. The results of the present experiments are consistent with a growing number of findings, in both experimental and applied settings, in which different-aged adults make different interpretations of the social world (e.g., Avolio & Waldman, 1989; Blanchard-Fields & Robinson, 1987; J. Erber, 1989). These interpretations appear to reflect underlying belief systems about people and the social world. More systematic research in this area is warranted if we are to identify the bases for the age-related changes in knowledge. For example, are they associated with general developmental changes in cognition or beliefs about the world, or are they context-specific variations associated with particular role transitions? In addition, more work is needed if we are to understand the impact of such knowledge variations not only on the supportive-organizational aspects associated with memory but also on people's goals, interpretation

of events, and motivation to engage in specific cognitive activities. References Avolio, B. J., & Waldman, D. A. (1989). Ratings of managerial skill requirements: comparison of age- and job-related factors. Psychology and Aging, 4, 464-470. Bargh, J. A., & Thein, R. D. (1985). Individual construct accessibility, person memory, and the recall-judgment link: The case of information overload. Journal of Personality and Social Psychology, 49, 1129-1146. Bern, S. L. (1974). The measurement of psychological androgyny. Journal of Consulting and Clinical Psychology, 47, 155-167. Blanchard-Fields, F. (1986). Attributional processes in adult development. Educational Gerontology, 12, 291-300. Blanchard-Fields, F, & Robinson, S. L. (1987). Age differences in the relation between controllability and coping. Journal of Gerontology, 42,497-501. Broverman, I. K., Broverman, D., Clarkson, F. E., Rosenkrantz, P. S., & Vogel, S. R. (1970). Sex-role stereotypes and clinical judgments of mental health. Journal of Consulting and Clinical Psychology, 34, 1-7. Erber, J. T. (1989). Young and older adults' appraisal of memory failures in young and older target persons. Journals of Gerontology: Psychological Sciences, 44, P170-P175. Erber, R., & Fiske, S. T. (1984). Outcome dependency and attention to inconsistent information. Journal of Personality and Social Psychology, 47, 709-726. Fiske, S. T, & Neuberg, S. L. (1990). A continuum of impression formation, from category-based to individuating processes: Influences of information and motivation on attention and interpretation. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 23, pp. 1-74). San Diego, CA: Academic Press. Gordon, S. E., & Wyer, R. S., Jr. (1987). Person memory: Category-setsize effects on the recall of a person's behaviors. Journal of Personality and Social Psychology, 53, 648-662. Hastie, R. (1984). Causes and effects of causal attribution. Journal of Personality and Social Psychology, 46, 44-56. Heckhausen, J., & Baltes, P. B. (1991). Perceived controllability of expected psychological change across adulthood and old age. Journals of Gerontology: Psychological Sciences, 46, P165-P173. Heckhausen, J., Dixon, R. A., & Baltes, P. B. (1989). Gains and losses in development throughout adulthood as perceived by different adult age groups. Developmental Psychology, 25, 109-121. Hertzog, C. K., & Rovine, M. (1985). Repeated-measures analysis of variance in developmental research: Selected issues. Child Development, 56, 787-809. Hess, T. M. (in press). Social cognition in adulthood: Aging-related changes in knowledge and processing mechanisms. Developmental Review. Hess, T. M., Donley, J., & Vandermaas, M. O. (1989). Aging-related changes in the processing and retention of script information. Experimental Aging Research, 15, 89-96. Hess, T. M., Follett, K. J., & Perry, P. M. (1992, November). Adult age differences in perceptions of trait organization. Paper presented at the annual meeting of the Gerontological Society of America, Washington, DC. Hess, T. M., & Tate, C. S. (1991). Adult age differences in memory and explanations for behavioral information. Psychology and Aging, 6, 86-92. Hess, T. M., & Tate, C. S. (1992). Direct and indirect assessments of memory for script-based narratives in young and older adults. Cognitive Development, 7,467-484. Hess, T. M., Vandermaas, M. O., Donley, J., & Snyder, S. S. (1987).

250

THOMAS M. HESS AND SHARON M. PULLEN

Memory for sex-role consistent and inconsistent actions in young and old adults. Journal of Gerontology, 42, 505-511. Hyde, J. S., & Phillis, D. E. (1979). Androgyny across the life span. Developmental Psychology, 15, 334-336. Kausler, D. H., & Puckett, J. M. (1981). Adult age differences in memory for sex of voice. Journal of Gerontology, 36, 44-50. Luszcz, M. A., & Fitzgerald, K. M. (1986). Understanding cohortdifferences in cross-generational, self, and peer perceptions. Journal of Gerontology, 41, 234-240. Rosch, E. (1978). Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition and categorization (pp. 27-48). Hillsdale, NJ: Erlbaum. Salthouse, T. A., & Babcock, R. L. (1991). Decomposing adult age differences in working memory. Developmental Psychology, 27, 763776. Skowronski, J. J., & Carlston, D. E. (1989). Negativity and extremity biases in impression formation: A review of explanations. Psychological Bulletin, 105, 131-142.

Smith, D. A., & Graesser, A. C. (1981). Memory for actions in a scripted activity as a function of typicality, retention interval, and retrieval task. Memory and Cognition, 9, 550-559. Srull, T. K.., & Wyer, R. S., Jr. (1983). The role of control processes and structural constraints in models of memory and social judgment. Journal of Experimental Social Psychology, 19, 497-521. Srull, T. K., & Wyer, R. S., Jr. (1989). Person memory and judgment. Psychological Review, 96, 58-83. Wechsler, D. (1981). Wechsler Adult Intelligence Scale—revised. New \brk: Psychological Corporation. Windle, M., & Sinnott, J. D. (1985). Psychometric study of the Bern Sex-Role Inventory with an older adult sample. Journal of Gerontology, 40, 336-343. Wyer, R. S., Jr., & Srull, T. K. (1989). Memory and cognition in Us social context. Hillsdale, NJ: Erlbaum.

Received March 5, 1993 Revision received September 20, 1993 Accepted September 22, 1993 •

P&C Board Appoints Editor for New Journal: Journal of Experimental Psychology: Applied In 1995, APA will begin publishing a new journal, the Journal of Experimental Psychology: Applied. Raymond S. Nickerson, PhD, has been appointed as editor. Starting immediately, manuscripts should be submitted to Raymond S. Nickerson, PhD Editor, JEP: Applied Department of Psychology Tufts University Medford, MA 02155 The Journal of Experimental Psychology: Applied will publish original empirical investigations in experimental psychology that bridge practically oriented problems and psychological theory. The journal also will publish research aimed at developing and testing of models of cognitive processing or behavior in applied situations, including laboratory and field settings. Review articles will be considered for publication if they contribute significantly to important topics within applied experimental psychology. Areas of interest include applications of perception, attention, decision making, reasoning, information processing, learning, and performance. Settings may be industrial (such as human-computer interface design), academic (such as intelligent computer-aided instruction), or consumer oriented (such as applications of text comprehension theory to the development or evaluation of product instructions).