The goal of most health-promotion campaigns is to

Transcription

1 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN Sherman et al. / HEALTH MES SAGES AND SELF-AF FIR MA TION Do Messages About Health Risks Threaten the Self? Increasing the Ac cep tance of Threatening Health Messages Via Self-Affir ma tion Da vid A. K. Sherman Leif D. Nelson Claude M. Steele Stanford University Two stud ies demonstrate that self-image maintenance pro cesses affect the acceptance of per son ally rel e vant health mes sages. Participants who completed a self-affir ma tion were less defensive and more accept ing of health infor ma tion. In Study 1, female par tic i pants (high vs. low rel e vance) read an arti cle link ing caf - feine con sump tion to breast can cer. High-rel e vance women rejected the information more than did low-relevance women; how ever, affirmed high-relevance women accepted the informa - tion and intended to change their behavior accordingly. In Study 2, sexually active participants viewed an AIDS educa - tional video; affirmed par tic i pants saw themselves at greater risk for HIV and pur chased con doms more often than did nonaffirmed participants. Results suggest that health mes sages can threaten an individual s self-image and that self-affirming techniques can increase the effec tive ness of health infor ma tion and lead to pos i tive health behav iors. man ner that may pre vent them from accept ing their risk for disease (Ditto & Lopez, 1992; Weinstein & Klein, 1995). If recip i ents of a health mes sage fail to accept the information, then they will be unlikely to change their risky behaviors. This problem is com pounded because health messages are often the least per sua sive among the indi vid u als for whom the issue is of high personal rele - vance ( Jemmott, Ditto, & Croyle, 1986; Kunda, 1987; Liberman & Chaiken, 1992; Morris & Swann, 1996). In try ing to under stand why per son ally rel e vant health messages pro duce defen sive ness and resistance to acceptance, this arti cle builds on theories of self-evalu - a tion (e.g., E. Aronson, 1969; Solomon, Greenberg, & Pyszczynski, 1991; Tesser & Cornell, 1991) and, in partic - ular, self-affirmation the ory ( J. Aronson, Cohen, & Nail, 1999; Steele, 1988). Per sonally relevant health commu - nications may link behaviors impor tant to an indi vid ual s The goal of most health-promotion campaigns is to change people s behav ior (Salovey, Rothman, & Rodin, 1998; Taylor, 1990). Whether the campaign aims at encour ag ing people to quit smoking, reduce caf feine intake, or prac tice safer sex, the goal is to change behav - ior that increases risk for disease or illness. To moti vate behavior change, many health campaigns attempt to increase perceptions of risk through information and appeals about the dis ease. For exam ple, AIDS education campaigns aimed at col lege stu dents (e.g., Fisher, Fisher, Misovich, Kimble, & Malloy, 1996) deliver information about AIDS and describe how stu dents have been infected as a means of increas ing perceptions of risk and promoting safer sexual behav iors. How ever, research in social psychology reveals a poten - tial obstacle for these health campaigns. Individu als often are moti vated to inter pret infor ma tion in a self-serving 1046 Au thors Note: We would like to thank Heejung S. Kim, Devra Kupor, Geoffrey Co hen, and Kelli Keough for their feedback through out the project and Eric D. Knowles, Emily Pronin, and David Fetherstonhaugh for commenting on earlier versions of this article. We would also like to thank Lauren Puryear and Edward Carstensen for serv ing as experi - ment ers in this research, Shelly Chaiken for providing the ma te ri als for Study 1, and Jeffrey Fisher for pro vid ing the AIDS ed u ca tional video used in Study 2. This re search was funded by a Stan ford University Grad u ate Re search Op por tu nities Grant and a Stan ford Uni ver sity Un der grad u ate Re search Op por tu nities Grant. David A. K. Sherman was sup ported by a Na tional Sci ence Foun da tion Predoctoral Fel low - ship while con duct ing this re search. Por tions of this ar ti cle were pre - sented at the 1998 West ern Psy cho log i cal As so ci a tion con fer ence in Al buquerque, New Mexico. Leif D. Nelson now at tends Princeton University. Cor re spon dence con cern ing this ar ti cle should be ad dressed to Da vid A. K. Sherman, who is now at the De part ment of Psy chol ogy, Uni ver sity of Cal i for nia, Los An geles, 1282A Franz Hall, 405 Hilgard Ave., Los An geles, CA ; sherman@psych.ucla.edu. PSPB, Vol. 26 No. 9, September by the Society for Personality and Social Psychology, Inc.

2 Sherman et al. / HEALTH MESSAGES AND SELF-AFFIRMATION 1047 self-image to disease and, hence, threaten that selfimage. An individual s self-image, con sist ing of impor - tant rela tion ships, val ues, expe ri ences, and behaviors, is an impor tant regulator of the individual s motivational and behav ioral systems (Baumeister, 1996; Steele, 1988). For individuals who seek to pre serve a self-image as posi - tive, moral, and adap tive, the self-image-threat en ing nature of these health messages may arouse defen sive - ness. This defen sive ness, in turn, may prompt indi vid u - als to attempt to restore their self-image by denying that they are at risk and in need of modifying their behav ior. This arti cle sug gests a strategy for reducing defen sive - ness and increas ing the accep tance of health messages. If defensiveness is prompted by threatening an impor - tant part of one s self-image (Steele, 1988; Tesser & Cornell, 1991), then affirm ing the self-image should decrease the defensive responses to the health informa - tion. Indeed, the research presented in this article sug - gests that affirm ing one s self-image by making salient impor tant val ues could increase the accep tance of threat en ing health mes sages and motivate participants to engage in pos i tive health behav iors. RELEVANCE AND DEFENSIVENESS Research link ing per sonal rel e vance to increased defensiveness and reduced acceptance of health mes sages has a long history. Early research ( Janis & Terwilliger, 1962) found that heavy smokers, relative to light smokers or those who did not smoke, were more likely to reject a health communication that linked smoking to cancer (also see Berkowitz & Cottingham, 1960). Personal rele - vance was associated with defensive ness and reduced accep tance of health com mu ni ca tions. How ever, in these early stud ies, high- and low-rel e - vance groups dif fered greatly in prior information and beliefs. Hence, more recent research has used novel issues to study the rela tion ship between rel e vance and defensive processing. Jemmott et al. (1986) created a fic - titious enzyme defi ciency and manip u lated whether par - ticipants believed they possessed the defi ciency. Those for whom the enzyme defi ciency was of high per sonal relevance rated the deficiency as less seri ous than did those for whom the deficiency was of low personal rele - vance. Kunda (1987) had heavy and light caffeine-con - sum ing participants respond to an article documenting the link between caffeine con sump tion and fibrocystic disease (a precursor to breast can cer). High-relevance individuals, that is, women (a group at risk for breast can - cer) who consumed heavy amounts of caffeine, believed the link between caffeine and fibrocystic disease less than did women who con sumed light amounts of caf - feine. Men (a group not at risk for breast cancer), who had the same prior beliefs as women, showed no such dif - ference between heavy and light caffeine consumers. Kunda (1987) argues that heavy caffeine-consuming women, who are moti vated to disbelieve that they are at risk for cancer, eval u ated information linking caf feine to can cer in a self-serving way. Liberman and Chaiken (1992) rep li cated the finding that a high-relevance group (female coffee-drinkers) was less accepting of an arti cle link ing caf feine consumption to breast cancer than was a low-relevance group (female non-coffee-drinkers). Moreover, Liberman and Chaiken (1992) found differences in cognitive processing between high- and low-relevance groups. High-relevance individ - ua ls exhib ited biased sys tem atic pro cess ing of the arti cle; they were more crit i cal of aspects of the arti cle linking caffeine to breast cancer than were low-relevance indi vid u als. The research sug gest ing that per sonal rel e vance increases defensiveness and reduces the accep tance of important health messages seems to con tra dict studies sug gest ing that increased personal rel e vance or issue involvement leads to greater per sua sion for high-quality messages (Petty & Cacioppo, 1979, 1986). How ever, drawing on the John son and Eagly (1989) distinction between value-relevant involvement (or, in our terms, self-image relevant) (Steele, 1988) and outcome-relevant involve ment sug gests that increased value rel e vance should lead to decreased persuasion (see also Eagly & Chaiken, 1993). When the issue is eval u at ing whether one s behaviors are associated with an increased risk for cancer, the self-image is relevant; consequently, biased defensive pro cess ing and reduced per sua sion will result (Giner-Sorolla & Chaiken, 1997). MOTIVATIONAL APPROACHES TO DEFENSIVENESS The research reviewed above (e.g., Giner-Sorolla & Chaiken, 1997; Liberman & Chaiken, 1992) has exam - ined the cog ni tive mechanisms in defensive responses to health information. However, cognitive strategies have been largely inef fec tive at changing defen sive responses to health information (e.g., Weinstein & Klein, 1995). In contrast, motivational strat e gies have been effec tive at reducing defen sive ness and increas ing accep tance of health messages. Research from the cog ni tive dis so nance perspective (e.g., Stone, Aronson, Crain, Winslow, & Fried, 1994) has used the self-threatening nature of AIDS in studies where college students were made mindful of their fail - ure to use condoms after pub licly advo cat ing safer sex. The dis so nance aroused by being hyp o crit i cal moti vated students to resolve their inconsistency by purchasing condoms. These find ings dem on strate that understand - ing the motivational fac tors leading to defen sive ness can result in positive health behav iors.

3 1048 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN Motivational strategies also have reduced vulnerabil - ity-deny ing defensive distortions. One study had partici - pants take a bogus test of their emotionality; experiment - ers told half of the participants that emotionality leads to early death and half that emo tion al ity leads to long life (Greenberg et al., 1993). Informing people that emo - tionality leads to early death attenuated reports of emo - tionality. However, this defen sive ten dency was reduced by a boost to self-esteem in the form of positive personal - ity feed back. tude in a dissonance-arousing, forced-compliance para - digm (Steele & Liu, 1983). Other studies have found that a self-image-affirming activity reduces defensive evalua - tion of stereotyped group members (Fein & Spencer, 1997), the defense of a strongly held belief in the face of opposing arguments (Cohen, Aronson, & Steele, 2000), and the biased processing of threat en ing health infor - mation (Reed & Aspinwall, 1998). In this article, we examine a spe cific pos tu late of self-affirmation the ory (Steele, 1988), namely, SELF-AFFIRMATION AND DE FEN SIVE NESS Research has shown that when an indi vid ual s selfimage is threat ened, the per son is more likely to respond in a self-serving way (Ditto & Lopez, 1992; Dunning, Leuenberger, & Sherman, 1995; Kunda, 1990). Many health messages simul ta neously threaten the self and pres ent impor tant infor ma tion. Thus, an analysis of the self-sys tem is cen tral to under stand ing why threat en ing health information prompts defen sive responses. Self-affirmation theory ( J. Aronson et al., 1999; Steele, 1988) pro poses that thoughts and actions are moti vated by a desire to maintain a self-image as moral, adap tive, and capa ble. 1 When people receive threat en ing health infor ma tion (Ditto & Lopez, 1992; Jemmott et al., 1986; Kunda, 1987), they respond defensively as a means of main tain ing their pos i tive self-image. However, self-affir - mation the ory predicts that if one s self-image can be affirmed through some other means, the need to respond defen sively to the threat en ing infor ma tion should be reduced. One study (Trope & Pomerantz, 1998) found that a successful experience with an unre - lated task increased interest in diag nos tic but poten tially threat en ing feedback, providing evi dence that positive expe ri ences bol ster the self and can prepare the individ - ual to confront neg a tive information. Other research finds that opti mis tic beliefs predict atten tion to health threats (Aspinwall & Brunhart, 1996) and that self-affirming activities could reduce stress and ill ness (Keough, Gar cia, & Steele, 1998), suggesting that cumu - lative pos i tive expe ri ences and affirming activities may buffer the self and help the indi vid ual confront potential health risks. Viewed in this light, the positive personality feedback pro vided to participants in the Greenberg et al. (1993) study reviewed above may have affirmed the self-image of par tic i pants in response to the threat of the dis ease. This affirmation, in turn, may have reduced the need to respond defen sively to the threatening health infor ma tion. This inter pre ta tion is plau si ble because self-affir ma - tions have reduced defen sive pro cess ing in many domains. Previous research has shown that affirm ing an important value reduces the need to change one s atti - Salient, self-affirming thoughts should make it eas ier to be objective about other, self-threatening information; they should reduce the pressure to diminish the threat inherent in this information. In this way, self-affirming thoughts may be an effective means of reducing thought distorting defense mechanisms such as denial and ratio - nalization. (p. 290) The the ory pro poses that affirming a cen tral value could reduce defensive responses to threatening health messages. Threatening health messages put an individ - ual s self-image at risk; the finding that high-relevance women are less accept ing than are low-rel e vance women of an arti cle link ing caffeine consumption to breast can - cer (Kunda, 1987; Liberman & Chaiken, 1992) is the result, we argue, of the self-image-threatening nature of this health information. Our hypothesis is that providing an alter na tive means to reduce the threat in this infor - mation via a self-affirmation should reduce the defen - sive ness of the relevant mes sage recip i ents and increase their acceptance of the mes sage. A more accepted mes - sage will then lead to greater perceptions of personal risk and greater movement toward positive health behav iors. OVERVIEW OF STUDIES The pres ent stud ies address whether self-image main - tenance pro cesses affect the defen sive pro cess ing of health messages by pro vid ing an opportunity to affirm the self as a means of reduc ing this defen sive ness. In Study 1, we use the design employed by Liberman and Chaiken (1992) to test whether affirmed cof fee-drink ers are more accept ing of the con tent of a high-threat mes - sage than are nonaffirmed coffee-drinkers. Study 2, in the domain of AIDS edu ca tion, extends this research from self-report attitudes to health-related behaviors. We test whether sexually active col lege stu dents who receive a self-affirmation prior to see ing an AIDS educa - tional video will perceive them selves as poten tially at greater risk for AIDS and be more likely to engage in safer sex ual behaviors such as purchasing condoms and obtaining AIDS educational brochures.

4 Sherman et al. / HEALTH MESSAGES AND SELF-AFFIRMATION 1049 STUDY 1 In Study 1, we presented high-rel e vance (i.e., cof feedrinking) and low-rel e vance (i.e., non-coffee-drinking) women with an article linking caf feine con sump tion to fibrocystic disease. After read ing the article, participants in the affirmation con di tion affirmed a cen tral value by filling out a val ues scale of their highest-ranked value, whereas those in the no-affirmation con trol condition filled out a values scale of their fifth-ranked value. All par tic i pants then completed mea sures assessing their accep tance of the arti cle and inten tions to change behavior. Following Kunda (1987) and Liberman and Chaiken (1992), we predicted that in the no-affirmation con di tion, high-rel e vance par tic i pants (the cof feedrinking women) would be more defensive and less accepting of the health infor ma tion than would the low-rel e vance par tic i pants (the non-cof fee-drink ing women). How ever, we predict that this ten dency would be reversed in the affirmation condition, with high-rele - vance par tic i pants accept ing the health information to a greater extent than would low-relevance participants. More over, we pre dict that this greater accep tance among the affirmed high-rel e vance par tic i pants would lead to greater inten tion to reduce caffeine intake. METHOD Par tic i pants and Design Sixty female stu dents at Stanford University par tic i - pated in the 30-minute exper i ment in exchange for $5. Relevance was determined prior to the experiment; as part of a preselection questionnaire, par tic i pants identi - fied themselves as either cof fee-drink ers (n = 28) or non-cof fee-drink ers (n = 32). Par tic i pants were ran - domly assigned to either the affirmation or no-affirma - tion condition. Thus, the exper i ment con sisted of a 2 (rel e vance: cof fee-drinker vs. non-coffee-drinker) 2 (affir ma tion sta tus: affir ma tion vs. no-affir ma tion) between-participants factorial design. Procedure Experimental ses sions were run in small groups of 2 to 5 by an experimenter who was unaware of participants cof fee-drink ing sta tus. The exper i menter instructed par tic i pants that they would be com plet ing two stud - ies, one mea sur ing per sonal val ues and the other attempt ing to assess stu dent opin ions of sci en tific arti - cles. After sign ing the con sent form for the per sonal val ues exper i ment, par tic i pants ranked a list of five per sonal val ues (social, polit i cal, reli gious, the o ret i - cal, aes thetic) (Allport, Vernon, & Lindzey, 1960) in order of importance. The caffeine-cancer article. Participants then signed the consent form for the scientific article experiment. Par - ticipants read the article, Caf feine and Women: A New Health Risk, which appeared to be photocopied from the Health Today News let ter. The content of the article was the same as the high-threat article in Liberman and Chaiken (1992). The article describes fibrocystic dis - ease, a precursor to breast can cer, and research that links caffeine intake with the dis ease. The article concludes by stating that caf feine poses a significant risk of breast can - cer for women. Affir ma tion manipulation. While the participants read the arti cle, the experimenter ran domly assigned them to either the self-affir ma tion con di tion or the no-affirma - tion condition. After read ing the article, participants com pleted the affir ma tion manipulation. The experi - menter, based on the participants rank ings of values, gave affirmation participants a scale concerning their high est-ranked value and no-affirmation participants a scale concerning their fifth (i.e., low est-ranked) value. The value scales, adapted from Allport et al. (1960) have been employed in other self-affirmation studies (Steele & Liu, 1983; Tesser & Cornell, 1991) and are the o rized to affirm the self by mak ing salient values that are cen tral to the indi vid ual s self-image (Steele, 1988). The value scales consist of 10 pairs of state ments. For participants in the affir ma tion condition, one statement of each pair was asso ci ated with their most impor tant value and the other state ment was filler. For par tic i pants in the no-affirmation condition, one statement was asso ci ated with their least impor tant value and the other statement was filler. Par tic i pants assigned 1 to 4 points to each state - ment, with greater points indi cat ing greater agreement with the statement. Dependent mea sures. Next, par tic i pants completed the depend ent measures. Two questions exam ined whether par tic i pants accepted the conclusions of the article, ask - ing on 9-point scales, To what extent do you agree or dis - agree that there is an association between caffeine and fibrocystic disease? and How impor tant do YOU think it is that women reduce their caffeine intake in order to avoid fibrocystic disease? Two other ques tions assessed par tic i pants intent to reduce caffeine consumption, asking on 9-point scales, To what extent do you think that you, personally, SHOULD reduce the amount of caf - feine you con sume? and To what extent do you think that you, per son ally, will ACTUALLY reduce the amount of caf feine you consume? Next, in an open-ended sec - tion, par tic i pants listed any thoughts they had while reading the article. Finally, participants reported their self-feel ings and mood, responding on 9-point scales to the ques tions, How do you feel about yourself? and What is your current mood?

5 1050 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN Debrief ing. After com plet ing the dependent measures, the exper i menter debriefed par tic i pants with a par tic u - lar emphasis on the false con tent of the arti cle and pro - vided a brief summary of research findings describing the relationship between caffeine and breast cancer. 2 RESULTS Checks on the Ex per i men tal Ma nip u la tion To exam ine the effect of the self-affirmation on selffeelings, participants responded to the ques tion, How do you feel about your self? on a 9-point scale anchored at poorly and extremely pos i tively. Affirmed participants (M = 6.94) felt better about them selves than did nonaffirmed par tic i pants (M = 6.07), F(1, 59) = 4.64, p <.05. Participants also responded to the ques tion, How would you describe your mood, right now? on a 9-point scale anchored at extremely bad mood and extremely good mood. This ques tion was uncorrelated with the self-feel - ing ques tion (r =.08, ns), indicating that it measured a con cep tu ally dis tinct state. There was no dif fer ence between affirmed par tic i pants (M = 5.84) and nonaffirmed par tic i pants (M = 6.10), F(1, 59) < 1.00, ns, sug gest ing that mood was not affected by the manip u la tion. For the values scale, the affir ma tion manip u la tion, the participants assigned 1 to 4 points to each of 10 pairs of values. The scale ranged from 10 to 40 points, with more points indi cat ing greater agree ment with the val - ues. As expected, affir ma tion participants assigned more points (M = 29.26) to their most impor tant value than the no-affir ma tion par tic i pants, who assigned fewer points (M = 23.10) to their least impor tant value, F(1, 59) = 42.31, p <.001. Both of these means dif fered from chance (25) in the pre dicted directions, t(30) = 6.79, p <.001, and t(28) = 2.69, p <.01, for the affirmation and no-affirmation conditions, respectively. Af fir ma tion and Ac cep tance of the Threat en ing In for ma tion Our hypothesis was that affirming a cen tral value would reduce defen sive ness and increase accep tance of the health infor ma tion. We measured acceptance by eval u at ing participant responses to the arti cle linking caffeine and fibrocystic disease. Par tic i pants rated To what extent do you agree or dis agree that there is an asso - ciation between caf feine con sump tion and fibrocystic disease on a 9-point scale anchored at strongly disagree and strongly agree and How important do YOU think it is that women reduce their caf feine intake in order to avoid fibrocystic disease? on a 9-point scale anchored at not at all impor tant and very important. These measures had high reliability (Cronbach s α =.71) and were aver - aged and ana lyzed as a single mea sure of acceptance of the arti cle s con clu sions, with higher num bers indicat - ing greater acceptance. A two-way ANOVA revealed that affirmed participants (M = 6.50) were more accept ing of the article s conclusions than were nonaffirmed partici - pants (M = 5.29), F(1, 59) = 8.04, p <.01. This main effect was qual i fied by the pre dicted Affir ma tion Relevance inter ac tion, F(1, 59) = 16.44, p <.001. As Fig ure 1 shows, within the no-affirmation condition, coffee-drinkers (M = 4.15) were less accept ing of the conclusions than were non-cof fee-drink ers (M = 6.22), F(1, 28) = 10.77, p <.01. How ever, in the affir ma tion condition, cof fee-drink ers (M = 7.23) were more accepting of the article s con clu - sions than were non-coffee-drinkers (M = 5.81), F(1, 30) = 5.85, p <.05. Exam ining the results some what differently shows that the effects of the affir ma tion were most beneficial to the relevant par tic i pants. For the non-coffee-drinkers, there was no difference between the affir ma tion condi - tion (M = 5.81) and the no-affir ma tion condition (M = 6.22), F(1, 31) < 1.00, ns. How ever, for the cof fee-drink - ers, the affir ma tion had clear ben e fi cial effects. Affirmed coffee-drinkers were much more accept ing of the arti - cle s conclusions (M = 7.23) than were nonaffirmed cof - fee-drink ers (M = 4.15), F(1, 27) = 33.26, p <.001 (see Fig - ure 1). Thus, for the cof fee-drink ing women, the self-affirmation reduced the defensive pro cess ing of the threat en ing message and persuaded them to accept that they should reduce their caf feine intake. Behavioral Intentions Would this greater acceptance of the conclusions of the arti cle by the affirmed cof fee-drink ers result in behav ioral change? Although we did not measure cof - fee-drinking behavior, two questions exam ined behav - ioral intentions among the cof fee-drink ers. 3 Par tic i pants were asked, To what extent do you think that you, per - sonally, SHOULD reduce the amount of caffeine you con sume? on a 9-point scale anchored at to no extent at all and to a very great extent and To what extent do you think that you, per son ally, will ACTUALLY reduce the amount of caffeine you con sume? on a 9-point scale anchored at to no extent at all and to a very great extent. These two items had high reli abil ity (Cronbach s α =.83) and were analyzed as a com pos ite measure, with higher numbers indi cat ing greater pre dicted reduc tion of caf - feine con sump tion. This combined mea sure was sub - jected to a one-way ANOVA among the coffee-drinking participants. The affirmed coffee-drinkers (M = 6.10) predicted much greater reduc tion in caf feine consump - tion than did the nonaffirmed cof fee-drink ers (M = 2.73), F(1, 27) = 25.89, p <.001. Thus, not only were the affirmed cof fee-drink ers more accept ing of the message that caf feine intake was related to fibrocystic disease but

6 Sherman et al. / HEALTH MESSAGES AND SELF-AFFIRMATION 1051 tance of the article s conclusions (r =.29, p <.05). Stronger sup port for the effect of the self-affirmation on increas ing the participants acceptance of the arti cle s conclusions comes from results indi cat ing that the rela - tionship between the thought-valence index and accep - tance of the article s conclusions only holds for the affirmed participants. That is, the correlation between the thought-valence index and acceptance of the arti - cle s conclusions is strong only for the affirmed partici - pants (r =.46, p <.01) but not the nonaffirmed partici - pants (r =.09, ns). Thus, for affirmed participants, the accep tance of the arti cle s conclusions is related to the positivity of their issue-rel e vant thoughts. 5 Figure 1 Acceptance of article s con clu sions (+ SE ) as a func tion of relevance (coffee-drinkers vs. non-coffee-drinkers) and self-affirmation condition. they also intended to reduce their caffeine intake as a result. Thought Listing Par tic i pants listed any thoughts rel e vant to the article in a free-response format. Two cod ers unaware of the par tic i pants condition or cof fee-drink ing status coded the thoughts. Coders rated the valence of the com - ments whether they were positive toward the issues raised by the article (e.g., The study raised impor tant issues about breast can cer ) or neg a tive toward the issues raised by the arti cle (e.g., The study was poor, it had no control group ). Interrater reliability was 80% (reli abil - ity was cal cu lated by number of agreements divided by total num ber of agreements and dis agree ments) and dif - ferences were resolved through discussion among the coders. A thought-valence index was created (following Killeya & John son, 1998) using the following formula: thought-valence index = (pos i tive issue-rel e vant thoughts + 1) / (total issue-rel e vant thoughts + 1). (The constant of 1 was added to the numer a tor and the denominator to prevent divisions by zero). Scores on the index approach ing 1.00 indi cate a predominantly posi - tive thought list, whereas scores approach ing 0.00 indi - cate a pre dom i nantly negative thought list. 4 The index was strongly related to both the number of positive thoughts (r =.59, p <.001) and neg a tive thoughts (r =.68, p <.001). Results indicate no main effects of either affirmation condition or coffee-drinking status or inter ac tions on the thought-valence index (scores range from.36 to.44 for the four cells, all Fs < 1.0). However, correlational anal y ses indi cate that the thought-valence index is related to the primary dependent measure, the accep - DISCUSSION Study 1 rep li cated the find ings of Liberman and Chaiken (1992) and Kunda (1987) in that within the no-affir ma tion con trol con di tion, cof fee-drink ing women were much less accept ing of the mes sage linking caffeine consumption to fibrocystic disease than were non-coffee-drinking women. However, Study 1 also found that with a self-affirmation, this defensive response pat - tern was not only attenuated but reversed. Affirmed cof - fee-drink ers were more accept ing of the article s conclu - sions than were affirmed non-cof fee-drink ers. Most inter est ingly, affirmed coffee-drinkers were significantly more accepting of the article s conclusions than were nonaffirmed coffee-drinkers, dem on strat ing not only that self-affirmation reduces the defen sive ness among high rel e vance par tic i pants (e.g., Reed & Aspinwall, 1998) but also that self-affirmation exerts a clear benefi - cial effect on increas ing the accep tance of threat en ing health messages. Indeed, affirmation also led affirmed coffee-drinkers to greater behavioral inten tions because they indi cated far greater inten tions to reduce their caf feine con sump tion than did nonaffirmed coffeedrinkers. Exam i na tion of the thought-list ing responses in Study 1 helps clarify the mechanism by which the affirmation reduced the defen sive pro cess ing and led to the increased accep tance of the threat en ing health mes - sages. Although par tic i pants did not dif fer by condition in the positivity of thoughts listed (as assessed by the thought-valence index), among affirmed participants (both cof fee-drink ers and non-coffee-drinkers) there was a high correlation between the thought-valence index and acceptance of the article s conclusions, whereas there was no correlation among the nonaffirmed partici - pants. This sug gests that the nonaffirmed participants were being defensive; relative to the affirmed partici - pants, they were much more reject ing of the arti cle s con clu sions, and this was unassociated with the positivity of the thoughts they generated. In con trast, the affirmed par tic i pants were much more accept ing of the article

7 1052 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN and had a strong cor re la tion between the thoughtvalence index and the accep tance of the arti cle s conclu - sions. This find ing sug gests that the affirmation may have reduced the defen sive ness observed in the nonaffirmed and led people to accept the article to the extent that they generated pos i tive thoughts about it. The strong correlation between the thought-valence index and accep tance of the article s conclusions among the affirmed also sug gests that the self-affir ma tion results are not due to positive mood (a point we will return to in the Gen eral Discussion). If the affirmation results were due to mood, we would expect minimal cor - relation between the thought-valence index and accep - tance of the arti cle s conclusions because regardless of con tent (pos i tive or neg a tive), they would be more accepting of the article. In contrast, we found a strong cor re la tion only among the affirmed participants, sug - gesting that it was not positive mood driving the results. Thus, Study 1 clarifies the mechanism by which affirma - tion leads to greater accep tance of threat en ing health messages, a find ing that extends the process find ings of Reed and Aspinwall (1998), which showed that highrelevance participants who are affirmed orient more quickly to risk-con firm ing infor ma tion. Study 1 found increased per sua sive ness of a health communication when the affirmation was admin is tered after the par tic i pants read the threat en ing infor ma - tion. This sug gests that it was not an on-line pro cess that medi ated the differences found in the con di tions (as in Reed & Aspinwall, 1998) but rather a recon sid er a tion of the evidence at the time the depend ent measures were asked. Study 2 examines whether an affirmation given prior to the threatening information would pro duce anal o gous results. As previous research (Steele & Liu, 1983; Steele, Spencer, & Lynch, 1993) has shown, self-affirmations can be effec tive at reduc ing defensive processing at either stage, oper at ing as either an inocula - tion or a cure against the threatening information. Two other con cerns with Study 1 moti vated our design of Study 2. First, Study 1 used the domain of caf - feine and fibrocystic dis ease, an issue that is probably not as salient to a col lege-age pop u la tion as are other health concerns. In Study 2, we generalize the find ings of Study 1 to a more central concern for our college stu dent partic - i pants AIDS; would a self-affir ma tion increase the accep tance of a poten tially threatening AIDS educa - tional mes sage? Sec ond, although the affir ma tion resulted in increased behav ioral intentions, Study 1 did not mea sure behavior. Thus, Study 2 examines whether a self-affirmation would allay threats provoked by an emo - tional mes sage about AIDS and whether this would trans - late into positive health behaviors. STUDY 2 In Study 1, affirming a central value among rel e vant par tic i pants increased the accep tance of a health com - mu ni ca tion link ing caffeine to breast can cer. Study 2 gen er al izes this finding to the domain of AIDS educa - tion. Col lege stu dents are typ i cally very sexually active, engaging in intercourse with multiple part ners; one sur - vey of col lege stu dents reported that 86% were sexually active, yet only 21% used a con dom every time they had inter course (Caron, Davis, Wynn, & Roberts, 1992). Stu - dents use risky strat e gies, adopt ing implicit personality theories about who is and is not at risk for AIDS to justify their con tin ued risky behavior (Williams et al., 1992). To com bat these beliefs, and change stu dents risky sex ual behaviors, intervention studies have attempted to make stu dents con front their risk for AIDS (Choi & Coates, 1994; J. D. Fisher & Fisher, 1992). Motivational fac tors such as atti tudes toward per for mance of AIDS-preventive behaviors and perceived personal vul - ner a bil ity to HIV are cen tral com po nents of many AIDS-inter ven tion strategies ( J. D. Fisher et al., 1996). For exam ple, the Infor ma tion-moti va tion-behav ioral Skills Model for AIDS intervention ( J. D. Fisher & Fisher, 1992; J. D. Fisher et al., 1996), an inter ven tion successful at mod i fy ing risky sex ual behavior in a col lege sample, has a moti va tion component that includes an educa - tional video depicting attrac tive young adults who had con tracted AIDS through sex ual con tact, describ ing life with the disease. Although this inter ven tion effectively changed behav - ior, the AIDS video was embed ded in a larger program of AIDS edu ca tion. When shown in iso la tion, how ever, these vid eos may have the con se quence of caus ing defensive reac tions. One study (Mor ris & Swann, 1996) showed virgins and nonvirgins an AIDS educational video featuring emotional appeals by young people with AIDS. Whereas the video prompted virgins to perceive them selves as being at greater risk for HIV (rel a tive to nonvideo con trols), the video prompted nonvirgins to perceive them selves as less at risk for HIV. That is, the AIDS educational video prompted a denial response among the most relevant people, a pattern of results similar to the nonaffirmed participants in Study 1. Thus, Study 2 has three pri mary goals. The first goal is to generalize the find ings from Study 1 that pro vid ing an oppor tu nity to affirm the self will increase the effec tive - ness of a poten tially threatening health mes sage. In par - ticular, we pre dict that an affir ma tion will reduce the defen sive responses among sex u ally active stu dents found in Morris and Swann (1996). The sec ond goal is to go beyond the self-report attitude mea sures in Study 1 to examine whether an affir ma tion cou pled with a poten -

8 Sherman et al. / HEALTH MESSAGES AND SELF-AFFIRMATION 1053 tially threatening health message would result in greater AIDS-preventive behav iors. Thus, in Study 2, partici - pants could pur chase con doms and obtain AIDS educa - tional bro chures, impor tant behaviors in the prevention of the dis ease. The third goal is to demonstrate that an affir ma tion could increase the persuasiveness of a threat - en ing health mes sage when the affirmation was adminis - tered before the presentation of the health information. METHOD Par tic i pants and Design Sixty-one Stan ford undergraduate students, 30 men and 31 women, participated in exchange for either course credit or $5. 6 On a pretest survey as part of a packet of ques tion naires, all students indi cated whether they had engaged in sex ual intercourse in their lifetimes. Only stu dents who engaged in sexual intercourse were recruited. 7 As part of the pretest sur vey, all stu dents also responded to the ques tions, Are you con cerned that you have been exposed to the HIV virus sex u ally? and How at risk do you think you are for HIV? The participants wrote an essay on either a central or unimportant value (affirmation manipulation) prior to watching a potentially threatening AIDS educational video. The exper i ment had two conditions: the affirma - tion con di tion and the no-affir ma tion con trol condi - tion. The dependent measures were AIDS-preventive behaviors (buy ing con doms and obtaining brochures), assess ments of the video, and perceptions of per sonal risk for AIDS. Procedure Via , par tic i pants signed up for the study, Eval - u ating AIDS Edu ca tional Materials. After the partici - pant arrived at the study, the experimenter explained that there would be three parts to the experiment. First, par tic i pants would fill out a brief questionnaire and complete a writ ing exer cise. Second, they would view an AIDS edu ca tional video. Third, the participants would answer a series of ques tions about the video as well as them selves. After admin is ter ing the consent form, the exper i menter explained that due to the per sonal nature of this experiment, all responses the participant pro - vided would be identified only by his or her participant number. In addi tion, the experimenter mentioned that the exper i ment may go over the allot ted 30 minutes but that the participant would be paid an addi tional $3 for his or her time. All par tic i pants agreed to this. Par tic i pants then ranked a list of 11 values and per - sonal char ac ter is tics (Harber, 1995) in order of per sonal impor tance. The list included such values or per sonal characteristics as athletics, artistic skills, cre ativ ity, rela - tions with friends/fam ily, spon ta ne ity, and phys i cal attractiveness. 8 Affir ma tion manipulation. After rank ing the values, par tic i pants completed a brief writ ing assignment. Par - ticipants opened a sealed envelope that contained the essay task, which kept the experimenter unaware of con - di tion. The task, adapted from pre vi ous affirmation studies (e.g., Fein & Spencer, 1997), served as the experi - men tal manip u la tion. Par tic i pants were ran domly assigned to one of two conditions. Those in the affirma - tion condition first indicated their most important value and then wrote an essay describ ing why the value was important to them and a time when it had been par tic u - larly important. Participants in the no-affirmation condi - tion first indicated their ninth most impor tant value and then wrote an essay describ ing why the value might be important to the aver age stu dent. Par tic i pants wrote for 5 minutes. AIDS educational video. After com plet ing the essay, par - ticipants viewed the AIDS edu ca tional video. The video, titled People Like Us ( J. D. Fisher, Fisher, & Marks, 1992), was part of the motivational com po nent of the Informa - tion-motivation-behavioral Skills AIDS intervention ( J. D. Fisher et al., 1996). The video, edited to 12 min utes for the pur poses of our experiment, con sisted of six people (four heterosexual women and two homosexual men, ages approx i mately 18 to 30) living with AIDS who described how they contracted the disease and what life has been like since they became aware of their infection. For example, one woman said, Having to tell my roommates was really dif fi cult.... One of my roommates didn t want to live with me anymore. The video concluded with one woman remark ing, No guy or girl, no mat ter how cute, is worth wak ing up with this every day. The video focused on increasing perceptions of personal vulnera - bil ity and emphasized the need to change stu dents AIDS-preventive behav ior and atti tudes. Dependent measures. We tested whether the affir ma tion would increase the effec tive ness of the AIDS educational video through both self-report mea sures assessing per - ceptions of the video and personal risk and behav ioral measures assess ing whether participants pur chased con - doms and took AIDS-related brochures. Par tic i pants rated, on 9-point scales, how sim i lar their own sex ual expe ri ence was to the people in the video and how accurately they thought the video rep re sented the like li hood of HIV infec tion for people like them. Then par tic i pants indi cated their per sonal risk for AIDS responding on a 9-point scale to the ques tion, How at risk are you for contracting HIV? After the ques tion naires, the experimenter paid par - ticipants $3 for the exper i ment going overtime. Fol low - ing the condom-purchasing methodology employed in

9 1054 PER SON AL ITY AND SO CIAL PSY CHOL OGY BUL LE TIN other studies (Stone et al., 1994; Stone, Wiegand, Coo - per, & Aronson, 1997), the par tic i pants filled out a receipt while the exper i menter com pleted some paper - work. The exper i menter told participants that AIDS edu - cators from the health center donated con doms and bro - chures for par tic i pants as part of the exper i ment. Par tic i pants in the exper i ment could buy con doms for the same price that they are sold at the health center, 10 cents each, and could take as many AIDS educational bro chures as they wanted. On the table was a large jar filled with 50 con doms of var i ous brands along with a cup con tain ing coins to make change. In addi tion, 10 copies each of three AIDS edu ca tional bro chures (one on HIV-test ing options, one about the AIDS virus, and one on how to use a condom). When the participants were finished buy ing con doms and/or taking brochures, the exper i menter returned and debriefed the par tic i pants. After the exper i men tal session, the exper i menter counted the num ber of con - doms and bro chures remain ing. RESULTS Pretest Covariate: Con cern About HIV All of the participants were sexually active. However, because being sexually active does not nec es sar ily mean one is at risk for AIDS, we asked participants in the pre - test, Are you con cerned that you have been exposed to the HIV virus sex u ally? on a 9-point scale anchored at not at all con cerned and extremely con cerned. We use this measure as a covariate on the self-report attitude ques - tions to control for pre test levels of con cern about AIDS. Ef fects of Self-Affirmation on Perceptions of the Video The pur pose of the AIDS educational video is to encourage stu dents to recognize their poten tial sim i lar - ity to peo ple with AIDS and the dan ger of risky sexual behavior. Two questions were related to the video: How similar is your sexual expe ri ence to any of the six people in the video? on a 9-point scale anchored at not at all sim - ilar to any person in the video and extremely similar to at least one person in the video and How accu rately do you think the video represented the likelihood of HIV for people like you? on a 9-point scale anchored at inaccurate percep - tion of risk and accurate per cep tion of risk. These measures had high reliability (Cronbach s α =.71) and, therefore, were summed into an index of similar risk, with higher numbers indi cat ing greater per ceived similarity in risk to the people in the video. An ANCOVA with pretest con - cern for HIV as the covariate, and similar risk as the depend ent vari able, revealed a mar ginal effect of gen - der, F(1, 60) = 3.51, p <.10 (the covariate met the assump - tion of homo ge ne ity for regres sion slopes across affirma - tion conditions and gen der). Women felt greater similar risk to people in the video (adjusted M = 6.18) than did men (adjusted M = 5.25). As noted before, the video con - tained four het ero sex ual women and two homo sex ual men; thus, it is reasonable that the (predominantly het - ero sex ual) men in the study saw their sex ual experience as less similar and the risk represented in the video as being less accu rate for peo ple like them. This main effect of gender was qualified by an Affirmation Gen - der inter ac tion, F(1, 60) = 5.61, p <.05; women who com - pleted a self-affir ma tion saw far greater sim i lar risk (adjusted M = 7.17) than did nonaffirmed women (adjusted M = 5.06), whereas men were not affected by the affirmation (adjusted Ms = 5.18 and 5.45 for affirmed and nonaffirmed men, respectively). Thus, for partici - pants who could see them selves as sim i lar to those in the video (female participants), the affir ma tion led to an increased perceived similarity of risk. Ef fects of Self-Af fir ma tion on Perceptions of Risk We pre dicted that the affirmation would reduce the need to respond defen sively in assess ing per sonal risk for AIDS. Par tic i pants responded to the ques tion, How at risk are you for HIV? on a 9-point scale anchored at very min i mal risk and very seri ous risk. After con trol ling for the pre test report of con cern about HIV, results indi cate that affirmed participants saw them selves as being at more seri ous risk for HIV (adjusted M = 4.02) than did nonaffirmed par tic i pants (adjusted M = 2.78), F(1, 60) = 6.62, p <.05. (Again, the covariate met the assumption of homo ge ne ity for regression slopes across the affirma - tion con di tions.) Anal o gous to Study 1, affir ma tion increased perceptions of per sonal risk after a threaten - ing health message. How ever, did the affirmation cou pled with the AIDS edu ca tional video increase the participants perceptions of risk or did the lack of an affirmation cou pled with the video decrease per cep tions of risk? Par tic i pants responded dur ing pre test to the same ques tion of how at risk they were for HIV. A paired t test indicates that affirmed participants increased their perceptions of risk from 3.16 to 3.97, paired t(31) = 2.25, p <.05, whereas nonaffirmed par tic i pants did not dif fer (Ms = 2.48 and 2.71 for pretest and posttest, respectively), t(30) = 0.83, ns. Thus, the affir ma tion cou pled with the AIDS video increased participants perceptions of risk. Did the affirmation coupled with the AIDS edu ca - tional video reduce defensiveness, as we found in Study 1? Based on Study 1, which showed the defen sive responding of high-relevance participants, and the pre - vi ous literature about assessments made about personal risk for AIDS (J. D. Fisher & Misovich, 1990; Mor ris &

10 Sherman et al. / HEALTH MESSAGES AND SELF-AFFIRMATION 1055 Swann, 1996; Williams et al., 1992), it appears that stu - dents are usu ally responding in a defen sive man ner when asked ques tions such as, How at risk are you for HIV? The pur pose of the AIDS video (J. D. Fisher et al., 1996) is to reduce this defen sive ness among stu dents and pro mote willingness to confront the potential link between their risky sex ual behav ior and HIV. For affirmed participants, this goal was achieved; the affir - mation coupled with the video reduced their pretest defensiveness and increased their perceived per sonal risk for HIV. For nonaffirmed participants, this goal was not achieved; they maintained their defensiveness and did not change their personal risk for HIV after seeing the video. Ef fects of Self-Affirmation on AIDS-Preventive Behaviors We hypothesized that affirm ing an impor tant value prior to watch ing a potentially threat en ing AIDS educa - tional video would elicit greater accep tance of the infor - mation in the video, result ing in greater AIDS-preventive behav iors. We exam ined the percentage of participants in each condition who purchased condoms. As shown in Figure 2, 50% of the affirmed participants pur chased condoms, whereas 25% of the nonaffirmed participants pur chased con doms, χ 2 (1, 60) = 4.03, p <.05. Among those who did purchase condoms, the aver age num ber of con doms pur chased did not differ between affirma - tion participants (n = 16, M = 4.88) and no-affir ma tion par tic i pants (n = 7, M = 5.00), F(1, 22) < 1.00, ns. In addi tion to pur chas ing con doms, par tic i pants could also take any of the three AIDS educational bro - chures. Each indi vid ual brochure-taking measure was highly cor re lated with the oth ers (Cronbach s α =.84); therefore, we mea sured the percentage of participants who took at least one AIDS educational brochure. As seen in Figure 3, in the affirmation condition, 78% of the par tic i pants took at least one brochure, whereas in the no-affirmation condition, 54% of the participants took at least one brochure, χ 2 (1, 60) = 4.09, p <.05. Thus, affirming a cen tral value prior to view ing the AIDS edu - ca tional video resulted in greater AIDS-pre ven tive behav iors; a greater per cent age of affirmed participants pur chased con doms and took AIDS educational bro - chures rel a tive to nonaffirmed participants. DISCUSSION Study 2 demonstrated that com plet ing a self-affirm - ing activ ity prior to viewing an AIDS edu ca tional video would increase perceptions of per sonal risk for HIV and affect AIDS-pre ven tive behav iors, such as purchasing condoms and obtain ing AIDS educational brochures. That affirmed female par tic i pants saw them selves as Figure 2 Pro por tion of par tic i pants (+ SE ) purchasing condoms as a func tion of self-af fir ma tion con di tion. Figure 3 Pro por tion of participants (+ SE ) tak ing at least one AIDS educational brochure as a func tion of self-af fir ma tion con di tion. more similar to the peo ple in the video in terms of their risky behavior, and that affirmed par tic i pants overall saw them selves as more at risk for HIV, sug gests that the affir - mation, coupled with the AIDS video, had a positive effect on pro mot ing par tic i pants awareness of personal risk for AIDS. Moreover, similar to Stone et al. (1994), Study 2 demonstrates that using moti va tional fac tors in conjunction with AIDS educational techniques can pro - duce pos i tive AIDS-preventive behaviors, such as pur - chasing con doms and obtain ing AIDS educational bro - chures. Finally, Study 2 pro vides evi dence that the affir ma tion could lead par tic i pants to accept poten tially threat en ing infor ma tion that they are at risk for HIV that would presumably put them in a worse mood, sug gest ing that affirmation results cannot be accounted for by mood.