Linguistic intergroup bias (LIB) is the tendency to describe positive in-group and negative out-group ... vs. environmentalists (Experiment 1, N = 160) and northern vs. southern Italians ...... The research conducted so far shows great differences.
Copyright 1996 by the American Psychological Association, Inc. 0022-3514/96/S3.00
Journal of Personality and Social Psychology 1996, Vol. 71, No. 3, 512-526
Linguistic Intergroup Bias: Evidence for In-Group-Protective Motivation Anne Maass, Roberta Ceccarelli, and Samantha Rudin University of Padova Linguistic intergroup bias (LIB) is the tendency to describe positive in-group and negative out-group behaviors more abstractly than negative in-group and positive out-group behaviors. Two experiments investigated the role of in-group-protective motives, by varying threat to ingroup identity of hunters vs. environmentalists (Experiment 1, N = 160) and northern vs. southern Italians (Experiment 2, N = 212). Participants whose in-group had or had not been threatened described positive and negative behaviors of in-group and out-group protagonists. In both experiments, the LIB was greater under identity threat. Experiment 1 also showed that LIB was positively related to postexperimental but not to preexperimental individual and collective self-esteem. Results suggest that the magnitude of LIB depends on in-group-protective motivation and that in-group-favoring language may be functional to self-esteem maintenance.
of behaviors, even when referring to a particular behavior in a particular situation. State verbs (SV) such as A hates B, describe enduring psychological states of a participant; although they are not limited to a specific time or place, they do refer to a specific object. Finally, located at the most abstract pole are adjectives (ADJ), such as A is aggressive, that describe a general tendency of the protagonist that generalizes across time, situation, and objects (for an extension of the linguistic category model, see Semin & Fiedler, 1992). Although descriptions at any level of abstraction may be equally adequate, they differ systematically in their psychological implications. Abstract terms such as state verbs or adjectives imply greater temporal and cross-situational stability, provide more information about the protagonist, and suggest a high likelihood that similar acts may be performed in the future (Maass et al., 1989; Semin & Fiedler, 1988, 1992). Consequently, the preference for relatively abstract terms when describing positive in-group and negative out-group behaviors may bolster existing stereotypes. The LIB has consistently been demonstrated in a wide range of controlled experiments. It has been shown to operate in very different intergroup settings, such as competing schools, sport teams, interest groups, and between sexes (Arcuri, Maass, & Portelli, 1993; Fiedler, Semin, & Finkenhauer, 1993; Maass, Milesi, Zabbini, & Stahlberg, 1995; Maass et al., 1989; for a recent overview, see Maass & Arcuri, 1996). The bias has also emerged in content analyses of mass communication regarding political or athletic events (Maass, Corvino, & Arcuri, 1993; Ng &Tait, 1994). Although there is now ample evidence for the existence of LIB, the mechanisms underlying such differential language use in intergroup contexts are less clear. There are at least two competing explanations of this phenomenon: one motivational, one cognitive. From a motivational perspective, one may suspect that the LIB is driven by in-group-protective motives. This hypothesis is loosely based on social identity theory (Abrams & Hogg, 1990; Brewer, 1979;Diehl, 1990; Hinkle & Brown, 1990; Tajfel & Turner, 1979) and hypothesizes that the LIB serves to enhance or protect one's social identity. Assuming that concrete
Recently, there has been increasing interest in the role that language may play in the transmission and maintenance of positive in-group perceptions and negative out-group perceptions (Graumann & Wintermantel, 1989; Hamilton, Gibbons, Stroessner, & Sherman, 1992; Maass & Arcuri, 1992; Maass, Salvi, Arcuri, & Semin, 1989). In particular, the linguistic intergroup bias (LIB) model developed by Maass et al. (1989) identifies a systematic bias in language use that may contribute in a subtle way to the maintenance of stereotypes. The main prediction states that positive behaviors are described in more abstract language terms when performed by an in-group member than when performed by an out-group member. The opposite holds for negative episodes that are described in more concrete terms when performed by an in-group member than when performed by an out-group member. For example, an in-group member offering help to a needy person may be described as "helpful" or "altruistic," whereas an out-group member engaging in exactly the same behavior may be described as "helping." In the case of aggressive behaviors, the in-group member may be described as "hurting somebody," whereas the out-group member may be described as "aggressive." The continuum from abstract to concrete terms is based on Semin and Fiedler's (1988, 1991, 1992) linguistic category model, which in its original version distinguishes four levels of language abstraction. At the most concrete end of the abstraction continuum are descriptive action verbs (DAV), such as A kicks B, which provide an objective description of a specific, observable event; there is at least one physically invariant feature shared by all actions to which the term applies. Interpretative action verbs (IAV) such as A hurts B describe a larger class Anne Maass, Roberta Ceccarelli, and Samantha Rudin, Dipartimento di Psicologia dello Sviluppo e della Socializzazione, Universita' di Padova, Padua, Italy. During the writing of this article, Anne Maass was visiting the Department of Psychology, University of California, Davis. Correspondence concerning this article should be addressed to Anne Maass, Dipartimento di Psicologia dello Sviluppo e della Socializzazione, Universita' di Padova, Via Venezia, 8, 35131 Padua, Italy. 512
LINGUISTIC 1NTERGROUP BIAS descriptions dissociate the actor from the act, whereas abstract descriptions imply that the behavior represents a stable and enduring property of the actor, one may argue that the LIB helps to portray the in-group in a favorable light while derogating the out-group. In other words, people may show LIB as one possible way of maintaining a positive image of their own group, which, in turn, enhances their self-esteem. As such, the LIB may fulfill much the same function as other strategies of in-group favoritism, including discriminatory reward allocation on minimal group matrices, and differential trait attribution. An entirely different but equally plausible interpretation comes from a cognitive point of view. From this perspective, LIB may derive from differential expectancies. One may argue that any expectancy-congruent behavior is described in abstract terms exactly because it is considered a stable and typical behavior tendency of the actor. If we expect Italians to be hospitable, then any hospitable behavior should be described in abstract terms. In contrast, unexpected or surprising behaviors are considered short-lived and atypical and, as such, should be described in concrete terms. The LIB simply derives from the fact that members of a given group usually expect more desirable and fewer undesirable acts from in-group than from out-group members (Howard & Rothbart, 1980). The main difference between the two explanations is that the expectancy-based explanation predicts abstract language use for expected and concrete language use for unexpected behaviors, regardless of the valence of the behavior, whereas the ingroup-protective explanation predicts abstract language use for positive in-group and negative out-group behaviors, regardless of stereotypic expectancy. Of course, in many intergroup settings the predictions of the two models will coincide simply because people tend to hold generalized negative expectations about the in-group and generalized positive expectations about the out-groups (Howard & Rothbart, 1980). But this is not always the case. Men and women, for example, tend to share highly differentiated stereotypes about both gender groups that include positive (e.g., independent for men, caring for women) as well as negative aspects (e.g., aggressive for men, dependent for women). For such cases, the cognitive and the motivational approach make distinct predictions. Recently, Maass et al. (1995) tested these explanations against each other and found evidence for the expectancy-based interpretation. In one of these studies (Experiment 1), participants viewed cartoons in which a protagonist coming from northern or southern Italy displayed behaviors that were either consistent with the stereotype of northern Italians (industrious, materialistic) or with that of southern Italians (hospitable, sexist); moreover, one half of these behaviors were of positive valence, the other half of negative valence. Results show that southern and northern Italian participants described stereotype-congruent behaviors of southern or northern protagonists in relatively abstract terms, whereas stereotype-incongruent behaviors were described more concretely, regardless of whether the behaviors were socially desirable or undesirable (Experiment 1). Similar results were obtained when the protagonist's membership had no personal relevance to the participants, as they were not part of either group, and when expectancies were induced experimentally (Experiment 3)—a situation in which self-protective motives could not possibly operate. Further sup-
513
port for the expectancy-based explanation comes from a study by Rubini and Semin (1994) showing that members of different parties (a Catholic and a Communist party in Italy) used abstract terms when explaining party-congruent behaviors of ingroup members but relatively concrete terms when explaining incongruent behaviors. Taken together, the results of these studies demonstrate that differential expectancies are sufficient to induce biased language use, even in the absence of motivational concerns. Although these data suggest that LIB is driven by differential expectancies and that in-group-protective motivation is not a necessary prerequisite for biased language use to occur, they do not prove that in-group-protective motivations are necessarily and always irrelevant. One may well imagine highly competitive, hostile, or in-group-threatening situations in which motivational considerations may become relevant and even override cognitive mechanisms. None of the studies reported above investigated such clearly hostile or in-group-threatening situations. Although northern and southern Italians tend to hold negative beliefs about the respective out-group (see Capozza, 1968; Capozza, Bonaldo, & Di Maggio, 1982), intergroup conflict rarely degenerates into open hostility. Imagine, instead, an openly aggressive intergroup setting in which in-group identity is at stake. If language use is, in part, driven by in-group-protective motivational needs, then it is plausible to argue that threat to in-group identity will enhance the LIB in much the same way as it has generally been found to enhance other in-groupfavoring strategies. Social identity theorists have long identified conditions that motivate people to use in-group-protective strategies as well as out-group discrimination (for a recent overview, see Brown, 1995). People seem particularly motivated to activate in-groupprotective strategies (a) when their social identities are threatened or devalued (Breakwell, 1978; Breakwell, Collie, Harrison, & Propper, 1984; Brown & Ross, 1982; Grant, 1992), (b) when they belong to groups that have an illegitimately low status (Ellemers, Wilke, & Van Knippenberg, 1993), (c) when the intergroup setting is highly competitive (Brewer & Silver, 1978; Kahn & Ryan, 1972; Rabbie, Benoist, Oosterbaan, & Visser, 1974; Ryan & Kahn, 1975; Worchel, Andreoli, & Folger, 1977; see also classical summer camp studies by Sherif, 1966; Sherif & Sherif, 1953; Sherif, Harvey, White, Hood, & Sherif, 1961), and (d) in the absence of overlapping social categories (Brown & Turner, 1979; Deschamps & Doise, 1978; Vanbeselaere, 1991). In the present article, we focus on three of these factors, namely threat to identity (Experiment 1), illegitimately low status (Experiment 2), and absence versus presence of an overlapping category (Experiment 2). In Experiment 1, we introduced a temporary threat to in-group identity in order to test whether this would affect the magnitude of the LIB in the same fashion as it has been shown to affect other in-group-protective strategies. In keeping with previous research (Brown & Ross, 1982), we manipulated in-group threat through an alleged derogatory (vs. friendly) message from the out-group, a manipulation that was expected to increase (vs. decrease) the participants' ingroup-protective motivation without affecting those long-term beliefs about in-group and out-group that are implicated in the expectancy-based process.
514
MAASS, CECCARELLI, AND RUDIN
Experiment 1 In Experiment 1, we took advantage of an intergroup setting that we knew from previous research to be highly conflictual and in which group members generally hold extremely negative views about the out-group: hunters versus environmentalists (Bencivenni, 1990). We either reinforced the existing antagonism between the groups through a derogatory, in-groupthreatening statement from the out-group or else presented a friendly and cooperative statement in which it was proposed to overcome traditional boundaries and to collaborate for the solution of environmental problems. We hypothesized that the LIB would be considerably stronger when an in-group-threatening rather than a friendly message was delivered. As in previous research, we used cartoons in which hunters or environmentalists displayed positive or negative behaviors and assessed language abstraction. For an independent control of the activation of in-group-protective motives, we also included allocation matrices generally used in social identity research (Brown, 1978). The rationale for including these matrices was the following: Because previous research (Brewer, 1979; Brown, 1978; Tajfel, Billig, Bundy, & Flament, 1971) has clearly proved the sensitivity of this instrument to motivational mechanisms, it provided the possibility to see whether the in-group-threatening versus friendly message affected language use in the same way as it affects reward allocation; if in-group threat increased in-group favoritism on both allocation matrices and LIB, it would seem likely that a common motivational mechanism is underlying both phenomena. Finally, social identity theorists have argued that in-group favoritism-out-group discrimination constitutes an effective strategy for protecting social identities (Lemyre & Smith, 1985; Oakes & Turner, 1980), althoughfindingsare not perfectly consistent in this respect (for an overview, see Abrams & Hogg, 1988). If the LIB is able to successfully meet in-group-protective needs, then it should have a positive effect on self-esteem. We therefore measured personal and collective self-esteem to test whether self-esteem was positively related to the LIB. Hypotheses 1. In line with the LIB model, we expected participants to describe positive in-group and negative out-group behaviors more abstractly than negative in-group and positive out-group behaviors. 2. More important, we predicted that this bias would be considerably stronger after receiving an in-group-threatening versus friendly message. 3. The third hypothesis refers to the strategies on reward allocation matrices predicting that the power of differentiation strategies favoring the in-group relative to the out-group (maximum differentiation = MD; maximum differentiation + maximum in-group profit = MD+MIP) would be stronger under high threat versus no threat whereas the contrary was expected for the strategy implying equal treatment (maximum joint profit = MJP). 4. If the LIB has an efficient role in self-esteem maintenance strategy, then it can be expected to correlate with postexperi-
mental (but not preexperimental) self-esteem. Thus, the greater the LIB shown in reaction to in-group threat, the higher we expected postexperimental self-esteem to be. If the LIB contributes uniquely to self-esteem maintenance, then this correlation should hold, even when controlling for in-group favoritism on reward allocation.
Method Participants. 80 hunters and 80 environmentalists from northern Italy participated in this study. Ages ranged from 20 to 72 years, with a mean of 40.9 years (hunters: 44 years; environmentalists: 37 years). The educational level was very similar in the two samples (hunters: 4% with elementary school degree, 58% with middle or high school degree, 38% with university degree; environmentalists: 61 % with middle or high school degree, 39% with university degree). The two samples differed only in sex composition, with 100% of the hunters and 59% of the environmentalists being men (note, however, that sex of participant has never been found to affect LIB in previous studies). Procedure. Participants were contacted in their clubs and associations. They were asked to participate in a study on environmental problems. Each participant received a booklet with the following written presentation: This questionnaire is part of a larger, nation-wide research project conducted by various Psychology Departments in order to know the attitudes regarding environmental issues of different social categories such as hunters, forest warders, environmentalists, farmers, and industrialists. This research is part of a larger project, initiated some time ago, with the scope to organise an educational program for environmental protection. None of the 169 participants originally addressed by the experimenter refused to participate, but 9 participants failed to complete the entire questionnaire and were therefore excluded from the analysis, resulting in a final sample of 160 participants. Participants responded individually to the questionnaire (time required: approximately 15-20 min) during their stay at the club house or association. Threat versus no threat. For each group, one half of the participants received a statement that was supposedly written by the out-group and that took a rather hostile stand against the in-group. Hunters, on one side, claimed "that they are right there in thefieldin order to intervene where needed against who litters or damages the environment without getting lost in endless political discussions (like the environmental groups." The environmentalists, on the other side, claimed "to have obtained positive results in the area of civil education regarding the environment and in their battles against categories that endanger the natural patrimony of all (such as the hunters)." Although not identical in content, both criticisms are actually quite typical of the reciprocal excuses between hunters and environmentalists in real life. The remaining participants received a friendly statement in which the out-group acknowledged the positive characteristics of the outgroup and encouraged a cooperative attitude for the solution of environmental problems. In each case, a majority (70%) of the out-group had supposedly claimed that the moment has come to start a collaboration of all categories that are truly interested in environmental protection, leaving behind old attitudes of rivalry that would only distract from the only important goal: salvaging and protecting the environment. From this prospective, it would be productive to join the forces and strength of the various groups, for example the capacity to increase environmental awareness which is typical of environmental organisations, and the presence in thefieldwhich is typical of hunters; the
LINGUISTIC INTERGROUP BIAS latter group could indeed intervene readily against who damages the environment. Cartoons. In line with previous research (Maassetal., 1989,1995), participants were then presented with eight single-frame cartoons, one half of which depicted hunters and the other half environmentalists engaging in either positive (e.g., dress well, study) or negative behaviors (e.g., litter, drink, and smoke). Hunters and environmentalists were matched equally often with each episode, and the eight cartoons were presented in a different random order for each participant. Four response alternatives were provided under each scene, corresponding to the four levels of abstraction in Semin and Fiedler's (1991) model (DAV, IAV, SV, ADJ) and presented in the same standardized order, starting from the lowest level of abstraction. For instance: "This person, a hunter and member of a hunting organisation: a) picks up paper (DAV), b) cleans up the wood (IAV), c) respects nature (SV), d) is conscientious (ADJ)." Participants were asked to observe each scene carefully and to select the phrase that described the scene best. Extensive pretesting had assured that the valence of the episodes was, indeed, perceived as intended by the authors, that each description was pertinent to the scene, and that none of the response alternatives was selected with extremely high or extremely low frequency (see Bencivenni, 1990). Language abstraction. On the basis of these responses, language abstraction scores were calculated for each participant, representing the main dependent variable. Following the procedure used in previous research (Maass et al., 1989, 1995), each response was scored by assigning a value from 1 (when participant chose DAV alternative) to 4 (when participant chose ADJ alternative)1; we then calculated the mean of each pair of vignettes (2 positive in-group, 2 negative in-group, 2 positive out-group, 2 negative out-group), resulting in four scores per participant. Moreover, a compound LIB index was calculated by subtracting the mean abstraction of negative in-group episodes from that of positive ingroup episodes and by subtracting the mean abstraction of positive outgroup episodes from that of negative out-group episodes. The LIB index is represented by the sum of the two values and has a range from - 6 (maximum out-group favoritism) to +6 (maximum in-group favoritism). Reward allocation matrices. The second dependent variable consisted of four matrices adopted from Brown (1978) used to measure intergroup differentiation versus cooperation. Participants were presented with a hypothetical task in which they were asked to allocate funds assigned by the environmental department for the implementation of an environmental protection program to environmental versus hunter organizations in different regions of Italy. For each of the matrices, participants were instructed to select the row that they thought was most appropriate for dividing the money between the two associations. The four matrices allow the identification of four strategies, (a) MD: tendency to maximize the relative difference between the two groups in favor of one's own group even at the expense of absolute in-group gain. People using this strategy prefer to renounce in terms of absolute ingroup profit rather than choosing an alternative that would allocate greater profit to the out-group than to the in-group. (b) MJP: This strategy maximizes the overall profit of in-group and out-group, suggesting that the participant considers the common profit as more important than the relative advantage of the in-group over the out-group, (c) MIP+MD: This strategy illustrates a rather egoistic tendency as it combines a maximum profit for the in-group both in absolute and relative terms, (d) MIP+MJP: represents a mixed strategy as it allows maximization of both ingroup profit and common profit but at the expense of the relative superiority of the in-group over the out-group. Following Brown's (1978) methodology,2 the strength of each strategy varies from - 4 to +4, with positive scores indicating the prevalence of the strategy under discussion. Self-esteem. Both collective and personal self-esteem were assessed.
515
In order to measure collective self-esteem, two subscales of Luthanen and Crocker's (1992) Collective Self-Esteem Scale, which seemed of particular relevance to the scope of our research, were translated into Italian—namely, Private Collective Self-Esteem and Importance To Identity. Participants responded with explicit reference to how they felt as members of the hunter or environmentalist category on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Although the two subscales have not yet been validated in Italy, they were found to have reasonable internal consistency (Cronbach's a = .72, when considering the two scales together). Personal self-esteem was assessed by a short form (4 items) of Rosenberg's (1965) Self-Esteem Inventory, using a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree), Cronbach's a - .70. Order ofpresentation ofdependent variables. One half of the participants were asked to respond to the self-esteem scales immediately before receiving the threat manipulation (preexperimental self-esteem); the remaining participants responded to the self-esteem items after having read the threat manipulation and after having responded to both LIB and matrices (postexperimental self-esteem). This established (a) whether preexperimental self-esteem was, indeed, distributed equally across competition conditions and across participant groups (hunters vs. environmentalists) and (b) whether the expression of in-group favoritism on LIB and matrices protects one's self-esteem. Also, the order of LIB and matrices was counterbalanced so that for one half of the participants under each of the conditions mentioned above (hunter vs. environmentalist, competition vs. cooperation, prevs. postexperimental assessment of self-esteem), the LIB preceded the matrices, whereas for the other half, the matrices preceded the LIB. Manipulation check. Two questions about the intergroup relation were introduced in order to check whether participants had paid attention to the manipulation and whether they did indeed perceive greater antagonism between the two groups after having received the threat manipulation, namely, "in your opinion, how opposed are the two groups?" and "in your opinion, how likely is it that a collaboration in environmental protection activities will be successful?" Participants responded on a 5-point scale ranging from 1 (very little) to 5 (very much). Debriefing. After completing the questionnaire, participants were informed about the true scope of the experiment and were offered the possibility of receiving further information about the outcome of the study.
Results Manipulation check. A 2 (participants' group membership: hunters vs. environmentalists) X 2 (threat vs. no threat) analy1 The assumption of interval scaling is obviously problematic, although non-parametric analyses have not yielded very different results in the past (Maass et al., 1989). Also, in a small-scale pilot study, six second-year psychology students unfamiliar with the linguistic category model viewed 24 cartoons (including the ones used in Experiments 1 and 2 in this article), together with the four descriptions presented in random order. They were asked to locate the four descriptions on a 10cm horizontal line byfirstfixingthe two extremes at the endpoints and then locating the two intermediate descriptions at any place on the dimension "from the most unstable to the most stable over time." The location on the stability dimension comes very close to an interval scale (DAV = 0.1, IAV = 3.3, SV = 6.7, ADJ = 9.7). Although stability is only one of the dimensions underlying abstraction (other dimensions remain to be tested), we felt justified in opting for the use of parametric statistics, especially considering their ability to deal with repeated measures. 2 The strategies are calculated by considering pairs of matrices together. An estimate of the strength of each strategy is obtained by comparing the participant's answers on the two matrices of each couple (for more information, see Brown, 1978).
516
MAASS, CECCARELLI, AND RUDIN
sis of variance (ANOVA) was conducted using the two items of the manipulation check as dependent variables. Results show a main effect of in-group threat for both items, F( 1, 156) = 9.29, p < .005, for thefirstitem, and F( 1, 156) = 60.37, p < .001, for the second, and no effect of participant's group membership, suggesting that the manipulation was, indeed, equally successful for both groups. Under high in-group threat, participants indicated greater contrast between the two groups (M = 4.24) and lesser chances for a successful collaboration (M = 2.24) than under no threat (M = 3.75 and M = 3.60, respectively). Level ofabstraction. A 2 (participants' group membership: hunters vs. environmentalists) X 2 (in-group threat vs. no threat) X 2 (order of self-esteem scales: pre vs. post) X 2 (order LIB-matrices vs. matrices-LIB) X 2 (protagonists' group membership: in-group vs. out-group) X 2 (favorability of behavior: positive vs. negative) ANOVA was conducted, with repeated measures on the last two variables. In line with the first hypothesis predicting a replication of the LIB, the expected interaction between the protagonists' group membership and favorability of behavior, F( 1,144) = 192.89, .p < .0001, emerged. Positive behaviors were described more abstractly when performed by in-group (M = 2.89) rather than out-group protagonists (M = 1.96), whereas negative behaviors were described more abstractly when performed by out-group (M = 2.96) rather than in-group protagonists (M = 1.78), with both desirable in-group and undesirable out-group behaviors differing significantly from undesirable in-group and desirable out-group behaviors (all t tests, p < .05). More important, the same analysis also showed the effect of the in-group-threat manipulation on the LIB as predicted by Hypothesis 2. The three-way interaction among the protagonist's category membership (in-group vs. out-group), the social desirability of the behavior, and the threat manipulation was, indeed, significant, F{\, 144) = 25.47, p < .001, and, note especially, was not modified by any additional variable (participant's group membership, order of self-esteem, order of matrices-LIB). As can be seen in Table 1, both groups used language in a way that favors the in-group, but this bias was considerably more pronounced under high in-group threat. The level of abstraction for positive in-group and negative out-group
Table 2 Reward Allocation Strategies (MD, MD+MIP, MJP) and LIB Index as a Function ofln-Group Threat in Experiment 1 No
Variable
In-group threat
threat
P
0.98