amiot e bouris, 2005

Upload: goosi666

Post on 07-Apr-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/3/2019 Amiot e Bouris, 2005

    1/18

    European Journal of Social Psychology

    Eur. J. Soc. Psychol. 35, 581598 (2005)

    Published online 11 November 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/ejsp.238

    Ideological beliefs as determinants of discrimination in positiveand negative outcome distributions

    CATHERINE E. AMIOT AND

    RICHARD Y. BOURHIS*

    Universite du Quebec a` Montreal, Canada

    Abstract

    Social identity theory proposes that discrimination contributes favourably to group members social

    identity. In minimal group paradigm (MGP) studies involving positive outcome distributions (e.g.

    money), discrimination is associated with a more positive social identity. But studies on the positive-

    negative asymmetry effect show that categorization leads to less discrimination when negative (salary

    cuts) than when positive outcomes (salary increases) are distributed. Using structural equation

    modelling, this study (N 279) tested whether discrimination involving negative outcome distribu-

    tions could contribute as much to group members positive social identity as discrimination on positive

    outcomes. The study also tested if ideological beliefs (i.e. social dominance orientation, authoritar-

    ianism), measured one month before the MGP experiment, could predict positive and negative

    outcome discrimination. While the fit of the hypothesized model was adequate, only social dominance

    orientation predicted both positive and negative outcome discrimination. Also, discrimination on

    positive outcomes but not on negative ones contributed to positive social identity. Copyright# 2004John Wiley & Sons, Ltd.

    It is one thing to explain why group members seek to benefit their own group more than out-groups,

    but it is quite another to account for why some people are interested in discriminating by imposing

    more harmful outcomes on out-group others. A growing number of researchers have proposed

    conceptual models combining intrapersonal, interpersonal, intergroup, and ideological levels of

    explanation to better account for different types of discriminatory behaviours (Doise, 1986; Taylor

    & Moghaddam, 1995). The present study combines two levels of analysis to explain discrimination on

    positive vs. negative outcome distributions: the intergroup level, using social identity theory (SIT), and

    the ideological level, via belief structures such as social dominance orientation (SDO) and right-wing

    authoritarianism (RWA). The main goal of this study was to explore the role of ideological beliefs and

    social identity processes as complementary accounts of discriminatory behaviours on positive vs.

    negative outcome distributions.

    Received 9 February 2004

    Copyright# 2004 John Wiley & Sons, Ltd. Accepted 25 June 2004

    *Correspondence to: Dr R. Bourhis, Departement de psychologie, Universite du Quebec a Montreal, PO Box 8888, Succ. CentreVille, Montreal, Quebec, H3C 3P8, Canada. E-mail: [email protected]

    Contract/grant sponsor: Social Science and Humanities Research Council of Canada (SSHRC).

  • 8/3/2019 Amiot e Bouris, 2005

    2/18

    As an intergroup level theory, SIT originated as an explanation of the discrimination effect obtained

    in the minimal group paradigm (MGP; Tajfel & Turner, 1986). In a typical minimal group experiment,

    members of two arbitrary groups allocate valued resources such as money to anonymous in-group and

    out-group members (Tajfel, Flament, Billig, & Bundy, 1971). There is no social interaction within or

    between the groups, no instrumental links between individuals responses and their personal self-

    interest, and no previous history or relations between the groups (Bourhis, Turner, & Gagnon, 1997).

    Although these procedures were designed to eliminate classic factors accounting for discrimination,research has shown that categorizing people into us and them is sufficient to foster discriminatory

    behaviour (Bourhis & Gagnon, 2001; Brewer, 1979; Diehl, 1990). This mere categorization effect

    provides the cognitive basis needed to trigger the motivational processes leading to discriminatory

    behaviours.

    Social categorization, social identification, social comparison, and the need for a distinct and

    positive social identity are the key psychological variables proposed by SIT to account for

    discriminatory behaviour within the MGP (Tajfel & Turner, 1986). Within this experimental situation,

    the arbitrary imposed us-them categorization provides participants with the opportunity to identify as

    group members. Degree of in-group identification, the cognitive component of social identity, refers to

    the strength of an individuals identification with the in-group, independently of whether the individual

    feels positive or negative about this identification (Gagnon & Bourhis, 1996). To the degree that

    individuals identify with their in-group in the MGP, desire for a distinct and positive social identity isachieved by seeking favourable comparisons between the in-group and the out-group on the only

    available dimension of comparison in the experiment, namely, the distributions of more valued

    resources to members of the in-group than to members of the out-group. In line with SIT, recent

    reviews have noted significant links between degree of identification to the in-group and discrimina-

    tory behaviour (Brown, 2000; Rubin & Hewstone, 1998).

    Quality of social identity can be defined as the affective evaluation, positive or negative, of an

    individuals group membership (Gagnon & Bourhis, 1996). In their MGP studies using positive

    outcome distributions, Perreault and Bourhis (1998, 1999) obtained results showing that quality of

    social identity was more positive after group members discriminated than before they engaged in the

    resource distribution task. Thus, by discriminating on positive outcome distributions, participants in

    MGP studies are able to achieve a more positive social identity. From an SIT perspective,

    discrimination within the MGP is seen as functional, fulfilling the need for differentiation from theout-group and contributing to group members positive social identity (Tajfel & Turner, 1986).

    THE POSITIVE-NEGATIVE ASYMMETRY EFFECT AND POSITIVE SOCIAL IDENTITY

    Brewer (1979) noted that the distinction between in-group favouritism and out-group derogation is not

    always present in classic MGP research so it is often ambiguous whether the comparison rests on

    enhancement of the in-group, devaluation of the out-group, or both (p. 321). Mummendey and Otten

    (1998) noted that most MGP studies have focused on group members distribution of positive

    outcomes such as giving money or symbolic points to in-group and out-group members. The question

    is whether the conditions sufficient to elicit discrimination in the distribution of positive outcomes can

    function equivalently for discrimination involving negative outcome distributions such as inflicting

    punishments, imposing extra hours of unpaid work, or enforcing salary cuts.

    Over 20 studies were conducted by the Mummendey research group to investigate the differential

    allocation of positive vs. negative outcome distributions (Mummendey & Otten, 1998). Generally,

    582 Catherine E. Amiot and Richard Y. Bourhis

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    3/18

    findings demonstrated that categorization per se leads to less discriminatory behaviour when negative

    outcomes are distributed than when positive outcomes are distributed (Buhl, 1999). This phenomenon

    was identified as the positive-negative asymmetry effect (PNAE). But beyond the absolute amount of

    discrimination manifested when positive vs. negative outcomes are distributed between group

    members, the underlying psychological processes involved when distributing positive and negative

    outcomes remain to be investigated. For instance, within the PNAE, can discrimination involving

    negative outcome distributions contribute as much to group members positive social identity as wasfound to be the case for discrimination involving positive outcomes? In his analysis, Buhl (1999) noted

    that SIT does not take into account whether the dimension of comparison contributing to positive

    social identity is made up of desirable or aversive outcome distributions (Mummendey, 1995).

    Different processes may account for discrimination on positive vs. negative outcome distributions.

    Based on an elaboration of the perceptual differentiation hypothesis first proposed by Doise (1976),

    one could expect group members to be mainly concerned with the distribution of outcomes in a

    manner that allows for the in-group and the out-group to receive different allocations, regardless of the

    valence of the outcomes being distributed. In line with this perceptual differentiation hypothesis, one

    could expect that discrimination on negative and positive outcome distributions may contribute

    equally to a perceptual differentiation between the in-group and the out-group, thus leading in both

    cases to a more positive social identity.

    However, discrimination on negative outcome distributions is often perceived as being even lesssocially desirable and less normatively appropriate than discrimination on positive outcome distribu-

    tions (Amiot & Bourhis, 2003; Blanz, Mummendey, & Otten, 1997). Harming the out-group more than

    the in-group involves an undesirable dimension of intergroup comparison, which is unlikely to

    contribute to positive social identity. Thus, according to the discrimination valence hypothesis,

    discrimination on negative outcome distributions should be less likely to contribute to positive in-

    group identity than discrimination on positive outcome distributions. Although recent MGP studies

    have shown that discrimination on positive outcome distributions are associated with a more positive

    social identity (Gagnon & Bourhis, 1996; Perreault & Bourhis, 1998), the effect of negative outcome

    discrimination on the quality of in-group identity remains to be investigated. The first goal of this study

    was thus to explore the relative merit of the perceptual differentiation vs. discrimination valence

    hypotheses as competing processes underlying discrimination on positive vs. negative outcome

    distributions.

    IDEOLOGICAL BELIEFS, DISCRIMINATION, AND THE PNAE

    Whereas studying the associations between positive and negative discrimination on group members

    quality of social identity allows one to tap into the consequences of these discriminatory behaviours,

    investigating the antecedent variables predicting the manifestation of positive and negative discrimi-

    nation is also of importance. In such a context, are some individuals more prone to discriminate on

    both positive and negative outcome distributions? As a second goal, the present investigation seeks to

    identify the ideological beliefs that predict both positive and negative discrimination.

    Recently, Duckitt (2001) as well as Guimond, Dambrun, Michinov, and Duarte (2003) argued that

    while a tradition of research has considered authoritarianism and SDO as stable personality traits,

    these intraindividual constructs can be more adequately understood in terms of ideological beliefs.

    Duckitt, Wagner, du Plessis, and Birum (2002) also propose that ideological beliefs are analogous to

    motivational goals schemas, which can become more or less salient depending on the specific context.

    Indeed, the RWA and Facism scales have been shown to be highly reactive to situational threat

    Ideological beliefs and discrimination 583

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    4/18

    manipulations (Duckitt et al., 2002; see also Guimond et al., 2003, for the moderating role of

    ideological beliefs in specific contexts). In the present context, both positive and negative outcome

    distributions may be likely to reflect ideological beliefs about how outcomes should be distributed

    between group members. Respondents endorsement of relevant ideological beliefs assessed one

    month priorto the conduct of the MGP study will be used to test whether such ideological beliefs can

    predict discriminatory behaviour on both positive and negative outcome distributions.

    Ideological beliefs such as authoritarianism (Adorno, Frenkel-Brunswik, Levinson, & Sanford,1950) or RWA (Altemeyer, 1998) as well as SDO (Pratto, Sidanius, Stallworth, & Malle, 1994) have

    been consistently linked to intergroup attitudes and behaviours. While RWA has been labelled as a

    submissive authoritarianism and SDO as a dominant authoritarianism, these two ideological beliefs

    show weak to moderate correlations with one another and have been associated with distinct correlates

    (for reviews, see Altemeyer, 1998; Pratto, 1999).

    As proposed by Altemeyer (1998), right-wing authoritarians believe in submission to established

    authorities and the social norms these authorities endorse, even if these norms involve aggression.

    According to Whitley (1999), people high in authoritarianism tend to organize their worldviews in

    terms of in-groups and out-groups and perceive members of out-groups as threatening the traditional

    values authoritarians hold dear (p. 126). There is also a strong body of research showing that RWA

    can predict racism and prejudice against visible and low-status out-groups (Adorno et al., 1950;

    Altemeyer, 1988, 1998; Duckitt, 2001; Duckitt et al., 2002; Duriez & van Hiel, 2002). Within anintergroup experiment using a modified version of the MGP, Altemeyer (1994) found that individuals

    scoring high on authoritarianism rated their minimal in-group more positively than did individuals

    scoring low on authoritarianism. There is also some evidence for the association between authoritar-

    ianism and discrimination involving negative outcome distributions, revealing that highly authoritar-

    ian individuals were more punitive and aggressive than low authoritarians (Christie, 1993). Elms and

    Milgram (1966) found that authoritarianism was the only ideological belief associated with destructive

    obedience, a result also obtained in a learning study conducted by Altemeyer (1988). Finally,

    according to Altemeyer (1998), submissive authoritarians tend to see the world as a dangerous place,

    which triggers fear and instigates aggression in them (see also Duckitt, 2001). Combined with their

    tendency to be self-righteous, submissive authoritarians would appear prone to manifesting both

    positive and negative outcome discrimination in the context of the MGP. We thus expect that the more

    group members endorse authoritarianism (RWA), the more they should discriminate on both positiveand negative outcome distributions in the MGP.

    SDO is defined as the tendency to consider ones own primary in-group to be better than, superior

    to, and dominant over relevant out-groups (Sidanius & Pratto, 1999). According to Pratto et al. (1994),

    high-SDO individuals are theorized to be relatively conservative, racist, ethnocentric, and prejudiced,

    and they should show little empathy for lower status others (p. 744). Studies have shown a strong

    relationship between SDO and both prejudice and racism (Sidanius & Pratto, 1999). Social dominance

    theory (SDT) has made explicit suggestions about how SDO could be linked to discrimination

    involving out-group derogation and negative outcome distributions. According to Sidanius, Pratto, and

    Mitchell (1994), the desire to actively oppress out-groups has been given a central role within SDT. In

    fact, SDT proposes that oppressive forms of intergroup behaviours (i.e. ethnic cleansings, lynchings)

    reflect more a desire to actively dominate and oppress out-groups than a motivation to benefit the in-

    group and improve positive in-group identity. Thus, SDT appears well-suited to account for the

    dominance motives underlying the distribution of both positive and negative outcome distributions.

    In line with such analyses, a number of questionnaire studies have shown that high SDO individuals

    have a propensity for cruelty (Altemeyer, 1998), while also being more likely to support punitive policies

    (Sidanius, Liu, Shaw, & Pratto, 1994). In a MGP study, Sidanius, Pratto et al. (1994) found that high and

    low SDO individuals displayed different affective reactions to the out-group. High SDO individuals

    584 Catherine E. Amiot and Richard Y. Bourhis

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    5/18

    reported a greater desire for social distance from out-group members while also claiming less willingness

    to cooperate with out-group than in-group members. In another MGP study, Federico (1998, Study 2)

    found that among high status group members, SDO was positively associated with evaluative favouritism

    for their in-group (i.e. using positive adjectives), while this relationship was not significant among low

    status group members. Although these MGP studies reveal that low and high SDO individuals differ in

    their affective and evaluative reactions toward minimal in-group and out-group members, SDO remains to

    be associated with actual discriminatory behaviour on both positive and negative outcome distributions. Inthis regard, we expect that the more group members endorse the SDO, the more they should discriminate

    on both positive and negative outcome distributions in the MGP.

    THE PRESENT STUDY

    The first aim of this study was to explore how discrimination on positive vs. negative outcome

    distributions could affect group members quality of social identity. According to the perceptual

    differentiation hypothesis, it is predicted that discrimination on both positive and negative outcome

    distributions should lead to a more positive social identity. However, the discrimination valence

    hypothesis predicts that discrimination on positive outcomes should contribute to positive socialidentity while discrimination on negative outcome distributions should not be related to a more

    positive social identity. The second goal of the study was to test whether different ideological beliefs

    assessed one month before the conduct of the MGP experiment can predict discrimination involving

    both positive and negative outcome distributions. The ideological beliefs hypothesis proposes that

    RWA and SDO should predict discrimination involving the distribution of both positive and negative

    outcome distributions.

    METHOD

    Participants

    Two-hundred-and-seventy-nine college students (194 women, 84 men, one participant did not report

    his/her gender) from a French language college (CEGEP) in Montreal participated in the two parts of

    the study. In Quebec, students must complete a two-year college programme at the CEGEP level

    before being eligible to apply as undergraduates at the university level. Most respondents were social

    science college students while none were enrolled in a business programme. The mean age of the

    students who took part in the study was 18.2 years old ( SD 0.72, range 1724 years old). None of

    the Quebec Francophone participants included in the study were members of a visible minority group.

    Design

    Part 1 of the study involved the completion of a questionnaire assessing participants ideological

    beliefs, while the second part, conducted a month later, involved a standard MGP experiment. In the

    first part of the study, CEGEP students were recruited during class time and were told that the

    questionnaire dealt with the validation of a number of psychological scales. Ostensibly, the study was

    presented as being conducted by a team of researchers from the Psychology Department of one of the

    Ideological beliefs and discrimination 585

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    6/18

    two Francophone universities in Montreal. The questionnaire consisted of various scales including

    SDO and authoritarianism (RWA). Just over a month after this first study, the MGP experiment was

    conducted during class time with the same students. This MGP experiment was allegedly unrelated to

    the previous study as participants were recruited by a female experimenter from the other Francophone

    university in Montreal. The experimenter presented herself as a graduate student in industrial/

    organizational psychology interested in the study of decision-making processes involved in the

    context of mergers and acquisitions. This procedure insured that respondents had no reason to surmisethat their participation in the MGP study was at all related to a study conducted by researchers from

    another university one month earlier.

    Three independent variables were manipulated in the MGP experiment. First, half the respondents

    distributed salaries, while the other half distributed hours of unpaid work, making this a between-

    subjects factor. Second, the valence of the outcomes distributed, a within-subjects factor, was

    manipulated. This was made possible by asking participants in the salary distribution condition to

    distribute both salary increases and salary cuts. Participants in the unpaid work condition distributed

    both increases and cuts in hours of work. Based on the taxonomy of discriminatory behaviours

    proposed by Mummendey (1995; Amiot & Bourhis, 2003), positive outcomes were conceptualized as

    either salary increases or cuts in unpaid hours of work. Negative outcomes were conceptualized as

    either cuts in salaries or increases in unpaid hours of work. Finally, the order of the outcomes

    distributed was manipulated so as to counterbalance the design. Group members in the order 1condition first allocated positive outcome distributions then allocated negative outcomes. Group

    members in the order 2 condition first allocated negative outcomes then allocated positive outcomes

    between group members. This third independent variable was a between-subjects factor.

    Procedures Used in the MGP Experiment

    Participants in the MGP experiment were instructed that for the purpose of the decision-making task, they

    would be randomly divided into two groups. Ostensibly, the outcome of a coin toss performed privately

    by each respondent in the presence of the experimenter determined their group membership in group K or

    group W of the newly merged corporation. Respondents were in fact randomly assigned to their group,

    and arrangements were made for group memberships to remain anonymous throughout the entire MGPexperiment. The experimenter explained that because the study was concerned with decision-making with

    minimum information, participants would allocate outcomes to anonymous in-group and out-group others

    but never to themselves. These procedures represent the standard method used to eliminate self-interest as

    a motive in minimal group experiments (Bourhis, Sachdev, & Gagnon, 1994).

    Before each outcome distribution, participants were told that they would be making decisions about

    the distribution of salary increases or reductions in hours of unpaid work (in the positive outcome

    allocation phase) and salary cuts or increases in unpaid hours of work (in the negative outcome phase).

    It was made clear that salary distributions did not involve variations in quantity or quality of work, and

    that hours of work distributions did not involve variations in salaries.

    The outcome distributions were justified by informing participants that such distributions took place in

    the context of a recent corporate merger. Participants were told that group K and group W were branches

    of the same merged organization and that, because of the organizational changes, they were asked to

    decide how outcomes should be distributed between these two branches. Negative outcome distributions

    were legitimized by specifying that when mergers are not well planned, a decrease in corporate efficiency

    and a drop in financial assets can occur. For the sake of insuring a successful merger, it becomes necessary

    for the corporation to impose salary cuts and increases in hours of unpaid work on its employees. Positive

    outcome distributions were justified by specifying that when mergers are well planned, an increase in

    586 Catherine E. Amiot and Richard Y. Bourhis

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    7/18

    efficiency and profit can follow, leading the corporation to provide employees with salary increases and

    reductions in hours of work with no loss in salary. Participants were instructed to imagine themselves as

    decision-makers within their respective corporate branch (K or W), which had just completed the

    corporate merger. Due to variations in profits following the merger, participants were asked to distribute

    resources to anonymous members of the K and W branches using the Tajfel matrices. At the very end of

    the minimal group study, respondents completed a scale measuring the quality of their social identity as

    members of their respective corporate branches.The instructional set was constructed to insure experimental realism, encourage the emotional

    involvement of the participants, and minimize the social desirability responses associated with the

    negative outcome distribution task. In both parts of the study, participants were informed that they were

    free to end their participation at any time. They were informed that there was no wrong or right answer

    when completing the questionnaires and their outcome distributions booklets. They were assured that

    their answers would remain anonymous and confidential and would serve for research purposes only. For

    matching purposes, participants were asked to indicate their date of birth on both the part 1 ideological

    beliefs questionnaire, and on the part 2 response booklets completed during the MGP experiment one

    month later. At the end of the second part of the study, participants were fully debriefed.

    Measures

    Ideological Beliefs

    In part 1 of the study, respondents completed the SDO and the RWA scales using a 7-point Likert scale

    ranging from do not agree at all to agree completely. Using a back translation procedure, the SDO

    scale (Pratto et al., 1994) was translated from English to French. Previous studies using the 16-item

    English language SDO scale have found it to be both reliable and valid (Pratto et al., 1994; Sidanius &

    Pratto, 1999). In the present study only the eight-item hierarchy-enhancing (i.e. group-based

    dominance) subscale was used given our interest in beliefs pertaining to group inequality and

    intergroup derogation. Doing so was justified on the basis of evidence from confirmatory factor

    analyses suggesting a factorial distinction between hierarchy-enhancing and hierarchy-attenuating

    (i.e. equality) components of the SDO scale (Jost & Thompson, 2000). In the present study, thereliability of the French language SDO hierarchy-enhancing subscale was acceptable (Cronbach

    alpha 0.78). Authoritarianism was assessed using a 16-item French-language version of the RWA

    scale (RWA; Altemeyer, 1998). For computation purposes, negatively-worded items were reversed so

    that all items indicate greater authoritarianism. The reliability obtained for this French language RWA

    scale was acceptable (Cronbach alpha 0.72).

    Tajfel Matrices

    Adapted Tajfel matrices were used to assess group members distributions of positive and negative

    outcomes in the MGP experiment. Although other matrices have been used to monitor resource

    distribution strategies (e.g. Gaertner & Insko, 2001, Study 2), the present study adopted the Tajfel

    matrices as they have been found to tap peoples social orientations in a valid, reliable and sensitive

    manner (Bourhis & Gagnon, 2001; Bourhis, Gagnon, & Sachdev, 1997; Brown, Tajfel, & Turner,

    1980; Diehl, 1990; Messick & Mackie, 1989; Turner, 1980, 1983). Four basic strategies are measured

    using the Tajfel matrices (for more complete information on the use and scoring of the Tajfel matrices,

    see Bourhis et al., 1994). Parity (P) represents a choice that awards an equal amount of outcomes,

    Ideological beliefs and discrimination 587

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    8/18

    whether they be positive or negative, to in-group and out-group recipients. Maximum in-group profit

    (MIP) represents a choice that awards the highest absolute amount of positive outcomes to in-group

    recipients regardless of awards made to out-group others. When distributing negative outcomes,

    opting for this strategy involves awarding the minimal amount of negative outcomes to in-group

    recipients regardless of negative outcomes distributed to out-group others. When distributing positive

    outcomes, maximum differentiation (MD) is a discrimination strategy that maximizes the difference in

    outcomes awarded to two recipients, the difference being in favour of the in-group member but at thecost of sacrificing absolute in-group profit. When distributing negative outcomes, MD is a strategy that

    maximizes the difference of negative outcomes distributed to the two recipients, the difference being

    in favour of the in-group member but at the cost of sustaining greater negative outcomes to the in-

    group. When used in positive outcome distributions, maximum joint profit (MJP) is a choice that

    maximizes the total amount of outcomes distributed to both in-group and out-group recipients. When

    used in negative outcome distributions, the MJP strategy minimizes the total amount of negative

    outcomes distributed to both in-group and out-group recipients.

    In this MGP experiment, three matrix types designed to assess discriminatory strategies were used.

    The first matrix type compared the strength of in-group favouritism (FAV) on P: FAVon P. FAV is made

    up of the strategies of MIP and MD: FAV MIPMD. The second matrix type pitted FAV against

    MJP: FAV on MJP. The third matrix compared MD against combined MIP and MJP: MD on

    MIPMJP. Each matrix type was presented once in its original form (in-group/out-group) andonce in its reverse form (out-group/in-group) to obtain pull scores. Each of these strategies was used in

    both positive and negative outcome distributions. In total, this resulted in six matrix presentations for

    the positive outcome distributions and another six matrix presentations for the negative outcome

    distributions. In the present study, each pull score had a theoretical range from 6 to 6. The order of

    presentation of each matrix within the booklet was randomized for each respondent. The Cronbach

    alphas obtained for these three discrimination matrices were satisfactory, both when positive (0.83)

    and when negative outcomes (0.80) were distributed.

    Given that positive and negative outcomes should have similar psychological weights, care was

    taken to achieve economic equivalence of salary distributions and hours of unpaid work distributions.

    Equivalence was achieved using national statistics pertaining to the average pay earned, as well as the

    average number of hours worked by employees whose education profile matched that of the

    respondents who took part in the experiment (Statistiques Canada, 1999). Based on these statistics,the Tajfel matrices were constructed so as to match salary matrices with hours of work matrices.

    Participants also completed a zero-sum distribution where outcomes were distributed to an in-group

    opposed to an out-group member. The zero-sum task completed by participants in the salary condition

    involved distributing either a salary increase or a salary cut of $100 in weekly pay, while participants

    in the hours of work condition distributed either a decrease or an increase of 10 hours in their weekly

    work schedule. Hours of work distributions were rescaled so as to match the 100-point scale used for

    salary distributions, and negative outcome distributions were reversed so that higher amounts

    indicated greater bias in favour of an in-group member as opposed to an out-group member.

    Quality of In-group Identification

    At the very end of the MGP experiment, respondents completed the quality of social identity scale

    using a 7-point Likert scale ranging from not at all to completely. As in previous MGP studies, this

    scale was composed of items measuring the extent to which respondents liked being a member of their

    own group, and how much they felt at ease and secure as a member of their own group (Gagnon &

    Bourhis, 1996; Perreault & Bourhis, 1998). Reliability for this three-item scale was acceptable

    (Cronbach alpha 0.91).

    588 Catherine E. Amiot and Richard Y. Bourhis

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    9/18

    Parcelling of the Items

    Parcels of items were formed on the ideological belief scale of authoritarianism and on the group-

    based dominance subscale of the SDO scale (Little, Cunningham, Shahar, & Widaman, 2002). Several

    reasons underlie the use of parcels for these instruments in the present study. First, in comparison to

    individual items, parcels present advantageous psychometric properties. Second, models based on

    parcelled data are more parsimonious, have fewer chances for residuals to be correlated, and lead toreductions in various sources of sampling error. Third, parcels can be recommended when the number

    of items included in a full model is high, thus allowing for a more optimal variable to sample size ratio

    and yielding more stable parameter estimates (see Duckitt et al., 2002, for an application with

    ideological beliefs measures). Finally, the present study focused primarily on the relations among

    latent variables rather than on the exact relations among the individual items.

    With respect to the authoritarianism measure, the procedure taken to regroup items into parcels

    involved including, within the same parcel, an equal number of items of different directions of

    wording. This procedure reduced systematic error due to method effects such as direction of wording

    (Altemeyer, 1988). Each parcel contained either two items (in the case of the group-based dominance

    subscale of the SDO scale) or four items (in the case of the authoritarianism scale).

    RESULTS

    Valence, Stimulus Type and Gender Effects on Resource Distributions in the MGP

    An outcome valence by stimulus type by order by sex Multivariate Analysis of Variance (MANOVA)

    was first conducted to test the effect of the three independent variables composing our experimental

    design and of sex on the discrimination matrices and the zero-sum distribution in the MGP part 2

    study. Results of this four-way repeated-measures MANOVA did not reveal any significant main or

    interaction effect for either order or sex (F ratios ranged from 0.46 to 2.01). The effects obtained for

    valence of outcome were not significant. The only significant effect to emerge was the stimulus main

    effect. In light of these results, and given the importance of the valence of outcome variable, anoutcome valence by stimulus type repeated-measures MANOVA was next conducted on the four

    discrimination measures. As can be seen in Table 1, the main effect for valence was not significant

    (F(4, 274) 1.10), suggesting the lack of a PNAE in the present study. While the valence by stimulus

    interaction was also non-significant (F(4, 274) 0.47), the main effect for stimulus type was

    significant (F(4, 274) 3.59, p< 0.01). At the univariate level, stimulus had a significant effect on

    each of the four discrimination measures, revealing a greater use of these strategies when salaries were

    distributed than when hours of unpaid work were distributed. Further interpretations for the lack of the

    PNAE and for the emergence of the stimulus main effect are presented in the discussion section.1

    The mean pull score obtained for each discrimination measure in each experimental condition was

    tested for significance. One-sample t-tests (two-tailed) comparing each pull scores on the Tajfel

    1Given the significant effect of stimulus type on discriminatory behaviour, hierarchical moderated multiple regressions(HMMRs) were conducted to test for the potential moderating role of stimulus type in each of the six associations tested inthe SEM model. To do so, HMMRs were conducted on each of the three dependent variables (i.e. positive outcomediscrimination, negative outcome discrimination, quality of social identity) (Aiken & West, 1991). Results revealed nosignificant interaction effect. However, stimulus type had a significant main effect in two of the regressions, revealing morepositive as well as more negative outcome discrimination when salary than when hours of unpaid work were distributed. Theseresults replicate those obtained in the MANOVA.

    Ideological beliefs and discrimination 589

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    10/18

    matrices from zero, and each zero-sum distribution from 50 (i.e. a score of P) were conducted to

    determine if these discriminatory strategies were used beyond chance level by participants. In nearly

    all experimental conditions, participants made significant use of each of the four discrimination

    strategies. Given that group members discriminated significantly on both positive and negative

    outcome distributions, we proceeded to investigate whether ideological beliefs could predict the use

    of these discriminatory strategies as well as for how discrimination on positive vs. negative outcomedistributions could contribute to positive social identity. These analyses were conducted using

    structural equation modelling (SEM).

    Descriptive Statistics for the Variables Included in the SEM Model

    Table 2 presents the correlations between computed aggregates of the manifest variables included in

    the SEM model as well as the means, standard deviations, skewness and kurtosis values for those

    computed variables. Univariate normality proved satisfactory and correlations among the variables

    suggest the absence of multicollinearity. The statistics for the entire set of manifest variables used in

    the model are available from the authors upon request.

    Structural Equation Modelling (SEM)

    Version 5.7 of the EQS/Windows program was used to test for the validity of the hypothesized SEM

    model. Doing so first involved the conduct of confirmatory factor analyses on each of the instruments

    Table 1. Mean pull scores of participants distribution strategies as a function of outcome valence and stimulus

    Outcome Stimulus Outcomevalence valence by

    Outcome valence stimulus

    Positive Negative

    Stimulus

    Salary Cuts in hours Salary Increases in Multivariate testsincreases of work cuts hours of work (4, 274) F

    n 133 n 146 1.10 3.59** 0.47

    Univariate tests (1, 277) F

    FAV on MJP 0.83*** 0.24* 0.76*** 0.29* 0.02 13.25*** 0.36MD on MIPMJP 0.79*** 0.29* 0.61*** 0.15 1.38 8.71** 0.03FAV on P 0.70*** 0.26* 0.71*** 0.46*** 1.00 6.08* 0.84Zero-sum distribution 54.55*** 51.23y 53.55*** 51.48* 0.43 9.43** 1.18

    Note: Tajfel matrices range from 6 to 6. Zero-sum distribution: Zero-sum distribution of either $100 or 10 hours of work/week distributed between an in-group and an out-group member. In order to make salary and hours of work comparable onzero-sum distributions, hours of work were multiplied by 10. To allow comparison between positive and negative outcomes,negative outcome distributions were inversed. Therefore, zero-sum scores represent bias in favour of the in-group. On the zero-sum scores, a score of 50 represents P (no bias), while scores departing from 50 and increasing represent a correspondingincrease in in-group bias.***p

  • 8/3/2019 Amiot e Bouris, 2005

    11/18

    included in the model. All fit indices and parameters obtained in these confirmatory factor analyses

    (CFAs) were acceptable. Detailed results of these analyses are available from the first author upon

    request.

    The maximum likelihood estimation procedure was used to estimate model fit. Given the diversity

    of the fit indices available, and in view of the controversies concerning measures of overall fit (Byrne,

    1994), several fit indices will be reported. As for the chi-square ratio, it is desirable to obtain a chi-

    square that is not significant. But when sample size is large, the chi-square can be significant even

    when the difference between the hypothesized and observed covariance matrices is minimal. While it

    is generally not valid to use the chi-square ratio to test the null hypothesis of the overall fit, it will be

    reported here for sake of convention as well as to allow the comparison of different models. The

    standardized RMR represents the average difference between the sample variances and covariances

    and the estimated population variances and covariances. The standardized Root Mean Square Residual(RMR) has a range of 0 to 1, where values less than 0.10 are desired. The Root Mean Squared Error of

    Approximation (RMSEA) assesses the estimated discrepancy, per degree of freedom, between the

    population covariance matrix and the model. Whereas the RMSEA value would be 0 if the fit was

    perfect, values smaller than 0.05 indicate a good fit, and values smaller than 0.08 represent a

    reasonable fit. The Comparative Fit Index (CFI) compares the chi-square of the estimated model with

    the chi-square of a null model. The values of the CFI range from zero to one, with a CFI value

    greater than 0.90 serving as the conventional lower cut-off of acceptable fit of the model to the data.

    Finally, the parsimony fit index (PCFI) adjusts the CFI to take complexity of the model into account

    and varies from 0 to 1. Values equal to or higher than 0.50 are considered acceptable.

    The Lagrange Multiplier Test (LM Test) was also used as a guide in identifying parameters initially

    constrained to zero that could contribute most to a significantly better fitting model if these parameters

    were freely estimated. Post-hoc model fitting was considered appropriate only when there was sound

    statistical, theoretical, and empirical justification to do so (Byrne, 1994). To assess the extent to which

    a respecified model represents an improvement in fit, the differences in 2 can be used to compare

    models. This variation is 2-distributed, with degrees of freedom equal to the difference in degrees of

    freedom of the two models. A significant difference in 2 indicates a substantial improvement in

    model fit. Finally, the Wald Test was used to identify non-significant parameters.

    Table 2. Correlation matrix, means, standard deviations, skewness, and kurtosis for the computed variables ofthe model

    Variables Mean SD SK KU 1 2 3 4 5 6

    1. Social dominance (SDO) 2.23 0.92 0.75 0.26 2. Authoritarianism (RWA) 3.74 0.68 0.17 0.06 0.22 Discrimination: Positive outcomes

    3. Discrimination Tajfel matrices 0.51 1.31 0.72 2.31 0.16 0.02 4. Zero-sum 52.82 8.87 1.20 3.38 0.16 0.02 0.77 Discrimination: Negative outcomes5. Discrimination Tajfel matrices 0.49 1.37 0.80 2.92 0.12 0.04 0.37 0.38 6. Zero-sum 52.47 8.84 1.13 3.61 0.11 0.05 0.35 0.41 0.74 7. Positive social identity 4.46 1 .36 0.61 0.09 0.05 0.01 0.19 0.15 0.10 0.04

    Note: Correlation coefficients above 0.12 are significant at p

  • 8/3/2019 Amiot e Bouris, 2005

    12/18

    The Hypothesized Model

    The hypothesized full model to be tested is presented in Figure 1. In line with the ideological beliefs

    hypothesis, it was anticipated that SDO and RWA would each predict discrimination on both positive

    and negative outcome distributions. The two ideological beliefs were also hypothesized to correlate

    with one another. The perceptual differentiation hypothesis predicted positive links between

    discrimination on positive and negative outcome distributions and positive social identity. Althoughthe discrimination valence hypothesis predicted the absence of a significant relationship between

    discrimination on negative outcome and positive social identity, this relation was nevertheless tested in

    the model to determine its exact strength. Finally, the residuals representing the error of prediction of

    all regression equations were hypothesized to be significant, and the covariance between these

    residuals was hypothesized to be zero.

    Goodness-of-fit values were as follows: 2 (145) 314.60, p< 0.001, CFI 0.929, RMR 0.099,

    RMSEA 0.065 (confidence interval 0.0550.074), PCFI 0.788. Examination of the multivariate

    LM 2 coefficients revealed improvement in model fit to be gained from the additional specification of a

    structural path between the two discrimination factors or by adding cross-loadings between parcels or

    items and factors to which they did not originally belong. Because of the already adequate fit of the model

    to the data, and because of the lack of theoretical and empirical evidence supporting the modifications

    proposed by the LM Test, it was decided not to add any parameters to the model. Doing so also enhancesthe potential replicability of the present model and minimizes the problem of capitalization on chance.

    Analyses of power within the context of analysis of covariance structure revealed adequate levels of

    power for the model tested (power 0.999; MacCallum, Browne, & Sugawara, 1996).

    Although presenting an acceptable fit to the data, this model presented three non-significant

    parameters, which were also identified by the Wald Test. As can be seen on Figure 1, authoritarianism

    was not significantly related to discriminatory behaviour on positive or negative outcome distributions.

    Only SDO predicted both positive and negative outcome discrimination, thus providing partial support

    for the ideological beliefs hypothesis. Also, the association between discrimination on negative

    outcome distribution and positive social identity was not significant, thus providing support for the

    discrimination valence hypothesis rather than for the perceptual difference hypothesis. A final model,

    which excluded these non-significant associations, was thus tested. Goodness-of-fit values for this final

    model were as follows: 2 (148) 314.97, p< 0.001, CFI 0.930, RMR 0.100, RMSEA 0.064(confidence interval 0.0540.073), PCFI 0.805. Although more parsimonious, this final model did

    not differ from the hypothesized model (2(3) 0.37, n.s.). The Wald Test did not reveal any non-

    significant associations in this final model.

    DISCUSSION

    The MGP results obtained in this study showed that group members discriminated as much on

    negative as on positive outcome distributions. By evoking the need to achieve a successful corporate

    merger within a competitive economic market, the present study provided participants with a plausible

    ideological justification for discriminating on both positive and negative outcomes (Amiot & Bourhis,

    2003). The profit motive guiding successful mergers may have made negative outcome distributions

    appear more acceptable and meaningful for group members in our study, thus eliminating the usual

    PNAE obtained by the Mummendey research group. As proposed by Mummendey and Otten (1998),

    discrimination involving negative outcomes requires additional aggravating conditions (Otten,

    Mummendey, & Blanz, 1996). In our study, the intergroup context of corporate mergers driven by

    592 Catherine E. Amiot and Richard Y. Bourhis

    Copyright# 2004 John Wiley & Sons, Ltd. Eur. J. Soc. Psychol. 35, 581598 (2005)

  • 8/3/2019 Amiot e Bouris, 2005

    13/18

    Figure1.

    Structu

    ralequationmodeltestingthehypothe

    sizedassociationsbetweenideological

    beliefs,positiveandnegativediscrimination,andqualityof

    socialidentification;*p