INGROUP BIAS AND SELF-ESTEEM: A META ANALYSIS
BY
CHRISTOPHER L. ABERSON
A Dissertation submitted to the Faculty of Claremont Graduate University in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Graduate Faculty of Psychology
Claremont, California 1999
Approved by:
______
Dr. Amy Marcus-Newhall © Copyright by Christopher L. Aberson, 1999 All Rights Reserved. We, the undersigned, certify that we have read this dissertation and approve it as adequate in scope and quality for the degree of Doctor of Philosophy.
Dissertation Committee:
______
Dr. Amy Marcus-Newhall, Chair
______
Dr. Dale E. Berger, Member
______
Dr. Ximena B. Arriaga, Member
______
Dr. Michele A. Wittig, Visiting Examiner Abstract of the Dissertation
Ingroup Bias and Self-Esteem: A Meta Analysis
by
Christopher L. Aberson
Claremont Graduate University, 1999
Social Identity Theory contains two seemingly incompatible predictions regarding the relationship between self-esteem and ingroup bias. The first focuses on low self- esteem as motivation for bias, predicting that low self-esteem individuals exhibit more ingroup bias. The second posits that high self-esteem results from exhibiting bias, thus, high self-esteem individuals exhibit greater bias.
A meta analysis examined the relationship between self-esteem and ingroup bias.
Additionally, the project examined methodological issues such as the lack of consistency in measurement of self-esteem, artificial dichotomization of self-esteem scores, classification of individuals as low self-esteem, and theoretical considerations such as the use of different ingroup bias strategies and the role of social category salience. Thirty- four studies yielding 102 effect sizes from 6660 subjects were included in the analysis.
Results indicated a consistent pattern whereby high self-esteem individuals exhibited more ingroup bias than did individuals with low self-esteem. However, this result was moderated by ingroup bias strategy. When using ingroup bias strategies that required ratings of ingroup superiority, high self-esteem individuals showed more ingroup bias than individuals with low self-esteem. However, when using "indirect" strategies, such as rating groups that the individual did not contribute to, differences between low and high self-esteem individuals were not found. This result leads to the conclusion that both groups exhibit ingroup bias; however, individuals with low self- esteem are limited in the types of bias they exhibit. The pattern of results held for all self-esteem measures except for the Collective Self-Esteem Scale (CSES). No differences between low and high self-esteem individuals were found when scores on the
CSES defined self-esteem. These results may however be an artifact of interactions with social identity salience.
Methodological shortcomings were found in the definition of low self-esteem.
Individuals were most commonly classified as "low self-esteem" based on median splits of self-esteem scores. This strategy resulted in classification of some individuals as low self-esteem despite relatively high self-esteem scores.
Results of the current analysis are interpreted as indicating that individual level phenomena such as self-esteem may be predictive of group level behaviors. Implications for social identity theories are discussed. Acknowledgments
Many people deserve acknowledgment for their roles in the completion of this project. First, I would like to acknowledge my committee and the faculty who have provided invaluable assistance. I want to thank Amy Marcus-Newhall and Ximena
Arriaga for their comments and more importantly, their generosity with their time.
Michele Wittig deserves thanks for her continued support and for the assistance she has provided long after any official commitment to me or the university had ceased. Dale
Berger deserves my heartfelt thanks for his friendship and mentoring. I cannot image having completed this project or the degree program without Dale. I will truly miss our regular meetings.
Official thanks go to the Dora and Randolph Haynes Foundation for their financial support of this dissertation and to the Claremont Graduate University's
Department of Psychology for Fellowship support. Also deserving of thanks are the individuals who provided information, encouragement, assistance, and, in some cases, data for the project at hand. Karen Long, Dominic Abrams, Miles Hewstone, Michael
Hogg, Caroline Seta, and Sean McCrea all contributed in this manner. I commend each for their collegiality. My coders, Michael Healy and Victoria Romero deserve thanks for their diligent work for little pay.
Of course, thanks go to my entire family. I specifically want to thank my
Grandparents Ernest and Hedy Schlosser for supporting my graduate career and always being very proud of me.
I have made several close friends during my stay at CGU. I want to thank Rob
Short, Lisa Magaña, Eric Emerson, Susan Kelso, and Dave Nalbone for their love and
vi support. The relationships I have developed in Claremont are something I hope to keep with me for the rest of my life.
Finally, acknowledgments would not be complete without thanking Nanda Prato and the Prato family. Nanda has loved, supported, assisted, and tolerated me throughout this difficult process.
vii TABLE OF CONTENTS
INGROUP BIAS AND SELF-ESTEEM: A META ANALYSIS...... i
CHRISTOPHER L. ABERSON...... i
Claremont, California...... i
Abstract of the Dissertation...... iv
Acknowledgments...... vi
TABLE OF CONTENTS...... viii
TABLE OF FIGURES...... xv
CHAPTER ONE: INTRODUCTION...... 1
Figure 1. Corollary 1: High self-esteem results from ingroup bias...... 2
The Role of Self-Esteem...... 4
Perspective One: Low Self-Esteem Leads to Greater Ingroup Bias...... 6
Perspective Two: High Self-Esteem Leads to Greater Ingroup Bias...... 7
Critiques...... 10
Conclusion...... 20
The Current Project...... 20
Defining the Project...... 21
CHAPTER TWO: METHODS...... 24
Overview...... 24
Collection and Screening of Literature...... 24
Coding of Study Characteristics...... 29
Effect Size Derivation...... 32
viii Data Dependency Assumptions...... 37
Data Analysis Strategies...... 39
Data Analysis (Testing Hypotheses)...... 41
Adjusting for Low Reliability and Other Artifacts...... 42
A Note on Effect Size Estimates...... 45
CHAPTER THREE: RESULTS...... 47
Descriptive Information...... 47
Definition of Low Self-Esteem...... 51
Ingroup Bias...... 54
Analysis of Data and Tests of Hypotheses...... 56
Study Characteristics, Effect Sizes, and Confidence Limits...... 57
Table 2 (Continued)...... 61
Correlation Between Salience and Effect Size Within Cells and Overall...... 66
Direct Bias...... 66
Indirect Bias...... 66
Overall...... 66
Collective...... 66
Self-Esteem...... 66
(-0.42 to 0.46)...... 66
(0.21 to 0.90)...... 66
(-0.19 to 0.57)...... 66
Rosenberg's...... 66
Self-Esteem...... 66
ix (-0.51 to 0.37)...... 66
(-.26 to .33)...... 66
All Other...... 66
(-0.73 to 0.08)...... 66
(-0.49 to 0.98)...... 66
(-0.61 to 0.17)...... 66
Overall...... 66
(-0.28 to 0.22)...... 66
(-0.11 to 0.50)...... 66
(-0.13 to 0.26)...... 66
Table 5...... 70
Median Effect Sizes by Salience Level, Self-Esteem Measure and Dependent Measure
Type...... 70
Analysis of Cell Means...... 71
Table 6...... 71
Mean Effect Sizes Within Cells and Overall...... 71
Direct Bias...... 71
Indirect Bias...... 71
Overall...... 71
Collective...... 71
Self-Esteem...... 72
(0.00 to 0.20)...... 72
(-0.13 to 0.13)...... 72
x (-0.02 to 0.14)...... 72
Rosenberg's...... 72
Self-Esteem...... 72
(-0.03 to 0.18)...... 72
(0.24 to 0.36)...... 72
All Other...... 72
(0.40 to 0.54)...... 72
(-0.04 to 0.31)...... 72
(0.36 to 0.49)...... 72
Overall...... 72
(0.33 to 0.42)...... 72
(-0.01 to 0.14)...... 72
(0.25 to 0.32)...... 72
Sub Analyses...... 73
Table 7...... 74
Correlation Between Salience and Effect Size Within Cells and Overall...... 80
Direct Bias...... 80
Indirect Bias...... 80
Collective...... 80
Self-Esteem...... 80
Reduced Data...... 80
-0.48...... 80 n = 5...... 80
xi (-0.96 to 0.69)...... 80
0.71...... 80 n = 5...... 80
(-0.46 to 0.97)...... 80
(-0.42 to 0.46)...... 80
(0.21 to 0.90)...... 80
Rosenberg's...... 80
Self-Esteem...... 80
Reduced Data...... 80
-0.03...... 80 n = 13...... 80
-0.57 to 0.53...... 80
(-0.51 to 0.37)...... 80
All Other...... 80
Reduced Data...... 80
-0.59*...... 80 n = 12...... 80
(-0.02 to -0.87)...... 80
0.75...... 80 n = 4...... 80
(-0.39 to 0.98)...... 80
(-0.73 to 0.08)...... 80
(-0.49 to 0.98)...... 80
xii Effectiveness of Bias Strategy...... 82
Analysis of Personal vs. Collective Measures of Self-Esteem...... 82
Additional Issues...... 85
CHAPTER FOUR: DISCUSSION AND CONCLUSIONS...... 89
Summary of Findings...... 89
Self-Esteem Measurement...... 95
Social Category Salience...... 99
Bias Strategies (Dependent Variable Type)...... 100
Implications for Theory...... 105
Future Research Directions...... 112
Concluding Comments...... 113
References...... 116
DC...... 125
Appendix A: Meta Analysis Codebook -- Study Qualification...... 132
STUDY NUMBER ______...... 132
Form 2: Document Characteristics...... 133
2) Name of First Author (Last, First) ______...... 133
Dependent Variables...... 135
Appendix B: Self-Esteem and Ingroup Bias -- Subjective Items...... 136
Fill out for each comparison – note, each question refers to the specific comparison at hand...... 136
Study Number ______...... 136
Section 1: Very Subjective Items...... 136
xiii This section is less subjective:...... 139
Appendix C: Objective Measures -- Methods and Dependent Measures...... 141
Methodological Variables...... 141
Complete for each comparison...... 141
Dependent Variables...... 143
A. Direct/Indirect? -- Why ______...... 143
Post-test Information...... 143
LOW SELF-ESTEEM GROUP (Skip if no data)...... 146
Note: Fill in as much as you can even if some calculation is required...... 146
HIGH SELF-ESTEEM GROUP (Skip if no data)...... 148
Note: Fill in as much as you can even if some calculation is required...... 148
Ingroup Ratings...... 150
2. No...... 152
Outgroup Ratings...... 153
Difference Ratings...... 154
xiv TABLE OF FIGURES
Figure 1. Corollary 1: High self-esteem results from ingroup bias...... 2 Figure 2. Corollary 2: Low self-esteem leads to ingroup bias...... 3 Figure 3. Ingroup bias integrating self-consistency and self-esteem...... 13 Figure 4: Correlation of effect size and salience by cell...... 67 Figure 5. Funnel plot: Effect size by sample size...... 86
xv CHAPTER ONE: INTRODUCTION
Social Identity Theory states that individuals define themselves in terms of their group memberships and seek to maintain a positive identity through association with positively valued groups and through comparisons with other groups (Tajfel, 1982; Tajfel
& Turner, 1979). In intergroup settings, individuals adopt comparison strategies that enhance differences between the groups in ways that favor the ingroup (Brewer &
Kramer, 1985). Social Identity Theory argues that the desire to maintain positive social identity leads to evaluations that bolster ingroups as a strategy to enhance or maintain self-esteem. The desire for a positive self-concept is believed to drive the need to evaluate one’s group positively in relation to other groups. The tendency to evaluate one's own groups more positively in relation to other groups is termed ingroup bias.
Hogg and Abrams (1990), in a critique of Social Identity Theory, highlighted two corollaries regarding the relationship of ingroup bias (outcomes) and self-esteem
(motivations/outcomes). The first corollary, represented by Figure 1, states that successful intergroup discrimination enhances self-esteem. The second corollary, depicted in Figure 2, argues that depressed self-esteem promotes ingroup bias. Both corollaries point toward a central role for self-esteem in Social Identity Theory.
However, the role of self-esteem is ill defined. Self-esteem hypotheses indicate self- esteem to be an outcome (Corollary 1) and a predictor (Corollary 2) of ingroup bias. A recent literature review supports the proposition that successful discrimination enhances certain dimensions of self-esteem (Rubin & Hewstone, 1998). However, the importance
1 of self-esteem as a predictor of ingroup bias is still in question. This project focuses on the role of self-esteem as a predictor of ingroup bias.
Behavior
Low Negative No Bias Self-Esteem Identity
Positive High Ingroup Bias Self-Esteem Identity
Biases enhances identity
Figure 1. Corollary 1: High self-esteem results from ingroup bias.
2 Behavior Outcome
Low Self-Esteem Ingroup Positive Bias Identity
High Self-Esteem No Bias Positive Identity
Figure 2. Corollary 2: Low self-esteem leads to ingroup bias.
In conceptualizing self-esteem as a predictor of ingroup bias, a fundamental issue is whether self-esteem is positively or negatively (or not at all) correlated with ingroup bias. The question most commonly asked by researchers is “who shows more bias, low or high self-esteem individuals?” Hogg and Abrams' (1990) statements argue that depressed self-esteem leads to greater motivation for ingroup bias. If low self-esteem motivates ingroup bias, then low self-esteem individuals should be more likely than individuals with high self-esteem to exhibit ingroup bias to make up for deficient self- concepts. Thus, predictions from this perspective argue for a negative correlation between self-esteem and ingroup bias. However, others argue (e.g., Luhtanen & Crocker,
1991) if ingroup bias produces positive self-esteem then those who exhibit the most bias should have the highest self-esteem, indicating a positive correlation between self-esteem and ingroup bias.
3 The research reviewed in this paper attempts to clarify the role of self-esteem and its relation to ingroup bias. Specifically, several perspectives focusing on the role of self- esteem are examined, methodological critiques and additional theoretical considerations are discussed.
The Role of Self-Esteem
In this section we will examine the relation of self-esteem to ingroup bias. Prior to this discussion, however, it is useful to draw distinctions between different dimensions of self-esteem, to clarify definitions, and to address status issues.
Global vs. domain specific self-esteem. One way to classify self-esteem measures is as global or domain-specific. Global self-esteem refers to stable aspects of self- concept. Domain-specific self-esteem comprises specific aspects of self-concept (e.g., math ability, athletic ability). Domain-specific esteem may covary with performances on related dimensions. For example, doing well on a mathematics examination in comparison to others could produce changes in math related self-esteem. However, it is unlikely that performance on a math examination will produce demonstrable changes in global self-esteem, as global esteem is the product of a lifetime of experiences. The belief that global self-esteem is consistent over time is supported by test-retest reliability data from the major measures of global self-esteem (e.g., Rosenberg Self-Esteem Scale,
Janis-Fields Feelings of Inadequacy Scale), all of which indicate strong reliability over time (Blascovich & Tomaka, 1991).
Recent intergroup studies draw distinctions between these two types of self- esteem (Hunter, Platow, Bell, Kypri, & Lewis, 1997; Hunter, Platow, Howard, &
4 Stringer, 1996). These studies found support for Corollary 1 (self-esteem is bolstered by ingroup bias) when examining pretest to posttest changes in domain-specific self-esteem following exhibition of ingroup bias. However, appreciable changes in global self- esteem were not found following intergroup discrimination. This distinction may be of use in differentiating between corollaries. Corollary 1, regarding changes in self-esteem, may be best conceptualized as referring to domain-specific self-esteem whereas the self- esteem dimension associated with lower ingroup bias may refer to global self-esteem
(Corollary 2; low self-esteem leads to ingroup bias). The current project examines the role of self-esteem as a predictor of ingroup bias; thus, the focus of this research will be global measures of self-esteem.
Social Identity Theory (Corollary 2) states that individuals with low self-esteem have greater motivation to use ingroup bias as a strategy to raise self-esteem and that the use of ingroup bias results in a positive identity (Abrams & Hogg, 1988; Hogg &
Abrams, 1990).1 Some researchers take this to mean that low self-esteem individuals will act in a more biased manner (e.g., Wills, 1981). Others, however, have argued that if this is the case then it should not be possible for some individuals to be chronically low in self-esteem (e.g., Luhtanen & Crocker, 1991). If people with low self-esteem are motivated to produce bias (Corollary 2) and bias raises self-esteem (Corollary 1), then exhibiting bias should turn low self-esteem individuals into high self-esteem individuals.
If this were the case, there would not be any individuals with chronically depressed self- esteem (Crocker & Luhtanen, 1990). Contrasting views on the role of self-esteem have led to a number of disparate studies that are discussed in the sections that follow.
1 This corollary has received considerable empirical attention. However, it has never been endorsed by the theory’s authors, Henri Tajfel and John C. Turner (Long & Spears, 1997). 5 Self-esteem vs. status. Another useful distinction is between self-esteem and group status. There is a long tradition of investigation of the effects of status on ingroup bias (see Mullen, Brown, & Smith, 1992). Often group status, defined as the relative standing of the group in relation to other groups, is equated with self-esteem. Several studies attempt to test Social Identity Theory corollaries regarding self-esteem through manipulation of status (e.g., Sachdev & Bourhis, 1984, 1985, 1987). However, several studies found that individuals who are members of low status or stigmatized groups are no more or less likely to have low self-esteem than members of high status or non- stigmatized groups (Crocker, Luhtanen, Blaine, & Broadnax, 1994; Crocker & Major,
1989). As such, it may be the case that negative aspects of the social identity (i.e., low status) are disassociated from self-esteem (Steele, 1997). Given these findings, status and self-esteem may be more accurately viewed as independent constructs. The current project focuses exclusively on issues of self-esteem. Sections that follow examine two perspectives on the role of self-esteem in relation to ingroup bias.
Perspective One: Low Self-Esteem Leads to Greater Ingroup Bias
Early theorization (e.g., Ehrlich, 1973) predicted that low self-esteem individuals exhibit greater prejudice than individuals with high self-esteem. According to this view, self-enhancement mechanisms are stronger for low self-esteem individuals. Deficient self-esteem acts as a stressor that prompts coping responses. High self-esteem individuals do not possess similar motivations as their positive self-concepts eliminate the need for coping responses (Wills, 1981, 1991). Low self-esteem individuals need to make up for poor self-concept and therefore they pick on others to raise deficient esteem,
6 whereas high self-esteem individuals do not need to bolster self-esteem (Fiske & Taylor,
1991).
As shown in Figure 2, people with low self-esteem produce positive identities through use of ingroup bias whereas high self-esteem individuals have positive identities and thus do not need to exhibit bias. Reviews of research from this perspective argue that poor self-concept predisposes individuals to racial prejudice (Ashmore & Del Boca,
1976). Those individuals with low self-esteem tend not to like themselves or others.
Thus, Ehlrich (1973) argues a negative attitude toward outgroups results from the generalization of poor attitudes about the self.
Several studies support these claims, reporting consistent, though often low to moderate, correlations between low self-esteem (or poor self-concept) and prejudice across various samples, including British school (Bagley, Verma, Mallick, & Young,
1979) and Hindu and Muslim high school students (Hassan, 1978 cited in Duckitt, 1992).
Longitudinal research indicates strong correlations between increases of self-esteem and positive changes in racial attitudes amongst U.S. high school students (Stephan &
Rosenfield, 1978) and adults (Rubin, 1967). Other studies, however, found only small, nonsignificant correlations between low self-esteem and prejudice (e.g., Ray & Heaven,
1984).
Perspective Two: High Self-Esteem Leads to Greater Ingroup Bias
Contrary to the predictions that low self-esteem individuals will exhibit greater prejudice, several studies found a pattern of bias that is stronger for high self-esteem individuals than for those low in self-esteem. These findings support the perspective that
7 ingroup bias allows high self-esteem individuals to create, bolster, and maintain positive social identities. Low self-esteem individuals have low self-esteem because they do not regularly engage in ingroup bias strategies. Research from this perspective contradicts motivational hypotheses of self-esteem (i.e., low self-esteem produces bias), instead favoring a self-esteem regulation model (i.e., high self-esteem individuals maintain positive self-concept through the use of ingroup bias).
A pair of studies by Crocker and colleagues (1987, 1990) asked participants to complete a social perception test. Participants were given feedback regarding either individual performance or group performance on the test, but not both, and then participants rated groups on a series of traits. Participants who received individual feedback completed personal self-esteem measures whereas those receiving group feedback completed collective self-esteem measures, employing a scale designed to tap the aspects of self-esteem resultant from group memberships (Luhtanen & Crocker,
1992). When rating minimally defined groups (those who had either succeeded or failed on the task), high personal self-esteem participants showed a pattern of ingroup- enhancing social comparisons whereas low personal self-esteem participants did not enhance their group (Crocker, Thompson, McGraw, & Ingerman, 1987). Similarly, when given feedback about group performances and using measures of collective self-esteem, participants high in collective self-esteem made ingroup-enhancing ratings whereas participants low in collective self-esteem did not enhance the ingroup (Crocker &
Luhtanen, 1990).
The authors of these studies concluded that low self-esteem individuals might not exhibit the ingroup serving biases predicted by Tajfel and Turner (1979). When judging 8 the ingroup, low self-esteem participants showed less ingroup bias than did high self- esteem participants. Crocker and Luhtanen (1990) took these results to indicate that
Social Identity Theory is most applicable to individuals with high collective self-esteem
(i.e., those with strong social identities).
Crocker and colleagues (1987, 1990) found individuals with high self-esteem to show greater bias than low self-esteem individuals only when categories were evaluative in nature (e.g., high vs. low scorers on a social intelligence task). However, when examining groups with true minimal boundaries (e.g., groups defined by lottery), they found no differences in biases exhibited by low and high self-esteem individuals.
Further, these results argue for the importance of group identity salience in such research.
Groups that have little meaning are not particularly salient to the individual and may not produce ingroup bias effects. When group membership is important, and thus more like real groups, effects of ingroup bias may be stronger.
Another interpretation of the Crocker studies focuses on the use of multiple group memberships. Participants in both studies rated two sets of ingroups and outgroups. One set was constructed by lottery procedure, whereas the other was a more meaningful grouping of individuals in terms of ostensibly important test results. Farsides (1995) and
Abrams (1993) both argued that predictions regarding ingroup bias are valid only for comparisons that do the most to enhance self-concept. This logic argues that when given opportunities to favor meaningless vs. meaningful ingroups, individuals will more likely favor meaningful ingroups. In the case of the Crocker studies, favoritism toward minimally defined ingroups did not occur when more attractive alternatives, such as favoring a more meaningful group, were available. 9 Whereas the studies by Crocker and her colleagues provide evidence that high self-esteem individuals may show greater ingroup bias, other studies examining naturally occurring social groups yield less consistent results. A study of Irish Catholics and
Protestants revealed no differences between high and low self-esteem individuals in attributional biases (Hunter, Stringer, & Coleman, 1993). A study of Jews and Arabs living in the United States showed little evidence for the influence of self-esteem on ratings of stereotype validity and ethnic humor (Ruttenberg, Zea, & Sigelman, 1996). An examination of South African school children provided moderate support for hypotheses regarding those with high self-esteem, though the dependent measures used in the study do not separate self-ratings from ratings of the ingroup (Robins & Foster, 1994). Another study examined ratings of several Canadian ethnic groups and found greater bias for individuals high in personal self-esteem and greater bias for individuals low in collective self-esteem (Lay, 1992).
Critiques
This section summarizes and extends critiques of the perspectives presented above. Included are a discussion of motivations and consistency needs of low and high self-esteem individuals, a discussion of identity salience, and comments on the measurement of self-esteem.
10 Critique: High and low self-esteem both show ingroup bias. Recent theorization argues for the compatibility of both of the above perspectives. Wills (1991) argued that different perspectives reflect different questions. First, studies finding individuals with high self-esteem to show greater bias through rating themselves or their groups (e.g.,
Crocker & Luhtanen, 1990) ask "Do low self-esteem individuals rate themselves as superior to others?" Logically, individuals who have low self-esteem will not rate themselves as superior to anyone else as low self-esteem consists of poor self-concept in relation to others. Studies that find people with low self-esteem more likely to engage in self-enhancement ask "Are individuals with low self-esteem motivated to enhance self- esteem?"
One primary critique can be leveled against the perspective that high self-esteem individuals exhibit greater ingroup bias. The idea that high self-esteem individuals elevate esteem by showing bias leads to an assumption that low self-esteem individuals are not motivated to favor their groups or themselves. This assumption characterizes low self-esteem individuals as having low self-esteem because they do not show ingroup bias.
High self-esteem people show bias because this is how they create and maintain their high self-esteem. Low self-esteem people do not show bias; if they did they would not have low self-esteem. This perspective does not allow for the possibility of low self- esteem individuals engaging in self-enhancement. Of course, other means to enhance self-esteem do exist. The focus of the current project is, however, exclusively on the relationship between ingroup bias and self-esteem.
Consistent with Wills’ (1991) thinking, low self-esteem individuals may not show ingroup bias because these individuals have a history of association with negative
11 outcomes. Along these lines, Brown (1993) argued that all individuals, including those with low self-esteem, experience a need to self-enhance; however, there also exists a need for self-consistency. As these individuals do not view themselves as superior to others, rating themselves or their ingroups as superior is inconsistent with experiences.
However, other definitions of ingroup bias, for example, rating similarity to successful ingroups, may be favored by those low in self-esteem as this type of measure does not require ratings of superiority (i.e., is not inconsistent with experience). The strategies used by low and high self-esteem individuals must reflect both needs. This is consonant with Hogg and Abrams’ (1990) observation that competing motivational forces (i.e., self- consistency needs) may mitigate the exhibition of ingroup bias. As shown in Figure 3, strategies meeting self-consistency needs allow low self-esteem individuals to exhibit bias and thus, enhance self-esteem, whereas measures incompatible with these requirements will not.
12 Self-Esteem Domain Behavior Outcome
Consistent Ingroup Bias Enhanced Identity
Low Self-Esteem
No Bias No Change in Identity Inconsistent
Consistent Ingroup Bias Enhanced Identity
High Self-Esteem
No bias, unlikely to occur No Change in Identity Inconsistent
Figure 3. Ingroup bias integrating self-consistency and self-esteem.
Empirical evidence. Traditional measures of bias in intergroup research take the form of point allocations or adjective ratings of groups. Ingroup bias is defined as rating the ingroup as superior to an outgroup. This strategy fits with self-consistency needs of high self-esteem individuals, but not for low self-esteem individuals. Low self-esteem individuals experience self-doubt about their abilities in relation to others so it is
13 inconsistent for them to indicate superiority. As low self-esteem individuals must use strategies that satisfy self-enhancement needs and self-consistency needs, traditional measures may ignore the strategies used by those low in self-esteem. Thus, low self- esteem and high self-esteem individuals may both show bias; however, they may do so using different strategies.
Several empirical studies examined the hypothesis that low and high self-esteem individuals use different ingroup bias strategies. A set of studies examined differences between ratings of individuals directly involved with the ingroup to those who are part of the ingroup but not involved with the group. A study by Brown, Collins, and Schmidt
(1988) distinguished between direct and indirect forms of ingroup bias. The authors defined direct bias as the bias shown by individuals comparing a group that they are active participants in to an outgroup. Indirect bias was operationalized as ratings by individuals assigned to the ingroup, but not participating in a group task, as compared to the outgroup. After participating or observing, participants rated the products of a group- brainstorming task. Those individuals high in self-esteem exhibited greater bias when they participated in the ingroup (direct bias), whereas low self-esteem individuals showed greater bias when not involved in the group task (indirect bias).
Three additional studies support these effects. One study examined the role of success versus failure feedback regarding ingroup and outgroup performance in addition to self-esteem and involvement (Seta & Seta, 1992). High self-esteem participants exhibited greater bias favoring the ingroup, but low self-esteem participants did not discriminate between groups. However, low self-esteem observers exhibited a pattern of ingroup bias consistent with the high self-esteem participants. High self-esteem
14 observers did not discriminate in this manner. Another study (Long, Spears, &
Manstead, 1994) allowed study participants to act as both participants and observers in the group task. Again, high self-esteem individuals exhibited bias when they were participants but when they were only observers, whereas low self-esteem individuals exhibited bias as observers but not as participants.
A more recent study using both group and individual performance feedback clarifies this effect. When ingroups were successful and outgroups failed, individual feedback mediated ingroup bias. High self-esteem individuals favored the ingroup regardless of whether they had individually succeeded or failed at the task. However, low self-esteem participants favored the ingroup only when given individual success feedback (Seta & Seta, 1996). One interpretation of these data is that low self-esteem individuals do not favor the ingroup as they feel they are not useful contributors. When they were told they made a valuable contribution (i.e., individual success feedback), they felt their role as an ingroup member was important and thus they favored the group.
These studies indicate one way in which those with low self-esteem can exhibit ingroup bias indirectly through enhancement of the ingroup when they are not directly associated with the group. Other strategies may exist such as basking in reflected glory of a group through enhancement of ratings of association and similarity to the ingroup when it is successful and minimizing association when the ingroup fails (e.g., Cialdini &
DeNicholas, 1989; Mummendy & Schreiber, 1983). One empirical investigation of this hypothesis found low self-esteem individuals to use this strategy when asked to indicate how similar they were to the ingroup and outgroup, whereas high self-esteem individuals did not use this strategy (Aberson, 1999).
15 As the studies discussed above constitute the whole of the literature examining direct and indirect strategies of ingroup bias (with the Seta & Seta studies never mentioning bias strategies), it is difficult to conclude definitively that different strategies of ingroup bias exist for low and high self-esteem individuals. However, these studies do point toward the complexity of the question. High self-esteem individuals seem to exhibit bias consistently. However, it seems certain conditions must be met for low self- esteem individuals to exhibit ingroup bias.
The above research may help to answer the questions posed by Wills (1991). Do individuals with low self-esteem rate selves as superior [to outgroup members]? No, people with low self-esteem do not rate themselves (or their groups) as superior because it is incompatible with self-consistency motivations. Are low self-esteem individuals motivated to self-enhance? Yes, but in manners that differ from individuals high in self- esteem.
Critique: Identity salience, which identity? Most of the studies discussed in this review take an “all-or-none” approach to the concept of group identity salience.
Individuals in groups are expected to view themselves as group members regardless of the nature of categorization or value or importance of the group. However, recent theoretical developments challenge this approach. Self Categorization Theory begins with the proposition that individuals normally function at three major levels of categorical abstraction 1) as Human Beings 2) as Social Group Members (e.g., religion, ethnicity, political orientation, sorority member) or 3) as Individuals (the “I” independent of the “we”) (Turner, 1985; Turner, Hogg, Oakes, Reicher, & Wetherall, 1987). The salience of the category determines whether individuals respond at a group level (show
16 bias for our own group) or at an individual level (do not show bias as group identity is not salient and does not govern responses). Thus, it is only when group membership is salient that individuals act as predicted by Social Identity Theory.
Salience of the group may vary between studies. To create optimal social category salience requires both fit and accessibility (Oakes & Turner, 1986). The best fitting social categories are those in which there exists a similarity in individual characteristics, attitudes, and behaviors that are consistent with the normative context of the categorization at hand. The most accessible categories are those that hold emotional and/or value significance (Oakes, 1987). Salient category memberships result in increased use of stereotypes as descriptors of ingroups and outgroups (Hogg & Turner,
1987; Oakes, Haslam, & Turner, 1994; Oakes, Turner, & Haslam, 1991), social influence
(McGarty, Haslam, Hutchinson, & Turner, 1994; Turner, 1987), and increases in ingroup bias (Turner, et al., 1987).2
Few of the studies reported in this paper examined the salience of category membership; a methodological circumstance that may affect findings as studies may differ greatly in the salience of group categorization. Two of the studies by Crocker and colleagues addressed this issue without explicit reference to salience. Both studies examined responses of participants when categorization was based on a lottery procedure
(i.e., low salience) and based on scoring high or low on a scale designed to measure social competence (i.e., high salience; Crocker & Luhtanen, 1990; Crocker, et al., 1987).
Results indicated that ingroup bias existed only when category salience was strongest.
2 Others (e.g., Gaertner & Schopler, 1998; Hamilton, Sherman, & Lickel, 1998) suggest that salience can be defined exclusively in terms intragroup interaction (e.g., entativity) without reference to intergroup factors. 17 Identity salience likely plays a central role in predicting ingroup bias. Brewer
(1991, 1993) argues, in her optimal distinctiveness theory, that individuals seek to distinguish ingroups from outgroups and to distinguish themselves from their ingroups.
That is, people seek dimensions which allow them to be members of groups that are different from other groups, but they also seek dimensions that allow them to be different from the ingroup. This seeming contradiction is handled nicely through use of the principle of identity salience. When group identity is salient, people respond as group members, seeking to distinguish the ingroup from relevant outgroups. When individual identity is salient, our goal is interpersonal distinction. Simply put, only when group identity is salient are group level goals (positive ingroup distinctiveness) and group level outcomes (ingroup bias) present.
Additionally, little is understood about the interaction of self-esteem and category salience. One possibility is low and high self-esteem individuals differ in the level of category salience necessary to produce ingroup bias. High self-esteem individuals may ignore poor performances in low salience groups, choosing to ignore group membership.
Low self-esteem individuals tend to be more reactive to self-relevant cues in the social environment or more responsive to threatened identity (Campbell & Lavallee, 1993;
Long & Spears, 1997). As such, individuals with low self-esteem may be more likely to interpret poor performances of a group that has little meaning as further evidence of inferiority, resulting in increased motivation to exhibit biases to bolster self-esteem.
Critique: Measurement of self-esteem. Measures of self-esteem used in the studies cited above range from measures of feelings about the self (e.g., Rosenberg Self-
Esteem scale) to feelings of inadequacy (e.g., Janis-Fields Feelings of Inadequacy Scale)
18 to social competence (e.g., Texas Behavioral Inventory) to aspects of identity derived from group memberships (e.g., Collective Self-Esteem scale). As these measures are conceptually distinct (Blascovich & Tamaka, 1991; Luhtanen & Crocker, 1992), there is some question as to the comparability across studies.
More problematic may be the definition of low self-esteem. Many studies employ median or tripartite splits to distinguish between high and low (and sometimes medium) levels of self-esteem. Tice (1993), in a review of self-esteem studies using such splits, concluded that “high scores are high, but low scores are medium, in an absolute sense (p.
40).” Some of the definitions used by the various authors seem likely to place participants with moderate to high self-esteem in the category of low self-esteem. For example, in the Crocker and Luhtanen (1990) study, the authors used a median split to classify participants as high or low self-esteem. Individuals classified as "low self- esteem" averaged a score of 5.5 or lower on four 7-point items. Individuals classified as having low self-esteem could have possibly indicated scores on the upper end of each item.
Research on the role of self-esteem may fail to clarify distinctions between low and high self-esteem. Whereas most methodological flaws serve to weaken effects detected in research, this observation may indicate that actual effects are larger than reported.
Type of self-esteem measured is another issue of note. Luhtanen and Crocker
(1991) argued that the self-esteem referred to by Social Identity Theory reflects esteem gained from social group memberships. As such, measures of personal esteem, such as the Rosenberg and Janis-Field scales, and measures of social competence, such as the
19 Texas Social Behavior Inventory, may not be predictive of collective enhancement, as these measures have no relation to social group membership. As a predictor of collective enhancement, the Collective Self-Esteem Scale may be the most appropriate measure.
Conclusion
Several of the perspectives examined have much to offer research on ingroup bias. However, the role of self-esteem appears oversimplified in much research. The need to enhance the ingroup may be mitigated by other competing motivations (Hogg &
Abrams, 1990). Motives such as self-consistency appear to be important factors in predicting ingroup bias. Salience of group identity and measurement of self-esteem also appear to have considerable import. It is clear that the focus of examination should shift from who shows bias toward consideration of different strategies of bias, and in what circumstances bias is strongest.
The Current Project
Several recent narrative reviews examined self-esteem as a predictor of ingroup bias. All reported inconclusive evidence. Rubin and Hewstone (1998) found little support for Corollary Two (low self-esteem leads to greater ingroup bias). However, the authors concluded that this finding was not surprising given the blurred distinctions regarding measurement of self-esteem, ingroup identification (salience), and different forms of discrimination. Hogg and Abrams (1990) argued that there is little effect for self-esteem and that too much emphasis has been placed upon it. Luhtanen and Crocker
(1992) argued that self-esteem is improperly measured in many studies. Another review
20 by Farsides (1995) argued that self-esteem hypotheses have never been accurately tested due to a lack of consideration of additional predictions made by Social Identity Theory regarding group composition and status. Other papers argued for a perspective that integrates self-consistency needs (Brown et al., 1988).
Clearly, narrative reviews have failed to clarify the link between self-esteem and ingroup bias. The primary value of these reviews has been to provide critiques of empirical studies of ingroup bias and self-esteem. Factors such as measurement of self- esteem (e.g., Luhtanen & Crocker, 1991), social identity salience (Oakes, 1987)3, and consistency between dimension of evaluation and self-concept (Brown et al., 1988), all may contribute to ingroup bias measurement in these studies. In the current study, meta analysis was used to clarify the relationship of these factors to ingroup bias.
Defining the Project
The specific goals of this meta analysis were to investigate the effects of self- esteem, dimensions of evaluation, social identity salience, and self-esteem measurement/self-esteem dimension on ingroup bias. This investigation concentrates on studies that conceptualize self-esteem as a predictor of ingroup bias or studies that include a pretest measure of ingroup bias. Several specific research questions are addressed:
1. Do low and high self-esteem individuals exhibit different amounts of ingroup bias?
2. Do low and high self-esteem individuals use different strategies to exhibit ingroup bias?
3 Oakes did not specifically address self-esteem. Oakes argued that identity salience determines how much ingroup bias occurs. 21 3. Do different measures of self-esteem (e.g., personal vs. collective) lead to different conclusions regarding the role of self-esteem?
4. What is the role of social identity salience?
5. Do the above factors interact?
The above questions provide a focus for this exploratory study. Additionally, several specific predictions are offered.
Hypothesis 1: High and low self-esteem individuals both exhibit ingroup bias.
Hypothesis 2: High self-esteem individuals will exhibit greater bias than low self- esteem individuals on measures of direct bias.
Hypothesis 3: Low self-esteem individuals will exhibit greater bias than high self- esteem individuals on measures of indirect bias.
Direct strategies are dimensions of evaluation requiring claims of ingroup superiority such as point or monetary allocation and adjective ratings. Indirect bias strategies include measures of similarity, group homogeneity, social distance, or ratings of groups that the participant is assigned but to which the individual does not contribute
(i.e., acts as an observer). This hypothesis follows from J. D. Brown et al. (1988) who argued that ingroup bias is moderated by consistency. Ratings of superiority (i.e., direct bias) are inconsistent with the prior experiences of individuals who have low self-esteem.
As such, low self-esteem individuals will exhibit less bias on these measures than individuals with high self-esteem. In contrast, indirect measures of ingroup bias do not require ratings of superiority. Thus, it is predicted that low self-esteem individuals will exhibit greater bias than high self-esteem individuals on indirect measures of bias.
22 Hypothesis 4: As social identity salience rises, ingroup bias will become stronger.
Ingroup bias will be strongest when categories fit and are accessible (Oakes, 1987).
Hypothesis 5: Following from the observation that low self-esteem individuals tend to be more reactive to self-relevant cues in the social environment or more responsive to threatened identity (e.g., Campbell & Lavallee, 1993; Long & Spears,
1997), an interaction is predicted whereby low self-esteem individuals will exhibit bias
(on indirect dimensions) regardless of level of social identity salience. High self-esteem individuals will exhibit ingroup bias (on direct dimensions) only when social identity salience is high. This prediction argues for an interaction between salience and self- esteem in predicting ingroup bias.
23 CHAPTER TWO: METHODS
Overview
This chapter describes methodological procedures used in the meta-analysis. The section details the following: (a) collection and screening of literature, including study inclusion criteria, (b) coding of study characteristics, (c) effect size derivation and calculation, (d) data dependency assumptions, (e) assessment and adjustment of reliability and other artifacts, (f) strategies for hypothesis testing and analysis of effect sizes, and (g) a discussion of effect size issues.
Collection and Screening of Literature
A detailed search of the CD_ROM databases yielded 86 studies for possible inclusion. The literature search began with a search of published materials using
PSYCLIT, PSYCINFO, and SOCIOFILE databases. The initial search used the following keyword/subject heading searches (self-esteem OR group identification) AND
(in(ter)group bias OR prejudice OR social comparison OR group dynamics OR collective behavior).
Initial screening of the 86 studies identified through database searches eliminated
58 of the references. Studies were eliminated if (a) they were duplicated from another search or (b) based on abstracts, had no bearing on the analysis. This left 28 studies for initial inclusion.
24 It was discovered that the PSYCLIT database does not map subject terms consistently. For example, two very similar studies (Brown, et al., 1988; Long, et al,
1994) examined the same concept, couched in the same theory with only minor methodological differences, but PSYCLIT used different subject terms for these two studies. One was identified as a study of ingroup bias, whereas the other was classified as a study of group participation. As such, it was clear that exclusive use of CD-ROM databases would not provide a complete and adequate representation of the research.
As suggested by Reed and Baxter (1994), the study employed other literature search strategies. First, relevant social psychology journals were hand searched. The table of contents was examined for every issue of the two most relevant journals,
European Journal of Social Psychology (the self-proclaimed ‘natural home for social identity research’; Van Avermaet, 1998) and the British Journal of Social Psychology.
Additionally, I examined all 1990-1998 issues of Journal of Personality and Social
Psychology, Personality and Social Psychology Bulletin, Journal of Experimental Social
Psychology, and the Journal of Social Psychology. The hand searches identified 20 additional studies.
To identify additional published research and unpublished studies (a.k.a.
"fugitive literature"), the study employed several strategies recommended by Rosenthal
(1994). First, published and unpublished works were identified through reference section searches of recent review articles and all studies were considered for inclusion in the analysis. Seventeen additional studies were identified in this manner. Next, searches examined recent conference programs from the American Psychological Society, the
American Psychological Association, the Society for Experimental Social Psychology,
25 and the Western Psychological Association. This search yielded eight studies. Another database search, using Dissertation Abstracts, yielded three additional studies. Finally, letters/email were written to seven authors of prominent articles in the areas of self- esteem and/or ingroup bias, requesting unpublished works.4 Several authors provided articles, of which one was included in the meta-analysis.
These searches identified 77 studies for possible inclusion. At this point, specific inclusion/exclusion criteria were created to determine study eligibility. Specific criteria follow:
1. Studies must include ratings of ingroup and outgroup members. Ratings may be of any form as long as ratings are taken for both groups. Seventeen studies were eliminated.
2. Studies must include a measure of self-esteem. Nine studies were eliminated.
3. Studies must include a pre-test measure of self-esteem. In order for self-esteem to be conceptualized as a predictor, a pre-test measure must be taken. Fifteen studies were eliminated.
4. Self-esteem must be global rather than domain-specific. Two studies were eliminated.
Application of these screening criteria identified thirty-four studies for inclusion in the meta analysis.
Rationale for inclusion criteria. This section examines the theoretical and practical rationale for the study inclusion criteria and provides clarification of these criteria. Criterion 1 required that ingroup and outgroup measures were present. Ingroup bias refers to ratings of ingroups in relation to outgroups. This cannot be assessed
4 Authors contacted: D. Abrams, M. Brewer, J. Crocker, M. Hewstone, M. Hogg, J. Jetten, and K. Long. 26 without ratings of both groups. Most studies eliminated based on this criterion only included evaluation of outgroups. As such, it was unclear as to whether bias toward outgroups was a form of ingroup enhancement or whether bias toward outgroups indicated a pattern of negative evaluations of both groups (e.g., Crocker & Schwartz,
1985).
The second criterion required measures of self-esteem. Studies eliminated based on this criterion most commonly manipulated self-esteem through esteem-enhancing or esteem-reducing feedback. Whereas these manipulations may constitute a threat to self- esteem (or a bolstering), these studies were judged inappropriate for the analyses at hand.
The primary rationale for exclusion of these studies is the observation that threatened self-esteem is not the same concept as low self-esteem. Furthermore, as discussed below, it may not be possible to manipulate global self-esteem.
The definition of self-esteem measures was expanded to include measures of group identification. Group identification measures included in this analysis assessed group identification through questions such as "I am a person who considers the group important", "….who feels strong ties with the group", "…who is glad to belong to the group" (Brown, Condor, Mathews, Wade, & Williams, 1986). These items were judged to be similar to the Collective Self-Esteem Scale. The Collective Self-Esteem Scale contains items such as "The social groups I belong to are an important reflection of who I am", "I feel good about the social groups I belong to", and "In general, belonging to social groups is an important part of my self-image." The inclusion of group identification measures as self-esteem measures is supported by a recent factor analysis
27 indicating that group identification measures and Collective Self-Esteem subscales load on similar factors (Jackson & Smith, 1999).
The third criterion required a pre-test measure of self-esteem. Only studies examining effects of self-esteem as a predictor of ingroup bias were included.
Researchers interested in the relationship between self-esteem and ingroup bias have conducted two general classes of studies. The first, which is the focus of the current research, examines the role of self-esteem as a predictor of ingroup bias. The second examines changes in self-esteem as a result of exhibiting ingroup bias. The current project does not include the latter studies. This in no way is meant to indicate that these studies are not worthy of examination, only that they ask different research questions and could constitute another distinct project.
The final criterion stated that studies must include a global self-esteem measure.
Global self-esteem measures tap stable and consistent aspects of the self-concept.
Contrariwise, domain-specific self-esteem refers to specific aspects of the self-concept
(e.g., math ability, athletic ability, etc.). The belief that global self-esteem is consistent over time is supported by test-retest reliability data from the major measures of global self-esteem (e.g., Rosenberg, Janis-Fields, Texas Behavior Inventory, etc.), all of which indicate strong reliability over time (Blascovich & Tomaka, 1991). Recent studies indicate that domain-specific measures of self-esteem are more appropriate for studies that conceptualize self-esteem as a dependent measure (e.g., Hunter, et al., 1996, 1997).
28 Coding of Study Characteristics
Many characteristics of each study were coded. For the sake of brevity, this section discusses only the primary measures used in the analyses that follow. Coded study characteristics fall into three general categories: objective measures, subjective measures, and effect size estimates. A full copy of the meta analysis codebook can be found in Appendices A, B, and C.
Subjective measures: Identity salience. Two coders (not including the author) rated subjective aspects of studies. The subjective nature of these items required interrater reliability measures. The coders were blind to hypotheses involving subjective items and generally unfamiliar with the ingroup bias literature. Subjective measures assessed various aspects of identity salience. Following from Oakes (1987), salience was defined as fit times accessibility (also see Bruner, 1957). Category fit was defined as follows: (Q1) correspondence of categorization to "real" traits held by group members;
(Q2) likelihood to use category outside of experimental setting; (Q8) similarity to ingroup; and (Q9) similarity to outgroup. Accessibility of category membership was defined in the following terms: (Q6) likelihood of using category during experiment and
(Q3) emotional importance of category. The full text of these items can be found in
Appendix B.
Coders received roughly four hours of training. Several studies which did not contain sufficient information to code effect sizes (and were subsequently left out of data analyses) were used for training purposes. The coders were instructed to imagine that they were participants in the study and asked to respond to each question as they would best imagine they would feel when exposed to the research situation. Coders worked on 29 the studies independently and later met, as a group, with the author. Coding of the training studies was discussed and discrepancies clarified through comparison with the author's ratings.
Initial reliabilities, defined as the correlation between rater's salience scores, proved adequate, however reliability was enhanced through several further steps. First, items in which disagreement between coders was high (a difference of three or more points on 7-point scale items) were flagged. The coders then individually discussed the coding rationale for these items with the author. In general, most large coding differences were the result of misreading or misunderstanding the studies rather than actual disagreement. When coding differed due to misunderstanding of experimental conditions, studies were recoded without reference to the original coding. When coders continued to disagree, no recoding was performed. Raters' scores on the revised composite salience measure were highly correlated (r = .80). The average of the two coder's scores comprises the salience measure used in analyses.
The validity of the use of subjectively coded items has been critiqued in the past
(Cialdini & Fultz, 1990). However, a survey of recent literature (e.g., effects Migdal,
Hewstone, & Mullen, 1998; Urban & Miller, 1998) supports the widespread use of such techniques as valuable, valid, and reliable tools for testing theory.
Objective measures. Objective measures, found in Appendix C, are measures taken directly from study documents that required no imputation or judgement. A single coder, the author, coded these measures. Though inter-rater reliability of coded materials is essential to meta-analysis, it was unnecessary for objective measures as no judgement was required. Measures in this section included type of self-esteem scale, type of
30 dependent measure, and effect size estimates (discussed in the next section). To assess the reliability of effect size estimates, a second coder derived effects for a random selection of 14 studies. The 14 studies included 50 effect size estimates. The coder's calculations all corresponded closely with the author's calculations. Some minor differences existed, however, this was attributed to rounding error. All differences were of inconsequential magnitude (i.e., .01).
Self-esteem measurement was assessed through the following questions. Which self-esteem scale was used? How were low and high self-esteem defined? What score defines low/high self-esteem? What is the average score on the scale? What is the scale reliability?
Dependent measures were coded as "direct" or "indirect" based on several criteria.
Direct measures of bias included measures requiring individuals to rate the ingroup as superior through use of adjective rating, point allocation, ratings of group products, and attributions. Indirect measures of bias included dependent measures that did not require ratings of superiority, such as perceptions of similarity (also referred to as "basking in reflected glory") and subtle measures of bias such as linguistic intergroup bias (e.g.,
Maass, Ceccarelli, & Rodin, 1996). Additionally, following from Brown, et al. (1988), comparisons that required participants to rate groups to which they directly contributed
(e.g., helped brainstorm on group project) were coded as direct. Comparisons requiring rating of groups to which participants were explicitly non-contributing members (e.g., observed task but not allowed to participate) were coded as indirect.
31 Effect Size Derivation
The primary goal of the meta analysis was to code comparisons of low vs. high self-esteem individuals. The general strategy for effect size coding was to derive an effect for comparisons of low vs. high self-esteem for the smallest possible codeable units. That is, instead of collapsing across factors, attempts were made to code cell differences within each extraneous factor. This allowed for additional coding of differences of aspects that may affect identity salience (e.g., failure vs. success of the ingroup).
For example, a study may contain a 2x2x2 design, examining self-esteem (high vs. low), ingroup performance (success feedback vs. failure), and outgroup performance feedback (success vs. failure). Instead of ignoring performance factors and coding a single low vs. high self-esteem effect based on scores collapsed across factors, each individual comparison was coded. This yields four comparisons, high self-esteem vs. low self-esteem in each of the following condition combinations; ingroup success -- outgroup success, ingroup success -- outgroup failure, ingroup failure -- outgroup success, ingroup failure -- outgroup failure. Ideally, coding in this manner increases specificity of the analysis.
In cases in which the data did not support these analyses, attempts were made to contact authors. Most authors proved reluctant to provide additional information or ignored requests. In these cases, computations were based on the most complete information available. Coding recorded the amount of estimation required for each effect.
32 Types of effect sizes. Five types of effect size estimates exist in the studies examined: (a) low self-esteem ratings of ingroup vs. outgroup; (b) high self-esteem ratings of ingroup vs. outgroup; (c) correlations between self-esteem and a single variable index of ingroup vs. outgroup rating (i.e., researchers provided a single item difference score for ingroup bias and assessed correlation of this measure with self-esteem); (d) high vs. low self-esteem ratings of the ingroup; and (e) high vs. low self-esteem ratings of the outgroup.
Additionally, an overall effect size, representing the difference in ingroup bias exhibited by high and low self-esteem individuals was calculated. The overall effect size for each comparison was defined in one of two manners, depending on the information available. Overall effect size was based on one of two estimates. For studies providing the correlation between raw scores on a self-esteem scale and a single variable index of ingroup bias, this correlation was converted to an overall effect size. For studies using dichotomized groups, effect size was defined as the (c) or high vs. low self-esteem ratings of the ingroup (d) minus ratings of the outgroup (e). Both procedures produced a single effect size value that was positive when high self-esteem individuals show greater ingroup bias, negative when low self-esteem individuals exhibit greater ingroup bias, and zero when groups showed the same amount of ingroup bias. All analyses use the overall effect size estimate. Effect size types (a) and (b) were not calculable for most comparisons. The section that follows discusses this problem and discusses specific computational strategies.
Computational specifics. Listed below are the strategies used to calculate effect size measures. Strategies are presented for between group and within group tests
33 separately, and strategies are listed from most to least accurate. In all cases, effect sizes were calculated using the most accurate strategy available given data constraints. Effects were calculated using D-STAT, a program for quantitative research synthesis and derivation of effect sizes. The author calculated effect sizes on two separate occasions.
Inconsistencies were addressed through recalculation until consistency was achieved. All formulae are taken from Johnson (1989).
Between group tests. 1. Calculations use descriptive statistics (mean, standard deviation, N's) when possible. This is the most direct manner of calculation, allowing for use of the classic effect size formula:
X X d h l (1) S p
where X h is the mean for the high self-esteem group, X l is the mean for the low self-
esteem group, and S p is the pooled within group standard deviation. The effect size was calculated separately for ratings of the ingroup (Strategy c) and ratings of outgroups
(Strategy d).
2. For studies not reporting standard deviations but providing cell means, mean- square error terms were derived from ANOVA statistics. When reports did not provide F values, effects were reconstructed from appropriate cell means. If an effect was not reported and cell means were not available, it was assumed that F = 1.0 for the effect.
For complex ANOVA statistics, variance and degrees of freedom from other between group factors were returned to the error term (Glass, McGraw, & Smith, 1981;
Johnson, 1989; Johnson & Eagly, in press). For example, a two way ANOVA yielding main effects for self-esteem (SS = 1, df = 1) and group performance (SS = 2, df =1), the 34 interaction (SS = 3, df = 1), and error (SS =10, df = 40, MS = 0.25) would return variance explained by extraneous factors (i.e., not self-esteem) to the error term. In this example, after returning variance, the error term would have SS = 15 (2+3+10) with df = 42
(1+1+40), yielding an mean square error of 0.36. This procedure assures that effects sizes are not inflated due to a reduction in error variance because of removal of variance associated with extraneous factors.
The reconstituted mean square error was used as the error term for contrasts using the formula below. Again, measures were calculated for ratings of both the ingroup and the outgroup. This procedure assumes homogeneity of variance.
X X d h l (2) MSerror
3. For studies using correlation/regression, effects were calculated directly from the r statistic using Formula 3 below. Zero-order correlations were used when available.
Partial correlations and r-square change values were used when zero-order statistics were not presented.
2r d (3) (1 r 2 )1/2
Within group tests. Within group/repeated measures tests present unique problems for meta-analysts. Within group tests for the analysis at hand consist of differences between ratings of the ingroup and outgroup.
1. Calculation from descriptive statistics using the general formula for within group effect sizes:
35 X X d I O (4) SDD
SDD is the standard error of the differences, defined by Formula 5,
X l = the ingroup mean, and X O = the outgroup mean.
2 2 SDD SDI SDO 2rSDI SDO (5)
However, it is rare to find the correlation between ingroup and outgroup ratings
(r) or the standard error of the difference reported. Thus, other techniques must be used.
This problem is prevalent in most reports, an issue discussed later in this paper.
2. Effect sizes were calculated from means and mean square error for within subjects effects or standard error of differences from paired t-test using Formula 2. This strategy also proved ineffective due to difficulties in reconstructing within subjects effects due to incomplete reporting of statistics.
In general, strategies used to calculate within subjects effects were ineffective due to statistically impoverished research reports. Strategies 1 and 2 for within subjects effects proved mostly untenable. Other possible strategies include use of denominators from multiple comparison tests and estimation of correlations between ingroup and outgroup measures (e.g., Dunlap, Cortina, Vaslow, & Burke, 1996; Rosenthal & Rubin,
1986). Unfortunately, adequate data were not provided to allow multiple comparison test denominators to be used. Another possibility was estimation of the correlation between ingroup and outgroup measures. However, this strategy required estimation for the majority of the effect sizes. Due to the large amounts of estimation required, the author judged this strategy inappropriate.
36 Several researchers present alternate strategies for calculating effect sizes based on within subjects ANOVA results (Dunlap, et al., 1996; Morris & DeShon, 1997).
These methods are more accurate than the procedures described above. However, these procedures also require complete statistical reporting (i.e., all F, MS, and r statistics) and the same problems were found as with Strategies 1 and 2.
Data Dependency Assumptions
Meta analysis assumes that effect size estimates are independent. This means that each effect size estimate should be based on different participants (Hedges & Olkin,
1985). Two types of studies can cause non-independence of measures. Multiple- treatment studies have two or more treatments that are compared to the same control group, or effects of several independent variables from the same subjects that are analyzed separately. Second, multiple-endpoint studies have participants who complete several dependent measures (Gleser & Olkin, 1994). Both types of non-independence were found in the current study, threatening the validity of results.
Studies that utilize more than one pre-test measure of self-esteem produce multiple effect size estimates that show multiple-treatment non-independence. For example, Long, et al. (1994) simultaneously examined effects of personal self-esteem and collective self-esteem, presenting means for high vs. low self-esteem groups for each measure. Thus, participants in these two comparisons overlap. Multiple-endpoint problems occur in studies with several dependent measures. For example, Kelly (1988) measured affect, evaluations, and perceived homogeneity, and analyzed each separately.
37 Thus, three dependent measures were based on the same experimental comparisons and with the same research participants.
A general strategy for addressing data dependency is to collapse all overlapping effect size measures into a single measure of effect. However, this strategy often proves to be too conservative and may ignore important study aspects (Matt & Cook, 1994).
Further, this strategy was incompatible with the goals of the current meta analysis.
Specifically, a goal for this analysis was to examine the effects of different types of dependent measures and different measures of self-esteem. Collapsing multiple measures within studies would result in a loss of information central to the analysis.
Several attempts were made to reduce the impact of non-independence of effect sizes. Collapsing of overlapping comparisons was used in situations where combination of overlapping effect sizes resulted in no loss of information (e.g., two dependent measures both coded as "direct" measures of bias such as adjective ratings and attributions.). For example, Gagnon and Bourhis (1996) presented a multiple-endpoint study measuring scores on three different point allocations scales. As the three strategies were all judged to be direct measures of bias and no other differences existed in the comparisons, the three effect size measures (as well as all other coded information) were collapsed into a single averaged comparison. This averaged comparison maintains integrity of information and reduces the impact of non-independence.
Several instances existed where collapsing into an overall effect size estimate was not used. Kelly (1988) presented a multiple-endpoint study examining affect, evaluation, and homogeneity. Affect and evaluation were coded as direct measures of bias.
Homogeneity was coded as an indirect measure of bias. The two direct measures were
38 collapsed, yielding a single effect size. Homogeneity was not collapsed with the two other dependent measures, as this would have combined measures that were qualitatively different. Further, this would interfere with the testing of a priori hypotheses regarding dependent variable type. This same logic holds for multiple-treatment studies using several self-esteem measures.
The collapsing strategies detailed above reduced the impact of data dependency but did not eliminate all non-independent effect size estimates. Whereas independence is an assumption of meta-analysis, often this assumption is violated to allow testing of a priori hypotheses (Johnson, 1989). This is the case with the current analysis. The next section discusses data analysis strategies and discusses strategies to further reduce the impact of data dependency.
Another strategy for dealing with overlapping effect size measures models the dependencies amongst the measures. This strategy requires derivation of correlation matrices so that the relationship between all measures can be assessed (Gleser & Olkin,
1994). Though this is the most elegant solution to the problem of overlapping effect size measures, derivation of correlation matrices from the studies included in the meta analysis was not possible. Only one of the studies included in the meta analysis contained a full matrix of correlations. Requesting this information from each study author was possible, but receiving an adequate response was judged unlikely.
Data Analysis Strategies
Initial analyses addressed hypotheses using data containing the overlapping effect size measures. Additional analyses reduced the effect of overlapping effect sizes. Some
39 analyses violated the assumption of data independence because they contained some overlapping effect size measures.
To address the issue of overlapping effect size estimates, the following data analysis strategies were used.
1. An initial analysis used all comparisons, collapsing only those measures that did not result in loss of information. This analysis and all that follow use data adjusted for attenuation. The sections that follow present information regarding attenuation adjustments.
2. Next, several sub-analyses examined effects of predictors and simultaneously addressed data dependency issues. The first sub-analysis split data based on measures of self-esteem. One analysis examined only the comparisons utilizing Collective Self-
Esteem, another examined Rosenberg's Personal Self-Esteem, and another examined all other measures. This allowed for testing of variable type and salience hypotheses for each measure. Another analysis split direct and indirect measures of bias into separate data sets and analyzed each individually. This eliminated the overlap in effect size estimates based on multiple-endpoint studies as overlapping dependent measures will have either been collapsed or segregated. This allowed for testing of self-esteem measurement and salience hypotheses for each variable type.
3. Another analysis removed all studies with overlapping effect size measures, yielding a data set that met strict meta analysis assumptions of independence.
Hypotheses were again tested using these data.
None of the strategies detailed above was ideal. Strategies 1 and 2 do not meet non-independence assumptions as some overlapping data remained, and Strategy 3 meets
40 assumptions but omits data from several studies. However, the accumulation of evidence from each of the analyses can provide convergent support for hypotheses.
Data Analysis (Testing Hypotheses)
Data analyses employed procedures analogous to regression analysis. Effect size was predicted from dependent variable type, self-esteem measure, salience, and all interactions between effects. This approach involved several steps. First, data were subjected to hierarchical weighted least squares analysis. Weights adjusted for sample size and attenuating factors (discussed in the next section). Next, the Q-statistic, defined as the weighted sums of squares for the addition of each block of variables, was calculated using Formula 6 below (Hedges, 1994; Hedges & Olkin, 1985).
QChange b(FChange )( MSError ) (6)
b = # of predictor variables, FChange=F-change statistic associated with addition of variable to regression model, and MS error = Error term for block.
The Q-statistic is distributed as Chi-Square with b degrees of freedom. In addition to providing a test of the contribution of each variable, this procedure also allows for a test of the fit of the linear model. The sum of squared deviations from prediction is also equivalent to a Q-statistic. This statistic tests the null hypothesis that the predictors fully explain the variance in effect sizes.
Categorical variables with more than two categories were recoded for further analysis. For example, a variable measuring the type of self-esteem measure utilized by studies in the meta analysis was classified into three categories, Collective Self-Esteem,
Rosenberg's Self-Esteem, and All Other measures. This information was coded into k-1
41 new variables (k refers to # of categories; 3 -1- = 2 new variables) as suggested by Cohen and Cohen (1983). A categorical variable measuring type of dependent measure utilized contained two categories (direct and indirect bias). As this variable contained only two categories, no recoding was necessary (i.e., it can be used directly as a predictor in regression analysis). Interaction effects were created through multiplication of variables.
For example, the interaction between dependent variable type and salience was created through multiplication of these two variables to create a single variable capturing the interaction. The interaction between self-esteem measure and salience required multiplication of each newly created variable individually with the salience variable, creating two new variables to capture this interaction.
Adjusting for Low Reliability and Other Artifacts
Several sources of variability exist that affect effect size measures. Hunter and
Schmidt (1990a, 1994) listed several causes of error including low reliability of independent and dependent measures, and artificial dichotomization of variables.
Reliability of self-esteem measures was estimated through several strategies.
Ideally, the document yielding the effect size estimate reported reliability data. For studies not reporting these data, information was taken from scale construction articles or a general review of the major scales (e.g., Blascovich & Tomaka, 1991). In the event that neither form of data was available, estimates were taken from studies reporting reliabilities for the same scale or studies reporting reliability for similar scales. In lieu of all of the above information, an average reliability score was taken from the entire data set and assigned to the variable.
42 Reliability of salience measures could be adjusted based on inter-rater reliability statistics. However, inter-rater reliability of salience items was a constant value because interrater reliability was the same for all studies rated by the same coders. The current study does not use adjustments for interrater reliability. The variances introduced by intercoder factors are not sources of variance introduced by the studies included in the meta analysis. Rather, the meta analysis procedures introduce this variance.
Reliability estimates for dependent measures used reported reliabilities where available. Research papers that described scale construction were not available for dependent measures. If information regarding reliabilities was not provided, reliability was taken from other studies using the same or similar measures. When none of the above information was provided, a mean value was assigned as above.
Two dependent measures, similarity and attribution, came from a single study that did contain reliability information. However, an overall average was used for these dependent measures as reliability information for similarity and attribution both came from a single study.
An additional artifact may affect effect size estimates. Artificial dichotomization of variables may reduce the effect size because of the loss of information resulting from collapsing continuous data into two categories (i.e., high and low self-esteem). This was judged to be a problem for many of the studies included in this meta analysis. Several studies used statistical designs requiring dichotomization of self-esteem scores. An adjustment for attenuation due to artificial dichotomization corrected for attenuation of effect sizes.
43 Statistical formulae for artifact adjustments. Reliability was adjusted through construction of an index of attenuation. The attenuation index provides an estimate of the reduction of the effect size (d). The index is provided below:
ar rxxryy (7)
rxx is the reliability of the self-esteem measure and ryy is the reliability of the dependent measure.
The effect size (d) was adjusted for attenuation through division using the formula below, yielding an effect size providing a more accurate estimation of the population effect size.
d da (8) ar
Correction for attenuation based on artificial dichotomization was achieved through application of another attenuating factor. Calculation of this attenuating factor is more complex than the reliability adjustments above. Hunter and Schmidt (1990b) presented Formula 9 for this adjustment.
(c) ad (9) pq
(c) Unit normal density function for z - transformed cutpoint p = proportion of scores defined as low self - esteem q = proportion of scores defined as high self - esteem
This new attenuation value shown below is included in the calculation of the attenuating value shown below, which is used in place of ar in Formula 8.
a arad (10)
44 If no artificial dichotomization was used, then the attenuating factor for dichotomization was set to 1.0. Several studies used tripartite divisions of groups into high, medium, and low self-esteem groups. No adjustment exists for this type of split.
Studies using a tripartite split were also assigned an attenuating value of 1.0.
A Note on Effect Size Estimates
The effect size estimate used in analyses was an index of the difference in ingroup bias between low and high self-esteem individuals. Predictor variables assessed the effect of variable type, self-esteem measure, and salience on the measured difference between low and high self-esteem individuals.
Another, perhaps more direct, way to assess hypotheses would be to extract an effect size estimate for high self-esteem and low self-esteem individuals independently.
Each effect size in this case would represent the amount of ingroup bias exhibited by the specific group. The overall effect size in this case would be an index of the size of the ingroup bias effect. Attempts were made to derive these effect size measures, however reporting of appropriate statistics to calculate these effects was uncommon. None of the studies furnished the correlation between ingroup and outgroup ratings, the error term for within group tests (i.e., standard error of the difference), or the information to calculate either. Thus, extraction of an overall effect size measuring ingroup bias in this manner proved impossible.
Another meta-analysis of ingroup bias effects by Mullen et al. (1992) used effect size estimates representing the differences between ingroup and outgroup ratings.
However, these authors treated within subjects tests as between group tests, and thus they
45 failed to adjust effect size estimates for the correlation between measures. This error in calculation caused effect sizes to be overestimated in this analysis (Dunlap et al., 1996).
Errors of this type are by no means limited to the studies of ingroup bias reported above. Dunlap and colleagues (1996) reported on several additional meta analyses that made the same errors. All of the studies cited used the Mullen and Rosenthal (1985) meta analysis software (BASIC Meta Analysis). This software treats all effects as between group tests.
Two recent analyses of ingroup bias effects (Migdal, Hewstone, & Mullen, 1998;
Urban & Miller, 1998) ameliorated these problems by constructing effect sizes from differences between means and mean square error terms. However, these syntheses were limited to a small number of studies with complete statistical information. As discussed in previous sections, this strategy was not tenable for the current analysis.
As reported in the above sections, several problems existed in using within subjects tests to calculate effect size measures. It was not possible in the current study to consistently construct effect size estimates based on within subjects tests. The author made efforts to contact study authors to obtain additional statistical information; however, those requests yielded very few responses. Thus, the effect sizes reported here are based on the differences between high and low self-esteem individual's ratings of ingroup and the differences between the groups in ratings of outgroup.
46 CHAPTER THREE: RESULTS
This chapter reports on the analysis of coded data. The chapter presents information describing studies and comparisons, statistical tests of hypotheses, exploratory analyses of data, and a discussion of issues such as data dependency and publication bias.
Descriptive Information
This section discusses all coded studies, including those excluded from tests of hypotheses due to lack of sufficient information to calculate effect size estimates.
Specifically, this section describes the documents selected, specific methods used to calculate effect size estimates, types of self-esteem measures, classes of dependent measures, and reliability information.
Document information. Thirty-four documents were included in the meta analysis. Of these documents, twenty-five were published or in press and nine studies were unpublished. Published articles came from several sources; Journal of Personality and Social Psychology (8 articles; 23.5%), British Journal of Social Psychology (3;
8.8%), The Journal of Social Psychology (3; 8.8%), European Journal of Social
Psychology (3; 8.8%), Personality and Social Psychology Bulletin (2; 5.9%), Journal of
Applied Social Psychology (1; 2.9%), Journal of Experimental Social Psychology (1;
2.9%), Journal of Occupational Psychology (1; 2.9%), Journal of Social Behavior and
Personality (1; 2.9%), Social Behavior and Personality (1; 2.9%), and South African
Journal of Social Psychology (1; 2.9%). Unpublished work included papers presented at
47 American Psychological Society conferences (2; 5.9%), dissertations (3; 8.8%), articles cited in other published work but not published or otherwise unavailable (2; 5.9%), a master's thesis (1; 2.9%), and a paper presented at the British Psychological Society (1;
2.9%). Though unpublished studies constituted 26.5% of the included studies, it should be noted that many of these unpublished studies or portions thereof are currently under review for publication.
The 34 studies represent the work of 29 different first authors. Two lead authors
(Karen Long and Jennifer Crocker) have three first-authored studies in the meta analysis, a third author has two lead author studies included (Anna Maass). No other first author's work appears more than once.
Effect sizes. The 34 studies selected for inclusion in the meta-analysis yielded
113 high self-esteem vs. low self-esteem comparisons. Of these 113 comparisons, 102
(90%) yielded effect size estimates. The remaining 11 comparisons did not include adequate information to code any effect size measures. These 11 comparisons without effect sizes were coded for all other study characteristics. The modal number of effect sizes per study was two (17 studies). One study yielded 16 effect size estimates; five yielded one effect size estimate. Three studies did not contain adequate information to code any measures of effect.
Most effect size estimates were calculated directly from descriptive statistics
(means and standard deviations or correlations). Further, most effect size estimates required little or no estimation. Thirty-nine (38.2%) of the effect sizes were calculated using means and standard deviations, 34 (33.3%) were calculated using correlations.
Twenty-four (23.5%) effect sizes were calculated using means and mean square error
48 measures. The remaining effect sizes were calculated either from probability values (4;
3.9%) or F-values (1; 1.1%). Most effects were judged as requiring no estimation or slight estimation (40 and 43 respectively, 81.4% total). Six effects (5.9%) required a small amount of estimation, eight (7.8%) required a moderate amount of estimation, and five (4.9%) required a high amount of estimation. Appendix C contains a detailed definition of estimation criteria.
Self-esteem measures. The majority of comparisons defined self-esteem using either Rosenberg's Self-Esteem Scale or the Collective Self-Esteem Scale (57 and 30 instances, 50.4% and 26.5%) respectively). Other measures used were group identification scales (13; 11.5%), the Texas Social Behavior Inventory (9; 8.0%), Janis-
Fields Feelings of Inadequacy Scale (2; 1.8%), and the Self-Descriptive Questionnaire III
(2; 1.8%). Of the 30 comparisons using the Collective Self-Esteem Scale, 8 (26.7%) used the entire scale, 8 (26.7%) used the Membership Self-Esteem Subscale, 8 (26.7%) used the Private Collective Esteem Subscale, 3 (10.0%) used the Public Collective Esteem
Subscale, and 3 (10.0%) were based on some other combination of scale items (i.e., combination of two subscales).
All self-esteem measures were at least moderately reliable. Reliability coefficients ranged from .63 to .93. The median reliability was .825. Fifty-five of the documents included reliability information. Forty-nine reliabilities were taken from scale construction information. The remaining nine reliabilities, all from studies using group identification measures, were assigned the mean reported reliability from group identification studies reporting reliability information (.85).
49 Self-esteem was used as an independent variable in three manners. Most commonly, researchers classified participants as high or low self-esteem based on a median split of self-esteem scores (64 comparisons, 56.6%). Twelve comparisons
(10.6%) used a tripartite split, classifying individuals as high, medium, or low self- esteem. Data for "medium" self-esteem groups was not included in the meta-analysis.
The remaining 37 comparisons (32.7%) did not split scores (i.e., used correlation/regression procedures).
Low/high self-esteem was most commonly defined using a median split. Scores falling below the 50th percentile were defined as low self-esteem using this procedure.
As shown in Table 1, scores falling below the 50th percentile do not necessarily represent the lower half of possible scale values. Many individuals who scored moderately high on the scale were classified as low self-esteem.
Examination of the percentage of the total scale score further points toward deficiencies in classification. Percentage of the total score is defined as the cut point for low self-esteem scores minus the lowest possible score divided by the highest possible score minus the lowest possible score. Ideally, low self-esteem would be defined as the lower half of possible scores. The percentage of total possible score defining low self- esteem in this case would be at or near 50%. This was rarely the case for the studies using artificial dichotomization. Furthermore, it is unlikely that scores on self-esteem scales were normally distributed. Given the concentration of scores on the high end, it is more likely that scale scores were negatively skewed.
50 Table 1
Definition of Low Self-Esteem
Measure Strategy Scale Definition of % of total
Range low self-esteem possible score
(Median Split) defining LSEa CSE Membership Regression (N=8) 4-28 N/A N/A
CSE Private Median Split (N=5); 4-28 22; 75%;
Regression (N=3) N/A N/A
CSE Public Median Split (N=1); 4-28 Not reported N/A;
Regression (N=2) N/A N/A
CSE Total Median Split (N=5); 16-112 69; 55%;
Regression (N=3) N/A N/A
CSE Combination of Regression (N=3) 8-56 N/A N/A
two subscales
Rosenberg Self- Median Split (N=33); 10-40 31; 70%;
Esteem Regression (N=10) N/A N/A
Rosenberg Self- Median Split (N=5); 10-70 53; 72%;
Esteem (Modified) Tripartite Split (N=9); 54 70%
Table 1 (continued)
Measure Strategy Scale Definition of % of total
Range low self-esteem possible
(Median split) score 51 defining
LSEa Janis-Fields Feelings Median Split (N=2) 18-126 Not reported N/A
of Inadequacy
Texas Social Median Split (N=6); 16-80 57; 55%;
Behavior Tripartite Split (N=3) Not reported N/A
Inventory
Group Identity Median Split (N=7); 1-5; 3.16 (1-5); 3.0 54%(1-5);
Regression (N=6) 1-7 (1-7) 33%(1-7)
Total N = 113
Note. CSE = Collective self-esteem. Definition of low self-esteem refers to median of cut points for all studies using the scale. a % of total possible = (LSE cut point - Minimum score )/ (Highest possible score - Minimum score) X 100. % of total possible yields a relative measure of the score that defines low self-esteem. N = number of comparisons.
52 One striking example involves scores on the Rosenberg scale. For studies using artificial dichotomization based on median splits, the cutting point separating high and low self-esteem was 31 on a scale ranging from 10 to 40. Rosenberg's Self-Esteem Scale consists of ten 4-point items, ranging from 1 to 4, four being highest. A score of 31 could represent the answers of an individual who chose 3 on nine of the items and 4 on the remaining item. That is, the individuals classified as "low self-esteem" could indicate scores on the upper end of each scale item. A score of 31 corresponds to 70% of the distance from the lowest possible to the highest possible score. Thus, scores classified as
"low" self-esteem may have been relatively high. Examination of results for others scales indicates that poor classification is common. Some of the individuals classified as low self-esteem may be better conceptualized as having medium or even high self-esteem.
Artificial dichotomization of self-esteem scores was widespread (64 out of 113 comparisons). As artificial dichotomization typically results in smaller effects, this practice contributes to an increase in Type II errors (Cohen, 1983). Further, classification of individuals as low self-esteem is clearly inadequate in many cases.
As the Janis-Fields Feelings of Inadequacy Scale, Texas Social Behavior
Inventory, Self-Descriptive Questionnaire II, and group identification measures comprised a small proportion of the self-esteem measures, these measures were collapsed into a single category for data analysis. This new category contains 23 comparisons.
This category is termed "All Others" for the remainder of this report.
Dependent Measures. Studies used several dependent measures. Adjective ratings were the most common (62; 54.9%), followed by ratings of group product (19;
16.8%), similarity (10; 8.8%), attribution (6; 5.3%), linguistic intergroup bias (4; 3.5%),
53 social distance (4; 3.5%), Tajfel Matrices (2; 1.8%), contribution to group (1; 0.9%), liking (1; 0.9%), and likelihood of cooperation (1; 0.9%). The three remaining comparisons used multiple categories of measures that were collapsed (3; 2.7%).
Of the 102 studies with codeable effect sizes, 35 (34%) reported reliabilities for dependent measures. The remaining reliabilities were imputed from other documents using the same or similar measures (36; 35%), an average based on other studies using the same measure (16; 16%), or an overall average if no other studies reported reliabilities for the specific measure (15; 15%). Dependent measure reliabilities ranged from .54 to .94, with a median value of .89. Using this strategy, 91 percent of the dependent measures were assigned a reliability of .80 or greater.
Normality. Though meta analysis techniques do not formally require normally distributed data, meta analyses are improved if the distribution of continuously scaled variables is normal (Raudenbush, 1994). Examination of histograms indicated that the distribution of effect sizes and of salience scores both adequately approximated normal distributions.
Ingroup Bias
An effect size estimate for differences between ratings of the ingroup and the outgroup could be used to assess the presence of a tendency to favor the ingroup. This test would provide information as to whether ingroup bias is consistently exhibited across studies. Calculation of this effect size estimate was not possible due to limitations of study reports (i.e., no data reflecting correlations between ingroup and outgroup ratings as discussed in the methods section).
54 Comparisons do allow for an alternate test that speaks to this question. The presence of certain types of ingroup bias, specifically, rating the ingroup more positively than the outgroup, were examined through frequency data. Ratings for ingroups and outgroups were compared and coded as either: a) ingroup favoritism; b) outgroup favoritism; c) no favoritism; or d) no information provided. This coding examined which group was rated more positively. For example, ingroup favoritism is defined as a higher mean rating of the ingroup compared to the outgroup. These data reflect examination of mean scores and do not reflect significant test statistics.
Of the 96 comparisons providing this information, 72 (75%) found the ingroup to be rated more positively, 23 (24%) found the outgroup to be rated more positively, and 1
(1%) found exactly equal ratings of both groups. Seventeen studies did not provide information as to which group was favored.
The 23 studies in which the outgroup was rated more positively than the ingroup seem to contradict expectations. However, rating the ingroup more positively than the outgroup is only one form of ingroup bias. Brewer (1979) reports that ingroup bias can take other forms when the ingroup is of low status or the ingroup fails while the outgroup succeeds. In these situations, it is unlikely to find ingroup ratings to be more positive than outgroup ratings. Rather, a more likely ingroup bias strategy is a minimization of differences in ratings between groups. Of the 23 studies finding outgroups to be rated more positively than ingroups, 11 were comparisons in which the ingroup fails and the outgroup is successful and another 6 were comparisons wherein the ingroup was the lower status group. Thus, of the studies finding outgroups to be favored, 17 of 23 comparisons represented situations wherein it is not likely to find ingroup favoritism.
55 Analysis of Data and Tests of Hypotheses
This section presents several analyses of data. First, analyses of all comparisons test predictor effects. The results that follow use effect size estimates that represent the difference between ingroup bias exhibited by low and high self-esteem individuals.
Differences in the positive direction indicate that high self-esteem individuals show more ingroup bias than individuals with low self-esteem. A negative effect indicates that low self-esteem individuals show more bias. This procedure allows differences to be assessed regardless of ingroup bias strategy utilized.
Analysis One: Full analysis. Table 2 presents the average overall effect size as well as effect sizes for individual studies. Of note is the relatively small, but clearly non- zero, overall raw effect size of 0.23 (95% CI: 0.19 to 0.26). Effect sizes range from
-1.55 to 2.09. Raw effect size is presented because it is the value used directly by the researchers in testing hypotheses. The effect size (Un) estimate refers to effects weighted for sample size, reliability, and dichotomization (i.e., unattenuated). An effect of this size and direction indicates that high self-esteem individuals show more bias than low self- esteem individuals. Given the median sample size of 53.5, many of these studies had insufficient power to detect this difference. Power calculation indicates the average study to have an 80% probability of committing a Type II error. Clearly, many of the initial studies were underpowered. However, claims of inadequate design should be tempered because many of the studies were not designed to test the relationship between self- esteem and ingroup bias. Several studies provided these tests as ancillary analyses.
56 Table 2
Study Characteristics, Effect Sizes, and Confidence Limits
NAME DV SE N SAL Raw LL UL ES ES ES ES (Un) LL UL Overall 58.5 225.0 0.23 0.19 0.26 0.28 0.25 0.32 Aberson (1999) 1 1 20 78.0 -0.25 -0.87 0.38 -0.42 -1.24 0.40 Aberson (1999) 1 1 23 78.0 0.60 0.01 1.19 1.03 0.25 1.81 Aberson (1999) 1 1 24 78.0 -0.01 -0.57 0.56 -0.01 -0.76 0.73 Aberson (1999) 1 1 10 78.0 0.41 -0.47 1.30 0.68 -0.49 1.85 Aberson (1999) 2 1 20 78.0 -1.05 -1.71 -0.38 -1.50 -2.30 -0.70 Aberson (1999) 2 1 23 78.0 -1.29 -1.93 -0.66 -1.86 -2.63 -1.09 Aberson (1999) 2 1 24 78.0 -0.15 -0.71 0.42 -0.21 -0.90 0.47 Aberson (1999) 2 1 10 78.0 0.01 -0.86 0.89 0.02 -1.04 1.08 Abrams (Unpublished; 1982)a 1 2 117 313.5 0.46 0.21 1.09 0.54 0.26 0.83 Abrams (Unpublished; 1982)a 2 2 117 313.5 0.22 -0.09 0.78 0.26 -0.02 0.54 Abrams (Unpublished; 1983)a 1 2 42 156.0 0.65 -0.04 0.48 0.76 0.28 1.23 Abrams (Unpublished; 1983)a 2 2 42 156.0 0.35 0.20 0.72 0.41 -0.06 0.88 Brickson (Unpublished) 1 5 294.0 Brickson (Unpublished) 1 2 285.5 Brickson (Unpublished) 1 5 294.0 Brickson (Unpublished) 1 2 285.5 Brockner & Chen (1996) 1 3 438 240.0 0.09 -0.05 0.22 0.11 -0.04 0.26 Brockner & Chen (1996) 1 3 188 232.5 1.03 0.82 1.25 1.27 1.03 1.51 Brown, Collins, & Schmidt (1988) 1 4 25 202.0 2.09 1.40 2.78 2.58 1.82 3.35 Brown, Collins, & Schmidt (1988) 2 4 26 216.0 -0.37 -0.91 0.18 -0.45 -1.07 0.16 Brown, Collins, & Schmidt (1988) 1 4 23 286.0 0.19 -0.39 0.77 0.24 -0.41 0.89 Brown, Collins, & Schmidt (1988) 2 4 27 305.0 1.04 0.47 1.60 1.28 0.64 1.91 Brown, Collins, & Schmidt (1988) 1 4 24 223.5 1.30 0.67 1.92 1.60 0.90 2.29 Brown, Collins, & Schmidt (1988) 2 4 21 223.5 -0.90 -1.54 -0.27 -1.11 -1.82 -0.40 Brown, Condor, Mathews, & Wade 1 5 126 379.5 0.23 -0.02 0.47 0.29 0.01 0.57 (1986) Crocker & Luhtanen (1990) 1 1 82 278.0 0.80 0.48 1.12 1.04 0.67 1.40 Crocker & Luhtanen (1990) 1 1 82 126.0 1.33 0.99 1.67 1.71 1.33 2.10
57 Table 2 (Continued)
NAME DV SE N SAL Raw LL UL ES ES ES ES (Un) LL UL Crocker & Luhtanen (1990) 1 1 43.5 Crocker & Schwartz (1985) 1 2 48.0 Crocker, Thompson, McGraw, & 1 2 42 273.0 0.93 0.48 1.38 1.09 0.60 1.58 Ingerman (1987) Crocker, Thompson, McGraw, & 1 2 42 200.5 1.82 1.31 2.33 2.13 1.58 2.68 Ingerman (1987) Crocker, Thompson, McGraw, & 1 2 15 369.5 -0.58 -1.31 0.15 -0.67 -1.46 0.12 Ingerman (1987) Crocker, Thompson, McGraw, & 1 2 19 369.5 0.68 0.02 1.33 0.78 0.07 1.49 Ingerman (1987) Crocker, Thompson, McGraw, & 1 2 47.5 Ingerman (1987) Donaldson (Unpublished; 1995) 1 2 180 49.5 -0.33 -0.53 -0.12 -0.45 -0.69 -0.20 Donaldson (Unpublished; 1995) 1 2 90 115.5 Donaldson (Unpublished; 1995) 1 2 90 77.0 Gagnon & Bourhis (1996) 1 5 47 75.5 0.56 0.15 0.98 0.74 0.26 1.21 Gagnon & Bourhis (1996) 1 5 47 55.0 1.25 0.80 1.69 1.64 1.13 2.14 Greenwood (Unpublished; 1996) 1 2 35 197.0 0.50 0.02 0.97 0.55 0.05 1.05 Greenwood (Unpublished; 1996) 1 2 29 263.5 0.19 -0.32 0.71 0.21 -0.34 0.76 Hinkle, Taylor, Fox-Cardamone, & Crook 1 5 62 214.0 0.65 0.29 1.01 0.72 0.34 1.11 (1989) Hunter, Stringer, & Coleman (1993) 1 53 399.0 0.14 -0.24 0.52 0.17 -0.25 0.59 Hunter, Stringer, & Coleman (1993) 1 54 399.0 0.11 -0.27 0.49 0.13 -0.28 0.55 Kelly (1988) 1 5 142 331.5 0.28 0.05 0.52 0.32 0.07 0.57 Kelly (1988) 2 5 142 331.5 0.03 -0.20 0.26 0.04 -0.21 0.28 Lay (1992) 1 2 64 304.5 0.04 -0.31 0.39 0.05 -0.33 0.43 Lay (1992) 1 1 64 304.5 0.00 -0.35 0.35 0.00 -0.39 0.39 Lay (1992) 1 1 64 304.5 -0.66 -1.01 -0.30 -0.83 -1.23 -0.43 Lay (1992) 2 2 64 304.5 0.52 0.17 0.87 0.62 0.23 1.01 Lay (1992) 2 1 64 304.5 0.11 -0.24 0.45 0.14 -0.26 0.53 Lay (1992) 2 1 64 304.5 -0.53 -0.88 -0.17 -0.67 -1.08 -0.27 Lindeman (1997) 1 5 181 348.0 0.66 0.45 0.87 0.87 0.63 1.12 Lindeman (1997) 1 5 132 268.5 0.15 -0.09 0.39 0.20 -0.08 0.47
58 Table 2 (Continued)
NAME DV SE N SAL Raw LL UL ES ES ES ES (Un) LL UL Lindeman (1997) 1 5 283.5 . . Long (Unpublished; 1997) 1 1 120 154.5 -0.95 -1.21 -0.68 -1.25 -1.56 -0.94 Long (Unpublished; 1997) 1 1 30 154.5 -0.39 -0.90 0.12 -0.52 -1.12 0.07 Long & Spears (1998) 1 2 56 351.5 1.27 0.86 1.67 1.61 1.15 2.06 Long & Spears (1998) 1 1 56 351.5 -0.48 -0.85 -0.10 -0.59 -1.01 -0.17 Long, Spears, & Manstead (1994) 1 1 55 388.5 0.43 0.05 0.81 0.55 0.12 0.98 Long, Spears, & Manstead (1994) 2 1 55 381.0 0.80 0.41 1.19 1.02 0.58 1.45 Long, Spears, & Manstead (1994) 1 2 55 388.5 -0.03 -0.41 0.34 -0.04 -0.47 0.39 Long, Spears, & Manstead (1994) 2 2 55 381.0 -1.49 -1.91 -1.07 -1.92 -2.40 -1.44 Long, Spears, & Manstead (1994) 1 1 54 306.0 0.22 -0.16 0.60 0.28 -0.15 0.71 Long, Spears, & Manstead (1994) 2 1 54 305.0 -0.38 -0.76 0.00 -0.48 -0.91 -0.05 Long, Spears, & Manstead (1994) 1 2 54 306.0 0.48 0.10 0.86 0.62 0.18 1.05 Long, Spears, & Manstead (1994) 2 2 54 305.0 0.87 0.48 1.27 1.13 0.68 1.58 Maass, Milesi, Zabbini, & Stahlberg 2 1 85 379.0 0.24 -0.06 0.54 0.30 -0.03 0.64 (1995) Maass, Milesi, Zabbini, & Stahlberg 2 1 73 379.0 -0.20 -0.53 0.12 -0.25 -0.62 0.11 (1995) Maass, Ceccarelli, & Rudin (1996) 2 2 80 370.5 0.10 -0.21 0.41 0.13 -0.22 0.48 Maass, Ceccarelli, & Rudin (1996) 2 1 80 370.5 0.14 -0.17 0.45 0.18 -0.17 0.53 Mallery (Unpublished; 1994) 1 2 72 202.0 -0.03 -0.35 0.30 -0.03 -0.41 0.34 Mallery (Unpublished; 1994) 1 2 72 240.0 0.63 0.30 0.97 0.82 0.44 1.21 Mallery (Unpublished; 1994) 2 2 72 240.0 0.44 0.11 0.77 0.59 0.20 0.97 Mallery (Unpublished; 1994) 2 2 72 232.5 -0.47 -0.80 -0.14 -0.63 -1.01 -0.25 McCrea, Crawford, & Hirt (Unpublished; 1 4 160 175.0 0.26 0.04 0.48 0.29 0.06 0.52 1998) McCrea, Crawford, & Hirt (Unpublished; 1 4 80 256.5 0.11 -0.20 0.42 0.12 -0.21 0.44 1998) McCrea, Crawford, & Hirt (Unpublished, 1 4 80 103.5 0.34 0.03 0.66 0.38 0.05 0.70 1998) Robbins & Foster (1994) 1 1 151 226.5 -0.16 -0.39 0.07 -0.21 -0.46 0.05 Ruttenberg, Zea, & Sigelman (1996) 2 1 40 399.0 0.06 -0.37 0.50 0.08 -0.41 0.56 Ruttenberg, Zea, & Sigelman (1996) 2 1 47 420.0 0.73 0.31 1.14 0.87 0.41 1.32 Seta, C. E. & Seta (1992) 1 2 14 129.5 -0.40 -1.15 0.35 -0.51 -1.36 0.35
59 Table 2 (Continued)
NAME DV SE N SAL Raw LL UL ES ES LL ES UL ES (Un) Seta, C. E. & Seta (1992) 1 2 14 191.0 0.53 -0.22 1.29 0.67 -0.19 1.53 Seta, C. E. & Seta (1992) 1 2 14 147.0 0.61 -0.15 1.37 0.77 -0.09 1.63 Seta, C. E. & Seta (1992) 1 2 14 103.0 -0.20 -0.94 0.54 -0.25 -1.10 0.59 Seta, C. E. & Seta (1992) 2 2 14 97.5 0.05 -0.69 0.79 0.07 -0.78 0.91 Seta, C. E. & Seta (1992) 2 2 14 157.5 -0.94 -1.72 -0.16 -1.19 -2.07 -0.30 Seta, C. E. & Seta (1992) 2 2 14 102.5 0.68 -0.08 1.44 0.85 -0.01 1.72 Seta, C. E. & Seta (1992) 2 2 14 96.5 -0.15 -0.90 0.59 -0.19 -1.04 0.65 Seta, J. J. & Seta (1996) 1 2 34 115.0 -0.23 -0.48 0.47 -0.30 -0.84 0.25 Seta, J. J. & Seta (1996) 1 2 34 238.0 1.78 -0.77 0.19 2.28 1.65 2.92 Seta, J. J. & Seta (1996) 1 2 34 115.5 0.19 -2.09 -1.01 0.24 -0.31 0.78 Seta, J. J. & Seta (1996) 1 2 34 132.0 -0.03 -0.83 0.12 -0.04 -0.58 0.50 Seta, J. J. & Seta (1996) 1 2 34 164.5 -0.45 -0.71 0.25 -0.58 -1.13 -0.04 Seta, J. J. & Seta (1996) 1 2 34 202.5 0.94 1.21 2.34 1.21 0.64 1.78 Seta, J. J. & Seta (1996) 1 2 34 222.5 0.31 -0.29 0.66 0.40 -0.14 0.95 Seta, J. J. & Seta (1996) 1 2 34 126.5 0.49 -0.51 0.44 0.63 0.08 1.18 Seta, J. J. & Seta (1996) 2 2 34 121.0 -0.30 -0.94 0.03 -0.38 -0.93 0.16 Seta, J. J. & Seta (1996) 2 2 34 213.0 0.85 0.44 1.44 1.09 0.53 1.66 Seta, J. J. & Seta (1996) 2 2 34 136.5 0.05 -0.16 0.79 0.06 -0.48 0.60 Seta, J. J. & Seta (1996) 2 2 34 125.5 1.61 0.01 0.98 2.07 1.45 2.69 Seta, J. J. & Seta (1996) 2 2 34 125.5 -0.01 -0.78 0.18 -0.01 -0.55 0.53 Seta, J. J. & Seta (1996) 2 2 34 205.0 -0.29 0.35 1.35 -0.37 -0.91 0.18 Seta, J. J. & Seta (1996) 2 2 34 169.5 -1.55 -0.43 0.52 -1.99 -2.61 -1.38 Seta, J. J. & Seta (1996) 2 2 34 115.5 -0.35 1.06 2.16 -0.46 -1.00 0.09 Sidanius, Pratto, & Mitchell (1994) 1 2 198 112.5 0.63 0.43 0.83 0.74 0.52 0.96 Sidanius, Pratto, & Mitchell (1994) 1 2 198 112.5 0.43 0.23 0.63 0.49 0.28 0.71 Smith & Tyler (1997) 1 1 83 378.0 0.27 -0.04 0.57 0.33 -0.01 0.67 Smith & Tyler (1997) 1 2 83 378.0 -0.02 -0.33 0.28 -0.03 -0.36 0.31 Thompson & Crocker (1990) 1 2 33.5 . . Verkuyten (1997) 1 1 118 380.0 0.48 0.23 0.74 0.64 0.34 0.94
60 Table 2 (Continued)
NAME DV SE N SAL Raw LL UL ES ES LL ES UL ES (Un) Wright & Wyer (Unpublished; 1997) 1 5 61 323.0 0.62 0.26 0.99 0.73 0.34 1.13 Wright & Wyer (Unpublished; 1997) 2 5 81 358.5 0.40 0.09 0.71 0.47 0.13 0.81
Note. DV = Dependent Variable: 1 = Direct Bias Strategies (Direct Dependent
Variable); 2 = Indirect Bias Strategy (Indirect Dependent Variable). SE = Self-Esteem
Measure: 1 = Collective Self-Esteem Scale; 2 = Rosenberg's Self-Esteem Scale; 3 = All
Other Measures. SAL = Salience: possible range is from 9 to 441. Raw ES = Raw effect size. Raw effect size is unweighted. LL, UL are 95% confidence limits around Raw ES.
ES(Un) is the effect size adjusted for attenuation due to reliability and artificial dichotomization. ES LL and ES UL are 95% confidence limits around ES(Un) Blank cells exist where effect sizes could not be calculated and Ns were unclear. aCited in
Hogg and Abrams (1990).
61 The first analysis explains differences between high and low self-esteem individuals' levels of ingroup bias (i.e., effect size) as a function of dependent variable type, self-esteem measure, salience, and all interactions between the variables. The preliminary analyses, and all that follow, use procedures analogous to least squares regression with weights adjusting for sampling error, reliability, and artificial dichotomization.
When adjustments are made to account for variance due to sample size, reliability, and artificial dichotomization, the average effect size for the 102 comparisons included in this analysis rises to 0.28 (95% CI: 0.25 to 0.32). As shown in Table 3, results indicate significant effects for dependent variable type with high self-esteem individuals exhibiting greater ingroup bias than low self-esteem individuals on direct measures of bias (M = 0.37, 95% CI: 0.33 to 0.42) but not on indirect measures (M = 0.06, 95% CI:
-0.01 to 0.14). Thus, individuals high in self-esteem show more bias than those with low self-esteem on direct measures. The two groups show comparable amounts of bias on indirect measures. This partially supports Hypotheses 2 and 3, the predictions that low self-esteem individuals are less likely those with high self-esteem to use direct strategies and individuals with low self-esteem are more likely than those with high self-esteem to use indirect strategies. However, this does not demonstrate that high self-esteem individuals only use direct strategies. Rather, low self-esteem individuals show comparable amounts of ingroup bias when using indirect strategies but show less ingroup bias using direct measures. More generally, this supports Hypothesis 1, the prediction that both groups will show ingroup bias.
62 Table 3
Analysis of Effect Sizes Predicted from Dependent Variable Type, Self-Esteem Measure,
Salience, and Interactions (Full Data Set)
Predictor df Q R-Square
Added Dependent Variable Type 1 31.1 *** .046
Self-esteem Measure 2 22.7 *** .034
Salience 1 6.4 * .010
Self-esteem X Dependent Variable 2 4.6 .007
Dependent Variable X Salience 1 6.4 ** .010
Self-esteem X Salience 2 11.6** .017
Three way interaction 2 18.3*** .027 Model 11 101.9 *** .151
Residual 90 567.2***
Note. Q is distributed as chi-square and provides a test for contribution of variables to prediction. The highly significant test for the residual value indicates that variance in effect sizes is not completely explained by the model.
*p<.05, **p<.01, ***p<.001.
63 These results, taken in conjunction with the observation that the majority of studies find an overall ingroup bias effect, point toward the conclusion that low self- esteem individuals are less likely to exhibit ingroup bias using direct measures of bias but do show comparable amounts of ingroup bias using indirect measures. Since most studies show an overall ingroup bias effect, any differences between individuals with high and low self-esteem would represent different amounts of ingroup bias. When high and low self-esteem individuals show comparable amounts of bias, it follows that both groups are showing ingroup bias.
Significant main effects exist for type of self-esteem measure and social identity salience. Self-esteem, as measured by the Collective Self-Esteem Scale (M = 0.06, 95%
CI: -0.02 to 0.14), showed the smallest effect. Rosenberg's Self-Esteem Scale (M = 0.30,
95% CI: 0.24 to 0.36), and All Other measures (M = 0.42, 95% CI: 0.36 to 0.49) showed a pattern whereby high self-esteem individuals exhibited greater bias than low self- esteem individuals. Salience effects, though small, were also statistically significant, r
(101) = .07, p<.05, indicating that higher salience was associated with more ingroup bias shown by high self-esteem individuals in relation to low self-esteem individuals.
Also producing significant effects were interactions between salience and dependent variable type, salience and self-esteem measure, and the three-way interaction.
The interaction between dependent variable type and measurement type did not improve prediction. The statistically significant three-way interaction serves to qualify the role of self-esteem measure and dependent variable type.
Due to the three-way interaction between categorical variables (dependent variable type and self-esteem measure) and a continuous variable (salience), it is useful to
64 examine correlations between salience and effect size at each combination of dependent measure type and self-esteem. Table 4 presents correlations between salience and effect size for each combination of dependent measure type and self-esteem measure. Figure 4 presents scatterplots including the least squares line and cell means for salience and effect size. This analysis yields six combinations of dependent measure type and self-esteem measure.
These results clarify Hypothesis 5 that predicted low self-esteem individuals would show bias regardless of level of salience whereas high self-esteem individuals would exhibit bias only when salience was high. In terms of the current data, this hypothesis would receive support if effect sizes for lower salience studies were negative, indicating greater favoritism by individuals with low self-esteem, and effects for higher salience studies were close to zero, indicating comparable amounts of bias exhibited by both groups. Overall, there was a small positive correlation between effect size and salience (r = 0.07) supporting this relationship. However, the three-way interaction, cell correlations, and cell means indicate a more complex relationship.
Findings regarding dependent measure type serve to further qualify relationships.
Low-self esteem individuals are more likely to use indirect strategies than they are to use direct strategies. As such, discussion of the relationship between salience and effect size will first focus on indirect measures.
65 Table 4
Correlation Between Salience and Effect Size Within Cells and Overall
Direct Bias Indirect Bias Overall Collective -0.05 0.68 * 0.26
Self-Esteem n = 16 n = 13 n = 29
(-0.42 to 0.46) (0.21 to 0.90) (-0.19 to 0.57) Rosenberg's 0.15 -0.09 0.06
Self-Esteem n = 30 n = 20 n = 50
(-0.22 to 0.48) (-0.51 to 0.37) (-.26 to .33) All Other -0.40 0.69 -0.26
n = 18 n = 5 n = 23
(-0.73 to 0.08) (-0.49 to 0.98) (-0.61 to 0.17) Overall -0.03 0.22 0.07
n = 64 n = 38 n = 102
(-0.28 to 0.22) (-0.11 to 0.50) (-0.13 to 0.26)
Note. *p<.05, **p<.01, ***p<.001. 95% confidence interval around in parentheses.
66 Figure 4: Correlation of effect size and salience by cell
Direct Bias Indirect Bias
Cell r = -0.05, n = 16 Cell r = 0.68*, n = 13
3 3
2 2
1 1 Effect Size Effect Size 0 CSE 0 (Cell = 0.09) (Cell = 0.00) -1 -1
-2 -2
-3 -3 0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450 Salience (Cell = 273) Salience (Cell = 229)
Cell r = 0.15, n = 30 Cell r = -0.09, n = 20
3 3
2 2
1 1 Effect Size RSE Effect Size 0 0 (Cell = 0.42) (Cell = 0.08)
-1 -1
-2 -2
-3 -3 0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450 Salience (Cell = 216) Salience (Cell = 198)
Cell r = -0.40, n = 18 Cell r = 0.69, n = 5
3 3
2 2
1 1 All Effect Size 0 Effect Size Other (Cell = 0.47) (Cell = 0.14) 0 -1 -1
-2 -2
-3 -3 0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450 Salience (Cell = 251) Salience (Cell = 259)
Note: For complete data, r = 0.07, Salience M = 231, Effect size M =0.28. Cell values refer to means (either effect size or salience) for the particular cell. CSE = Collective
Self-Esteem Measure. RSE = Rosenberg's Self-Esteem.
67 As seen in Table 4 and Figure 4, correlations between salience and effect size are the strongest for indirect measures using Collective Self-Esteem and All Other measures
(r's = 0.68, p<.05 and 0.69, ns, respectively), but slightly negative for Rosenberg's Self-
Esteem (r = -0.09, ns). A positive correlation indicates that high self-esteem individuals showed more ingroup bias as salience rose (or that low self-esteem individuals showed more bias when salience was low). Thus, when using indirect measures with the
Collective Self-Esteem Scale and All Other measures, a pattern was found whereby individuals with low self-esteem were more likely to exhibit ingroup bias in low salience situations. When social identity salience rose, low self-esteem individuals showed less ingroup bias in relation to individuals with high self-esteem. This may indicate that groups that are more important are favored less by individuals with low self-esteem. This conclusion and those that follow based on these correlations should be tempered as only one significant difference existed between any of the correlations (Direct -- All Other vs. indirect -- Collective Self-Esteem, p<.05). As the confidence intervals in Table 4 show, population correlations may differ greatly from the sample values.
Results for direct bias measures show small correlations between salience and effect size for Collective Self-Esteem (r = -0.05, ns) and Rosenberg Self-Esteem (r = -
0.15, ns). Also of note is the negative relationship found for the All Other measures cell
(r = -0.40, ns). This negative relationship indicates that low self-esteem individuals showed more bias as salience rose, a result that is opposite findings discussed in the previous paragraph. However, removal of a single data point (Salience = 55, ES = 1.63) reduces this correlation to -0.29. The substantial change in the correlation with the removal of the score indicates that the negative correlation depends heavily on one
68 comparison. Interpretation of this finding is further complicated by the fact that effect sizes in this cell range from 0.11 to 2.58. This is the only cell in which no negative effect size estimates are found. Furthermore, only three of the comparisons have salience scores of less than 155. As there are very few low salience scores in this cell and no negative effect sizes, restriction of range problems may exist. As such, results for direct bias measures provide little conclusive evidence regarding the relationship of salience to effect size. However, results do suggest that salience has its strongest effects when indirect bias measures are utilized.
The Indirect bias -- Rosenberg's Self-Esteem cell also contains an outlying data point that may affect the correlation (Salience = 381, ES = -1.92, r = -0.09). Removal of this data point results in a correlation of 0.15 for the remaining data. Though the outlying point does affect the correlation, the magnitude of the correlation indicates no relationship between salience and effect size for this cell.
Another way to examine this relationship is to divide salience scores into three equal groups (i.e., a tripartite split). Table 5 presents the median effect sizes for each of these groups. Low salience is defined as scores of 154.5 and below. Medium salience is defined as scores of 155 to 304.5. High salience is defined as scores of 305 and above.
Median scores are presented as some cells have small n, making mean scores especially sensitive to outliers.
69 Table 5
Median Effect Sizes by Salience Level, Self-Esteem Measure and Dependent Measure
Type
Collective Rosenberg's All Other Overall Self-Esteem Self-Esteem Measures Direct Bias Salience Low 0.68 0.10 0.74 0.44 n = 5 n = 10 n = 3 n = 18
Medium -0.36 0.71 0.29 0.55 n = 4 n = 12 n = 9 n = 25
High 0.28 0.30 0.30 0.29 n = 7 n = 8 n = 6 n = 21
Indirect Bias Salience Low -0.85 0.03 N/A -0.10 n = 4 n = 8 n = 0 n = 12
Medium N/A -0.37 -0.78 -0.45 n = 0 n = 7 n = 2 n = 9
High 0.14 0.26 0.47 0.18 n = 9 n = 5 n = 3 n = 16
These data further suggest that low self-esteem individuals will show more bias when using indirect measures compared to direct measures, with the pattern most prominent for Collective Self-Esteem and All Other measures. This effect is limited to
70 low and medium salience conditions (Mdn's = -0.10, -0.45, respectively). When salience
is high, individuals with high self-esteem exhibit more bias than individuals with low
self-esteem regardless of the type of dependent measure used (Mdn = 0.18).
Direct measures exhibit relatively consistent median scores across salience
conditions with overall medians ranging from 0.29 to 0.55. Of course, the relatively
small sample size in each of the observed cells makes these conclusions tenuous.
Analysis of Cell Means
Table 6 presents means and confidence intervals for ingroup bias at each
combination of dependent measure type and self-esteem measure. Though the interaction
explored in this table is not statistically significant, the means presented further clarify
patterns of relationships and allow for follow-up tests to examine differences between
direct and indirect bias for the three types of self-esteem measures. Overall, effect sizes
are higher for Rosenberg's Self-Esteem and All Other measures. Rosenberg's Self-
Esteem Scale and All Other measures show a pattern whereby effect sizes are higher for
direct bias, indicating that high self-esteem individuals exhibit more ingroup bias than
individuals with low self-esteem. Post hoc comparison tests indicate this differences is
statistically significant for Rosenberg's Self-Esteem Q(1) = 25.5, p<.001 and All Other
measures, Q(1) = 11.7, p<.05, but the difference between bias types did not attain
statistical significance for Collective Self-Esteem measures, Q(1) = 1.3, ns.
Table 6
Mean Effect Sizes Within Cells and Overall
Direct Bias Indirect Bias Overall Collective 0.09 0.00 0.06
71 Self-Esteem n = 16 n = 13 n = 29
(0.00 to 0.20) (-0.13 to 0.13) (-0.02 to 0.14) Rosenberg's 0.42 0.08 0.30
Self-Esteem n = 30 n = 20 n = 50
(0.34 to 0.49) (-0.03 to 0.18) (0.24 to 0.36) All Other 0.47 0.14 0.42
n = 18 n = 5 n = 23
(0.40 to 0.54) (-0.04 to 0.31) (0.36 to 0.49) Overall 0.37 0.06 0.28
n = 64 n = 38 n = 102
(0.33 to 0.42) (-0.01 to 0.14) (0.25 to 0.32)
Note. Overall averages cannot be directly computed from cell averages due to weighting
of effect sizes. 95% confidence interval in parentheses.
72 Sub Analyses
The three sections that follow present analyses of data (a) by self-esteem measurement, (b) by dependent variable type, and (c) for only studies with no overlapping effect size measures. These analyses allow an examination of violation of data dependency assumptions. Examining effects broken down by dependent measure reduces the impact of multiple endpoint studies and examining effects for each self- esteem measure reduces the impact of multiple-treatment studies. A third analysis removing all overlapping effects provides the most conservative analysis. Analyses of data by dependent measure type and self-esteem measure clarify interaction effects through simple effects/simple interaction tests.
Analyses by self-esteem measure. Table 7 presents analyses for the Collective
Self-Esteem Scale, Rosenberg's Self-Esteem Scale, and All Other measures.
Dependent measure type remained a significant predictor for all measures except the Collective Self-Esteem Scale. Salience effects are statistically significant only for
Collective Self-Esteem measures. The interaction between dependent measure and salience remains statistically significant for all measures except Rosenberg's Self-Esteem
Scale.
73 Table 7
Analysis of Effect Sizes for Data Separated by Self-Esteem Measure Predicted from
Dependent Variable Type, Salience, and Interaction
Self Esteem Measure Variable df Q R-Square Added Collective DV Type 1 0.5 .003 Self-Esteem (N = 29) Salience 1 17.5 *** .091 DV Type x Salience 1 7.9 ** .041 Model 3 25.9 *** .135 Residual 22 166.3 ***
Rosenberg DV Type 1 17.8 *** .057 Self-Esteem (N = 50) Salience 1 2.1 .007 DV Type x Salience 1 1.9 .006 Model 3 21.8 *** .070 Residual 43 291.7 ***
All Others DV Type 1 7.5 ** .058 (N = 23) Salience 1 2.0 .016 DV Type x Salience 1 5.1 * .084 Model 3 10.8 * .157 Residual 16 109.2 *** Note. Q is distributed as chi-square and provides a test for contribution of variables to prediction. The highly significant test for the residual value indicates that variance in effect sizes is not completely explained by the model.
*p<.05, **p<.01, ***p<.001.
74 Results for significance tests differed slightly from the overall analysis. However, as presented in Table 6, patterns of mean effect size estimates remained the same. For collective self-esteem measures, the average effect size for direct measures (M = 0.09) and indirect measures (M = 0.00) did not differ significantly though the pattern of results were consistent with those in the previous analysis. Rosenberg self-esteem measures and
All Other self-esteem measures found significant differences between direct (M's = 0.42,
0.47 respectively) and indirect measures (M's = 0.08, 0.14, respectively).
The dependent measure type by salience interaction was not statistically significant for studies that used Rosenberg's Self-Esteem, but it did attain significance for
Collective Self-Esteem and All Other measures. This result further clarifies the three- way interaction found in the full analysis. As shown in Table 4, when self-esteem was measured by the Collective Self-Esteem Scale, correlations differ between salience and effect size (r = -0.05) for direct bias measures and for indirect bias measures (r = 0.68).
A similar pattern is found for All Other measures wherein correlations when using direct bias measures (r = -0.40) are smaller than indirect bias measures (r =0.69). However, when self-esteem is measured using Rosenberg's Self-Esteem Scale, correlations for direct bias (r =0.15) and indirect bias (r = -0.09) are similar.
Again, it should be noted that the highly significant residual values for analyses of each self-esteem measure type indicate that variance in effect sizes is not completely explained by the model.
75 Analysis by variable type. To examine results by dependent measure type, two separate analyses examined studies using dependent measures classified as direct bias and indirect bias. As shown in Table 8, comparisons using direct measures showed effects for self-esteem and the interaction between self-esteem measurement and salience.
Comparisons using indirect measures showed effects for salience and the interaction.
Salience effects were statistically significant for indirect measures (r = 0.22) but not for direct measures (r = -0.03). This result points toward a relationship between salience and ingroup bias for indirect bias measures. This result is further clarified by the prior discussion of the three-way interaction.
Analysis without overlapping data. A separate analysis was conducted on the reduced data set that was produced when all studies with overlapping effects were omitted. All studies using multiple treatments (e.g., same participants in 2 or more comparisons, usually multiple self-esteem measures) or multiple endpoints (e.g., two or more dependent measures analyzed separately but representing different constructs and thus not allowing for collapsing) were removed for this analysis. This left 64 comparisons, only 53 of which included effect size estimates. This technique greatly reduced power.
76 Table 8
Analysis of Effect Sizes for Each Dependent Measure Type Predicted from Self-Esteem
Measure, Salience, and Interaction
Dependent Measure Variable df Q R-Square Added Direct SE 2 26.7 *** .064 Measures (N = 64) Salience 1 0.8 .002 SE x Salience 2 10.1 ** .024 Model 3 37.6 *** .090 Residual 25 381.5 ***
Indirect SE Measure 2 1.0 .004 Measures (N = 50) Salience 1 11.6 *** .053 SE x Salience 2 19.8 *** .091 Model 3 32.4 *** .148 Residual 46 185.7 *** Note. SE = Self-Esteem Measure. Q is distributed as chi-square and provides a test for contribution of variables to prediction. The highly significant test for the residual value indicates that variance in effect sizes is not completely explained by the model.
*p<.05, **p<.01, ***p<.001.
77 Table 9 summarizes results from this analysis and compares results to the initial analysis using the full data set. The pattern of R-squared added as indicated by a comparison of to the initial analysis is very similar. Main effects for self-esteem measure, dependent variable type, and the three-way interaction remain statistically significant. None of the two-way interactions retains statistical significance, though the size of each R-square added is similar to the corresponding R-square added in the analysis with full data. The reduction in power because of the smaller sample size accounts for the loss of statistical significance. The fit of the model to the data for the analysis with no overlapping cases is comparable to the fit for the analysis with the full data set that included overlapping comparisons and studies in which the same participants were tested on multiple dependent measures. Given the similar pattern of R-square added values, differences in statistical significance between analyses are due largely to differences in sample size.
Though the magnitude of effects (i.e., R-square added) are similar for this analysis and the analysis of the full data, this does not assure that identical effects exist.
For example, it is possible that effects take different forms but coincidentally explain the same amount of variance. Table 10 examines correlations between salience and effect size within levels of self-esteem and dependent variable type. No significant differences between correlations in the reduced set of data and the full data set were found. Only one of the cells, direct bias with Collective Self-Esteem found differences of a magnitude greater than .20. For this cell the a correlation of -0.05 was found for the full set of data and the reduced data set showed a correlation of -0.48. This difference was not statistically significant.
78 Table 9
Analysis of Effect Sizes Predicted from Dependent Variable Type, Self-Esteem Measure,
Salience, and Interactions (Data Set Without Overlapping Data Compared to Full Data)
Predictor df Q R-Square Q R-Square Added Added (Full Data) (Full Data) Dependent Variable 1 13.0 *** .040 31.1 *** .046 Type Self-esteem Measure 2 10.8 ** .033 22.7 *** .034
Salience 1 2.4 .007 6.4 * .010
Dependent Variable 2 0.4 .001 4.6 .007 X Self-esteem Dependent Variable 1 1.8 .005 6.4 ** .010 X Salience Self-esteem X Salience 2 5.2 .016 11.6** .017
Three way interaction 2 18.1 *** .055 18.3*** .027 Model 11 51.7 *** .157 101.9 *** .151
Residual 47 276.5*** 567.2***
Note. Q is distributed as chi-square and provides a test for contribution of variables to prediction. The highly significant test for the residual value indicates that variance in effect sizes is not completely explained by the model.
*p<.05, **p<.01, ***p<.001.
79 Table 10
Correlation Between Salience and Effect Size Within Cells and Overall
Direct Bias Indirect Bias Collective Reduced Data -0.48 0.71
Self-Esteem n = 5 n = 5
(-0.96 to 0.69) (-0.46 to 0.97) Full Data -0.05 0.68 *
n = 16 n = 13
(-0.42 to 0.46) (0.21 to 0.90) Rosenberg's Reduced Data 0.19 -0.03
Self-Esteem n = 20 n = 13
(-0.26 to 0.58) -0.57 to 0.53 Full Data 0.15 -0.09
n = 30 n = 20
(-0.22 to 0.48) (-0.51 to 0.37) All Other Reduced Data -0.59* 0.75
n = 12 n = 4
(-0.02 to -0.87) (-0.39 to 0.98) Full Data -0.40 0.69
n = 18 n = 5
(-0.73 to 0.08) (-0.49 to 0.98)
Note. *p<.05, **p<.01, ***p<.001. 95% confidence interval around in parentheses.
80 Also of interest is a comparison of values for main effects between the two analyses. The pattern of results for dependent measure type remained similar to the results from the full data set. Results for direct bias measures for the reduced data set (M
= 0.40, 95% CI: 0.34 to 0.47, n = 37) did not differ significantly from the full data set (M
= 0.37, 95% CI: 0.33 to 0.42, n = 64). Similarly, indirect measures showed small but, nonsignificant differences between the reduced data set (M = 0.12, 95% CI: 0.01 to 0.23, n = 22) and the full data set (M = 0.06, 95% CI: -0.01 to 0.14, n = 38). The pattern of results for self-esteem measures did change. Self-esteem, as measured by the Collective
Self-Esteem Scale differed significantly between the reduced data set (M = 0.38, 95% CI:
0.26 to 0.49, n = 10) and the full data set (M = 0.06, 95% CI: -0.02 to 0.14, n = 29).
Differences between Rosenberg's Self-Esteem Scale for the reduced data (M = 0.13, 95%
CI: 0.04 to 0.23, n = 33) compared to the full data (M = 0.30, 95% CI: 0.24 to 0.36, n =
50) were statistically significant as well. All Other measures did not differ significantly for the reduced data (M = 0.45, 95% CI: 0.37 to 0.54, n = 16) as compared to the full data
(M = 0.42, 95% CI: 0.36 to 0.49, n = 23).
Most of the results of this conservative analysis point toward a pattern of results similar to effects found in the full analysis. Self-esteem measures do exhibit a different pattern of results. However, differences between main effects for self-esteem measures do not strongly affect study conclusions as the three-way interaction involving these measures retained a pattern of results similar to those found in the full data set. However, interpretation of differences between self-esteem scales should be tempered given the possibility that these effects result from data dependency. As the R-squared added values and the pattern of correlations and means for the three-way interaction and dependent
81 measure type from this conservative analysis match the corresponding terms from the analysis of the full data set, it can be concluded that these observed effects are not artifacts produced by violations of data dependency assumptions.
Effectiveness of Bias Strategy
Analyses point toward a pattern of results whereby high self-esteem individuals use both direct and indirect strategies to exhibit bias and low self-esteem individuals are likely to use only indirect bias. However, it is unclear as to whether high self-esteem individuals use both strategies equally or if there exists a preference for one strategy over another. An interesting question is whether high self-esteem individuals prefer direct or indirect strategies. One way this question could be addressed would be to examine results from studies where participants were allowed to exhibit bias on both direct and indirect dependent measures. For example, if high self-esteem individuals given the opportunity to exhibit bias using direct dependent measures did not subsequently exhibit bias on indirect dependent measures, the result could be taken to indicate that direct measures effectively satisfy the need to exhibit ingroup bias. However, the subset of studies that allowed participants to use both types of dependent measures also used counterbalancing procedures. This analysis strategy was inappropriate to uncover differences in effectiveness of bias strategies as that information regarding order of strategy presentation would be essential to this test and was not available.
Analysis of Personal vs. Collective Measures of Self-Esteem
Self-esteem measurement data was also coded as reflecting "personal" or
"collective" self-esteem measures and effects of dependent measure type, self-esteem
82 measure, and salience were assessed. The rationale behind the current analysis follows from Crocker and Luhtanen (1990) who argued that group-based measures of self-esteem are most appropriate for group-level phenomena such as ingroup bias whereas personal self-esteem measures are most appropriate when examining individual or interpersonal phenomena. For the purposes of this analysis, personal self-esteem measures were defined as Rosenberg's Self-Esteem, Janis-Field's Feelings of Inadequacy Scale, Self-
Descriptive Questionnaire, and Texas Social Behavior Inventory. Collective measures were defined as the Collective Self-Esteem Scale and group identification measures.
As shown in Table 11, self-esteem measures defined as personal (M = 0.33, 95%
CI: 0.28 to 0.38) or collective (M = 0.22, 95% CI: 0.16, to 0.28) did not differ. Low and high self-esteem individuals showed the same pattern of results regardless of whether measures were classified as personal or collective. Main effects for dependent variable type and salience, as well as the three-way interaction and the interaction between dependent variable type and salience remained statistically significant.
Differences between self-esteem results for this analysis and the previous analysis are most likely due to the definition of "collective" measures of self-esteem. Collective measures in this analysis were defined as Collective Self-Esteem and Group
Identification measures. Collective Self-Esteem measures (M = 0.09, 95% CI LL = 0.00,
UL =0.20, n = 29) and Group Identification measures (M = 0.47, 95% CI LL =0.38, UL
=0.57, n = 10) exhibited very different mean effect sizes. Given these differences between effects based on the two scales, collapsing in this manner may not be appropriate.
83 Table 11
Analysis of Effect Sizes Predicted from Dependent Variable Type, Personal vs.
Collective Self-Esteem, Salience, and Interaction (Full Data Set)
Predictor df Q R-Square Added
Dependent Variable Type 1 31.1 *** .046
Self-esteem (Personal vs. Collective) 1 3.2 .005
Salience 1 6.5 * .010
Self-esteem X Dependent Variable 1 0.9 .001
Dependent Variable X Salience 1 5.9* .009
Self-esteem X Salience 1 2.5 .004
Three way interaction 1 11.0*** .017 Model 7 61.1 *** .091
Residual 94 567.2***
Note. Q is distributed as chi-square and provides a test for contribution of variables to prediction. The highly significant test for the residual value indicates that variance in effect sizes is not completely explained by the model.
*p<.05, **p<.01, ***p<.001.
84 Additional Issues
This section addresses additional methodological and statistical issues such as publication bias and amount of explained variance.
Publication bias. A possible alternate explanation for statistically significant findings in any meta analysis is the presence of publication bias. Published materials tend to report statistically significant results. Research that does not produce statistically significant results often remains in the "file drawer" and subsequently is not published
(Rosenthal, 1979). Several suggestions exist for assessing the possible effects of publication bias. The current study uses "funnel" plots and calculation of the "file- drawer" statistic assess possible publication bias effects.
Figure 5 plots study size versus raw effect size. Ideally, the data will represent an inverted funnel shape, where effect size becomes less disperse as the sample size rises.
This is the case with these data. These data clearly group around the raw mean effect size of 0.23. The funnel plot presents raw effect size because it is the value used directly by the researchers in testing hypotheses. Another useful application of the funnel plot is to uncover exclusion of non-significant results. If there are few cases grouped around a raw effect of 0.0, then problems with publication bias may be present (i.e., results with non- significant results and subsequently unpublished). There does not appear to be exclusion of null results. The funnel plot shows that it is unlikely that exclusion of null results have inflated the study effect sizes. This result may be attributed to the fact that many of the studies at hand provided tests of self-esteem hypotheses as ancillary results.
Additionally, data were often used to test hypothesis different from the hypotheses tested
85 in the original documents. In both cases, null results would not have prevented publication.
Figure 5. Funnel plot: Effect size by sample size
500
450
400
350
300
n 250
200
150
100
50
-2.50 -2.00 -1.50 -1.00 -.50 .00 .50 1.00 1.50 2.00 2.50 Effect Size
86 Calculation of "file-drawer" statistics produced impressive results supporting the funnel plot findings. Based on the file-drawer calculations, it would take 4166 studies with null results to reverse the statistical significance of this analysis. The file-drawer statistic however may not be appropriate in this case (or most others). The file-drawer statistic indicates the number of null results necessary to reverse statistical significance.
However, it does not account for the likelihood of results in the opposite direction canceling out effects (Begg, 1994). The effect size in this study is positive. Whereas it would take 4166 studies with effect sizes of 0.0 to reverse this statistically significant result, it would take far fewer studies with negative effect sizes to reverse this effect.
The funnel plot and file drawer statistic both indicate that the effects of this study are not likely to be artifacts of publication bias.
Percent of variance explained. Though the amount of variance explained by significant factors in this analysis may appear small, Hedges (1994) warns that interpretation of r-square values for meta analysis is limited in utility. R-square represents the percentage of total variance explained. However, effects such as nonsystematic bias within studies are unexplainable. Since much of the variance between studies is unexplainable, rather than simply unexplained, r-square statistics are poor indicators of explained variance. A better statistic would tap the percentage of explainable variance accounted for by predictors.
An important value of meta analysis is in uncovering small effects. Hunter (1997) argues for use of meta analysis in cases wherein poor power may lead to frequent Type II errors. Given the mean raw effect size estimate of 0.23 combined with the mean sample size of 53.5, the power associated with this test is 0.20. Given the low power for the
87 average test, it is clear that most studies are not sensitive enough to detect effects. If the amount of variance explained by factors was large, it is likely that most, if not all, empirical studies would reach similar conclusions. In these situations, meta analysis would not have much added value.
88 CHAPTER FOUR: DISCUSSION AND CONCLUSIONS
The purpose of this study was to test specific hypotheses regarding ingroup bias and to explore a set of related research questions. A priori hypotheses regarding dependent variable type and interactions between salience and ingroup bias strategies received some support. Additionally, results clarified the role of self-esteem measurement, identity salience, and interactions between these factors and dependent measure type. The sections that follow summarize findings and discuss results in terms of hypotheses and research questions addressed. Additional sections discuss implications for self-esteem measurement and salience. Another section includes a discussion of ingroup bias strategies and how findings regarding these strategies relate to self-esteem research in other fields and social identity theories. A final section suggests future research directions.
Summary of Findings
The following section discusses the results of the study in terms of hypotheses and addresses additional research questions.
Tests of hypotheses. Five a priori hypotheses were proposed. The first hypothesis predicted that both high and low self-esteem individuals would both exhibit ingroup bias. This hypothesis was supported and is further clarified by the second and third hypotheses. Hypotheses Two and Three predicted that high self-esteem individuals would show more ingroup bias than low self-esteem individuals on direct bias measures and low self-esteem individuals would show more ingroup bias than high self-esteem individuals when using indirect bias measures. These hypothesis received some support.
89 High self-esteem individuals did show more ingroup bias than low self-esteem individuals on direct bias measures. However, low and high self-esteem individuals showed comparable amounts of bias on indirect measures. Statistically significant differences were found in effect size between direct and indirect bias measures. Direct measures were found to produce larger differences in ingroup bias wherein high self- esteem individuals showed more bias. Indirect measures found little differences between high and low self-esteem individuals. The differences indicated that high self-esteem individuals showed more ingroup bias than low self-esteem individuals using direct measures as compared to indirect measures.
Though this result does not correspond exactly to the a priori hypothesis, it does support the proposition that low self-esteem individuals do not use direct bias strategies and that low self-esteem individuals only use indirect bias strategies. It should be noted that the data do not support analyses that directly assess the level of bias exhibited by low self-esteem individuals for direct and indirect dependent measures. Data of this form would allow for a comparison of levels of bias exhibited by low self-esteem individuals on direct/indirect dependent variables and allow for a test of the difference between the two types of variables. However, this type of test would require data reflecting the correlation between ingroup and outgroup ratings. These data were not available (see discussion of effect sizes in Chapter Two).
Regardless of these limitations, data do support the proposition that low self- esteem individuals favor indirect bias strategies. Data show that a majority of studies find ingroup bias effects in the form of more positive ratings of the ingroup compared to the outgroup. Most of the cases in which more positive ratings of the ingroup were not
90 found correspond to situations in which ingroup bias strategies reflect a minimization of differences in ratings between the ingroup and a "superior" outgroup (i.e., ingroup is low status, ingroup fails while outgroup succeeds). As ingroup bias is a consistent finding across studies, any differences in ratings between high and low self-esteem individuals reflect differential use of ingroup bias. If ingroup bias is shown in all studies, and high and low self-esteem individuals exhibit the same amount of bias in studies using indirect measures of bias, then both groups must be exhibiting ingroup bias. Given the differences between results for direct and indirect bias results, it is reasonable to conclude that low self-esteem individuals are more likely to use indirect bias whereas high self- esteem individuals will use either strategy..
Direct bias strategies such as rating the ingroup as superior on a set of adjectives may conflict with the poor self-concept of low self-esteem individuals. Regardless of this conflict, there still exists a need to self-enhance (Brown, 1993). In the intergroup setting, self-enhancement takes the form of ingroup bias. Given the need to self-enhance and the inconsistency between low self-esteem and claims of superiority, individuals with low self-esteem are unlikely to exhibit bias on "direct" measures (e.g., may not rate the ingroup as superior). The indirect measures of bias discussed in this study, such as favoring groups to which the individual did not contribute, may allow low self-esteem individuals to exhibit bias, and thus bolster self-concept. Indirect strategies may allow individuals with low self-esteem to bolster self-concept without creating conflict with previous experiences and past association with negative outcomes.
High self-esteem individuals used both direct and indirect strategies to bolster self-esteem. This also supports a self-consistency argument. Using indirect strategies
91 does not conflict with the self-concept of an individual with high self-esteem. This point provides some clarification for the first three hypotheses. Concerning bias strategy, it seems that low self-esteem individuals are more likely to use indirect strategies whereas high self-esteem individuals will use all available strategies to bolster self-esteem.
Results found for different bias strategies speak to arguments made by Crocker and colleagues (1987, 1990). Crocker and her colleagues argue that only high self- esteem individuals use ingroup bias to bolster positive social identities. Low self-esteem individuals have low self-esteem because they do not regularly engage in ingroup bias strategies. Whereas it is likely true that low self-esteem individuals do not engage in as much ingroup bias as high self-esteem individuals, it should be noted that this may be due to inconsistencies between certain bias strategies and the individual's depressed self- esteem. That is, there are fewer bias strategies available to individuals with low self- esteem as compared to those with high self-esteem. Low self-esteem individuals do engage in ingroup bias, however, they engage in fewer strategies. Indirect bias strategies also may not be as effective in bolstering self-esteem as are direct bias strategies.
The fourth and fifth hypotheses contained two predictions regarding the role of identity salience. The first prediction was that as social identity salience became stronger, ingroup bias would rise. This prediction was not directly testable due to constraints regarding the use of repeated measures tests to calculate effect sizes (again, see the section on effect sizes in Chapter Two). The second prediction stated that low self-esteem individuals would exhibit greater bias on indirect bias measures regardless of level of identity salience and high self-esteem individuals would exhibit ingroup bias on direct measures only when salience was highest. This hypothesis received some support.
92 Though low-self esteem individuals used indirect strategies more than direct strategies, low self-esteem individuals were less likely to exhibit bias in high salience situations.
This may be taken to indicate that groups that are more important are favored less by individuals with low self-esteem. This further supports a self-consistency argument.
Highly salient groups are viewed as more important and more emotionally valued.
Favoring such a group may conflict with the self-concept of individuals with low self- esteem. Individuals with low self-esteem may not expect to be associated with such groups. Further, others have suggested that individuals with low self-esteem may not favor such groups as they doubt their ability to contribute to the group or their right to belong in such a group (e.g., Seta & Seta, 1996). This pattern exists only for self-esteem defined by the Collective Self-Esteem Scale and All Other measures (though the effect was not significant for All Other measures).
High self-esteem individuals did show more bias on direct measures. However, this result was unaffected by identity salience. Low and high self-esteem individuals did not differ in bias shown on indirect measures. When using indirect strategies, high self- esteem individuals tended to show more bias in situations where salience was highest.
Again, it should be noted that this pattern was found only for Collective Self-Esteem
Scale with the same, albeit non-significant pattern found for All Other measures.
This relationship may indicate that the measures used in these cells are especially sensitive to social identity salience. First, the effect is limited to indirect bias. Also, effects are limited to Collective Self-Esteem and All Other measures. For this cell
(indirect bias -- All Other measures), the scales included were Group Identification and the Texas Social Behavior Inventory. Group Identification items are similar to Collective
93 Self-Esteem Scale items (as discussed in the methods section). As such, these measures, both focusing on group memberships, may be most sensitive to group-level phenomena such as social identity salience. Under conditions wherein both low and high self-esteem individuals to exhibit comparable amounts of ingroup bias (i.e., indirect bias strategies) and group-level esteem is measured (i.e., Collective Self-Esteem or Group Identification), effect are present whereby individuals with low self-esteem are likely to exhibit more ingroup bias than high self-esteem individuals when social identity salience is lower and high self-esteem individuals are more likely than those with low self-esteem to exhibit ingroup bias when identity salience is higher. The complexities of these results speak to the need for more research on the effects of social category salience and the relationship between salience and different types of self-esteem.
Additional research questions. Five additional research questions were examined.
Each is addressed individually below.
First, do low and high self-esteem individuals exhibit different amounts of ingroup bias? Yes, overall results indicate that high self-esteem individuals do show more ingroup bias. However, this result is clarified by the second question, do low and high self-esteem individuals use different strategies to exhibit ingroup bias? Yes, differences existed on direct bias measures, indicating that high self-esteem individuals direct bias strategies more than low self-esteem individuals. Both groups showed comparable amounts of bias on indirect measures, indicating that both low and high self-esteem individuals exhibited ingroup bias on those measures.
Do different measures of self-esteem lead to different conclusions regarding the role of self-esteem? Yes, differences exist between self-esteem measures. Self-esteem
94 measured by the Collective Self-Esteem Scale shows no effect for differences between high and low self-esteem. However, Rosenberg's Self-Esteem Scale and All Other measures do find differences between high and low self-esteem individuals. Differences between measures were not found when measures were classified as personal or collective.
What is the role of social category salience? The role of social identity salience is still unclear. Initially, analyses were planned to examine the correlation between social identity salience and ingroup bias. However, incomplete reporting of results made this task impossible. The role of social identity salience was clarified to some extent by the final question. Are there interactions between the above factors? Yes, as addressed in the section on hypothesis tests, salience interacted with both self-esteem measurement and dependent measure type (bias strategy).
Self-Esteem Measurement
Three primary issues regarding self-esteem measurement were uncovered. First, artificial dichotomization of self-esteem scores limits research results. Second, treatment of self-esteem data may be flawed. Third, differences between specific self-esteem measures were found.
Dichotomization. Analyses of self-esteem scale data indicated a tendency to dichotomize self-esteem scale scores into high and low self-esteem categories based on median scores. Sixty-four of 113 comparisons examined used artificial dichotomization into high and low self-esteem groups. Much of this dichotomization resulted in questionable classification of individuals as "low self-esteem." These findings are similar
95 to those of Tice (1993) who, in a review of self-esteem studies, concluded that individuals classified as "low self-esteem" would more accurately be termed "medium self-esteem."
The effects of artificial dichotomization of self-esteem measures would most likely result in a reduction of raw effect sizes and a subsequent increase in the probability of Type II errors. This methodological shortcoming likely results from over dependence on ANOVA (or t-test) procedures.
This meta analysis included adjustments for the effects of dichotomization. This strategy does provide adjustment for reduction in effect sizes. However, it would be preferable if primary research used correlation/regression data analysis strategies that did not require dichotomization. Further, this strategy would allow for tests of a wider range of hypotheses (e.g., presence of non-linear trends).
Normality. The adjustment made for artificial dichotomization assumes a normal distribution of self-esteem scores. None of the studies included in this analysis addressed normality issues for self-esteem data. However, given the relatively high median scores, it is unlikely that normal distributions were approximated. Further, Blascovich and
Tomaka (1991) report that some self-esteem measures tend to be skewed in certain populations. If the self-esteem scores in the current analysis were in fact skewed, it is possible that the adjustment for attenuation used in this study provided an extremely conservative estimate of attenuation.
Another issue concerns treatment of these data. It was unlikely that self-esteem scale scores achieved normality. No studies reported the use of data transformations to achieve normality. This could have affected results of studies that used
96 correlation/regression procedures as lack of normality in variables can lead to violation of other specific regression assumptions (e.g., normality of residuals; Tabachnick & Fidell,
1989; but also see Wainer, 1976). Studies using ANOVA techniques and those using correlation/regression both suffer from statistical shortcomings.
Type of measure. Differences existed between results for self-esteem defined by the Collective Self-Esteem Scale, Rosenberg's Self-Esteem Scale, and All Other self- esteem measures. For Rosenberg's Self-Esteem and All Other measures, there was a clear pattern whereby high self-esteem individuals exhibited greater amounts of bias than did individuals with low self-esteem. This result was not present for comparisons involving the Collective Self-Esteem Scale.
This finding is worthy of comment in that the development of the Collective Self-
Esteem Scale and results from a prominent study by Crocker and Luhtanen (1990) introducing this scale are at the heart of much of the controversy over the relationship between self-esteem and ingroup bias. Crocker and Luhtanen distinguished between personal self-esteem (usually measured by Rosenberg's Self-Esteem Scale) and group- based self-esteem (measured by the Collective Self-Esteem Scale). The authors suggested that relations between groups can be predicted by group-based self-esteem whereas personal self-esteem is better conceptualized as a predictor of interpersonal relations. The current study found that Rosenberg's self-esteem does predict ingroup bias. Further, self-esteem measured by the Collective Self-Esteem Scale does not appear to be directly related to ingroup bias. However, interactions with salience do point towards the importance of Collective Self-Esteem. Crocker and Luhtanen also argued that Social Identity Theory predictions are most applicable to individuals with high self-
97 esteem. The results of this meta-analysis do not support this claim. Social Identity predictions appear to apply to individuals with high or low self-esteem. Both groups exhibit ingroup bias.
When analyses classified measures as personal or collective, both classes of measures show that high self-esteem individuals exhibited more ingroup bias. This result may be due to the inclusion of group identification measures as measures of collective self-esteem. A recent factor analysis indicated that while group identification measures and Collective Self-Esteem subscales do load on similar factors, the group identification measures may be better conceptualized as measures of group attraction rather than group- based self-esteem (Jackson & Smith, 1999).
Interactions further qualify conclusions regarding the Collective Self-Esteem
Scale and may actually point toward the importance of collective self-esteem. Self- esteem measurement, social identity salience, and dependent measure type interact in predicting differences in ingroup bias. A strong correlation between social identity salience and effect sizes in studies using indirect measures drives the interaction. It may be the case that differences between the amount of ingroup bias exhibited by low and high self-esteem individuals are moderated by salience. To date, the relationship between self-esteem and identity salience has received no empirical attention and would likely benefit from future investigation.
Another possible explanation for lack of an overall effect for studies that measured self-esteem using the Collective Self-Esteem Scale results from possible deficiencies in the utilization of the scale. Conceptually, the Private Self-Esteem
Subscale may be the most appropriate measure of self-esteem derived from group
98 memberships as this subscale measures internalized perceptions of the value of group memberships (Luhtanen & Crocker, 1991; also see Long & Spears, 1997). However, studies varied in the use of the Collective Self-Esteem Scale and Collective Self-Esteem
Subscales. Many studies used overall scale values whereas others presented various subscales. As the Private Self-Esteem Subscale is reported to be the most appropriate measure of group-based self-esteem, it is surprising that many study authors use other subscales as predictors of ingroup bias. It seems that much of the use of the Collective
Self-Esteem Scale does not follow theoretical rationale.
Social Category Salience
Social category salience effects are difficult to interpret. An overall pattern existed whereby high self-esteem individuals showed greater ingroup favoritism as social category salience rose. This could be interpreted as low self-esteem individuals exhibiting less ingroup bias when salience was high. In situations wherein low self- esteem individuals showed comparable amounts of ingroup bias (i.e., indirect bias measures), a pattern was found for self-esteem measured by the Collective Self-Esteem
Scale and All Other measures that indicated salience was positively correlated with differences between high and low self-esteem ingroup bias. In this case, a positive correlation indicates higher salience to be associated with more bias shown by individuals with high self-esteem. That is, low self-esteem individuals showed more bias when salience was low and high self-esteem individuals showed more bias when salience was high. However, it is clear from interaction effects that salience is an important consideration warranting future investigation.
99 Some empirical attention has been paid to social category salience. Most research has focused on aspects of category salience such as category fit (e.g., Hogg & Turner,
1987; Oakes, et al., 1991). However, no investigations tested issues of social category salience interactions related to self-esteem. A question for future studies in this area is whether low/high self-esteem individuals respond differently to different levels of social identity salience.
One conclusion stemming from the current study would be that low self-esteem individuals might exhibit inhibitions against exhibiting ingroup bias under certain circumstances. Favoring a highly salient ingroup is a means of enhancing the self, but low self-esteem individuals may be less likely to engage in this strategy because they do not see themselves as effective members of such groups. Ingroups that are the most salient may be those groups that are the most important and the most positively valued by others. Low self-esteem individuals may not favor these groups as it is inconsistent with their negative self-image to be associated with groups that are positive. Additionally, these individuals may imagine themselves as unimportant members of these groups, limiting their ability to gain esteem from association (Seta & Seta, 1996). Further, low self-esteem individuals may not identify as members of such groups or even view the group less positively due to their association with the group (Seta & Seta, 1992).
Bias Strategies (Dependent Variable Type)
The most definitive findings of this meta analysis concern the role of dependent measure type. Findings show a clear pattern whereby individuals with high self-esteem show more bias than low self-esteem individuals when using direct bias strategies
100 whereas both groups show comparable amounts of bias when using indirect strategies.
This may indicate that high self-esteem individuals use all available strategies but those with low self-esteem prefer strategies that are indirect such as rating themselves as more similar to the ingroup (but not "better" than the outgroup), favoring groups to which they did not contribute, or basking in reflected glory. The difference in measures taps important concepts regarding how individuals with low and high self-esteem exhibit ingroup bias. The positive effect size for direct measures indicates that individuals with high self-esteem show more bias than individuals with low self-esteem when using these measures. Direct measures were defined as those requiring claims of superiority such as rating the ingroup better on a set of positive adjectives. This result indicates that individuals with high self-esteem show a greater tendency to claim ingroup superiority.
When examining indirect bias measures, only small differences were found between low and high self-esteem individuals. The near zero effect in this case indicates that high and low self-esteem individuals did not exhibit differences in the use of this strategy. This indicates that when using indirect measures of bias, low self-esteem individuals exhibit levels of ingroup bias comparable to individuals with high self- esteem.
The results of the current analysis support a growing body of literature that finds low/high self-esteem (or low/high status) individuals to exhibit different patterns of self and group favoring behaviors. The sections that follow discuss similarities between this finding and findings regarding self-esteem and status from the fields of social comparison, personality, and intergroup relations.
101 Social comparison studies. Social comparison studies also point toward a moderating role for self-consistency regarding the relationship between self-esteem and downward comparison. Gibbons and McCoy (1991) distinguished between active and passive downward comparisons (DC) as strategies preferred by high and low self-esteem individuals respectively. Active DC refers to the tendency to rate the self (or the group) as better off or the tendency to derogate a worse off target. Passive DC refers to utilization of DC opportunities for mood enhancement such as the continued choice of worse off targets for comparisons. A pair of studies (Gibbons & McCoy, 1991; Wood,
Giordano-Beech, Taylor, Michela, & Gaus, 1994) demonstrated that high self-esteem individuals preferred active strategies such as evaluations of selves as superior to failing others. Low self-esteem individuals did not use this strategy. When passive comparison strategies, such as indicating preferences for future comparisons with failing others, were available, low self-esteem individuals showed greater tendencies toward utilization.
Personality studies. Research from the field of personality psychology characterizes high self-esteem individuals as using self-enhancing strategies whereas low self-esteem individuals make use of self-protective strategies. Self-enhancing strategies are those which directly benefit the self, such as a willingness to accept risks, a focus on one’s own good qualities, or calling attention to the self. Self-protection strategies focus on an avoidance of risk taking, a focus away from bad qualities, and a reluctance to draw attention to the self (Baumeister, Tice, & Hutton, 1989). In general, self-enhancing strategies allow high self-esteem individuals to feel better about themselves whereas self- protective strategies help low self-esteem individual not feel any worse about themselves.
102 Several studies demonstrated use of these strategies. One study examined reactions of study participants to successful performances. After succeeding at a task, high self-esteem individuals were more likely than low self-esteem individuals to spend additional time working on the task (Baumeister & Tice, 1985). The authors interpret this as a willingness on the part of high self-esteem individuals to risk failure in order to succeed again whereas low self-esteem individuals were unwilling to attempt the task again as they did not want to take the chance of failing and thus reducing the positive impact of the initial success.
Further research by Tice (1991) indicates that individuals with high self-esteem use self-handicapping strategies, such as limiting practice time, to enhance attributions of ability when success occurs. Low self-esteem individuals use self-handicapping when expecting failure to enhance attributions of lack of effort and downplay ability as a reason for failure. Another study of failures found both groups attempted to bolster selves after a failure, but those low in self-esteem only enhanced themselves in a private forum whereas individuals with high self-esteem bolstered themselves regardless of forum
(Brown & Gallagher, 1992).
A study examining self-esteem in romantic relationships found low self-esteem individuals to be more likely to self-enhance indirectly through bolstering ratings of their partners whereas high self-esteem individuals preferred to bolster ratings of themselves
(Schütz & Tice, 1997), indicating that low self-esteem individuals enhance through
“basking in reflected glory” of their partners. Individuals with low self-esteem attempted to gain positive self-images through association with positively valued individuals whereas high self-esteem individuals simply rated themselves as better than others.
103 Studies of status effects. Similar to the high self-esteem individuals in this analysis, members of high status groups exhibit ingroup bias effects in a more consistent manner than do members of low status groups (Mullen, Brown, & Smith, 1992). Blanz,
Mummendy, and Otten (1995a, 1995b) argued that members of low status groups might use dimensions of ingroup bias not typically measured in ingroup bias studies. Ingroup bias studies tend to focus on distribution of positive stimuli such as positive attributes and benefits to the ingroup and outgroup. This ignores the distribution of negative stimuli such as negative attributes and punishments. Blanz, et al. (1995a, 1995b) found that members of low status groups favor the use of negative evaluations whereas members of high status groups favor use of positive evaluations. An implication of these findings is that the bulk of studies examining the relationship between status and ingroup bias focus on dimensions favored by high status groups. Thus, findings indicating that high status groups exhibit greater ingroup bias result from the type of ingroup bias strategy measured.
Summary of findings from other fields. Social comparison, personality, and status research all support the proposition that low self-esteem individuals engage in different types of strategies than do high self-esteem individuals to favor themselves and their groups. The findings of each field also support the proposition that individuals with low self-esteem favor strategies that are indirect and possibly less effective. Low self- esteem individuals' responses seem to reflect less direct attempts to increase the self- image such as bolstering association with positively valued others or choosing to limit social comparisons to those whom they believe are worse-off. Individuals with low self- esteems do not engage in self-aggrandizing statements of their own abilities.
104 The similarities between these finding and the results of the current study paint a clear picture of low self-esteem individuals exhibiting different self-enhancement strategies at both the interpersonal and the intergroup levels.
Implications for Theory
This section focuses on implications of the current study to Social Identity Theory
(specifically, Corollary 2 -- low self-esteem leads to greater ingroup bias) and discusses implications for other social identity theories such as Self-Categorization Theory and
Social Self-Regulation Theory.
Social Identity Theory. Social Identity Theory posits that low self-esteem leads to increased ingroup bias (Hogg & Abrams, 1990). This perspective argues that individuals with low self-esteem need to pick on others to raise deficient self-esteem whereas high self-esteem individuals do not need to bolster self-esteem and thus do not engage in such behaviors.
Influential studies by Crocker and colleagues (1990, 1993) challenge this perspective, arguing that high self-esteem individuals are more likely to exhibit ingroup bias. Another perspective suggests that low and high self-esteem individuals exhibit ingroup bias in different manners (e.g., Brown, et al., 1988). Results of this meta analysis support this perspective. Both low and high self-esteem individuals exhibit ingroup bias.
However, low self-esteem individuals are limited by self-consistency needs (Brown,
1993). When ingroup bias strategies conflict with self-concept, low self-esteem individuals exhibit less bias. When ingroup bias strategies do not conflict with self-
105 concept, low self-esteem individuals show the same amount of ingroup bias as do individuals with high self-esteem.
This finding relates particularly to the type of bias exhibited by individuals low in self-esteem. When bias strategies require individuals to rate themselves or their groups as superior, those with low self-esteem do not show bias. They do not expect themselves to be superior and will not make such claims. Two studies by Seta and Seta (1992, 1996) examining the role of participant vs. observer status within groups point toward the validity of this view. In these studies, participant status was defined as actively contributing to group products whereas observer status was defined as being a non- contributing member of the group. Low self-esteem individuals only exhibited bias when observing. That is, when these individuals were active contributors, they did not favor the ingroup, possibly indicating that they do not expect to be associated with positive outcomes. The authors' later study clarified this effect by giving those low in self-esteem feedback regarding their specific performances. When given failure feedback, low self- esteem individuals exhibited the same pattern of results. However, when the low self- esteem individuals were told that they (individually) were successful, they were more likely to exhibit ingroup bias in all situations (even those categorized as direct bias for this study). As results were similar for no feedback and failure feedback conditions, it can be assumed that those low in self-esteem expect to do poorly and will act in accordance with that expectation. Relating these results specifically to the analysis at hand, it appears that individuals low in self-esteem will not favor the ingroup using direct bias strategies such as rating the ingroup as superior as these individuals do not expect association with positive outcomes. One implication of this result is that "traditional"
106 ingroup bias measures fall short when considering the case of the individual with low self-esteem.
Self-Categorization Theory and Social Self-Regulation Theory. Two additional social identity perspectives, specifically, Self-Categorization Theory and Social Self-
Regulation theory advance Social Identity Theory predictions. Self-Categorization
Theory introduces the role of identity salience and normative influence. Abrams' (1993,
1994) Social Self-Regulation theory introduces self-attention as a factor in determining intergroup behavior. Social Self-Regulation is unique in that it also attempts to explain situations in which ingroup bias is not exhibited.
Both perspectives posit a self-monitoring component to the expression of ingroup bias. Aspects of Self-Categorization Theory focus on normative influence. Turner
(1987) argues that many behaviors at the group level involve reference to group norms
(prototypes) for ideas about how to respond to stimuli. As there is a general norm to prefer one's own groups, ingroup bias occurs. However, normative factors moderate the use of ingroup bias. This perspective dovetails with modern theories of racism (e.g.,
Dovidio & Gaertner, 1991; Sears, 1988) in that both argue that expression of bias is determined by normative factors.
As many studies of ingroup bias use minimal groups, it would seem that self- monitoring concerns do not apply. However, if self-consistency needs are taken into consideration, it becomes clear that expression is determined by a different type of self- monitoring concern. This could be termed "internal self-monitoring." The individual looks toward internal norms to determine action in this case. Rather than looking exclusively to others to determine normative behaviors, individuals with low self-esteem
107 use self-perceptions to determine the most appropriate strategy of bias. Though the focus of Self-Categorization Theory is external factors, reactions to outgroups may be affected by both external and internal factors.
Social Self-Regulation conceptualizes ingroup bias as an automatic process regulated by identity salience and self-awareness. Self-awareness is defined as either public or private. Public focus refers to being aware of and conforming to what others expect. Private self-focus refers to conforming to self or ingroup-defining characteristics.
Private self-awareness results in a focus on self-defining or group defining characteristics, depending on the salience of personal or social identities. In intergroup settings, private self-awareness results in ingroup bias. When public self-attention is high individuals tend to respond in manners that may not correspond to ingroup favoritism.
That is, individuals often respond in non-biased manners to achieve specific goals. One specific goal fostered by public awareness is conformity to a general norm of egalitarianism, as would be posited by modern racism theories.
Social Self-Regulation Theory posits that only group-level concerns affect intergroup behavior. That is, when social identity is salient, all aspects of behavior are controlled by reference to group-level criteria. For example, when the social group is salient, individuals act exclusively as group members, referencing group norms and group goals as guides for action. The results of this study indicate that self-esteem predicts ingroup bias. Just as conformity to public norms affects ingroup bias, conformity to personal norms and expectations may affect ingroup bias as well. Personal norms and expectations may have roots in self-esteem. If an individual has low self-
108 esteem, rating an ingroup as superior may be unlikely, giving the appearance of non- biased responding.
The presence of individualistic factors that predict responses to groups conflicts with Self-Categorization and Social Self-Regulation perspectives. These theories propose an all or nothing view of identity salience. Personal identity or social identity (or human identities) are all possible but only one identity is salient at any given time. From this perspective, the group is an extra-individual entity that is greater than the sum of the individual parts (Turner, et al., 1994). This perspective is unique in that it conceptualizes the group as being more than a simple collection of individuals. However, this rejection of individual-level phenomena may limit explanations.
The exclusively group-based propositions of Self-Categorization and Social Self-
Regulation Theories conflict with the results of the current study. Aspects of the self such as self-esteem prove important predictors of ingroup bias. As such, individual factors such as self-esteem may function simultaneously with group-level perceptions.
Given these findings, future theorization may benefit from inclusion of the possibility of multiple levels of identity acting simultaneously.
Social vs. individual level explanations. Social identity theories such as Self-
Categorization Theory and Social Self-Regulation are completely social explanations for all behavior. Much of the social identity research focuses on group-level phenomena.
Both theories place the social group at the center of all explanations. Individual phenomena exist but are viewed exclusively as predictors of interpersonal process and otherwise ignored.
109 This perspective is in sharp contrast to North American psychological views.
North American psychology primarily reflects a definition of groups as collections of interdependent individuals (Cartwright, 1969). Turner and colleagues (1987) take the opposite view, arguing that the group is a socially constructed entity with specific properties that group members use for self-definition. A limitation of both views is that they largely ignore the other. Explanations are either all group or all individual.
The social identity perspectives developed by Tajfel and colleagues shifted the focus of psychological investigation of group processes from the self to the collective.
This shift represents a rejection of investigations of how people behave within groups in favor of exploration of how groups (or the psychological representation of the group) function within the individual (Miller & Prentice, 1994). This focus represents movement away from explanations centering on individual level processes. Social
Identity Theory and its progeny focus on the importance of collective processes.
Turner, Abrams, and others extended the initial statements of Social Identity
Theory to the development of Self-Categorization Theory and Social Self-Regulation
Theory. Both of the "new" social identity theories focus on the definition of identity as comparative, fluid, and context dependent (e.g., Turner et al., 1994). Depending on context, one of three salient identities is present; (a) individual, (b) social, or (c) human.
Leaving aside categorization of the self in the most general category (human), a fundamental issue is that of identity dichotomy. Social identity theories argue that when social identity is salient individuals behave exclusively as group members. When individuals do not act as group members (i.e., do not exhibit ingroup bias) personal
110 identity is assumed to be salient. As such, the new social identity theories allow little room for individual variation in behavior when social identity is salient (Abrams, 1993).
Recent theorization and research in intergroup relations has recognized the existence and simultaneous functioning of multiple social identities (Brewer, 1997). At any one time, an individual can respond to two or more salient social identities, often affecting ingroup bias responses (see reviews by Migdal, Hewstone, & Mullen, 1998;
Urban & Miller, 1998). If an individual can respond to multiple social identities then it must be the case that the individual has the capability to respond to and form multiple identities based on multiple comparative dimensions in multiple contexts. Given the capacity for multiplicity of social identities, it is likely that both social identity and individual identity also work simultaneously. If individuals can respond to multiple social categories, then it logically follows that individuals can respond to social and personal identities at the same time.
The proposition that individuals respond either as group members or as individuals (but not both) appears to be increasingly untenable. Results of the current study indicate that individual level characteristics such as self-esteem do predict group- level phenomena such as ingroup bias. Group level self-esteem measures did not predict group-level responses any more than did individual-level measures. In some cases, group-level self-esteem measures were less predictive of ingroup bias than were personal self-esteem measures. Given these results, it appears that group and individual identities do act simultaneously.
Future theorization may benefit from consideration of the possibility of simultaneously activated personal and social identities. Social Self-Regulation, on the
111 surface, appears to integrate some individual-level concerns. However, the increased individualistic approach is only nominal. Social Self-Regulation integrates self- awareness concerns. However, self-awareness concerns are limited by salient categorization (individual or social). When social identity is salient, self-awareness refers to awareness of the self as a representative of the group and not to individual level characteristics.
Given the shortcomings of social identity theories in explaining the relationship of self-esteem to ingroup bias, it is clear that future theorization may benefit from integration of individualistic explanations. Specifically, factors such as individual self- esteem may improve predictions of social identity theories. This is ironic in that this relationship "may come to be explained by recourse to the very same individualistic explanations (basic cognitive processes, personality, etc.) which social identity theorists eschew" (Abrams, 1993, p.68).
Future Research Directions
This study has clarified some aspects of the relationship between self-esteem and ingroup bias. However, many additional research questions remain. One fundamental issue is that of the complex relationship between ingroup bias and social identity salience.
Whereas theorists predict that ingroup bias rises with identity salience, the relationship may be more complex, wherein identity salience interacts with other factors. Another salience issue is the relationship between individual and social identity. Previous research and theorization conceptualized individual and social identity as mutually
112 exclusive categories. Future studies may benefit from conceptualizing these as independently functioning categories.
Another issue is the effectiveness of different bias strategies. One conclusion drawn from these data is that high self-esteem individuals use a broader range of ingroup bias strategies. If ingroup bias is a path to positive self-concept, then it is reasonable to assume that high self-esteem individuals use strategies that are more effective. Low self- esteem individuals rely on indirect strategies. These strategies may be less effective in bolstering self-esteem, perhaps perpetuating the individual's low self-esteem. It may also be the case that individuals low in self-esteem do not bolster self-esteem, as they use fewer bias strategies and hence have fewer esteem raising opportunities. Future studies could allow individuals to choose social comparison dimensions or specifically test whether use of strategies varies based on order of strategy presentation.
Concluding Comments
This meta analysis demonstrates that ingroup bias is affected by individual-level factors such as self-esteem, group-level factors such as social identity salience, and methodological factors such as scale of measurement and type of dependent measure utilized. The effects of each of these factors have implications for the study of ingroup bias and more generally, the study of prejudice and racism.
Self-esteem effects indicate that individual-level factors do affect group level processes such as ingroup bias. This result suggests that responses to groups are at least partially moderated by individual concerns. This represents a departure from recent
113 theorization in the social identity tradition. As such, predictions of social identity theories may benefit from acceptance of individual level explanations.
Different classes of dependent measures produced different effects. One interpretation of these results is that findings show a pattern whereby individuals with high self-esteem show more bias than low self-esteem individuals when using direct bias strategies whereas both groups show comparable amounts of bias when using indirect strategies. Low self-esteem individuals engaged in fewer ingroup bias strategies than did individuals with high self-esteem, only exhibiting bias on "indirect" measures. High self- esteem individuals exhibited bias on all measures. These results add to a growing body of literature from the fields of personality and social comparison that indicate low and high self-esteem individuals differ in use of self-enhancement strategies.
Measurement of self-esteem remains fecund for future investigation. The current study found Collective Self-Esteem least predictive of ingroup bias. However, these results were qualified by interactions with social identity salience. Future studies should clarify the predictive power of different self-esteem measures. More troubling was the treatment of self-esteem scores. Artificial dichotomization of self-esteem scores was widespread, often resulting in questionable classification of individuals as "low self- esteem." This methodological shortcoming could be ameliorated through use of correlation/regression procedures instead of ANOVA designs.
Finally, the role of identity salience was investigated. A complex relationship was found whereby salience was related to ingroup bias for comparisons using group based self-esteem measures and indirect bias. This may indicate a pattern of results wherein low self-esteem individuals show less bias as salience rises. Many questions
114 remain regarding salience of identities. Specifically, ideas about the possibility of the existence of complex relationships between self-esteem, bias strategy, and salience would benefit from future investigation.
115 References
References marked with an asterisk indicate studies included in the meta analysis.
*Aberson, C. L. (1999). Low self-esteem and ingroup bias. Social Behavior and
Personality, 27, 17-28.
Abrams, D. (1993). Processes of social identification. In G. Breakwell (Ed.),
Social psychology of identity and the self-concept (pp. 57-99). London: Surrey
University Press.
Abrams, D. (1994). Social self-regulation. Personality and Social Psychology
Bulletin, 20, 473-483.
Abrams, D., & Hogg, M. A. (1988). Comments on the motivational status of self-esteem in social identity and intergroup discrimination. European Journal of Social
Psychology, 18, 317-334.
Ashmore, R. D., & Del Boca, F. K. (1976). Psychological approaches to understanding intergroup conflict. In P. Katz (Ed.), To wards the elimination of racism
(pp. 73-123). New York: Pergamon Press.
Bagley, C., Verma, G., Mallick, K., & Young, L. (1979). Personality, self- esteem, and prejudice. Westmead, England: Saxon House.
Baumeister, R. F., & Tice, D. M. (1985). Self-esteem and responses to success and failure: Subsequent performance and intrinsic motivation. Journal of Personality, 53,
450-467.
116 Baumeister, R. F., Tice, D. M., & Hutton, D. G. (1989). Self-Presentational motivations and personality differences in self-esteem. Journal of Personality, 57, 547-
579.
Begg, C. B. (1994). Publication bias. In H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 399-410). New York: Russell Sage Foundation.
Blanz, M., Mummendey, A., & Otten, S. (1995a). Perceptions of relative group size and group status: effects on intergroup discrimination in negative evaluations.
European Journal of Social Psychology, 25, 231-247.
Blanz, M., Mummendey, A., & Otten, S. (1995b). Positive-negative asymmetry in social discrimination: The impact of stimulus valence and size and status differentials on intergroup evaluations. British Journal of Social Psychology, 34, 409-419.
Blascovich, J., & Tomaka, J. (1991). Measures of self-esteem. In J.P.
Robinson, P.R. Shaver, & L.S. Wrightsman (Eds.), Measures of personality and social psychological attitudes (pp. 115-160). San Diego, Ca: Academic Press, Inc.
Brewer, M. B. (1991). The social self: On being the same and different at the same time. Personality and Social Psychology Bulletin, 17, 475-482.
Brewer, M. B. (1993). The role of distinctiveness in social identity and group behavior. In M. A. Hogg & D. Abrams (Eds.), Group motivation: Social psychological perspectives (pp. 1-16). London: Harvester Wheatsheaf.
Brewer, M. B. (1997). The social psychology of intergroup relations: Can research inform practice? Journal of Social Issues, 53, XX - XXX
Brewer, M. B., & Kramer, R. M. (1985). The psychology of intergroup attitudes and behaviors. Annual Review of Psychology, 36, 219-244.
117 *Brickson, S. L. Correlation between collectivism, self-esteem, and ingroup bias: Costa Rica and the United States. Unpublished master's thesis, Carleton College,
Northfield, Minnesota.
*Brockner, J., & Chen, Y. (1996). The moderating roles of self-esteem and self- construal in reaction to threat to the self: Evidence from the People's Republic of China and the United States. Journal of Personality and Social Psychology, 71, 603-615.
Brown, J. D. (1993). Motivational conflict and the self: The double-bind on low self-esteem. In R. F. Baumeister (Ed.), Self-Esteem: The puzzle of low self-regard
(pp.117-130). New York: Plenum Press.
Brown, J. D., & Gallagher, F. M. (1992). Coming to terms with failure: Private self-enhancement and public self-effacement. Journal of Experimental Social
Psychology, 28, 3-22.
*Brown, J. D., Collins, R. L., & Schmidt, G. W. (1988). Self-Esteem and direct versus indirect forms of self-enhancement. Journal of Personality and Social
Psychology, 55, 445-453.
*Brown, R., Condor, S., Mathews, A., Wade, G., & Williams, J. (1986).
Explaining intergroup differentiation in an industrial organization. Journal of
Occupational Psychology, 59, 273-286.
Bruner, J. S. (1957). On perceptual readiness. Psychological Review, 64, 123-
51.
Campbell, J. D., & Lavallee, L. F. (1993). Who am I? The role of self-concept confusion in understanding the behavior of people with low self-esteem. In R. F.
118 Baumeister (Ed.), Self-Esteem: The puzzle of low self-regard (pp. 3-20). New York:
Plenum Press.
Cartwright, D. (1979). Contemporary social psychology in historical perspective.
Social Psychology Quarterly, 42, 82-93.
Cialdini, R. B., & DeNicholas, M. E. (1989). Self-presentation by association.
Journal of Personality and Social Psychology, 57, 626-631.
Cialdini, R. B., & Fultz, J. (1990). Interpreting the negative mood-helping literature via "mega"-analysis: A contrary view. Psychology Bulletin, 107, 210-214.
Cohen, J. (1983). The cost of dichotomization. Applied Psychological
Measurement, 7, 249-253.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral science. Lawrence Erlbaum Associates: New Jersey.
*Crocker, J., & Luhtanen, R. (1990). Collective self-esteem and ingroup bias.
Journal of Personality and Social Psychology, 58, 60-67.
Crocker, J., Luhtanen, R., Blaine, B., & Broadnax, S. (1994). Collective self- esteem and psychological well-being among White, Black, and Asian college students.
Personality and Social Psychology Bulletin, 20, 503-513.
Crocker, J., & Major, B. (1989). Social stigma and self-esteem: The self- protective properties of stigma. Psychological Review, 96, 608-630.
*Crocker, J., & Schwartz, I. (1985). Prejudice and ingroup favoritism in a minimal intergroup situation: Effects of self-esteem. Personality and Social Psychology
Bulletin, 11, 379-386.
119 *Crocker, J., Thompson, L., McGraw, K.M., & Ingerman. C. (1987).
Downward comparison, prejudice, and evaluation of others: Effects of self-esteem and threat. Journal of Personality and Social Psychology, 52, 907-916.
*Donaldson, S. J. (1995). Factors affecting the self-esteem hypothesis: Self- serving biases in the intergroup situation. Dissertation Abstracts International, 56 (05B),
2939. (University Microfilms No. AAI9531831).
Dovidio, J. F., & Gaertner, S. L. (1991). Changes in the expression and assessment of racial prejudice. In H. J. Knopke, R. J. Norrell, & R. W. Rogers
(Eds.), Opening doors: Perspectives on race relations in contemporary America (pp.119-
148). Tuscaloosa, AL: University of Alabama Press.
Duckitt, J. (1992). Psychology and prejudice: A historical analysis and integrative framework. American Psychologist, 47, 1182-1193.
Dunlap, W. P., Cortina, J. M., Vaslow, J. B., & Burke, M. J. (1996). Meta- analysis of experiments with matched groups or repeated measures designs.
Psychological Methods, 1, 170-177.
Ehlrich, H. J. (1973). The social psychology of prejudice. New York: Wiley.
Farsides, T. (1995). Why social identity theory's self-esteem hypothesis has never been tested -- and how to test it. Unpublished manuscript.
Fiske, S. T., & Taylor, S. E. (1991). Social cognition. New York: McGraw-
Hill.
Gaertner, L., & Schopler, J. (1998). Perceived ingroup entativity and intergroup bias: An interconnection of self and others. European Journal of Social Psychology, 28,
963-980.
120 *Gagnon, A., & Bourhis, R. Y. (1996). Discrimination in the minimal group paradigm: Social identity or self-interest. Personality and Social Psychology Bulletin, 22,
1289-1301.
Gibbons, F. X., & McCoy, S. B. (1991). Self-Esteem, similarity, and reactions to active versus passive downward comparison. Journal of Personality and Social
Psychology, 60, 414-424.
Glass, G. V., McGaw B., & Smith, M. L. (1981). Meta analysis in social research. Beverly Hills, CA: Sage.
Gleser, L. J., & Olkin, I. (1994). Stochastically dependent effect sizes. In H.
Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 339-356). New
York: Russell Sage Foundation.
*Greenwood, L. (1996). Effects of social categorization, repressive coping, and differential status on intergroup discrimination (Doctoral dissertation, Lehigh University,
1996). Dissertation Abstracts International, 57, 5977.
Hamilton, D. L., Sherman, S. J., & Lickel, B. (1998). Perceiving social groups:
The importance of the entativity continuum. In C. Sedikides, J. Schopler, & C. A.
Insko (Eds.) Intergroup cognition and intergroup behavior. New Jersey: Lawrence
Erlbaum Associates.
Hedges, L. V. (1994). Fixed effects models. In H. Cooper & L. Hedges (Eds.),
The handbook of research synthesis (pp. 285-299). New York: Russell Sage Foundation.
Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis.
Orlando, FL: Academic Press.
121 *Hinkle, S., Taylor, L. A., Fox-Cardamone, D. L., & Crook, K. F. (1989).
Intragroup differentiation and intergroup differentiation: A multicomponent approach.
British Journal of Social Psychology, 28, 305-317.
Hogg, M. A., & Abrams, D. (1990). Social motivation, self-esteem, and social identity. In D. Abrams & M. A. Hogg (Eds.), Social Identity Theory: Constructive and critical advances. New York: Springer-Verlag.
Hogg, M. A., & Turner, J. C. (1987). Intergroup behaviour, self-stereotyping and salience of social categories. British Journal of Social Psychology, 26, 325-340.
Hunter, J. A., Platow, M. J., Bell, L. M., Kypri, K., & Lewis, C. A. (1997). intergroup bias and self-evaluation: Domain-specific self-esteem, threats to identity, and dimension of importance. British Journal of Social Psychology, 36, 405-426.
Hunter, J. A., Platow, M. J., Howard, M. L., & Stringer, M. (1996). Social identity and intergroup evaluative bias: Realistic categories and domain-specific self- esteem in a conflict setting. European Journal of Social Psychology, 26, 631-647.
*Hunter, J. A., Stringer, M., & Coleman, J. T. (1993). Social explanations for self-esteem in Northern Ireland. The Journal of Social Psychology, 133, 643-650.
Hunter, J. E. (1997). Needed: A ban on the significance test. Psychological
Science, 8, 3-7.
Hunter, J. E., & Schmidt, F. L. (1990a). Methods of meta analysis. Newbury
Park, CA: Sage.
Hunter, J. E., & Schmidt, F. L. (1990b). Dichotomization of continuous variables: The implications for meta-analysis. Journal of Applied Psychology, 75, 334-
349.
122 Hunter, J. E., & Schmidt, F. L. (1994). Correcting for sources of artificial variation across studies. In H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 323-336). New York: Russell Sage Foundation.
Jackson, J. W., & Smith, E. R. (1999). Conceptualizing social identity: A new framework and evidence for the impact of different dimensions. Personality and Social
Psychology Bulletin, 25, 120-135.
Johnson, B. T. (1989). DSTAT: Software for the meta-analytic review of research literatures. Lawrence Erlbaum Associates: Hillsdale, New Jersey.
Johnson, B. T., & Eagly, A. H. (in press). Quantitative synthesis of social psychological research. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social psychology. London: Cambridge University Press.
*Kelly, C. (1988). Intergroup differentiation in a political context. British
Journal of Social Psychology, 27, 319-332.
*Lay, C. H. (1992). If personal self-esteem predicts personal enhancement, will collective self-esteem predict collective enhancement? Journal of Social Behavior and
Personality, 7, 189-200.
*Lindeman, M. (1997). Ingroup bias, self-enhancement, and group identification. European Journal of Social Psychology, 27, 337-355.
*Long, K. M. (1997, September). Identification and discrimination in response to threat: The influence of personal and collective self-esteem. Paper presented at meeting of the British Psychological Society, UK.
Long, K. M., & Spears, R. (1997). The self-esteem hypothesis revisited:
Differentiation and the disaffected. In R. Spears, P. J. Oakes, N. Ellemers, and S. A.
123 Haslam (Eds.), The social psychology of stereotyping and group life (pp. 296-317).
Oxford: Blackwell.
*Long, K. M., & Spears, R. (1998). Opposing effects of personal and collective self-esteem on interpersonal and intergroup comparisons. European Journal of Social
Psychology, 28, 913-930.
*Long, K. M., Spears, R., & Manstead, A. S. R. (1994). The influence of personal and collective self-esteem on strategies of social differentiation. British Journal of Social Psychology, 33, 313-329.
Luhtanen, R., & Crocker, J. (1991). Self-Esteem and intergroup comparisons. In
J. Suls and T.A. Mills (Eds.), Social comparison: Contemporary theory and research
(pp.211-236). New Jersey: Lawrence Erlbaum Associates
Luhtanen, R., & Crocker, J. (1992). A collective self-esteem scale: Self- evaluation of one’s social identity. Personality and Social Psychology Bulletin, 18, 302-
318.
*Maass, A., Ceccarelli, R., & Rudin, S. (1996). Linguistic intergroup bias:
Evidence for in-group-protective motivation. Journal of Personality and Social
Psychology, 71, 512-526.
*Maass, A., Milesi, A., Zabbini, S., & Stahlberg, D. (1995). Linguistic intergroup bias: Expectancies or in-group protection? Journal of Personality and Social
Psychology, 68, 116-126.
*Mallery, P. (1994). Social comparison choices: The top people in the bottom group (Doctoral dissertation, University of California, Los Angeles, 1994). Dissertation
Abstracts International, 55, 3064.
124 Matt, G. E., & Cook, T. D. (1994). Threats to validity of research synthesis. In
H. Cooper & L. Hedges (Eds.), The handbook of research synthesis (pp. 503-520).
New York: Russell Sage Foundation.
*McCrea, S. M., Crawford, M. T., & Hirt, E. R. (1998, May). Group meaningfulness and self-esteem influence type of self-enhancement and ingroup bias.
Poster presented at meeting of the American Psychological Society, Washington,
DC.
McGarty, C., Haslam, S. A., Hutchinson, K. J., & Turner, J. C. (1994). The effects of salient group membership on persuasion. Small Group Research, 25, 267-293.
Migdal, M. J., Hewstone, M., & Mullen, B. (1998). The effects of crossed categorization on intergroup evaluations: A meta-analysis. British Journal of Social
Psychology, 37, 303-324.
Miller, D. T., & Prentice, D. A. (1994). The self and the collective. Personality and Social Psychology Bulletin, 20, 451-453.
Morris, S. B., & DeShon, R. P. (1997). Correcting effect sizes computed from factor analysis of variance for use in meta-analysis. Psychological Methods, 2, 192-199.
Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance, and status: An integration. European Journal of Social Psychology,
22, 103-122.
Mullen, B., & Rosenthal, R. (1985). Basic meta-analysis: Procedures and programs. Hillsdale, NJ: Erlbaum.
125 Mummendy, A., & Schreiber, H. J. (1983). Better or just different? Positive social identity by discriminating against or differentiating from outgroups. European
Journal of Social Psychology, 13, 389-397.
Oakes, P. J. (1987). The salience of social categories. In J. C. Turner, M. A.
Hogg, P. J. Oakes, S. D. Reicher, & M. S. Wetherell (Eds.), (Eds.), Rediscovering the social group: A self-categorization theory (pp. 117-141). Oxford: Basil Blackwell.
Oakes, P. J., Haslam, S. A., & Turner, J. C. (1994). Stereotyping and social reality. Oxford: Blackwell.
Oakes, P. J., & Turner, J. C. (1986). Distinctiveness and the salience of social category memberships: Is there a perceptual bias toward novelty? European Journal of
Social Psychology, 16, 325-344.
Oakes, P. J., Turner, J. C., & Haslam, S. A. (1991). Perceiving people as group members: The role of fit in the salience of social categorizations. British Journal of
Social Psychology, 30, 125-144.
Raudenbush, S. W. (1994). Random effects models. In H. Cooper & L.
Hedges (Eds.), The handbook of research synthesis (pp. 301-322). New York: Russell
Sage Foundation.
Ray, J. J., & Heaven, P. C. L. (1984). Conservatism and authoritarianism among urban Afrikaners. The Journal of Social Psychology, 122, 163-170.
Reed, J. G., & Baxter, P. M. (1994). Using reference databases. In H. Cooper
& L. Hedges (Eds.), The handbook of research synthesis (pp. 71-83). New York:
Russell Sage Foundation.
126 *Robins, E. M., & Foster, D. (1994). Social identity versus personal identity: an investigation into the interaction of group and personal status with collective self- esteem on ingroup favouritism. South African Journal of Psychology, 24(3), 115-121.
Rosenthal, M. C. (1994). The fugitive literature. In H. Cooper & L. Hedges
(Eds.), The handbook of research synthesis (pp. 85-94). New York: Russell Sage
Foundation.
Rosenthal, R. (1979). The "file-drawer problem" and tolerance for null results.
Psychological Bulletin, 86, 638-641.
Rosenthal, R., & Rubin, D. B. (1986). Meta-analytic procedures for combining studies with multiple effect sizes. Psychological Bulletin, 99, 400-406.
Rubin, I. M. (1967). Increased Self-Acceptance: A means of reducing prejudice.
Journal of Personality and Social Psychology, 5, 233-238.
Rubin, M., & Hewstone, M. (1998). Social identity theory's self-esteem hypothesis: A review and some suggestions for clarification. Review of Personality and
Social Psychology, 2, 40-62.
*Ruttenberg, J., Zea, M. C., & Sigelman, C. K. (1996). Collective identity and intergroup prejudice among Jewish and Arab students in the United States. The Journal of Social Psychology, 136, 209-220.
Sachdev, I., & Bourhis, R. Y. (1984). Minimal majorities and minorities.
European Journal of Social Psychology, 14, 35-52.
Sachdev, I., & Bourhis, R. Y. (1985). Social categorization and power differentials in group relations. European Journal of Social Psychology, 15, 415-434.
127 Sachdev, I., & Bourhis, R. Y. (1987). Status differentials and intergroup behavior. European Journal of Social Psychology, 17, 277-293.
Schütz, A., & Tice, D. M. (1997). Associative and competitive indirect self- enhancement in close relationships moderated by trait self-esteem. European Journal of
Social Psychology, 27, 257-273.
Sears, D. O. (1988). Symbolic racism. In P. A Katz & D. M. Taylor (Eds.),
Eliminating racism: Profiles in controversy (pp. 53-84). New York, NY: Plenum Press.
*Seta, C. E., & Seta, J. J. (1992). Observers and participants in an intergroup setting. Journal of Personality and Social Psychology, 63, 629-643.
*Seta J. J., & Seta, C. E. (1996). Big fish in small ponds: A social hierarchy analysis of intergroup bias. Journal of Personality and Social Psychology, 71, 1210-
1221.
*Sidanius, J., Pratto, F., & Mitchell, M. (1994). In-group identification, social dominance orientation, and differential intergroup social allocation. The Journal of
Social Psychology, 134, 151-167.
*Smith, H. J., & Tyler, T. R. (1997). Choosing the right pond: The impact of group membership on self-esteem and group-oriented behavior. Journal of Experimental
Social Psychology, 33, 146-170.
Steele, C. M. (1997). A threat in the air: How stereotypes shape intellectual identity and performance. American Psychologist, 52, 613-629.
Stephan, W. G., & Rosenfield, D. (1978). Effects of desegregation on racial attitudes. Journal of Personality and Social Psychology, 36, 795-804.
128 Tabachnick, B. G., & Fidell, L. S. (1989). Using multivariate statistics (3nd ed.). New York: Harper Collins.
Tajfel, H. (1982). Social psychology of intergroup relations. Annual Review of
Psychology, 33, 1-30.
Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict.
In W. G. Austin and S. Worchel (Eds.), The social psychology of intergroup relations
(pp.33-47). Monterey, CA: Brooks/Cole.
Tice, D. M. (1991). Esteem protection or enhancement? Self-Handicapping motives and attributions differ by trait self-esteem. Journal of Personality and Social
Psychology, 60, 711-725.
Tice, D. M. (1993). The social motivations of people with low self-esteem. In
R. F. Baumeister (Ed.), Self-Esteem: The puzzle of low self-regard (pp37-54). New
York: Plenum Press.
*Thompson, L. L., & Crocker, J. (1990). Downward social comparison in the minimal intergroup situation: A test of a self-enhancement interpretation. Journal of
Applied Social Psychology, 20, 1166-1184.
Turner, J. C. (1985). Social categorization and the self-concept: A social cognitive theory of group behavior. In E. J. Lawler (Ed.), Advances in group processes:
Theory and research (Vol. 2). Greenwich, CT: JAI Press.
Turner, J. C. (1987). The analysis of social influence. In J. C. Turner, M. A.
Hogg, P. J. Oakes, S. D. Reicher, & M. S. Wetherell (Eds.), Rediscovering the social group: A self-categorization theory (pp. 68-88). Oxford: Basil Blackwell.
129 Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S.
(1987). Rediscovering the social group: A self-categorization theory. Oxford: Basil
Blackwell.
Urban, L. M., & Miller, N. (1998). A theoretical analysis of crossed categorization effects: A meta-analysis. Journal of Personality and Social Psychology,
74, 894-908.
Van Avermaet, E. (Ed.). (1998). Notes to contributors. European Journal of
Social Psychology, 28.
*Verkuyten, M. (1997). Intergroup evaluation and self-esteem motivations: Self- enhancement and self-protection. European Journal of Social Psychology, 27, 115-119.
Wainer, H. (1976). Estimating coefficients in linear models: It don't make no nevermind. Psychological Bulletin, 83, 213-217.
Wills, T. A. (1981). Downward comparison principles in social psychology.
Psychological Bulletin, 90, 245-271.
Wills, T. A. (1991). Similarity and downward comparison. In J. Suls & T.
Wills (Eds.), Social comparison: Contemporary theory and research (pp. 51-78).
Hillsdale, NJ: Erlbaum.
Wood, J. V., Giordano-Beech, M., Taylor, K. L., Michela, J. L., & Gaus, V.
(1994). Strategies of social comparison among people with low self-esteem: Self- protection and self-enhancement. Journal of Personality and Social Psychology, 67, 713-
731.
130 *Wright, C. L., & Wyer, N. A. (1997, May). Identification moderates intergroup biases. Poster presented at meeting of the American Psychological Society,
Washington, DC.
131 Appendix A: Meta Analysis Codebook -- Study Qualification Form 1: Study Qualification and General Characteristics
STUDY NUMBER ______Does this study meet all of the following criteria (If NO to any the study does not qualify) YES NO Is this a study in which individuals rate their group compared to another group (i.e., ingroup-outgroup design). Studies which do not contain these measures usually look at attitudes about the self and do not qualify. YES NO Is there a measure of self-esteem? There must be a direct measure of self-esteem such as the Rosenberg or Collective self-esteem scale. Measures of ‘group identity’ qualify – these are usually very similar to collective self-esteem measures. If the study indicates a measure of self-esteem but, does not report analyses of ingroup bias DVs using these measures, make a note of it and I will contact the authors. YES NO Regarding the research design: Is self-esteem used as a predictor of ingroup bias (e.g., pre-test levels of self-esteem taken and amount of bias compared between self-esteem groups)? Other studies look at self-esteem as a DV and try to assess how much self-esteem changes based after exhibiting ingroup bias – note these studies below but code as ineligible. Some of the studies looking at changes in self-esteem may be codeable as predictors – pay close attention here, try to find some way to make these studies fit the coding scheme before disqualifying. YES NO Is the self-esteem measure global rather than domain-specific? Several authors have argued for the use of stable measures to predict bias and for domain-specific measures as indication of change in self-esteem. (E.g., Hunter et al., 1996, 1997 studies). We are only coding the stable measures.
A) Is this a study of 1. Self-Esteem differences in ingroup bias (e.g., High vs. Low) 2. Changes in self-esteem 3. Status differences in ingroup bias
132 Form 2: Document Characteristics
1) ID NUMBER ______2) Name of First Author (Last, First) ______3) Type of Publication: 1. Journal 2. Book Chapter 3. Unpublished paper 4. Dissertation 5. Conference Paper 0. Other (Specify) ______
4) Institutional Affiliation of Author ______
5) Country where research conducted ______
6) If the study is published, indicate the journal: 1. Journal of Personality and Social Psychology 2. Personality and Social Psychology 3. European Journal of Social Psychology 4. British Journal of Social Psychology 5. Journal of Social Psychology 00. Other (Specify)______
133 List all the comparisons for this study. For each comparison, list the name of the dependent variables on the DV worksheet FOR EVERY COMPARISON, CODE EACH AVAILABLE DV – Designate each comparison with a number (e.g., comparison 1, 2, 3, and each DV with a number that is consistent for each comparison). There may be many comparisons done in a particular study. Only code those that yield high vs. low self-esteem groups. If a study has multiple comparisons based on multiple types of self-esteem (e.g., study breaks subjects in to low and high collective self-esteem and low and high personal self-esteem and compares both), then you would code 2 comparisons, low vs. high personal self-esteem and low vs. high collective self-esteem. Don’t code low personal vs. high collective, keep the classification consistent for each comparison. Also, if there are status differentials between groups, try to code each status condition as a different comparison – e.g., if you have a 2 x 2 design with status x self-esteem, instead of collapsing, look at low status – high vs. low self-esteem, high status - high vs. low self-esteem.
Success/Failure studies should be coded IG success vs. OG Success, IG Success vs. OG Failure, IG failure vs. OG success, and IG fail vs. OG Fail (4 comparisons).
Enter a description for each comparison (e.g., low self-esteem vs. high collective self-esteem) For studies with comparisons of real groups, try to code each group as a separate comparison – Studies using minimal groups usually collapse group (i.e., do not give data for blue vs. green group but, rather for ingroup vs. outgroup) so try to code as 1 comparison. Comparison 1: Comparison 2: Comparison 3: Comparison 4: Comparison 5:
134 Dependent Variables
For each comparison, list the DV’s to be coded: Comparison # ______DV# Name
1
2
3
4
5
6
7
8
If there are more, just use another copy of this sheet and change the DV#’s – If all of your comparisons use the same DV’s just note that above instead of filling out one of these each time. Experimental Comparison Number ______The below are to be filled out for each experimental comparison Description of Comparison ______
Does this comparison contain the same participants as another comparison (e.g., subjects divided as to CSE and then examined in terms of PSE). 1. Yes 2. No If yes – Indicate the comparison numbers for which there is overlap ______
135 Appendix B: Self-Esteem and Ingroup Bias -- Subjective Items
Fill out for each comparison – note, each question refers to the specific comparison at hand.
Study Number ______Comparison Number ______
Section 1: Very Subjective Items
This section centers on items that are “subjective” in nature. For the section, try to put yourself in the place of the study participant and rate each item according to how you would feel if you were a subject in the research situation.
1) Do real traits correspond to group assignment? This is your perception of how reasonable the groups are to be accepted even if the traits described are bogus. E.g., randomly assigning individuals would be low correlation, feedback on performance rated more toward the middle (higher scores indicating based on more important traits or more justification for categorization), the highest scores indicating phenotypic or political or historical bases to the categorization. Continuum -- Low to high scores -- Random with no explanation, Given test no explanation, Given test with explanation as to what group means, Real groups based on important traits.
1 2 3 4 5 6 7 -9 (cannot tell) Low High 2) How important is group membership? This is your perception of how important the category would be at the point and time of the study. Ignore factors such as whether groups are real or not and focus on the impact of category membership at that specific point in time. Focus on cognitive rather than affective factor – the next question gets at affect. Main issue here is how likely would you be to look at another individual during the experiment and categorize them in terms of the categories studied in the experiment.
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Extremely 3) How emotionally important is group membership? This is your perception of how emotionally significant the category would be at the point and time of the study. Ignore factors such as whether groups are real or not and focus on the impact of category membership at that specific point in time. Focus on affective rather than cognitive factors. Main issue -- does it matter that someone is in the same or different group? For the most part, real groups score high, but focus on the meaning given to the category.
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Extremely
136 4) How focused would you be on your group identity? This question tries to assess how much attention is focused on the group as opposed to the individual. Think about things such as how “group oriented” you would be. Things to consider here are the amount of feedback regarding group performance, explicit references to the group and group identity, etc. in relation to reference to the individuals (i.e., references to your group rather than you). Main issue -- "groupiness"
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
5) How focused would you be on your individual identity? This question tries to assess how much attention is focused on the individual as opposed to the group. Think about things such as how “individually oriented” you would be. Same as above except individual and group are reversed. (Note: do not assume a binary relationship between individual and group identities). Main issue -- "How did I do?"
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
6) How likely would you be to use the group identity outside of the experimental setting? Main issue -- stability of group identity.
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very Likely
7) If groups perform tasks, how important would your group's performance be to you? That is, how much would you care to see your group does well? Main issue -- How much would you want to perform well? Does it matter if you do well?
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
8) How similar to ingroup members would you perceive yourself? Main issue -- "Are these other people like me?"
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
9) How similar to outgroup members would you perceive yourself ? Main issue -- "Are they different?"
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
137 10) How much competition would you feel existed between groups? Main issue -- Do you want to do better than the outgroup? Can you perform better than another group?
1 2 3 4 5 6 7 -9 (cannot tell) None A great deal
11) How much disagreement would you feel existed between groups? This item refers to differences in opinion or values between groups (either real or constructed). For example, real groups can differ in terms of things like political orientation. Main issue -- are there real dimensions on which the groups disagree?
1 2 3 4 5 6 7 -9 (cannot tell) None A great deal
12) How meaningful would your group membership after a week? (answer this question without regard to debriefing – i.e., ignore whether participants were later informed that group membership was bogus). Main issue -- would you go around using your group identity as a source of self-definition?
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
13) If the groups are performing tasks: How important would your group’s success be to you on this task? That is, how happy would you be if your group performed well. This item tries to get at the more emotional aspects of task importance. I.e., would you care either way if your group performed well or not.
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very 14) How relevant would you feel the group task was to you individually ? That is, would you care how you personally did either way?
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very 15) How relevant would you feel the task is? That is, could you easily ignore performances or does the task measure or purport to measure characteristics that could be considered important?
1 2 3 4 5 6 7 -9 (cannot tell) Not at all Very
138 This section is less subjective:
16) How much contact do participants have with ingroup members in the experiment? 1. None – subjects assigned but don’t know who other group members are 2. None – don’t know who others are, but led to believe interaction will occur 3. Some minimal interaction – know who members are, no interaction 4. Some interaction – know who group members are, possibly sit together 5. Interact – sit with (or communicate with) other group members 6. Interdependent – work together on group tasks (<30 minutes) 7. Highly interdependent – spend at least 30 minutes working together on a task
17) How much contact do participants have with outgroup members in the experiment?(note: few of these studies should use higher numbers) 1. None – subjects assigned but don’t know who other group members are 2. None – don’t know who others are, but led to believe interaction will occur 3. Some minimal interaction – know who members are, no interaction 4. Some interaction – know who group members are, possibly sit together 5. Interact – sit with (or communicate with) other group members 6. Interdependent – work together on group tasks (<30 minutes) 7. Highly interdependent – spend at least 30 minutes working together on a task
18) Are the groups divided physically? 1. Yes 2. No 3. Can’t Tell
19) Is membership reinforced? Many studies reinforce group membership with things like name tags, etc. 1. Yes 2. No 3. Can’t Tell
20) Is there group history? Are participants lead to expect future interaction? 1. No, truly minimal contact, no history of interaction or expectation of future interaction 2. No history but led to expect future contact in the experimental setting 3. No history but led to expect a great deal of future contact 4. History no expectation of contact in future 5. History and contact in future
21) Does the group create a product that is evaluated by both groups? This is meant to distinguish studies where the groups work in projects and then rate each other’s output. 1. Yes 2. No 3. Not Sure
139 22) If yes above, are the subjects making the ratings contributors to this product? 1. Yes 2. No 3. Not Sure 23) Is there a threat to group status? Examples of threat are failure on a task, etc. 1. Yes 2. No 3. Not Sure
24) Is status fixed or can it change? If status is fixed, then exhibiting bias against the outgroup may do no good. E.g., the poor hating the rich doesn't change status, but the rich hating (or helping) the poor can change status. The main issue here is whether groups have the power to exhibit bias or not. Often low power groups “give-up” and don’t show bias against powerful outgroups. 1. Fixed 2. Not fixed 3. No idea
25) Do individuals have opportunities to rate themselves (not the group) in relation to other ingroup members? 1. Yes 2. No 3. No idea
26) Do individuals have opportunities to rate themselves (not the group) in relation to other outgroup members? 1. Yes 2. No 3. No idea
27) Do subjects get feedback as to their own performances (rather than just how the group did)?
1. Yes 2. No 3. Don't perform tasks
28) Does the individual feedback conflict with group performance (that is, is the individual successful while group fails, etc.).
1. Yes 2. No 3. Don't perform tasks
140 Appendix C: Objective Measures -- Methods and Dependent Measures
Methodological Variables
Complete for each comparison Study Number ______Comparison Number ______1) How are subjects assigned to each group? This question tries to get at random assignment issues. 1. Randomly 2. Non randomly, but with matching 3. Non random, no matching 4. Haphazard 5. Pre-existing groups 6. Can’t tell
2) How are subjects selected for inclusion in the study? ______Enter how included and codes will be built. 3) Who are the subjects? 1. College Students 2. School Children 3. Other ______
Rate the following methodological quality issues from 1 to 3. 1 = poorly conducted or no information available, 2 = some problems, 3 = well conducted. The examples for Well, Some Problems, and Poor below are meant to guide and not to serve as absolute definitions. 4) Sampling of study population: Well conducted = random sampling of a large population, Some problems = subject pool, Poor = someone’s intro psychology class takes part as a whole. Poorly Some Problems Well Conducted Can’t tell Conducted 1 2 3 0 5) Comparability of groups: Well = randomly assigned or pre-existing groups with matching, Some problems = no matching but groups relatively homogenous (e.g., German vs. Turkish University Students), Poor = Haphazard assignment of heterogeneous groups. Poorly Some Problems Well Conducted Can’t tell Conducted 1 2 3 0
141 6) Reliability: Well = All DV’s have reliability data (if scales), Some problems = Reliability mentioned but not elaborated, Poor = No information. Poorly Some Problems Well Conducted Can’t tell Conducted 1 2 3 0
7) Statistical power: Well = N of 60 or more per group or power analysis conducted to show adequate power existed, Some problems = N less than 60 per group and researchers reported confidence intervals, Poor =N less than 60 per group, no other information Poorly Some Problems Well Conducted Can’t tell Conducted 1 2 3 0
8) Correct or appropriate statistical analysis: Well = use of inferential statistics with sensitivity to assumptions (e.g., don’t use t-test for ordinal data; adjust for inflated alpha when doing multiple test), Some problems = use inferential statistics but, Can’t tell if data fits assumptions, Poor = use of inferential statistics when assumptions obviously violated. Poorly Some Problems Well Conducted Can’t tell Conducted 1 2 3 0
9) What is the author’s definition of ‘ingroup bias?’ ______
142 Dependent Variables
Several different manners for ES calculation (these do overlap) will be used: 1. Low vs. High self-esteem -- Ingroup Ratings 2. Low vs. High self-esteem -- Outgroup Ratings 3. Low self-esteem -- Ingroup vs. outgroup ratings (repeated measure) 4. High self-esteem -- Ingroup vs. outgroup ratings (repeated measure) 5. Low vs. High self-esteem -- Ingroup vs. Outgroup (difference) Ratings
Study number ______Comparison # ______Dependent Variable Number ______Name of DV ______
Detailed description of the DV (what scale, how divided etc.) ______
A. Direct/Indirect? -- Why ______Post-test Information 1) What is the status of the groups in this comparison (try to isolate by status in comparisons) 1. Low status 2. High status 3. Equal status 4. Can’t tell 5. Mixed 2) What is the status of subjects? (try to code low status and high as different comparisons) 1. Low 2. High 3. Medium (or equal) 4. Can’t tell – but no differences between groups 5. Can’t tell 6. Mixed 3) How did the ingroup perform? (If performing tasks – these studies are coded as having several comparisons) 1. Success 2. Failure 3. Neutral 4. Can’t tell/NA
143 4) How did the outgroup perform? (If performing tasks – these studies are coded as having several comparisons) 1. Success 2. Failure 3. Neutral 4. Can’t tell/NA
5) How is self-esteem measured? 1. Collective Self-Esteem Scale Specify Subscale 1. Membership 2. Private 3. Public 4. Identity 5. Used entire scale 6. Used combination of items but, not entire scale 0. Not specified 2. Rosenberg Self-Esteem Scale (often called ‘personal self-esteem’) 3. Janis-Fields Feelings of inadequacy scale 4. Texas Social Behavior Inventory (TBI) 5. Measure of group identification (Specify scale name) ______0. Other (Specify) ______6) Reliability Index ______7) How is reliability measured (skip if 6 blank)? 1. Alpha 2. Internal consistency (other) 3. Kappa 4. Percent agreement 5. Split-Half 6. Test-retest (note -- can get this info from other sources as needed) 8) How is low self-esteem defined? 1. Lower 1/2 of Median Split of scores 2. Lower 1/3rd of scores 3. Used classification criteria from scale (e.g., clinical criteria) 4. Does not apply – used regression 0. Not Specified 9) What score classifies an individual as Low Self-Esteem (i.e., what is the highest score that a LSE can have) ______
144 10) How is high self-esteem defined? 1. Upper 1/2 of Median Split of scores 2. Upper 1/3rd of scores 3. Used classification criteria from scale (e.g., clinical criteria) 4. Does not apply – used regression 0. Not Specified
11) What score classifies an individual as High Self-Esteem (i.e., what is the lowest ______score that a HSE can have)?
12) What is the lowest possible score on the self-esteem scale? ______
13) What is the highest possible score in the self-esteem scale? ______
14) How many items are on the scale? ______
15) What is the range on each item? ______Low ______High
15b) What is measure of CT (mean / median) ______
16) Page number where scale information can be found. ______
145 LOW SELF-ESTEEM GROUP (Skip if no data) Note: Fill in as much as you can even if some calculation is required. 1) Measure of central tendency for Posttest of Low Self-Esteem Group – RATING OF INGROUP ______2) N ______3) Measure of central tendency for Posttest of Low Self-Esteem Group – RATING OF OUTGROUP ______4) N ______5) Measure of central tendency for Posttest of Low Self-Esteem Group – Difference between INGROUP and OUTGROUP ______6) N ______7) Measure of dispersion for Posttest of Low Self-Esteem Group RATING OF INGROUP ______8) Measure of dispersion for Posttest of Low Self-Esteem Group – RATING OF OUTGROUP ______9) Measure of dispersion for Posttest of Low Self-Esteem Group – Difference between INGROUP and OUTGROUP ______10) Reliability Index ______11) How is reliability measured? 1. Alpha 2. Internal consistency (other) 3. Kappa 4. Percent agreement 5. Split-Half 6. Test-retest 0. Not Specified
12) Who does the group favor more? (i.e., which group is rated more positively - this on the raw numbers rather than the statistical test – the purpose of the question is to establish the direction of the relationship.) 1. Ingroup 2. Outgroup 3. Neither (exactly equal) 4. Can’t tell
146 13) Type of statistical test used for testing differences between the groups for INGROUP ratings: 1. t, F, z, or r (parametric, no partialing or variance adjustment) 2. Parametric, variance adjusted by covariate (ANCOVA, covariate partialed from r) 3. Parametric, variance adjusted by pretest (Repeated measures, pretest partialed from r, regression with pretest entered first) 4. Chi-Square 5. Other nonparametric 6. Can’t tell 7. None
14) What is the test statistic value (i.e., what is the value of t, F, etc.) ______15) What are the degrees of freedom between groups? ______16) What are the degrees of freedom within groups? ______17) What type of test statistic is represented in question 14? 1. t-test 2. F value 3. r 4. multiple or partialed r 5. chi square 6. some other non parametric 7. Can’t tell 8. None
18) Given the alpha level selected by the researcher, is the difference between groups: (note: if no alpha specified assume .05) 1. Significant 2. Non-significant 3. Can’t tell 19) What is the effect size? Indicate to 4 digits. Also, indicate whether positive or negative – Positive favors ingroup, negative outgroup. Use the raw-g from d-stat. ______20) Indicate your confidence in the effect size measure. This is an indication of how much estimation was required. Descriptions are just rough guides here. 1. Highly estimated: Have N and crude p value (e.g., p<.05). 2. Moderate estimation: Have complex statistics that are complete like beta weights or F from a multifactor ANOVA 3. Some estimation: Have exact p level only 4. Slight estimation: Must use significance test statistic (or r) rather than descriptive statistic 5. No estimation: Have mean and standard deviations 0. Not applicable, was unable to get an effect size (refer these to Chris)
147 21) Method of calculation of effect size (when calculating try to use the lowest numbered strategies first) 1. Descriptive statistics 2. t-test 3. F-test 4. Correlation 5. Chi-square 6. Significance levels 7. Other (specify) ______8. Descriptive with MS error (specify MS error) ______
9. Descriptive with SD (specify) ______0. Not applicable
21) Is this a correlated means or repeated measures test? 1. Yes IF yes, note r (-8 for not given) ______2. No
22) Page(s) statistical information for ES found______
HIGH SELF-ESTEEM GROUP (Skip if no data) Note: Fill in as much as you can even if some calculation is required. 1) Measure of central tendency for Posttest of High self-esteem Group – RATING OF INGROUP ______2) N ______3) Measure of central tendency for Posttest of High self-esteem Group – RATING OF OUTGROUP ______4) N ______5) Measure of central tendency for Posttest of High self-esteem Group – Difference between INGROUP and OUTGROUP ______6) N ______7) Measure of dispersion for Posttest of High self-esteem Group RATING OF INGROUP ______8) Measure of dispersion for Posttest of High self-esteem Group – RATING OF OUTGROUP ______9) Measure of dispersion for Posttest of High self-esteem Group – Difference between INGROUP and OUTGROUP ______10) Reliability Index ______
148 11) How is reliability measured? 1. Alpha 2. Internal consistency (other) 3. Kappa 4. Percent agreement 5. Split-Half 6. Test-retest 0. Not Specified 12) Who does the group favor more? (i.e., which group is rated more positively - this on the raw numbers rather than the statistical test – the purpose of the question is to establish the direction of the relationship.) 1. Ingroup 2. Outgroup 3. Neither (exactly equal) 4. Can’t tell 13) Type of statistical test used for testing differences between the groups for INGROUP ratings: 1. t, F, z, or r (parametric, no partialing or variance adjustment) 2. Parametric, variance adjusted by covariate (ANCOVA, covariate partialed from r) 3. Parametric, variance adjusted by pretest (Repeated measures, pretest partialed from r, regression with pretest entered first) 4. Chi-Square 5. Other nonparametric 6. Can’t tell 7. None 14) What is the test statistic value (i.e., what is the value of t, F, etc.) ______15) What are the degrees of freedom between groups? ______16) What are the degrees of freedom within groups? ______17) What type of test statistic is represented in question 29? 1. t-test 2. F value 3. r 4. multiple or partialed r 5. chi square 6. some other non parametric 7. Can’t tell 8. None
18) Given the alpha level selected by the researcher, is the difference between groups: (note: if no alpha specified assume .05) 1. Significant 2. Non-significant 3. Can’t tell
19) What is the effect size? Indicate to 4 digits. Also, indicate whether positive or negative – Positive favors ingroup negative outgroup. Use the raw-g from d-stat. ______
149 20) Indicate your confidence in the effect size measure. This is an indication of how much estimation was required. Descriptions are just rough guides here. 1. Highly estimated: Have N and crude p value (e.g., p<.05). 2. Moderate estimation: Have complex statistics that are complete, like beta weights or F from a multifactor ANOVA 3. Some estimation: Have exact p level only 4. Slight estimation: Must use significance test statistic (or r) rather than descriptive statistic 5. No estimation: Have mean and standard deviations 0. Not applicable, was unable to get an effect size (refer these to Chris)
21) Method of calculation of effect size (when calculating use try to use the lowest numbered strategies first) 1. Descriptive statistics 2. t-test 3. F-test 4. Correlation 5. Chi-square 6. Significance levels 7. Other (specify) ______8. Descriptive with MS error (specify MS error) ______9. Descriptive with SD (specify) ______0. Not applicable
22) Is this a correlated means or repeated measures test? 0. Yes IF yes, note r (-8 for not given) ______1. No
23) Page(s) statistical information for ES found ______
Ingroup Ratings
1) Who favors the ingroup more? (i.e., who rates the group more positively - this on the raw numbers rather than the statistical test – the purpose of the question is to establish the direction of the relationship. 1. Low self-esteem group 2. High self-esteem group 3. Neither (exactly equal) 4. Can’t tell
150 2) Type of statistical test used for testing differences between the groups for INGROUP ratings: 1. t, F, z, or r (parametric, no partialing or variance adjustment) 2. Parametric, variance adjusted by covariate (ANCOVA, covariate partialed from r) 3. Parametric, variance adjusted by pretest (Repeated measures, pretest partialed from r, regression with pretest entered first) 4. Chi-Square 5. Other nonparametric 6. Can’t tell 7. None 3) What is the test statistic value (i.e., what is the value of t, F, etc.) ______4) What are the degrees of freedom between groups? ______5) What are the degrees of freedom within groups? ______6) What type of test statistic is represented in question 29? 1. t-test 2. F value 3. r 4. multiple or partialed r 5. chi square 6. some other non parametric 7. Can’t tell 8. None
7) Given the alpha level selected by the researcher, is the difference between groups: (note: if no alpha specified assume .05) 1. Significant 2. Non-significant 3. Can’t tell
8) What is the effect size? Indicate to 4 digits. Also, indicate whether positive or negative – Positive favors high self-esteem and negative high self-esteem. Use the raw-g from d-stat. ______9) Indicate your confidence in the effect size measure. This is an indication of how much estimation was required. Descriptions are just rough guides here. 1. Highly estimated: Have N and crude p value (e.g., p<.05) 2. Moderate estimation: Have complex statistics that are complete, like beta weights or F from a multifactor ANOVA 3. Some estimation: Have exact p level only 4. Slight estimation: Must use significance test statistic (or r) rather than descriptives 5. No estimation: Have mean and standard deviations 0. Not applicable, was unable to get an effect size (refer these to Chris)
151 10 10) Method of calculation of effect size (when calculating use try to use the lowest numbered strategies 11 first) 12 1. Descriptive statistics 2. t-test 3. F-test 4. Correlation 5. Chi-square 6. Significance levels 7. Other (specify) ______8. Descriptive with MS error (specify MS error) ______9. Descriptive with SD (specify) ______0. Not applicable
11) Is this a correlated means or repeated measures test? 1 1. Yes IF yes, note r (-8 for not given) ______2. No 12) Page(s) statistical information for ES found ______
152 Outgroup Ratings
1) Who favors the outgroup more? (i.e., who rates the group more positively - this on the raw numbers rather than the statistical test – the purpose of the question is to establish the direction of the relationship. 1. Low self-esteem group 2. High self-esteem group 3. Neither (exactly equal) 4. Can’t tell
2) Type of statistical test used for testing differences between the groups for OUTGROUP ratings: 1. t, F, z, or r (parametric, no partialing or variance adjustment) 2. Parametric, variance adjusted by covariate (ANCOVA, covariate partialed from r) 3. Parametric, variance adjusted by pretest (Repeated measures, pretest partialed from r, regression with pretest entered first) 4. Chi-Square 5. Other nonparametric 6. Can’t tell 7. None
3) What is the test statistic value (i.e., what is the value of t, F, etc.) ______4) What are the degrees of freedom between groups? ______5) What are the degrees of freedom within groups? ______6) What type of test statistic is represented? 1. t-test 2. F value 3. r 4. multiple or partialed r 5. chi square 6. some other non parametric 7. Can’t tell 8. None 7) Given the alpha level selected by the researcher, is the difference between groups: (note: if no alpha specified assume .05) 1. Significant 2. Non-significant 3. Can’t tell
8) What is the effect size? Indicate to 4 digits. Also, indicate whether positive or negative – Positive favors high self-esteem and negative low self-esteem. Use the raw-g from d-stat. ______
153 9) Indicate your confidence in the effect size measure. This is an indication of how much estimation was required. Descriptions are just rough guides here. 1. Highly estimated: Have N and crude p value (e.g., p<.05). 2. Moderate estimation: Have complex statistics that are complete, like beta weights or F from a multifactor ANOVA 3. Some estimation: Have exact p level only 4. Slight estimation: Must use significance test statistic (or r) rather than descriptive statistic 5. No estimation: Have mean and standard deviations 0. Not applicable, was unable to get an effect size (refer these to Chris)
10) Method of calculation of effect size (when calculating use try to use the lowest numbered strategies first) 13 14 1. Descriptive statistics 15 2. t-test 3. F-test 4. Correlation 5. Chi-square 6. Significance levels 7. Other (specify) ______8. Descriptive with MS error (specify MS error) ______9. Descriptive with SD (specify) ______0. Not applicable
______11) Is this a correlated means or repeated measures test? 0. Yes IF yes, note r (-8 for not given) ______1. No
12) Page(s) statistical information for ES found
Difference Ratings
Some studies will just report a single score, representing differences between the ratings of the ingroup and outgroup – Also use this section for any single measure rating (e.g., an index of ethnocentrism). Try to calculate difference measures when possible.
1) Differences between scores favor for Difference/Single measure ratings: (i.e., who rates the group more positively - this on the raw numbers rather than the statistical test – the purpose of the question is to establish the direction of the relationship.) 1. Low self-esteem group 2. High self-esteem group 3. Neither (exactly equal) 4. Can’t tell
154 2) Type of statistical test used for testing differences between the groups for Difference/Single measure ratings: 1. t, F, z, or r (parametric, no partialing or variance adjustment) 2. Parametric, variance adjusted by covariate (ANCOVA, covariate partialed from r) 3. Parametric, variance adjusted by pretest (Repeated measures, pretest partialed from r, regression with pretest entered first) 4. Chi-Square 5. Other nonparametric 6. Can’t tell 7. None 3) What is the test statistic value (i.e., what is the value of t, F, etc.) ______4) What are the degrees of freedom between groups? ______5) What are the degrees of freedom within groups? ______6) What type of test statistic is represented? 1. t-test 2. F value 3. r 4. multiple or partialed r 5. chi square 6. some other non parametric 7. Can’t tell 8. None
7) Given the alpha level selected by the researcher, is the difference between groups: (note: if no alpha specified assume .05) 1. Significant 2. Non-significant 3. Can’t tell
8) What is the effect size? Indicate to 4 digits. Indicate whether positive or negative – Positive favors high self-esteem and negative low self-esteem. ______
9) Indicate your confidence in the effect size measure. This is an indication of how much estimation was required. Descriptions are just rough guides here. 1. Highly estimated: Have N and crude p value (e.g., p<.05). 2. Moderate estimation: Have complex statistics that are complete like beta weights or F from a multifactor ANOVA 3. Some estimation: Have exact p level only 4. Slight estimation: Must use significance test statistic (or r) rather than descriptive statistic 5. No estimation: Have mean and standard deviations 0. Not applicable, was unable to get an effect size (refer these to Chris)
155 10) Method of calculation of effect size (when calculating use try to use the lowest numbered strategies first) 16 1. Descriptive statistics 17 2. t-test 18 3. F-test 4. Correlation 5. Chi-square 6. Significance levels 7. Other (specify) ______8. Descriptive with MS error (specify MS error) ______9. Descriptive with SD (specify) ______0. Not applicable
11) Is this a correlated means or repeated measures test? 0. Yes IF yes, note r (-8 for not given) ______1. No
12) Sample size? ______
13) Page(s) statistical information for ES found ______
156