Contribution
The first review since Abrams and Hogg (1988) to take old and new evidence on the self-esteem hypothesis together. It gives a taxonomy of self-esteem, eight types from three distinctions, and uses it to sort the evidence. It concludes that the hypothesis as stated is not supported, and argues that the fault lies in its formulation and in measures that do not fit it. It ends with a reformulation and a list of what a fair test would need.
Key insight
The self-esteem in question is a group member’s, not an individual’s. Social identity theory says members discriminate to make their group positively distinct, and so are “striving for positive self-esteem as ingroup members, rather than as individuals” (pp. 5–6). To test it with personal self-esteem “would be to recast social identity theory as an individualistic theory of intergroup behavior” (p. 6). Most tests did just that, and the motivational half failed: “intergroup discrimination leads to an increase in self-esteem but is not motivated by a need for self-esteem” (p. 22).
Assumptions
- The hypothesis is Hogg and Abrams’s (1990) two corollaries of Turner’s (1982) account (pp. 3–4). A conclusive test examines both together: low self-esteem promotes discrimination and that discrimination then raises self-esteem (p. 4).
- Status is not self-esteem. Evidence for corollary 2 that relies on low-status groups is excluded, since members of low-status groups can hold their group in high esteem (p. 7; nine studies named as excluded).
- Competitive versus normative discrimination. The hypothesis applies only to discrimination that seeks positive distinctiveness against intergroup norms, not to discrimination prescribed by history or status (pp. 7–8). Evidence where normative discrimination was possible is kept but reported separately.
- The type of self-esteem does not decide the verdict. The taxonomy is used to compare, not to rule evidence in or out (fn. 5, p. 31).
- A vote count. Given the variety of designs, the review is qualitative, not quantitative (fn. 7, p. 31). Each experiment counts as supportive or unsupportive; effect sizes are not pooled.
Key results
- Eight types of self-esteem (pp. 4–6; Table 1, p. 32). Global or specific (the whole self-image or one aspect), trait or state (over time or now), personal or social (esteem for oneself or for the shared self-image of an in-group). Example: specific social state is “At the moment, I feel good about my gender.” The personal–social distinction is “the most important from a theoretical perspective” (p. 5).
- The measures misfit (pp. 6–7). Most scales measure global personal trait self-esteem (Rosenberg’s RSES and others). The only social scale, Luhtanen and Crocker’s Collective Self-Esteem Scale, is global and trait, and only its 4-item private subscale fits the theory’s concept. “No purpose-built measures of either specific social trait or specific social state self-esteem exist” (p. 7).
- Competitive discrimination (pp. 8–15; Table 2, pp. 33–34).
Corollary 1: 9 experiments supportive, 3 unsupportive. Corollary 2: 3
supportive, 16 unsupportive.
- Lemyre and Smith (1985), the best controlled test: discrimination predicted posttest self-esteem, but mere categorization had first lowered it, and discrimination raised it only back to the uncategorized level. Pretest self-esteem did not predict discrimination (p. 14).
- Gagnon and Bourhis (1996): discrimination predicted specific social state self-esteem, only insofar as participants identified with the group (p. 11).
- Platow et al. (1997, Expt. 2): those with high personal and low social self-esteem discriminated more than those high in both (pp. 11–12).
- High personal self-esteem went with more discrimination in several studies; for instance, “the higher the subjects’ level of trait self-esteem, the greater the difference in perceived competence” (Sidanius et al. 1994, as quoted, p. 13).
- Competitive and/or normative discrimination (pp. 15–21; Table 3,
pp. 34–36). Corollary 1: 9 supportive, 10 unsupportive. Corollary 2: 4
supportive, 12 unsupportive.
- Branscombe and Wann (1994), the one study to support both corollaries (p. 19). High identifiers with America who saw an American boxer lose to a Soviet one had lower collective self-esteem and derogated Russians more. Among them, high derogators showed a greater rise in self-esteem. Low identifiers showed neither effect.
- Long et al. (1994): high personal with low social self-esteem gave the most discrimination (p. 18).
- Totals (pp. 21–22). Corollary 1: 18 supportive, 13 unsupportive. Corollary 2: 7 supportive, 28 unsupportive. One of the eight full tests supports both. Nine studies find people with high global personal trait self-esteem the most discriminatory, which is “congruent with the idea that individuals who possess generally low self-esteem lack the self-confidence” to self-enhance this way (p. 22).
- By type of measure (Table 4, p. 37). For both corollaries, specific, social and state measures give a higher share of support than global, personal and trait ones. Corollary 1: social 4 for and 1 against, personal 14 and 12; state 12 and 4, trait 6 and 9. Corollary 2: specific 6 and 6, global 1 and 22.
- The verdict (p. 22). The theory’s explanation is motivational, so it “fails unless either both corollaries … receive convincing support or corollary 2 receives support”. “The majority of evidence does not support social identity theory’s self-esteem hypothesis in its full and unqualified form.”
- The reformulation (p. 22). (1) Successful competitive discrimination increases the specific state self-esteem associated with that identity. (2) Low specific social state self-esteem promotes competitive discrimination. The only two studies that measured specific social state self-esteem (Gagnon & Bourhis 1996; Platow et al. 1997, Expt. 2) both support it.
- Two cautions for such tests (pp. 22–23). Explicit state items (“at the moment, I feel good about my group”) invite demand effects. And trait ratings of in- and out-group are “directly tapping the relative esteem in which subjects hold their own group” (Abrams & Hogg 1988, as quoted, p. 23), so a bias measure can be an esteem measure. Behavioural allocation measures are advised.
- Further distinctions needed (pp. 23–25): achieving, maintaining and enhancing self-esteem; threatened versus actually low self-esteem; successful versus unsuccessful discrimination; in-group favouritism versus out-group derogation; and identification versus social self-esteem, which current scales mix although the theory predicts opposite correlations with bias (positive for identification, negative for pretest self-esteem).
- The theory’s later course (p. 25). Extensions of social identity theory either ignore the hypothesis (self-categorization theory) or replace it with other motives (Brewer’s optimal distinctiveness; Hogg and Abrams’s uncertainty reduction).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Social identity theory’s self-esteem hypothesis concerns specific social state self-esteem, but most tests measured global personal trait self-esteem | strong as a reading of the theory and of the measures used; the tables show it | pp. 4–7, Tables 2–3 |
| C2 | Intergroup discrimination tends to raise self-esteem (corollary 1) | weak to moderate: 18 experiments for, 13 against, in a vote count without effect sizes; some gains only restore what categorization lowered | pp. 21–22, p. 14 |
| C3 | Low self-esteem does not, in general, motivate discrimination (against corollary 2) | moderate: 7 for, 28 against; nine studies find the reverse for personal self-esteem | pp. 21–22 |
| C4 | The hypothesis is not supported in its full and unqualified form | moderate: follows from C2 and C3, and only one of eight full tests supports both | p. 22 |
| C5 | A version in specific social state self-esteem may hold | weak: two studies, and the authors call it “tentative” | p. 22 |
| C6 | Measures of identification and social self-esteem, and some measures of bias and of esteem, overlap, which may have hidden effects in both literatures | argued, with examples; not tested | pp. 23–24 |
Concepts
- self-esteem hypothesis: social identity theory’s claim, in Hogg and Abrams’s two corollaries, that discrimination raises self-esteem and that low self-esteem motivates it (pp. 3–4).
- social self-esteem: “the esteem in which [members] hold the shared self-image that constitutes their social psychological ingroup” (p. 5).
- specific social state self-esteem: esteem for one particular in-group, at the moment; the type the authors say the theory implies.
- competitive versus normative discrimination: discrimination that seeks positive distinctiveness against intergroup norms, versus discrimination the situation already prescribes (pp. 7–8).
- categorization-depression: the finding that mere minimal categorization lowers self-esteem (Lemyre & Smith 1985; p. 9).
Connections
- Hornsey 2008 (NOTE-1253) and Reimer et al. (NOTE-1251). The LIT was filed to settle their disagreement. Hornsey’s reading is right on the facts: more support for “bias raises self-esteem” than for “low self-esteem causes bias”, and criticism for measuring personal self-esteem to predict a collective outcome. But “dead end” is Hornsey’s verdict, not the review’s. Rubin and Hewstone call for a narrower hypothesis and better measures. Reimer et al.’s “extensive empirical support” is not what the review found: even for corollary 1 the count is 18 to 13, and the review’s own conclusion is negative. Both readings call it a meta-analysis; it is a qualitative vote count (fn. 7). NOTE-1251’s reconciliation, that the two agree on corollary 1 and differ in tone, is fair, but its C2 should stay weak.
- Self-categorization theory (THEORY-1114). The review confirms the THEORY’s statement that the self-esteem strand is social identity theory’s and not part of self-categorization theory: the latter “ignore[s]” the hypothesis (p. 25).
- The minimal group paradigm (NOTE-1301). The review restates its design: participants are “not given the opportunity to award points to themselves”, which is meant to remove self-interest (p. 3). Otten’s finding that participants given that chance pay themselves most is the natural complement.
- Identity fusion (NOTE-1256). High personal with low social self-esteem gives the most discrimination (Long et al. 1994; Platow et al. 1997). The gloss is that low social self-esteem is “a threat to positive personal identity” and a motive to improve the social one (p. 24). That parallels fusion’s finding that a challenge to the personal self is answered through the group. The link is the record’s, not the authors’.
- Leach et al. (NOTE-1262). Their split of identification into self-definition and self-investment answers, ten years on, the review’s call to separate identification from social self-esteem.
Bearing on the record
- The owner’s thesis (CLAIM-491). The self-esteem hypothesis is
social identity theory’s own version of the thesis: group-serving bias
is pursued for the esteem of a social self, “as ingroup members, rather
than as individuals”. The review is the best evidence the record holds
on whether that motive operates. These are inferences.
- The benefit is found more often than the motive. Discrimination tends to raise self-esteem; a need for self-esteem does not, in general, drive it. A gain to the self that follows group-serving action without motivating it does not show the action aimed at that self’s good. In Stich’s terms (NOTE-1282) it is a side effect, not an ultimate or instrumental aim.
- The motive was mostly tested at the wrong level. Only two studies measured esteem for the group itself, and both support the motive. So the review does not refute a group-level self-interest. It shows that by 1998 it was nearly untested.
- The personal self keeps showing. The most discriminatory people had high personal self-esteem, and the strongest motivated pattern is high personal with low social self-esteem. The group’s standing seems to matter as a threat to a personal self that already includes it. That fits “larger self” (the group added to a kept personal self) better than “changed self”.
- The triviality objection (CLAIM-492). The review states, for measures, the objection’s demand. Identification scales contain esteem items and esteem scales contain identification items, though the theory predicts opposite effects for them. It proposes measuring identification as importance and self-esteem as evaluation (p. 24). It also warns that bias ratings can be esteem ratings (p. 23), so “the self’s good” and “the group-serving act” can be one measurement. A thesis that cites this literature needs the two kept apart.
- Batson’s objection (CLAIM-494). Not addressed: no helping, no empathy.
- Proposed corrections to two readings. On NOTE-1253, C4’s strength should say “one cited qualitative review (a vote count)”, not “meta-analysis”. On NOTE-1251, the open question that names this review is answered: the review supports Hornsey’s description, not “extensive empirical support”. These are proposed dated notes, not edits.
- Dated note. Proposed on CLAIM-491 (Section 3 of the draft).
- No THEORY filed. The hypothesis is social identity theory’s and contested, and the record files only self-categorization theory’s core.
- No ML instruction.
Limitations
- A vote count. Experiments are counted as for or against, with no effect sizes, no weighting by sample size and no test of heterogeneity. Some studies appear in more than one cell of the tables.
- The classification is the authors’. Whether a study is “competitive” or “normative”, or counts as supportive “with qualifications”, is a judgement, made in the text case by case.
- 1998. Later work, including implicit measures the authors call for and the later meta-analysis Reimer et al. cite (Aberson, Healy & Romero 2000), is not covered.
- Mostly laboratory and student samples, many minimal groups.
- An internal inconsistency. The text lists Crocker and Luhtanen (1990) among eight experiments in which high self-esteem went with more discrimination (p. 12), but then reports that study as finding “no significant relationship” (p. 13), and Table 2 does not mark it as a reverse finding.
- Author’s final version. Not compared with the published text.
- Typos and reference slips (manuscript pages).
- “Azjen” for Ajzen (p. 5).
- “operationlized”, in a quotation from Crocker et al. (p. 12).
- “This lead us” (p. 15).
- “Vanbeseleare” (p. 16) and “Nascimeto-Schulze” (p. 17).
- “Maass & Acuri” for Arcuri (p. 20 and references).
- “St. Claire” in the text (p. 9), “St. Clare” in the references; “Schiffmann” in the text, “Schiffman” in the references.
- Rosenberg et al. (1995) is “Rosenberg, Schooler, Schoenbach, & Rosenberg” in the text (p. 4) and “Rosenberg, Schoenbach, Schooler, & Rosenberg” in the references.
- Sachdev and Bourhis (1984) and St. Clare and Turner (1982) are both given as EJSP 12, 307–314; one of them is wrong.
- Long et al. (1994): pages “313-219”.
Open questions
- Does the reformulated hypothesis hold? A test would measure specific social state self-esteem implicitly, before and after a behavioural allocation, with identification measured separately. The record holds no later test.
- What did Aberson, Healy and Romero (2000) find? It is the quantitative successor Reimer et al. cite and is not held.
- Is the high-personal, low-social pattern a case of the group in the self? A study crossing it with identity fusion would show whether those who discriminate most under it are the fused.