NOTE-1280: Group Identity, Social Influence, and Collective Action Online: Extensions and Applications of the SIDE Model

Contribution

A handbook chapter that states the SIDE model (the Social Identity model of Deindividuation Effects) and reviews twenty-five years of work on it. It does three things the record lacked. It sets out the model’s two dimensions and the experiments behind each. It corrects readings the authors call mistaken: SIDE does not endorse deindividuation, and it is “not about anonymity” as such. And it extends the model to visible media, online communities and collective action.

Key insight

The crowd member has not lost a self; a group has come into it. “Whereas deindividuation implies a loss of self in the group, depersonalization refers to the emergence of group in the self: the tendency to see others but also oneself in group terms” (p. 27). So anonymity does not loosen regulation. It moves regulation to the group level: “behavior becomes more social or socially regulated, not less so” (p. 26). What anonymity amplifies depends on which self is salient. When personal identity is salient, anonymity amplifies that instead.

Assumptions

• The social identity approach (social identity theory and self-categorization theory) is the theory. SIDE is “a model rather than a theory”: a more specific analysis of technological features within it (p. 31).
• Two aspects of anonymity, taken from deindividuation theory (pp. 26–27). Deindividuation theory had two routes to antinormative behaviour: anonymity cutting accountability, and submergence in the group cutting private self-awareness (Prentice-Dunn & Rogers 1989). SIDE keeps the distinction but changes both:
  • anonymity of others within the group bears on which identity is salient (the cognitive dimension);
  • anonymity to an audience bears on what can safely be expressed (the strategic dimension).
• Salience is set in the experiment by instructions. In the first study, participants were told the researchers were interested in them as psychology students, or in personality differences in communication style (p. 28).

Key results

The chapter describes designs and directions of effects. It gives no statistics.

• Deindividuation theory (pp. 25–26). From Le Bon through Festinger et al. (1952), Zimbardo (1969), Diener (1980) and Prentice-Dunn & Rogers (1989): “anonymity in the mass promotes a loss of self-awareness, which reduces inhibitions and increases the likelihood of antisocial behavior”. Early work on computer-mediated communication borrowed it to explain “flaming” and extreme group decisions (Kiesler et al. 1984).
• Group polarization as conformity (p. 25). The prototypical position of a discussion group reflects both its mean and what distinguishes it from others. With no explicit out-group, the other end of the scale serves as an implicit one, so conformity to the prototype is polarization.
• The cognitive dimension (pp. 27–29).
  • Reicher (1982, 1987) argued that crowd members switch from individual to group identity and become more, not less, sensitive to norms. Crowd behaviour “is not antinormative and disinhibited … but rather it is the result of conformity to explicit or implicit norms” (p. 27).
  • Reicher (1984): science and social science students dressed in overalls and hoods conformed more to in-group norms than students in their own clothes (p. 27).
  • Spears, Lea & Lee (1990): first-year psychology students discussed attitude topics, with the group norm made known. Group or personal identity salience was crossed with sitting in separate rooms or face to face. Conformity to the group norm was greatest with group identity salient and members isolated. “When personal identity was salient, however, participants actually contrasted their views away from the group norm under anonymity” (p. 28).
  • Postmes et al. (2001), two studies (pp. 28–29). Efficiency or patient-care norms were primed with a scrambled-sentence task before a discussion of hospital policy. Group influence was stronger when members were anonymous than when they were shown by portrait. When two of four members were primed, the norm spread to the two unprimed members, again mainly in anonymous groups.
  • Lee (2004) found that identification mediated the effect. Lee (2008) found that identifiable users attended to argument strength, while for anonymous users the group norm mattered more (p. 29).
• Against deindividuation’s process (p. 29).
  • Postmes & Spears (1998) found that anonymity in the group “actually enhances conformity to local or specific group norms, rather than increasing antinormative or antisocial behavior”.
  • In computer-mediated groups, anonymity does not reduce personal self-awareness and tends, if anything, to increase public self-awareness. Neither kind of self-awareness predicts disinhibition (Adrianson 2001; Lee 2007; Matheson & Zanna 1988; Yao & Flanagin 2006).
• Conceptual recap (pp. 29–31).
  • “Deindividuation” in the name is descriptive. It names the effects to be explained, and the model does not endorse “a state of deindividuation”.
  • The name is not “Depersonalization Effects” because anonymity does not always bring a shared identity (p. 30).
  • Neither technology nor the social context determines the outcome; they interact.
  • “SIDE is not about anonymity but about the online representation of individuals and groups” (p. 31). Uniform representations of members have the same effect as anonymity.
• The strategic dimension (pp. 31–33).
  • SIDE makes “no blanket assumption that people will always act in line with individual self-interest when anonymous” (p. 32). What unaccountable people do depends on the salient identity.
  • Reicher & Levine (1994a, b) sorted student behaviours into three kinds: normative for students and acceptable to staff (missing the occasional lecture); normative but punishable (copying from another student’s essay); and neither (cheating in an exam). Students identifiable to staff endorsed the second kind less, and students anonymous to staff endorsed it more. Copresence of other students had the same effect. All endorsed the first kind and none the third (p. 32).
  • Spears et al. (2002): visual anonymity raised influence on the unpunishable items (the cognitive effect). Being able to communicate by computer raised support for the punishable items (the strategic effect), since students could then gauge each other’s support (p. 32).
  • Women more often concealed their gender online (Flanagin et al. 2002). They were less likely to choose female avatars on topics of male expertise, and less likely to reveal gender in Italy than in the UK or the Netherlands (Spears et al. 2011; p. 33).
• Extensions (pp. 34–37).
  • Depersonalized online groups showed more identification, cohesion, trust, satisfaction, knowledge sharing and, for high identifiers, performance. Members were more likely to perceive the group or dyad “as one” (Sassenberg & Postmes 2002; Tanis & Postmes 2008; p. 34).
  • Lea, Spears & Watt (2007). In mixed groups of British and Dutch men and women, depersonalization mediated influence under anonymity on nationality topics but under visibility on gender topics. Gender is visually cued, and visibility gives it comparative fit (p. 35).
  • Postmes & Spears (2002): adding individuating profiles weakened the salience and power differences of gender in text chat (p. 35).
  • Ren et al. (2007, 2012): in a movie database, users identifiable only as team members formed common-identity groups. These were the “stickiest”, with the strongest commitment (p. 37).
  • Farzan et al.: Tetris players assigned to teams played more and returned more often (p. 37).
  • The authors speculate that environments that block credit-taking “foster more prosocial and community-oriented (even altruistic?) norms” (p. 37).
• Collective action (pp. 37–41).
  • Claims that Twitter and Facebook organized the 2011 London riots were disproven (Postmes et al. 2013). Egypt’s uprising went on through an Internet and SMS shutdown (p. 38).
  • Brunsting & Postmes (2002) surveyed non-activists, sympathizers, hardcore activists and pressure-group members. The predictors of online and offline action intentions showed “no huge differences” (p. 38). Expected effectiveness counted slightly more online.
  • Three predictors are assumed: shared social identity, efficacy, and anger and injustice (Van Zomeren, Postmes & Spears 2008; p. 39).
  • Chan (2010): church groups gave more when a call for donations came by email than face to face, “particularly when Christian identity was salient, and particularly among low identifiers” (p. 39).
  • The authors reject the view that online action becomes personal identity expression (Bennett & Segerberg 2012): “This is not a view we share” (p. 39).
  • A Facebook event by three individuals drew an estimated 300,000 people across 10 cities in Portugal on 12 March 2011, around the shared identity of the “endangered generation” (p. 40).
  • Rumours of police mistreating a 16-year-old girl circulated for a day and a half at up to 70 tweets an hour. Clearly false rumours were quickly quashed (pp. 40–41).

Claims

id	claim	strength	support
C1	Group immersion and anonymity increase conformity to local group norms rather than releasing antinormative behaviour	moderate: one meta-analysis and several experiments, cited without effect sizes	pp. 27–29
C2	Anonymity within a group amplifies whichever identity is salient: toward the group norm under group salience, away from it under personal salience	moderate: one crossed experiment described (Spears et al. 1990); others replicate the group-salient half	pp. 28–30
C3	Depersonalization is not loss of self: anonymity does not reduce personal self-awareness, and self-awareness does not predict disinhibition	weak to moderate: four studies cited in one sentence	p. 29
C4	Anonymity to a powerful out-group, and the support of in-group members, free behaviour that is normative for the in-group but punishable by the out-group	moderate: three experiments described (Reicher & Levine 1994a, b; Spears et al. 2002)	pp. 31–32
C5	For visually cued categories, visibility rather than anonymity raises group salience and influence	weak to moderate: one study (Lea et al. 2007)	p. 35
C6	Online and offline collective action share their psychological predictors	weak: one survey study and a research synthesis by the authors	pp. 38–39

Concepts

• deindividuation (theory): loss of self-awareness under anonymity and immersion in a group, releasing antinormative behaviour. The model rejects it.
• depersonalization: seeing oneself and others in terms of group norms and stereotypes; “heightened social or group-level self-regulation” (p. 27).
• cognitive SIDE (“anonymity of”): anonymity within the group raising the salience of the group identity.
• strategic SIDE (“anonymity to”): anonymity to an audience changing what can be expressed without sanction.
• pseudonymity: traceable but not recognizable in the situation; the usual condition online (p. 30).
• common-identity versus common-bond groups: groups held together by a shared identity or by bonds between members (p. 37).

Connections

• Reicher, Spears & Postmes 1995 (LIT-1744), unread. Cited as a founding statement of the model and of its strategic dimension (pp. 23, 31). The chapter gives the model’s content and the experiments that preceded and followed it, but not the 1995 paper’s own argument.
• Postmes & Spears 1998 (LIT-1731), unread. Given one sentence (p. 29).
• Turner et al. 1994 (NOTE-1284) and Hornsey 2008 (NOTE-1253). Turner et al. define depersonalization as a shift in self-perception (p. 455), which NOTE-1284 glosses as “not a loss of self”, but they do not discuss crowds. Hornsey gives SIDE one sentence (p. 210). This chapter supplies the model’s structure, the strategic dimension (absent from both), and the evidence that anonymity does not reduce self-awareness.
• Skorich (NOTE-1295, THEORY-1112). Skorich separates category effects, such as out-group homogeneity, from group-entity effects, such as crowd behaviour, and holds that accentuating likeness gives interchangeability, not unity. SIDE reports that depersonalized members see the group “as one” (p. 34). Its polarization account also relies on an implicit out-group (p. 25).
• Spears 2021 (read alongside). It covers SIDE more briefly and influence more widely, and cites this chapter as the fuller review.
• León (THEORY-1115). SIDE’s anonymous groups are the opposite of León’s close dyads. Their members’ individuating features are hidden, and that is what makes the “we” salient.

Bearing on the record

• The owner’s thesis (CLAIM-491). This chapter answers the question the thesis needs answered about crowds. Is the crowd member selfless, or acting on a larger self? These are inferences; the authors do not discuss selflessness.

  • Not selfless. Deindividuation theory is the “selfless crowd”: a loss of self-awareness and of the self’s regulation. SIDE rejects it on evidence. Anonymity does not lower personal self-awareness, and crowd and online behaviour follows the group’s norms (p. 29). The crowd member acts as a self, “the group in the self” (p. 27).
  • Changed, more than larger. Reicher’s word is a “switch” from individual to group identity (p. 27). Spears et al. (1990) show that anonymity amplifies whichever identity is salient, and pushes people away from the group under personal salience. That is a choice of one self over another, not an enlargement. It confirms the thesis’s own caveat that depersonalization is “a change of self, not plainly a larger one”. Fusion (NOTE-1256) remains the record’s case for “larger”.
  • Strategic action on a group self. Unaccountable students did what their group’s norm allowed and staff would punish, and nothing the norm forbade (p. 32). The authors say this means people do not always act “in line with individual self-interest when anonymous”. The calculation of what one can get away with is kept, but it serves the salient self’s norms. That fits “self-interest on a larger self-identity” well, provided the larger self is the salient one.

• The triviality objection (CLAIM-492). SIDE experiments fix the self’s level by instruction, not by what participants care about, and that level predicts the direction of influence (p. 28). That is the kind of independent criterion the objection asks for. Chan’s donors are a sharper case. Salience raised giving “particularly among low identifiers” (p. 39), people who by their own report are less attached. If this holds, a salience criterion predicts prosocial action beyond measured attachment. But caring was not measured separately, the outcomes are attitudes and one field donation study, and all of it is at second hand.

• Batson’s objection (CLAIM-494). Not addressed. Nothing here concerns empathic concern for an individual.

• Self-categorization theory (THEORY-1114). The implicit out-group in the polarization account (p. 25) bears on the THEORY’s proposed test against Skorich, as noted from Spears 2021. Lea et al. 2007 is a candidate primary study, from the authors’ own lab.

• CASE-104 and QUESTION-119. The strategic dimension gives a reading of fan crowds. Chants and boycott calls, made anonymously or among many other fans, express in-group norms that the club or police would sanction, so they show the norm, not each fan’s view. Stated as a derivation, not a finding about fans.

• A dated note is added to CLAIM-491. Proposed text:

  Note of 2026-10-29: the crowd member is not selfless. Spears and Postmes (2015), read in place of Reicher et al. 1995 and Postmes & Spears 1998, state the SIDE model’s case against deindividuation: anonymity and immersion do not remove the self but bring “the group in the self” (p. 27), and behaviour follows the group’s norms more closely, not less. So the thesis’s denial of selflessness holds for crowds. Its word “larger” does not: anonymity amplifies whichever identity is salient, which is a switch of self. Experimental salience manipulations, which fix the self’s level apart from caring, are the kind of criterion CLAIM-492 asks for.

• No THEORY filed. SIDE would be a THEORY in its own right. Its founding statement and every primary study are unread.

• No ML instruction.

Limitations

• By the model’s authors. Every study described in detail is theirs or their collaborators’. The chapter does not report failures to replicate.
• No statistics. Designs and directions only; no sample sizes or effect sizes, including for the 1998 meta-analysis.
• Laboratory attitudes, mostly. Influence is measured as attitude change in discussion groups. The behavioural outcomes are a few field results (Chan 2010; Ren et al.; Farzan et al.).
• Typos.
  • “in Manchester the mid-1980s” (p. 24); “tends be more extreme” (p. 25); “might otherwise we sanctioned” (p. 27); “when the discussion topic were relevant” and “the Dutch were predicted be more critical” (p. 35); “connected though interpersonal relations” (p. 39); “a number extensions” (p. 41).
  • “Tanis & Postmes, Tanis & Postmes, 2005, 2007” (p. 34).
  • A full reference printed in the running text for Spears et al. 2002 (p. 40).
  • Citations that disagree with the reference list. Spears, Lea & Lee is 1990 in the text and 1992 in the list, with the 1990 volume. Matheson & Zanna is 1998 in the text and 1988 in the list. Farzan et al. is 2012 in the text and 2011 in the list. Postmes et al. 2013 has “Van de Sande” in the text and “Van der Sande” in the list.
  • “Postmes, 2000” and “Spears & Lea, 2002” (p. 29) are not in the reference list.
  • In the references, “Anonymous (1998)” has the page range “3481–407”. Turner 1982’s title lacks “social” (“Towards a cognitive redefinition of the social group”, as Spears 2021 gives it). Farzan et al. carry a library proxy URL.

Open questions

• Is the depersonalized member’s personal self lowered, or only outranked? A study measuring personal self-concept clarity and group identification together, under anonymity, would tell “switch” from “addition”. That bears on whether SIDE’s self is changed or larger.
• Does salience predict sacrifice, not only attitude change, when identification and caring are held fixed? Chan’s design, with a measure of concern for the group’s welfare added, would answer CLAIM-492.