Workplace bullying as predicted by non-prototypicality, group identification and norms: a self-categorisation perspective

Research and theory on deviance in work groups suggest that nonprototypical members risk devaluation and mistreatment by their peers. Drawing on the self-categorisation theory, we propose and test a contextual model to explain workplace bullying from a target perspective, using non-prototypicality as a predictor and social identification and anti-bullying norms at the work group level as twoand three-way cross-level moderators. Multilevel modelling and a sample of employees from the university sector in the Low Lands (n = 572) was employed. In line with our first hypothesis, we found that risk of exposure to workplace bullying is particularly high for non-prototypical work group members. We also hypothesised that work group social identification would facilitate bullying of non-prototypical members, while anti-bullying norms would buffer it. Results showed, however, that both conditions acted as buffering moderators on the main association. Lastly, the combination of the two also entailed a cross-level three-way interaction effect, showing that non-prototypicality is associated with bullying only in work groups characterised by low levels on both moderators. These novel and partly unexpected results demonstrate the relevance and significance of group level explanations for workplace bullying, holding significant implications for scholars and practitioners. ARTICLE HISTORY Received 19 February 2019 Accepted 3 January 2020


Introduction
Workplace bullying affects employees and organisations in every corner of the world , denoting systematic negative interpersonal treatment of employees by other organisational members in a situation where the target has problems defending against the behaviour due to a real or perceived power imbalance (Einarsen, Hoel, Zapf, & Cooper, 2011). Regarded by some as the single most devastating psychosocial work environment stressor (Hauge, Skogstad, & Einarsen, 2010), workplace bullying yields significant detrimental outcomes among targeted individuals (Nielsen & Einarsen, 2012) and organisations alike (Hoel, Sheehan, Cooper, & Einarsen, 2011). Consequently, bullying is now widely acknowledged as a major and pervasive work environment problem, which in turn has spurred a need for a fuller understanding of its antecedents, and in particular, the mechanisms by which it commences and escalates (Rai & Agarwal, 2018). This especially applies for mechanisms residing at the group level (Ramsay, Troth, & Branch, 2011), even though scholars have made a strong case for the significance of the group process perspective when explaining bullying, harassment and other forms of mistreatment in the workplace (Escartín, Ullrich, Zapf, Schlüter, & van Dick, 2013;McCluney & Cortina, 2017).
In keeping with recommendations to explore contextual and group level explanations for workplace mistreatment (e.g. McCluney & Cortina, 2017), the key ambition of this paper is twofold. Firstly, we aim to disentangle the association between non-prototypical work group membership and risk of exposure to bullying. As a backdrop for this aim, we draw on the tenets of the self-categorisation theory (SCT; Turner, Hogg, Oakes, Reicher, & Wetherell, 1987) as well as on previous research demonstrating that category based minority status (e.g. Cortina, 2008) and vulnerable in-group positions associated with nonprototypicality (e.g. Marques & Paez, 1994) are related to rejection and risk of mistreatment. However, we also expand on these contributions by assessing non-prototypicality per se, as opposed to any specific individual indicators of non-prototypicality, and doing so with respect to exposure to bullying behaviours in work groups. This entails a novel approach with implications for practice and further theoretical development and integration. Secondly, drawing on the broader social identity theory-framework (SIT; Tajfel & Turner, 1985) and related theories such as social rules theory (SRT; Argyle, Henderson, & Furnham, 1985), we test how social identification and anti-bullying norms at the work group level moderates this relationship, both independently and in combination, employing multilevel analyses. Ultimately, the aim is to establish whether specific group level conditions that are under the influence of organisations may protect non-prototypical work group members from exposure to workplace bullying behaviours.

The self-categorisation framework
As a derivate of SIT, SCT posits that in social contexts, individuals partially understand and experience self and others as functions of group membership. Specifically, people tend to categorise self and others in social groupings, and subsequently to derive a part of their personal identity from the common identity of the group to which they belong (an in-group), through assimilation to the groups' prototype. A prototype in this regard, is a shared representation of the most typical, or "ideal" group member (Hogg & Terry, 2000), serving important psychological functions once established as a source of social identity. According to the self-esteem hypothesis (Abrams & Hogg, 1988), drawing on Festinger's (1954) theory of social comparison, individuals constantly seek self-enhancing social comparisons in order to achieve a positive self-image. Consequently, a salient group identity formed around a valued prototype serves a self-enhancing function for individual members of a group. In addition, according to the uncertainty reduction hypothesis (Hogg, 2000), group members' assimilation to prototypes may also serve to fulfil the basic human need for stability and predictability, as prototypes inform group members about which attitudes and behaviours are expected and accepted in the group (Hogg & Terry, 2000). Thus, a prototype represents the core features of a social group and renders group life a safe and predictable source of positive identity for in-group members. However, this also implies that group members are motivated to eliminate threats to the salience and positive nature of the prototype, suggesting that non-prototypical group members may be at increased risk of rejection and mistreatment. In an organisational context, this is relevant because organisations house a myriad of formal and informal groups, including the organisation itself, its units, divisions, professions and demographic sub-groups, all with different social identities and group prototypes (Hogg & Terry, 2000). Hence, to the degree that group dynamics govern bullying mechanisms within organisational life, the framework of SCT should hold significant predictive and explanatory power with respect to antecedents.
In the following, we outline a rationale for a potential link between non-prototypicality and workplace bullying. Note that whereas we mainly employ the term non-prototypicality as derived from SCT, the literature on mistreatment of non-prototypical individuals refer to a number of terms similar to or overlapping with this construct, including minority status, marginalisation and in-group deviance.

Non-prototypicality and workplace bullying
The key premise in the assumption that non-prototypicality is related to workplace bullying rests on the notion that members experience group prototypes as functional and constructive sources of a positive identity and of interpersonal cohesion. When all members of a group are highly prototypical, a general climate of social attraction is likely (Hogg & Terry, 2000). When certain members stray too far from the prototype, however, the social identity of other group members is threatened. This also hampers any associated selfesteem-and uncertainty reduction effects (Hogg, 2001), implying that non-prototypicality may entail negative consequences in terms of how one is regarded and treated by fellow in-groupers (Hogg, 2005). For example, non-prototypical members are typically liked less, have less influence on other members and are less trusted than are prototypical members (Hogg, 2005;Hogg, Fielding, & Darley, 2005). Research on the so-called black sheep-effect goes far in supporting such a notion, showing that in-group members tend to hold greater disdain against non-prototypical group members than against actual out-group members (Marques & Paez, 1994;Marques, Yzerbyt, & Leyens, 1988). Related to this, experimental studies suggest that in-group deviants (i.e. non-prototypical members) are devaluated as a step towards ejection from the group (Eidelman, Silvia, & Biernat, 2006), which mirrors Hogg's notion, that non-prototypical members are likely targets of rejection and vilification, often being labelled as outcasts (Hogg, 2005). Alternatively, non-prototypical members may fulfil the role of a scapegoat (Hogg, 2005), denoting a vulnerable and unprotected position where one is likely to be blamed with any frustration or misfortunes evident in the group, and to be treated accordingly (see e.g. Thylefors, 1987).
Based on this rationale, a link between work group non-prototypicality and exposure to bullying behaviours is plausible. Surprisingly, though, empirical support for such a position remains scattered, represented only by studies that mirror the mechanisms outlined, but without addressing them directly. One study, for instance, investigating unequal gender distribution and conflicts in work groups, found that gender identity salience (i.e. proneness to describe the group in terms of an unequal gender distribution) contributed to higher conflict levels (Randel, 2002). In another study, gender dissimilarity (i.e. representing a gender minority) within work groups in the Dutch university sector was shown to reduce one's degree of social inclusion (Jansen, Otten, & van der Zee, 2017). Furthermore, work group heterogeneity in general has been found to represent a condition for higher conflict levels (Pelled, Eisenhardt, & Xin, 1999). Scholars have also found that within work groups, peers tend to disproportionally victimise both high performers (Kim & Glomb, 2014) and low performers (Jensen, Patel, & Raver, 2014), possibly reflecting a similar tendency involving perceptions of group member non-prototypicality.
Additionally, there is a large body of research showing that minority statusimplying non-prototypical in-group membershiprepresents a risk factor for exposure to workplace bullying and other forms of mistreatment (e.g. Cortina, 2008). For example, racial/ethnical minorities (Fox & Stallworth, 2005), gender minorities (Eriksen & Einarsen, 2004;Kabat-Farr & Cortina, 2014) and physically disabled workers (Fevre, Robinson, Lewis, & Jones, 2013) are more frequently exposed to bullying behaviours than are others. Importantly, when individuals hold several minority statuses simultaneously, interaction effects involving mistreatment incidence that go over and beyond the isolated risk associated with each minority status put together have also been found (Berdahl & Moore, 2006;Shaw, Chan, & McMahon, 2012). This may indicate that the social connotations of marginalisation is in fact an active ingredient in the relative increased vulnerability for bullying and other forms of mistreatment among group members with a minority status, thus further supporting our notion that non-prototypicality is at the heart of the risk of being targeted with bullying among in-group "deviants".
Although our study may be the first to assess non-prototypicality directly with respect to workplace bullying or similar forms of workplace mistreatment, it is evident from the above review that being marginal, different or atypical comprise a widely recognised conglomerate of related risk factors relating to non-prototypicality. Hence, based on our outline of the literature so far, we propose our first hypothesis as follows: H1: Non-prototypicality is associated with an increased risk of exposure to bullying behaviours in work groups The role of group level social identification and norms Following the tenets of SCT, we have proposed that a relationship between non-prototypicality and workplace bullying may exist because groups place high value on their social identity, and because the presence of in-group "deviants" threatens this identity by decreasing the clarity of the prototype (Abrams, Marques, Bown, & Henson, 2000;Hogg et al., 2005). In this reasoning, the salience of the group's identity denotes the "stakes", i.e. what the group and its members risk losing by allowing the prototype threat to remain part of the group. Hence, the salience of a work group's social identity should itself play a key role in group members' inclination to reject or mistreat non-prototypical co-workers. Possibly, in a work group with a vague or ambiguous social identity, the identity threat associated with non-prototypical members is less important, because group members derive less of their individual social identity from that of the group. In a work group with a salient identification, on the other hand, the stakes are higher, because members acknowledge the group's social identity as a significant source of their personal identity, and possibly as an important source of uncertainty reduction (Hogg, 2000) and self-esteem (Abrams & Hogg, 1988). This would be in line with Hogg (2005Hogg ( , 2016, who suggests that the group dynamical implications of prototypicality are especially pronounced in groups with a salient identification, because highly identified members are more attentive to prototypicality than others. Additionally, it is also possible that high identifiers experience self in terms of the group prototype (i.e. self-stereotype) to a larger extent when the group's distinctiveness is under threat (Spears, Doosje, & Ellemers, 1997). If so, salient group social identification should further increase the perceived gap between prototypically central and peripheral members, thereby reinforcing any existing inclination to reject and mistreat group members low on prototypicality.
This rationale is also in line with Ramsay et al. (2011) and Escartín et al. (2013), proposing that a salient work group identification may act as a facilitating moderator in the relationship between non-prototypicality and workplace bullying, with non-prototypical members at an even higher risk of bullying when other group members are highly identified. Based on these notions, we propose our first cross-level two-way interaction hypothesis as follows: H2. Work group social identification strengthens the relationship between non-prototypicality and exposure to bullying behaviours Thus far, we have based our argumentation mainly on the theoretical premise that prototypes are sources of distinct and unambiguous social identities, leading to the presumption that non-prototypical group members face a higher risk of exposure to bullying behaviours, and particularly so in highly identified groups. However, as stipulated in SCT, prototypes are sources not only of distinct, but also positive social identities (Abrams & Hogg, 1988;Hogg & Terry, 2000). Hence, as workplace bullying denotes an inherently negative phenomenon, there should also be exceptions to the rules suggested in H1 and H2, at least to the degree that a work group appraises its social identity as incompatible with such behaviour. In order to account for that possibility, we draw on the notion that group identification engender adherence to the norms of the group (Ashforth & Mael, 1989). Generally, norms or social rules (cf. Argyle et al., 1985) guide group behaviour, and acting in accordance with group norms reinforces one's position in the group as well as the social identity of the group and its members. Thus, in order to fully understand the behavioural implications of social identity processes, such as those presently proposed, the role of norms must be taken into account (Marques, Abrams, & Serôdio, 2001).
In this study, we employ the term work group anti-bullying norms to account for a work group's shared and agreed-upon perception that bullying behaviours are illegitimate and in conflict with prevailing social rules. We hence understand the construct as a source and sustenance of a positive social identity, and propose that it may affect the previously outlined associations in terms of two mechanisms. Firstly, because these norms are in conflict with any predominant inclination to mistreat non-prototypical members, they may moderate the proposed main association simply by heightening the threshold to act on those intentions due to the general impact of social rules (cf. Argyle et al., 1985). Importantly, as norms are often deeply embedded in the prototype, anti-bullying norms may even render enactment of bullying behaviour itself a threat to the group's social identity. Prevailing anti-bullying norms may thus enforce other, more constructive ways of responding to and relating to non-prototypical work group members. For instance, anti-bullying norms may represent a condition where group members more readily construe non-prototypicality as a source of favourable diversity and broadened group competence and competitive ability. Drawing on Brewer's (e.g. 1996) optimal distinctiveness theory, non-prototypical work group members may even contribute to a positive sense of balance between group identity and individual identity among other members, insofar as the group is characterised by a pro-social climate tolerant of diversity. Hence, as a first exception to the rule, anti-bullying norms should represent a protective mechanism where non-prototypical work group members otherwise would have risked exposure to bullying behaviours from fellow group members (see also Ramsay et al., 2011). Secondly, as norms and identification are intertwined, the buffering effect of anti-bullying norms should ordinarily be stronger when social identification is salient, implying a possible three-way interaction. However, as we argued with respect to H2, social identification may itself denote a risk factor for bullying among non-prototypical group members. Hence, rather than to presume that salient identification and strong anti-bullying norms combine to yield a buffering mechanism over and above that of anti-bullying norms alone, we suggest that norms eliminate the added risk of bullying among non-prototypical members in highly identified groups, and present our third and fourth hypotheses as follows: H3. Work group anti-bullying norms weaken the relationship between non-prototypicality and exposure to bullying behaviours H4. Anti-bullying norms buffer the facilitating role of group social identification in the relationship between non-prototypicality and exposure to bullying behaviours For an overview of the proposed conceptual model including all study variables, see Figure 1.

Procedure and sample
The present study employs data collected by an independent work environment assessment company in June-July 2017 among employees of a university in the Low Lands. They submitted electronic questionnaires to 4,836 employees, which yielded a response rate of 26%. For the present analyses, we extracted a sub-sample of all employees belonging to work groups with at least five members in order to accommodate the sample for group level aggregation of certain variables. This resulted in a final sample of 572 employees nested in 55 work teams. The majority of the employees in this sample was female (63%). Whereas 55% occupied an academic position, 45% occupied an administrative position. The mean age was 42 years (SD = 12.18). Fifty-six percent worked full-time. Among the part-time employed, 18% worked ≥ 80%, 12% had 60-70% positions, 6.1% had a 50% position, and finally, 7.7% worked less than half-time.

Measures
Workplace bullying was measured using a standardised nine-item short version of the Negative Acts Questionnaire-Revised (Notelaers, Van Der Heijden, Hoel, & Einarsen, 2018). The scale comprises a list of common bullying behaviours, such as "Silence or hostility as a response to your questions or attempts at conversation". Each item was evaluated by the respondents in terms of frequency of exposure over the past six months, using a five-point Likert-type scale ranging from (1) never to (5) daily. Reliability assessment revealed adequate internal consistency for this scale (α = .88).
Non-prototypicality was measured using a five-item scale from Van Kleef, Steinel, Van Knippenberg, Hogg, and Svensson (2007), with each item referring to individual perceptions of being a more or less typical for and/or central within a particular group. In the present study, we specified referral to the group in question to the respondents' work groups in each item, e.g. "How similar do you feel to your fellow work group members?" The respondents evaluated each item using a six-point Likert-type scale (adapted from the original nine response categories for the purpose of this study), ranging from (1) not at all to (6) completely. Reliability assessment revealed adequate internal consistency for this scale (α = .88).
Work group social identification was measured using an adapted version of Mael and Ashforth (1992) organisational identification scale. In accordance with a study by Escartín et al. (2013), the six items of the scale were rephrased to (1) include a reference to the respondents' respective work group, and to (2) refer to the respondents' assessment of the identity perceptions of the group as a whole in order to allow aggregation to the group level. As an example, one item was "When people in my work group talk about this work group, they usually say 'we' rather than 'they'". Each item was evaluated by the respondents using a five-point Likert-type scale ranging from (1) totally disagree to (5) totally agree. Reliability assessment revealed adequate internal consistency for this scale (α = .8).
Work group anti-bullying norms were measured using a five-item scale developed for the purpose of this study, inspired by previous work on measuring violence climate (Kessler, Spector, Chang, & Parr, 2008). The complete list of items were: (1) "In my work group, bullying is condemned", (2) "In my work group, workplace bullying would be reported (if it occurred)", (3) "In my work group, you can easily get away with bullying" (reversed item), (4) "In my work group, someone can be bullied as a disciplinary measure", and (5) "In my work group, bullying is tolerated" (reversed item). The respondents assessed each item using a five-point Likert-type scale ranging from (1) totally disagree to (5) totally agree. Reliability assessment revealed adequate internal consistency for this scale (α = .88).
Covariates included administrative vs academic position, sexual harassment, violence and group conflict. We controlled for administrative vs academic position because of earlier reports of workplace bullying among university employees (Björkquist, 1992), in order to rule out any effects attributable to professional roles. We also controlled for violence, conflicts and sexual harassment, knowing that these phenomena correlate particularly high with workplace bullying at low or discrete levels (Notelaers, Van der Heijden, Guenter, Nielsen, & Einarsen, 2018). These constructs were measured using five-point Likert-type scales. We measured violence using a four-item measure based on the "physical intimidation"-subscale of the full NAQ-R (Einarsen, Hoel, & Notelaers, 2009), adding one item on "verbal violence" (α = .72). We measured conflicts using seven items based on Jehn's (1995) Intragroup Conflicts Scale (α = .91), and sexual harassment using a five-item measure based on the Bergen Sexual Harassment Scale (see Nielsen, Bjørkelo, Notelaers, & Einarsen, 2010) (α = .8).

Aggregation statistics
Prior to aggregating individual responses to the group level, we assessed within-group agreement for anti-bullying norms and social identification using the rwg(J)-index (James, Demaree, & Wolf, 1984) and a uniform null-distribution. Furthermore, we calculated the average deviation index (AD M ) proposed by Burke, Finkelstein, and Dusig (1999). A value of .70 or higher is often held to suggest strong agreement (George, 1990;Lebreton, Burgess, Kaiser, Atchley, & James, 2003), while values below .50 implies disagreement. For group identification we obtained a rwg(J) of .84 and an AD M index of .41. For anti-bullying norms we obtained a rwg(J) of .56 and an AD M of .69. Next, we computed the intra-class correlation coefficient ICC(1) (Bliese, 2000) to examine the relative consistency of responses among group members. ICC(1) for work group anti-bullying norms was .06, denoting a small to medium effect. For work group social identification, ICC(1) was .11, denoting a medium to strong effect. Altogether, the present ICC (1) values suggest that group membership influenced members' perceptions of social identification and anti-bullying norms to a degree that justified the use of multi-level analyses (LeBreton & Senter, 2008). We further estimated the reliability of the group mean by assessing the ICC(2) (Bliese, 2000). ICC(2) was .79 for group social identification and .87 for anti-bullying norms. Both are well above the critical value of .70, indicating that mean scores can be used to reliably differentiate between groups (Klein & Kozlowski, 2000). Finally, we carried out a one-way analysis of variance (ANOVA) to ascertain meanlevel differences between groups in terms of anti-bullying norms and social identification. The F value was statistically significant for both work group social identification (F(54, 517) = 2.07, p < .01) and work group anti-bullying norms (F(54, 517) = 1.61, p < .01). Together, the aforementioned indices provided justification for aggregation.

Hypothesis tests
The hypotheses were tested in MLwiN version 3.02, using stepwise estimation of different models (Charlton, Rasbash, Browne, Healy, & Cameron, 2017). To account for the nested dependent nature of the measurements at the lower level, we employed a hierarchical linear modelling approach (Hox, 2002). Both the independent variable (non-prototypicality) and the dependent variable (exposure to workplace bullying behaviours) comprise individual level (Level 1) variables, while the moderators (social identification and antibullying norms) comprise group level (Level 2) variables. Following the recommendations of Hofmann and Gavin (1998), we centred all predictors except the control variable administration vs academic position. We centred the Level 1 predictors around the group mean and the Level 2 predictors around the grand mean (see Hofmann & Gavin, 1998) to reduce possible problems with multicollinearity (Enders & Tofighi, 2007;Hofmann & Gavin, 1998).
First, we built an intercept-only model (a null-model) (Rasbash, Steele, Browne, & Goldstein, 2009), subsequently including covariates and predictors to the equation in a stepwise fashion. Covariates were entered in the following order: Administrative vs academic position (model 1), sexual harassment, violence and group conflict (model 2). Direct effect predictors were entered in the following order: Non-prototypicality (model 3), work group anti-bullying norms (model 4) and work group social identification (model 5). Interactions were entered in the following order: Non-prototypicality*work group social identification (model 6), non-prototypicality*work group anti-bullying norms (model 7), and finally, anti-bullying norms*social identification as well as non-prototypicality*anti-bullying norms*work group social identification (model 8). Hence, we also used the between-group interactions as a control for a more clean-cut assessment of the multilevel relationships evident in the data (Enders & Tofighi, 2007;Hofmann & Gavin, 1998) Pseudo-R 2 s were calculated after each step, directed on the within and the between part of the variance explained by adding a variable in a step (Snijders & Bosker, 1994). We tested the multivariate significance of effects in each step by computing the increase in model fit compared with the previous step, represented by decrease in the Δ−2 × log statistic. This statistic follows a X 2 distribution, and with one degree of freedom from between the models, a Δ−2 × log statistic of −3.84 yields a significant difference and improvement of fit from one model to the next. We plotted interactions at two levels of the moderator (i.e. +1 SD and −1 SD; Bauer & Curran, 2005), using Jeremy Dawson's tool (see Dawson, 2014), while testing simple slopes with Preacher's hierarchical linear modelling tools (see Preacher, Curran, & Bauer, 2006), available online (http://www.quantpsy.org/interact/ hlm3.htm).

Results
An overview of descriptive statistics and correlations is presented in Table 1, and results from the main analysis are presented in Table 2.
From the null model, 4.7% of the variance of exposure to bullying behaviours can be attributed to the difference between work groups. Model 1 shows that reports of bullying were not significantly associated with working as an administrative or academic member of the university. However, as evident from Model 2, reporting sexual harassment, violence and conflicts were. In Model 3, a significant association between non-prototypicality and workplace bullying was found (B = 0.11, p < .05), in support of H1. In Model 4, the main effect of work group anti-bullying norms similarly entailed a significant negative effect on bullying (B = −.12, p < .05) while the overall fit of the model improved (Δ −2*loglikelihood of 5.8). Model 5 also revealed a significant negative relationship between group identification and exposure to bullying behaviours (B = −.18, p < .05), again improving the fit of the overall model (Δ−2*loglikelihood of 6.04). Thus, from the direct effects assessed, we conclude that non-prototypicality increases the risk of exposure to workplace bullying, while work group anti-bullying norms and social identification represent contextual protective factors against workplace bullying.
As for the first two-way interaction, Model 6 showed that the interaction between work group social identification and non-prototypicality was negative and significant (B = −0.15, p < .05). Specifically, as visualised in Figure 2, in groups with stronger identification, a weaker relationship between non-prototypicality and exposure to workplace bullying exists. Furthermore, simple slope tests revealed that the main association was only significant under conditions of low group identification (B = .11, p < .001) and not for high group identification (B = −0.05, p = .503). In addition, the interaction term explained an additional 1.69% of the variance while the Δ−2*loglikelihood was significant. Hence, with respect to H2, the results revealed a tendency opposite of what we anticipated, with social identification significantly weakening rather than strengthening the main relationship. Model 7 showed that the interaction between work group anti-bullying norms and non-prototypicality was also negative and significant (B = −0.15, p < .05). Specifically, as visualised in Figure 3, in groups with stronger anti-bullying norms, a weaker relationship between non-prototypicality and exposure to workplace bullying exists. Moreover, simple slope tests revealed that the main association was significant for weak (B = .11, p < 0.05) rather than for strong norms (B = -0.05, p = .466). Model 7 also showed that the additional variance in individual exposure to bullying explained by the cross-level interaction was 0.39% and that the overall fit improved as the Δ−2*loglikelihood was 8.5, with one degree of freedom. Hence, H3 stating that work group anti-bullying norms buffers the relationship between non-prototypicality and bullying was supported.
Before assessing the three-way interaction, we estimated the between-level interaction for work group anti-bullying norms and group level social identification. Model 8 showed that both interactions accounted for an additional 0.78% of the variance of individual level    exposure to workplace bullying. Furthermore, the three-way interaction was found to be significant, explaining an additional proportion of the variance and improving the fit of the overall model (Δ−2*loglikelihood with one degree of freedom = 4.57). According to tests of the simple slopes depicted in Figure 4, the relationship between non-prototypicality and exposure to bullying was positive when group identification was low and anti-bullying norms were weak (B = 0.18, p = .002). Additionally, non-significant simple slopes  were found in groups where group identification was low and anti-bullying norms were strong (B = 0.03, p = .654), where group identification was high and anti-bullying norms were weak (B = 0.08, p = .307), and where group identification was high and anti-bullying norms were strong (B = 0.08, p = .172). H4 held that that anti-bullying norms would eliminate the facilitating role of work group social identification in the relationship between non-prototypicality and exposure to bullying. Yet, as identification unexpectedly proved to constitute a buffering moderator, support for this hypothesis was logically impossible. However, in showing that only work groups low on both group level moderators tend to enact bullying behaviours against non-prototypical members, the three-way interaction is still of high theoretical and practical interest, as discussed below.

Discussion
The present study has shown that exposure to bullying behaviours is significantly more common among non-prototypical work group members as compared to members with prototypically central positions in the group (H1). Furthermore, we have shown that work group social identification and anti-bullying norms buffer the general incidence of bullying behaviour in work groups, both alone and in combination. Lastly, we have shown that the same group level constructs buffer the specific incidence of bullying behaviour in work groups otherwise directed at non-prototypical group members, again both alone (H2, H3) and in combination (H4). Hence, while non-prototypicality does indeed represent a significant risk factor with respect to workplace bullying, work group social identification and anti-bullying norms can be construed as protective factors, both in general and specifically concerning non-prototypical work group members. . Three-way cross-level interaction between non-prototypicality, work group social identification and work group anti-bullying norms. Note. Only the slope representing the combination of vague social identification and weak anti-bullying norms is statistically significant.
Our finding concerning H1 aligns with key tenets of SIT and SCT, as well as with previous findings regarding workplace mistreatment of individuals with minority status (Cortina, Kabat-Farr, Leskinen, Huerta, & Magley, 2013;Shaw et al., 2012) and with research and theory on "scapegoats" (Thylefors, 1987) and "black sheep" (Marques & Paez, 1994). However, as far as we know, this is the first study to demonstrate the significance of within-group deviance with respect to bullying based on a direct and non-specific assessment of non-prototypicality. In supporting H1, the present study thereby discloses a fundamental, but previously merely presumed social psychological antecedent of workplace bullying, and demonstrates a tendency with wide applicability across the whole range of non-prototypical statuses existing within organisational sub-unites such as work groups and teams. Importantly, we have also shown that contextual factors that are subject to the influence of organisations can alleviate this added risk. Specifically, work group social identification and anti-bullying norms are demonstrated as effective buffers of workplace bullying of non-prototypical workers, via two-and three way cross-level interaction effects. Interestingly, only anti-bullying norms was hypothesised to entail such a buffer effect. For work group social identification, we anticipated an opposite tendency based on the notion that groups with a salient identification are more inclined to penalise and reject any threats to the prototype, in line with recent theoretical suggestions (Escartín et al., 2013;Ramsay et al., 2011). We based this assumption on key tenets of SCT, such as the motive for self-enhancement (Abrams & Hogg, 1988) and uncertainty reduction (Hogg, 2000), as a clear prototype is needed for both to be met. However, these motives were not originally hypothesised with respect to bullying, a severe and inherently negative form of interpersonal mistreatment. If behaviour aimed at eliminating a prototype threat can reach a critical level in terms of its negative connotations, a possible explanation for this finding could be that salient identification works to prevent the most severe and overt forms of rejection of non-prototypical members from escalating to a point where it becomes characteristic of the groups' social identity. This would further suggest that with respect to interpersonal mistreatment, the motive for self-enhancement could lead social identification to yield different effects depending on the severity of the behaviour involved. It also suggests that the motive for self-enhancement can take precedence over the motive for uncertainty reduction, which would be more dependent on prototypical clarity and group cohesion. Yet another explanation can be found in the self-categorisation theory of stress (SCS; Haslam, 2004), which posits that stressful events tend to prompt social support in groups with a salient identification. Possibly, as non-prototypical members face negative treatment from certain work group colleagues, a simultaneous process of increased social support from others may thus commence, reversing the mistreatment process before it escalates. These accounts remain theoretical for the time being, and further research on the associations between non-prototypicality, work group social identification and workplace bullying is needed to determine the mechanisms involved. However, our finding does offer an important corrective to the notion that group level social identification per se denotes a vulnerability factor for mistreatment and rejection of deviant group members.
The last finding of this study revealed that the inclination to enact bullying against nonprototypical work group members prevails only in work groups low on both social identification and anti-bullying norms. Apparently, in the vaguely identified groups, anti-bullying norms suffice to prevent bullying among those low on prototypicality. One explanation could be that workplace bullying conflicts with behavioural standards shared at and derived from a higher level, such as that of the organisation or society (Power et al., 2013). Moreover, as posited in social rules theory (Argyle et al., 1985), people readily create and alter social rules in a manner that allows goal attainment, a notion that should reasonably apply in work groups, where task completion and goal attainment is normally the shared objective. In a similar manner, in groups with weak norms, a salient identity was found to yield an equivalent buffering effect. This finding offers additional support to our alternative explanation for H2, that cohesive groups value positive identities to the degree that their identification protects non-prototypical members against mistreatment.
We may also add that both moderators entailed direct negative effects on the general prevalence of workplace bullying within the work groups of our sample. These findings imply that salient group identities and strong anti-bullying norms represent important contextual protective factors against bullying regardless of the distribution of prototypicality positions, and replicate previous findings and assumptions (Aquino & Lamertz, 2004;Escartín et al., 2013;Ramsay et al., 2011). Importantly, when combined, the two-way interaction between these group level constructs also entailed a general contextual protective condition over and above the isolated effect of the two put together. Hence, high identification and strong norms appear to be a constructive combination regardless of the prototypicality of the work groups' members. More generally, in this study, we go a long way in showing that contextual factors such as group level processes are indeed significant and relevant in the area of bullying and harassment, as previously suggested (see e.g. McCluney & Cortina, 2017). However, we also demonstrate the importance of putting established theoretical suggestions and notions concerning such factors to the test.

Strengths, limitations and directions for future research
The use of multilevel analyses represents the principal methodological strength of this paper, as it allows for conclusions about context or group level constructs and their impact on individual level risk factors. The multilevel design with cross-level moderators also largely avoids the pitfalls of common method variance (see Podsakoff, MacKenzie, Lee, & Podsakoff, 2003) by drawing on information collected from numerous sources rather than from self-reports only (Evans, 1985;McClelland & Judd, 1993). In addition, we hold that measuring each construct directly represents another significant strength, particularly with respect to non-prototypicality. As far as we can see, the present paper is the first to report on non-prototypicality per se with respect to workplace bullying or other forms of interpersonal mistreatment in organisations. Thus, the paper theoretically contextualises a large body of scholarly and empirical work on bullying and harassment among specific minorities and protected groups (e.g. Fevre et al., 2013;Fox & Stallworth, 2005;Kabat-Farr & Cortina, 2014;Shaw et al., 2012).
Among the study's limitations, we would first highlight that that we employed data collected at a single time point. Hence, we were unable to predict stability adjusted exposure to workplace bullying, and by extension, to make causal inferences. On the one hand, the possibility that workplace bullying can instigate perceptions of being non-prototypical thus remains. On the other hand, there is a large body of theoretical literature suggesting and empirical literature demonstratingan increased risk of workplace mistreatment such as bullying among employees demonstrably low on prototypicality (e.g. genderand ethnic minorities; high-and low performers), leading us to assume that bullying should nonetheless be regarded as an outcome in this relationship. In addition, if bullying was indeed a more significant predictor of non-prototypicality than the reverse, we could reasonably have assumed a non-existent or even opposite interaction pattern for the moderating role of anti-bullying norms. Specifically, bullying targets would likely feel even more non-prototypical if members of a group where other members are unlikely targets (i.e. in groups strong on anti-bullying norms), in contrast to the buffering effect presently demonstrated. Hopefully, these interpretations may be assessedand our findings replicatedin future studies employing cross-lagged designs. Second, although we primarily employ standardised scales in our study, the measure of anti-bullying norms has not been validated in previous studies. At the same time, the scale has a high internal consistency and appears to work adequately in our analyses, leaving us confident that the employment of the scale has not negatively influenced the results. Third, although the total sample holds adequate statistical power for the present analyses, the response rate of 26% is somewhat lower than the average of 36% for surveys assessing sensitive issues in organisations (Greco, O'Boyle, & Walter, 2015), possibly due to the fact that the respondents were asked to provide information about rather infrequent behaviours (see Armstrong & Overton, 1977). Finally, as respondents were sampled from the university sector only, there is a chance that sector-specific circumstances such as academic rivalry may have affected the results. Hence, one should yet remain cautious with respect to generalising the results to the general working force.
With these limitations in mind, we invite others to replicate our study, preferably employing samples in different types of organisations. We would also recommend the use of prospective data to address the unfolding of bullying scenarios over time, preferably using several time lags in order to allow for more precise estimates of the strength and dynamics of the associations. In addition, future research could benefit from addressing the perspective of actual or potential perpetrators with respect to social identity, e.g. by addressing the association between employees' perceptions of other work group members' non-prototypicality and any intended or actual enactment of bullying behaviours.

Implications and conclusion
In terms of theoretical implications, this paper contextualises existing research showing that bullying and other forms of mistreatment disproportionally affect minorities and protected groups (see Cortina, 2008;Lewis, Giga, & Hoel, 2011). Specifically, by demonstrating that non-prototypicality per se is an active social psychological ingredient in workplace mistreatment such as bullying, our results go far in exposing the fundaments of victimisation of workplace "deviants". Importantly, this also further justifies the use of prevailing social psychological theories to account for, explain, prevent and tackle bullying at work by showing that social identity and identification mechanisms denote a significant part of the psychological fundament of workplace mistreatment (McCluney & Cortina, 2017).
Secondly, in showing that group level social identification buffers mistreatment regardless of prototypicality, this study suggests that measures to boost identification may not be the dual edged sword it has previously been deemed (see e.g. Ramsay et al., 2011). Although members of cohesive groups are highly invested and may have much to lose in terms of individual social identity when group identification is obscured, the need to keep identification positive may appear to take precedence when threats to identification are met. Whether certain sub-groups low on prototypicality are still at increased risk compared to others under conditions of salient group identification remains to be seen. However, based on our findings, we would hold that identification is essentially constructive and positive in and of itself, even in the presence of non-prototypical members.
In terms of practical implications, we underscore the significance of our main finding that non-prototypical members risk exposure to bullying behaviours to a higher degree than do other group members. This tendency appears to exist as a general mechanism, implying that it may be relevant regardless of what specifically characterises a prototypical versus a non-prototypical member. Whether it is professional background, tenure, sexual orientation, gender, ethnicity or any other characteristic alone or in combination, prototypicality seems to be relevant in and of itself. Practitioners may thus benefit from assessing the mistreatment susceptibility of teams, work groups and other organisational consortia though the lens of this insight, for instance by monitoring the interpersonal behaviour patterns in heterogeneous groups and paying particularly close attention to groups where one or more members appear to differ significantly from the prototype. Importantly, as suggested by our findings, one should also remain particularly aware of the significance of social identification and anti-bullying norms. Specifically, in showing that anti-bullying norms and social identification can in fact eliminate the risk of mistreatment of non-prototypical members, practitioners may greatly benefit from evaluating and exercising influence on organisational sub-units in terms of these constructs. With respect to identification, this may include working to maximise conditions for a healthy and positive social identity, ensuring that work groups strive for common goals and employ as much of the beneficial aspects of their own diversity as possible, for example by exploiting the resources, competency, background and experience of all members. Interestingly, research has shown that work group diversity is beneficial to the performance of a group when biases related to categorisation are countered or eliminated (van Knippenberg & Schippers, 2007;van Knippenberg, De Dreu, & Homan, 2004). Hence, fostering diversity tolerance in organisational sub-units may yield a double gain, preventing mistreatment and its disparaging consequences on the one hand and promoting work group efficiency and productivity on the other. With respect to anti-bullying norms, practitioners should first and foremost attempt to establish proactive norms against workplace bullying and other forms of mistreatment. This may include promoting awareness and explicitly defining the boundaries between acceptable and unacceptable interpersonal behaviour. It may also include establishing routines for reporting unwanted conduct to the appropriate responsible body, such as an immediate supervisor or HR.
In conclusion, in line with suggestions and recommendations from scholars in the workplace mistreatment research field (e.g. McCluney & Cortina, 2017;Ramsay et al., 2011), the present study demonstrates that group level psychological mechanisms hold significant explanatory potential with regard to workplace mistreatment such as bullying. Along with recent work (e.g. Escartín et al., 2013), this study contributes in bridging a gap in the literature where the significance of groups and the social context have largely been ignored up until now. As for the specific contribution of this paper in this regard, we conclude that non-prototypical work group members are more at risk of exposure to bullying than are others, in line with the tenets of SCT. Additionally, we conclude that social identification and anti-bullying norms at the work group level denotes effective buffering conditions for such a risk, reinforcing a climate that largely tolerates within-group diversity.

Disclosure statement
No potential conflict of interest was reported by the authors.