Leaders’ influence on collective action: An identity leadership perspective

Publisher's copyright statement: This is the peer reviewed version of the following article: Khumalo, N., Dumont, K. B. & Waldzus, S. (2022). Leaders’ influence on collective action: An identity leadership perspective. Leadership Quarterly. 33 (4), which has been published in final form at https://dx.doi.org/10.1016/j.leaqua.2022.101609. This article may be used for non-commercial purposes in accordance with the Publisher's Terms and Conditions for self-archiving.

The present research aimed at addressing this question theoretically from the perspective of the model of identity leadership (Haslam, Reacher, & Platow, 2011, Reacher, Haslam, & Platow, 2018, which proposes that leaders influence followers through social identity processes. As social identity, like efficacy beliefs, is key for collective action, we consider the identity leadership approach as most appropriate to address the how question.
Besides examining the social identity processes involved in leaders' influence on followers to act collectively, we argue and provide empirical evidence that leaders' influence through social identity depends on the group context in which the relationship between leader and followers is embedded.
Collective action and identity processes 7 contexts. For instance, there are social contexts that according to Simon et al. (1998), De Weerd andKlandermans (1999), Stürmer, Simon, Leowy and Jörger (2003), and van Stekelenburg and Klandermans (2013) produce politicised social identities, which by definition oblige members to advance group interests and goals through social actions aiming for social change. These politicised social identities correspond with distinct ideologies and orientations that prevail within and predominate the context of partisan groups and that are different from other contexts such as workgroups (Brown et al., 1992). As partisan groups, such as political parties, trade unions or NGOs, only have meaning because they either compete with relevant outgroups such as other political parties or they are set up against powerful adversaries such as employers or governments, they are most likely to define their ingroup and its outcomes with reference to other relevant groups and challenges; that is to say, a relational orientation is most likely to dominate (Brown et al., 1992). In contrast, workgroups can be conceived as task-based that is not necessarily related either objectively or psychologically to other groups. Therefore, workgroups might be better described in their orientation as autonomous as they tend to evaluate their ingroup and its outcomes with reference to some abstract standards such as efficiency and professionalism that typically do not implicate other groups (Brown et al., 1992). Thus, different group orientations and consequently different social identities might determine that the same degree of ingroup identification evokes different group outcomes.
Essentially, we argue that collective action to advance group interests and goals is in some social groups, such as partisan groups, part and parcel of the shared social identity. Or, to put it differently, to act collectively for the interest of the group is more normative in some social groups than in others. It is, however, important to note that we are not implying that only members of politicised groups engage in collective actions aiming at social change. As countless examples teach us, most politicised groups such as trade unions resulted from collective actions of non-politicised group members such as factory workers. We are, however, implying that the importance of social identity processes for the effectiveness of identity leadership to mobilise followers might vary depending upon the social context in which the relationship between leader and followers is embedded.
Consequently, although we assume that social identity influences collective action as one of its key predictors (e.g., social identity theory, Tajfel & Turner, 1979: selfcategorisation theory, Turner et al., 1987; and social identity model of collective actions, van , we propose that the social group context of the leader-followers relationship determines the role that shared social identity plays in the relationship between the perception of identity leadership and the compliance to the request by the leader to act collectively. More specifically, we propose that the relationship between perceived identity leadership and followers' compliance to the leader's request to act collectively should involve the mobilization of identification with the social group in some social contexts more than in others. We expect identification with the group to be more important in social groups for which collective action is normative (e.g., partisan groups) than in social groups for which collective action is less normative (e.g., workgroups). Therefore, we hypothesise: H2. The indirect effect of identity leadership on collective action through ingroup identification is moderated by group context. More specifically, the link between ingroup identification and collective action intentions is stronger in the partisan group context than in the workgroup context (see Figure 1).
We tested these two hypotheses in two studies using a quasi-experimental survey design. We acknowledge the limitations of this approach as it captures the individual subjective perceptions of identity experiences rather than the shared and lived experiences themselves (Fukuyama, 2018). To capture the latter other methods involving direct observations would be necessary as they are often applied in the study of dynamics in large-9 scale movements (e.g., participant observation, Drury & Reicher, 2000; ethnographic designs, Drury & Reicher, 2005;and/or interviews, Stevenson, Reicher, Pandey, Shankar, Tewari, & Hopkins, 2020). Studying such dynamics, however, would go beyond the scope of the current research as we aimed to examine how identity leadership experienced by an individual can motivate her or him to engage in collective action and how the role that social identification plays in this process varies across different group contexts. Such individual-level research, which is common in identity leadership studies in the context of workgroups (e.g., Steffens, Haslam, Kerschreiter et al., 2014) or in the context of sport and exercise (e.g., Fransen, Steffens & Haslam, 2015;Stevens et al., 2020), is a necessary complement to the indispensable observational approaches that are able to capture directly the group-level phenomena and historically contextualised intergroup dynamics.

Study 1 Participants
In total, 293 students from the University of South Africa took part in the study. The University of South Africa is a distance learning university. Thus, most students work fulltime and study part-time. The sample consisted of 231 females and 56 males (six participants did not indicate their gender). The age of the participants ranged from 18 to 65 years with a mean age of 29.6. Of all participants, 129 indicated to be employed and 99, not to be employed (65 did not answer the question).
As our database of Study 1 included 23 missing values on the items of the dependent variable, collective action (0.8%), we analysed whether they were missing completely at random (MCAR). Little's MCAR test was significant, Chi-Square = 369.16, df = 315, p = .019, indicating that these missing values were not missing completely at random. However, removing one item of the ingroup identification scale from the analysis rendered the MCAR 10 test insignificant, Chi-Square = 344.246, df = 305, p. = .060, that means that the nonrandomness was mainly due to the dependency on certain aspects of ingroup identification.
At the same time, these results suggest that it might be reasonable to consider that the less demanding (though not directly testable) assumption that missing values were missing at random was met. Using AMOS 28, we, nevertheless, applied different techniques of dealing with the missing values such as listwise deletion, maximum likelihood estimation, regression imputation (which for each case, regresses the unobserved values on the observed values, assuming that the population means and covariances of all variables are equal to their maximum likelihood estimates) and multiple stochastic regression imputation (which draws at random "from the conditional distribution of the missing values given the observed values, with the unknown model parameters" again "set equal to their maximum likelihood estimates", Arbuckle, 2021, p. 477). The latter technique generated ten data output files with imputed missing values. Apart from negligible differences in the estimations, the overall results were the same in all these analyses (apart from the fact that AMOS does not allow bootstrapping when missings are dealt with by maximum likelihood estimation, see Supplemental Material, Tables S14-16). For analyses including missing values, we therefore report in this paper only the pooled results of the multiple stochastic regression imputation, applying Rubin's rules (Rubin, 1976) and the Degree-of-Freedom approximation suggested by Lipsitz et al. (2002). All datasets including the 10 stochastic regression imputations are available on the OSF repository (osf.io/zeuv6).

Procedure
The study was conducted using the online platform Qualtrics. Participants were invited via email and informed that the study aims at understanding people's perceptions about leaders of various groups and organizations in South Africa. After providing consent, 11 participants were randomly assigned to one of two different social group contexts: partisan (political party: n = 121, trade union: n = 61) and workgroup 1 (n = 111) contexts.
Participants, randomly assigned to the partisan context, were either presented with a list of political parties currently occupying most seats in the South African parliament (i.e., African National Congress, Democratic Alliance, Economic Freedom Fighters, Inkatha Freedom Party, and National Freedom Party) or with a list of South African trade unions (i.e.,

Association of Mineworkers and Construction Union, the National Education Health and
Allied Workers Union, the South African Democratic Teachers Union, the National Union of Metalworkers South Africa and the South African Transport and Allied Workers Union).
Each participant chose one political party/trade union she or he identified with. Subsequently, participants were presented with a picture of the leader matching the political party/trade union they chose and were asked to briefly think about this leader before they completed the measures assessing identity leadership, ingroup identification, and collective action. Due to the low number of participants in the trade union condition, we decided to merge the political party and trade union conditions as both present patisan contexts. More precisely, we propose that relational orientation (Brown et al., 1992) dominantes in both contexts as they derive meaning from competing with or standing up against relevant outgroups or powerful institutions. Moreover, participants in these two conditions did not differ in their responses with regard to identity leadership, ingroup identification, and collective action intentions (see summary statistics in Supplementary Material, Table S1).
Participants, randomly assigned to the workgroup context, were told to think about their current workgroup and workgroup leader (n = 68). In situations where participants were not employed (n = 41), they were asked to think about a past or imagined workgroup and its leader. Participants then proceeded to complete the aforementioned measures.

Measurements
Identity leadership was assessed through the identity leadership inventory  which measures four dimensions: identity prototypicality (e.g., "This leader embodies what the group stands for"), identity advancement (e.g., "This leader promotes the interests of members of the group"), identity entrepreneurship (e.g., This leader creates a sense of cohesion within the group"), and identity impresarioship (e.g., "This leader creates structures that are useful for the group members").
Ingroup identification was assessed through ten items that were selected from the ingroup identification scale by Leach et al. (2008): "I feel a bond with my group", "I feel committed to my group", "I think that my group has a lot to be proud of", "It is pleasant to be a member of my group", "The fact that I am a member of this group is an important part of my identity", "Being a member of this group is an important part of how I see myself", "I have a lot in common with the average member of my group", "I am similar to the average person in my group", "Members of my group have a lot in common with each other", and "Members of my group are very similar to each other".
Collective action was assessed by participants' attitudes towards (i.e., support items) and intentions to engage in actions (i.e., participation items) for the interest of the group when instructed by the group's leader (adapted from van Zomeren et al., 2010). More precisely, the following instruction was provided. If the leader of the group asks you to take part in various actions for the sake of the group, to what extent would you agree or disagree with the following statements: " I would support future demonstrations of fellow group members", " I would support raising a collective voice as a group", "I would support doing something with fellow group members", "I would support those who participate in some form of action for the group", "I would participate in a future demonstration with fellow group members", "I 13 would participate in raising our collective voice as a group", "I would do something together with fellow group members", and "I would participate in some form of action for the group".
All measures were rated on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). The items of the identity leadership inventory were presented in a fixed order as proposed by ; while the items within the ingroup identification and collective action measures were presented in randomized order to each participant. The order of the measures was the same in both group contexts.

Preliminary Analysis
We conducted a confirmatory factor analysis to assess discriminant validity for the four-dimensional identity leadership inventory and the ingroup identification measure. This analysis was necessary to ascertain that these measures represented distinct constructs.
Accordingly, three competing models were tested. The first model was a one-factor model (Model 1: identity leadership items and ingroup identification items loading on a single factor), the second model was a two-correlated factors model (Model 2: identity leadership items and ingroup identification items loading on two separate factors), and the third model was a five-correlated factors model (Model 3: identity leadership items loading on four separate factors representing the theorized dimensions, and ingroup identification items loading on a fifth factor).
The Chi-Square test was statistically significant for all three models, implying that none of the models fitted the data well. Nevertheless, the fit indices and the chi-square difference tests indicated that Model 2 (i.e., two-correlated factors), χ 2 (274) = 1134.17, p < .001; CFI = .88; NFI = .85; RMSEA = .10, RMSEA CIs [.10, .11], showed a better fit to the data than Model 1 (i.e., single factor), χ 2 (275)  Again, however, the results of the chi-square difference tests should be interpreted carefully, given that they are based on models that might be misspecified. Although the results suggested that the five-correlated factors model (i.e., Model 3) represents the comparably best fit to the data, the intercorrelations between the four dimensions of identity leadership indicated some potential multicollinearity (rs > .81). According to , p. 1009, strong intercorrelations between the leadership dimensions suggest that participants perceive prototypicality, advancement, entrepreneurship and impresarioship as overlapping in their leaders. As the aim of the present study was to examine the overall role of identity leadership for collective action, we decided to base our further analyses on one underlying latent factor that captured the shared variance of the items of all four dimensions.
Insert Table 1 and Table 2 around here Insert Figure 1 around here

Hypotheses testing
We tested whether the indirect effect of identity leadership on collective action through ingroup identification (Hypothesis 1) is conditional upon the group context in which the leader-follower relationships are embedded (Hypothesis 2) through multiple group comparison in structural equation analyses using AMOS 27. The model included the direct paths from identity leadership and ingroup identification on collective action, and the indirect path from identity leadership on collective action through ingroup identification ( Figure 1).
One measurement weight of each latent construct was set to be equal to one in order to define the scale of the construct. For all parameter estimates we applied bootstrapping with 2000 iterations calculating 95% percentile confidence intervals.
In a first step, we examined the context-dependency of the relationships between the theoretical constructs by comparing the data of the partisan and the workgroup contexts through model specification (see Table 2). We first compared the totally unconstrained model that allowed between-group differences in all estimated parameters (Model 1), with the limited intercept model that assumed measurement weights for all three variables and item intercepts for identity leadership and collective action to be equal between the two group contexts (Model 2). The results of the model comparisons as depicted in Table 2 (see under well the data as the less parsimonious Model 1. Restricting intercepts of the ingroup identification items to be equal between the two contexts would have significantly worsened the model fit. The assumption that the intercepts of the ingroup identification items differ between the two group contexts is reasonable, given that item difficulties in the two contexts might be influenced by different social norms (e.g., in terms of loyalty expressions). Thus, the subsequent models kept the assumptions of Model 2 but accounted for the partial measurement non-equivalency of ingroup identification by allowing the item intercepts to differ. Secondly, we compared a limited structural weight model, which assumes that only the effect of identity leadership on ingroup identification is group context invariant (Model 3), with the totally unconstrained model (Model 1) and the limited intercept model (Model 2).
The most parsimonious Model 3 fitted equally well the data as Model 1 and Model 2. Thus, the path from identity leadership on ingroup identification was assumed to be equivalent between the two contexts in all subsequent models. Thirdly, we compared three additional structural weight models assuming either the path from identity leadership on collective action (Model 4) or the path from ingroup identification on collective action (Model 5) or both paths (Model 6) as group context invariant. Each of these models fitted the data worse than Model 1, Model 2, and Model 3; and Model 6 turned out to be the worst fitting model in comparison to almost all other models (see under Model Comparisons Δ Chi-Square in Table   2). It is, however, important to note that even though the model-fit of Model 3 was satisfying according to conventional criteria, the significant Chi-Square test indicated again substantial misfit of the model to the data (Hayduck, 2014). Comparisons of models that fail their Chi-Square test of fit need to be treated with caution (Yuan & Bentler, 2004). Most of the misspecification in our models was due to the fact that they did not account for pairwise covariation of item-specific residuals, particularly of some items of the identity leadership 17 inventory and the ingroup identification scale. Such specific covariations can exist for various reasons such as similarity in wording or references to similar sub-dimensions of the constructs. In the Supplementary Material (Tables S2 to S10), we provide a documentation of the sources of misfit of the final models of both Study 1 and Study 2. Table 3 reports the direct and indirect paths between identity leadership, ingroup identification and collective action for the limited structural weight Model 3. In the partisan context identity leadership explained 25.6% of the variance of ingroup identification and both identity leadership and ingroup identification explained together between 29.2% and 30.2% (varying across the 10 imputations) of the variance of collective action intentions. The explained variance in the workgroup context ranged between 38.5% and 38.9% for ingroup identification and between 15.1% and 16% for collective action. The path between identity leadership and ingroup identification was statistically significant in both the partisan and workgroup contexts. However, the path between ingroup identification and collective action was only statistically significant in the partisan group context, but not in the workgroup context, whereas the direct path between identity leadership and collective action was only statistically significant in the workgroup context but not in the partisan group context. Moreover, the indirect path between identity leadership and collective action through ingroup identification as predicted in Hypothesis 1 only reached statistical significance in the partisan group context, but not in the workgroup context. Supporting Hypothesis 2, the difference in the indirect effects was statistically significant, Estimate = 0.244, SE = 0.117, 95% CI [0.014; Adding four independent factors to the model, each one capturing the specific variance of one of the four leadership dimensions, did not change the predicted results for the general identity leadership factor. Coefficients of this model, χ 2 (1000) = 1874.31, p < .001, are reported in Table S11 in the Supplementary Material.
In sum, Study 1 presented first empirical evidence that leaders' influence on followers' tendencies to act collectively through social identity processes does indeed depend on the group context in which the leader-followers relationship is embedded (Reicher et al., 2018, see also van Dick et al., 2018). More specifically, we found evidence for Hypothesis 1 only in the partisan context, that is, only in this context was social identity the key intervening variable for followers to act collectively when instructed by their leader they perceive as exercising identity leadership. The same was not the case in the workgroup context, in which identity leadership was correlated with ingroup identification as well, but was directly related to collective action, independent of its relation to ingroup identification.
One possible reason for such a direct path of identity leadership could be that leadership has an impact on group efficacy perceptions. Self-categories, social identity and leadership are not just about "being" but also about "becoming" (Spears, Jetton, & Doosje, 2001;Reicher, Spears, & Haslam, 2010). Unsurprisingly, research on collective action predicted and showed that social identity co-varies with the belief of group members to be capable of transforming a situation through coordinated actions, which makes group efficacy another key predictor of collective action (van Zomeren, Postmes, & Spears, 2008, p. 507).
We acknowledge the unresolved debate of the directional relationship between group identification and group efficacy (see van Zomeren et al., 2010;Thomas, Mavor, & McGarty, 2012). For instance, previous research reasoned that highly identified individuals put more emphasis on collective than individual efficacy (Mummendey, Kessler, Klink, & Mielke, 1999, p. 238), that social identity inspires people's self-definitions as empowered individuals 19 (e.g., through expressing their confidence in the followers) through strengthening followers' collective sense of "us" (Fransen et al., 2015). Others assumed the reversed effect and showed experimentally that group efficacy beliefs influence ingroup identification through collective action tendencies (van Zomeren et al., 2010). Without denying the complexity of the relationship between group efficacy and social identity, we propose for the current research that beliefs in a group's capability to achieve change through coordinated action are key for the mobilisation of social identity and therefore should positively influence group members' ingroup identification. Our proposal is informed by social identity theory's reasoning that general efficacy beliefs as part of a social change belief system render intergroup relations insecure (Tajfel & Turner, 1979), which has been shown to positively affect ingroup identification (Bettencourt, Charlton, Dorr, & Hume, 2001). More specifically, we argue that leaders do not only shape the shared social identity by defining "who we are" and "what we stand for" but also "what can be done" by portraying the current social context as changeable and the group as capabale to execute these changes. Consequently, we extended Hypothesis 1 and further assumed: H1a. The more followers perceive their leader to exercise identity leadership the more they experience group efficacy and identify with their ingroup, which in turn makes them more likely to respond with intentions to act collectively when instructed by the leader (see Figure 2). For the same reasons as for Hypothesis 1, this extended mediation is conditional upon group context (Hypothesis 2).

Study 2 Participants
In total, 338 students from the same university as in Study 1 participated in Study 2 of which 239 indicated to be currently employed (1 participant did not answer the question).
The majority of participants were female (n =298; male: n = 39; one participant did not 20 indicate his/her gender). The age of the participants ranged from 18 to 70 years with an average age of 35.5. Similar to Study 1, the database of Study 2 included 21 missing values (0.6%), one on an item of the group efficacy scale and 20 on the items of the dependent variable, collective action. Again Little's MCAR test was statistically significant, Chi-Square = 496.13, df = 382, p < .001, but when the ingroup identification items were removed from the analysis it was not significant any more, Chi-Square = 291.27, df = 272, p = .202. Thus, although missingness depended on ingroup identification, the assumption that values are missing at random might be plausible. We applied the same multiple techniques to treat the missing values as in Study 1. Like in Study 1, we report in this paper the pooled estimates of ten multiple stochastic regression imputations (available on the OSF repository: osf.io/zeuv6) as well as the result of alternative missing value treatment techniques in the Supplemental Material (Tables S14-S16). Again the results were robust across these different analyses.

Procedure
The procedure of Study 2 was similar to Study 1. After participants provided consent to participate in the study, they were randomly assigned to one of two group contexts, namely the partisan context (n = 121) and workgroup context (n = 217). Different from Study 1, the partisan context referred to Non-Governmental Organizations (NGOs) rather than to political parties and trade unions. Participants in this context were provided with a list of South African NGOs (i.e., Save South Africa, Treatment Action Campaign, International Federation of Red Cross, Black Management Forum, and Nelson Mandela Children's Fund) and asked to select the NGO they identify mostly with. After selecting the NGO, participants were provided with a picture of the NGO's leader. The instructions for employed (n = 157) and not employed participants (n = 59) allocated to the workgroup context were the same as in Study and collective action. 2

Measures
As in Study 1, all measures were rated on a seven-point Likert scale, the items assessing identity leadership were presented in a fixed order, the items within the group efficacy, ingroup identification and collective action measures were presented in random order, and the order of the measures was the same in the two contexts. Identity leadership and ingroup identification were measured as in Study 1.
Group-efficacy was assessed with three items proposed by van Zomeren et al. (2010): "I think together we are able to change the situation of the group", "I think together as group members we are able to stand up for our rights", and "I think group members are able to influence the decisions of the group".
Collective action assessment used the same instruction as in Study 1 and the four participation items. Different to Study 1, the four items assessing attitudes towards collective action (i.e., support items) were replaced by four items assessing concrete behavioural intentions (Tausch et al., 2011) in order to capture a more proximal predictor of actual behavioural tendencies: "I would participate in discussion meetings with fellow group members", "I would participate in plenary meetings with fellow group members", "I would write flyers on behalf of the group", and "I would sign a complaint with fellow group members".

22
We again first conducted a confirmatory factor analysis to assess the discriminant validity of the four-dimensional identity leadership inventory, the ingroup identification and the group-efficacy measures by comparing three competing models. The first model was a one-factor model (Model 1: identity leadership, group-efficacy and ingroup identification items loading on a single factor), the second model was a three-correlated factors model indicating the presence of potential misspecifications. Also similar to Study 1, the intercorrelations between the identity leadership dimensions were above .83. Similar to Study 1 we decided to base our further analyses on one underlying latent factor that captured the shared variance of the items of all four dimensions. However, as in Study 1, we will also report in the Supplementary Material the results for a model in which we added four independent factors, each one capturing the specific variance of one of the four leadership dimensions. Table 4 reports the means, standard deviations, and intercorrelations of identity leadership, group-efficacy, ingroup identification and collective action separately for the 23 partisan and workgroup contexts as well as the internal consistency for each construct. Like in Study 1, gender, -1.11 < ts(335) < -0.36, .269 < ps < .724, was not related with identity leadership ( Table 4 and Table 5 around here Insert Figure 2 around here

Hypotheses testing
As in Study 1, we tested the conditional effects of identity leadership on collective action through group efficacy and ingroup identification using multiple group comparison in structural equation analyses (AMOS 27). The model of Study 2 included the path from identity leadership on group-efficacy, the paths from identity leadership and group-efficacy on ingroup identification, and from identity leadership, group efficacy, and ingroup identification on collective action; as well as the indirect paths from identity leadership on collective action through group efficacy, from identity leadership on collective action through ingroup identification; and from identity leadership on collective action through group 24 efficacy and ingroup identification ( Figure 2). We set again one measurement weight of each latent construct equal to one to define the scale of the construct. For all parameter estimates we applied bootstrapping with 2000 iterations calculating 95% percentile confidence intervals.
We again examined in an initial step the context-dependency of the relationships between the theoretical constructs by comparing the data of the partisan and the workgroup contexts through model specification. We first compared the totally unconstrained model that allowed between-group differences in all estimated parameters (Model 1) with a limited intercept model that assumed measurement weights for group efficacy and collective action and item intercepts for collective action to be equal between the two group contexts (Model 2). Table 5 reports the model comparisons Δ Chi-Square. Like in Study 1, the significant Chi-Square tests indicated substantial misfit of the model to the data (see Supplementary Material, Tables S2 to S10). Thus, the results of model comparisons should be treated with caution.
Model 2 fitted equally well the data as the less parsimonious Model 1. Restricting the measurement weights for identity leadership, ingroup identification or the item intercepts of the identity leadership or ingroup identification or group efficacy items to be equal between the two group contexts would have significantly worsened the model fit. We, therefore, kept in the subsequent models the assumptions of Model 2 and allowed these latter parameters to differ between group contexts. Secondly, we compared the limited structural weight model that corresponded to the best fitting model in Study 1 assuming the effects of identity leadership on ingroup identification, both the direct and the indirect via efficacy, as well as the direct path from efficacy on collective action intentions to be group context invariant Model 2. Thirdly, we compared another limited structural weight model assuming additionally the direct path from identity leadership on collective action (Model 4) as group context invariant with the previous models. This most parsimonious Model 4, in which the effect of ingroup identification on collective action was the only structural weight assumed to differ between group contexts, fitted equally well the data as Models 1, 2 and 3. Finally, we compared Model 4 with Model 5 assuming an invariant group context for all structural paths.
Model 5 fitted the data worse than the previous models (see under Model Comparisons Δ Chi-Square in Table 5). Consequently, we report for Study 2 the direct and indirect effects for the limited structural weight Model 4.
In the partisan context identity leadership explained 17.8% of the variance of group efficacy, identity leadership and group efficacy explained between 37.0% and 37.2% of the variance of ingroup identification (varying across the 10 imputations), whereas identity leadership, group efficacy and ingroup identification explained together between 44.5% and 45.4% of the variance of collective action. The explained variance in the workgroup context was 36.4% for group efficacy, 59.9% to 60% for ingroup identification and 29.5% to 30% for collective action. The direct and indirect paths between identity leadership, group efficacy, ingroup identification, and collective action are reported in Table 3. The direct paths between identity leadership and group efficacy, between identity leadership and ingroup identification, between group efficacy and ingroup identification, and between group efficacy and collective action were statistically significant in both contexts. Consistent with Hypothesis 2, and replicating the findings of Study 1, the path between ingroup identification and collective action was only significant in the partisan group context, but not in the workgroup context.
The direct path between identity leadership and collective action was negative and weak, but reached statistical significance in both the partisan context and the workgroup context. and collective action through ingroup identification (predicted by Hypothesis 1 and bypassing efficacy) reached statistical significance only in the partisan group context, but not in the workgroup context. The differences between these two indirect paths was significant, Estimate = 0.12, SE = 0.06, 95% CI [0.000; 0.241]. Moreover, the extended path linking identity leadership with collective action through group efficacy and ingroup identification (Hypothesis 1a) was only significant in the partisan context, but not in the workgroup context. Again, supporting Hypothesis 2, the difference between these two indirect paths reached statistical significance, Estimate = 0.10, SE = 0.05, 95% CI [0.003; 0.193].
Adding the four independent factors into the model, χ 2 (1179) = 2653.26, p < .001, did not change the predicted results for the general identity leadership factor (Supplementary Material, Table S12).
As we found in Study 2, unlike to Study 1, that the participants' age and status of employment were related to identity leadership, ingroup identification and group-efficacy, we estimated the same limited structural weight Model 4 while controlling for age and employment status (see Table S13 in the Supplementary Material). The inclusion of these covariates did neither influence the findings concerning the indirect paths supporting Hypotheses 1 and 1a nor the conditionality of these indirect paths supporting Hypothesis 2.
Overall, the results of Study 2 replicated the findings of Study 1 that the given group context in which the leader-followers relationship is embedded matters for leadership exercised through social identity processes. Moreover, the results of Study 2 imply that identity leadership nurtures followers' beliefs in being capable to change a situation or the destiny of the ingroup , which also makes the ingroup matter for followers. Group efficacy and its implication for collective action was, however, group context invariant.
The overall aim of the present research was to address the question what makes followers respond to a leader's instruction to act collectively. More specifically, we were interested in the group context-dependency of the hypothesized relationships between identity leadership, group efficacy, ingroup identification, and intentions to act collectively when instructed by the leader. In two studies we found that the link between identity leadership and collective action through ingroup identification is indeed conditional upon the group context in which the leader-followers relationship is embedded. More specifically, we found that the statistical indirect effect of identity leadership on collective action through ingroup identification was evident in the partisan group but not in the workgroup context (Study 1 and 2). Moreover, the chain mediation capturing the relationship between identity leadership and collective action intentions through the sequence of group efficacy and ingroup identification was context-dependent as well. As predicted, it was only significant in the partisan group and not in the workgroup context, and the difference between these two effects was statistically significant (Study 2). Interestingly, however, is that this group context-dependency only applied to effects that involved a link between ingroup identification and collective action.
Relationships between identity leadership and ingroup identification (Studies 1 and 2), and between group efficacy and all three other variables (Study 2) did not differ between group contexts, nor did the indirect effect through efficacy that bypassed ingroup identification. In conclusion, our findings imply that the group context in which the leader-follower relationship is embedded matters for collective action evoked by ingroup identification due to identity leadership but not for collective action evoked by group efficacy due to identity leadership.

28
Our results demonstrate that identity leadership is a promising approach not only for leadership in business and work contexts but also for leadership in the contexts of politics and volunteer work (van Dick et al., 2018). More specifically, our results are consistent with the idea that identity leadership does not only influence workplace-specific behaviors such as job satisfaction and work performance but also group behavior that aims at advancing group interests and goals. Accordingly, leaders are able to evoke not only personal and organizational change but also large-scale social change through social identity processes.
Secondly, our results support the notion that the group context in which the leaderfollower relationship is embedded matters for leadership when exercised through social identity processes; although, it does so in a particular way. Whereas group efficacy and ingroup identification were related to perceived identity leadership irrespective of whether participants were allocated to partisan or workgroup contexts, ingroup identification was only directly linked to intentions to act collectively in the partisan context. It indicates that ingroup identification due to identity leadership is more important for collective action in partisan than in workgroup contexts.
Our results suggest that the shared identity plays a more important role for the mobilization of followers in groups, such as partisan groups, in which they seem to be by definition obliged to act collectively in the interest of the group. From a broader point of view, this particular context effect can be explained by a more general principle, namely by the fact that in such groups collective action has a better normative fit to the behavior that is expected from a prototypical member (Turner et al., 1987). According to this principle, our results do not imply that a leader who exercises identity leadership can mobilize followers to act collectively for or against just anything. Even in partisan groups, the collective actions that leaders ask followers to take need to be appropriate according to the norms and values of the social group. For instance, a trade union functionary might ask his or her followers to sign 29 a petition against the introduction of affirmative action in the composition of decision boards.
This request is likely to be appraised as inappropriate if members of this group share feminist and even post-feminist norms and values because they might appraise affirmative action as an effective tool to increase gender equality in the organisation. Consequently, the leader requesting his followers to sign such a petition is rather unlikely to find broader support among his followers. Our reasoning related to group normativity has at least two implications, which should be addressed by future research. First, while social identification in workgroup contexts might be less important for identity leadership effects on collective action as measured in the present research, it might play a stronger role in mobilising followers for work-related collective efforts (e.g., extra shifts of medical teams to treat Covid-19 patience) or within liberal-leaning organisations in which collective action of employees is accepted part of the organisational culture (Gupta & Briscoe, 2020). Secondly, although collective action forms part of the social identity in partisan group contexts, shared social norms and values will have a crucial function in regulating whether and which kind of collective action is appropriate and can be expected in a certain situation (e.g., Haslam, Reicher, & Birney, 2016).
Finally, our results showed that group-efficacy evoked by identity leadership, ingroup identification evoked by identity leadership and group efficacy, and collective action evoked by group efficacy due to identity leadership were group invariant. These results imply that irrespective of the group context, the perception of the leader as being one of us and as doing it for us fosters followers' beliefs that they can act as and achieve something for the group, which is linked to their ingroup identification and their readiness to engage in collective action. Moreover, our results suggest that different to ingroup identification effects, the effects of followers' beliefs about their capabilities to act as a group are not necessarily shaped by social group context and probably less dependent on whether these acts are 30 actually appropriate or inappropriate or fit a certain identity. By nurturing followers' group efficacy, leaders actually exercise identity entrepreneurship by defining and influencing what the group is able to do, which is, however, different from defining and influencing what the group stands for. Nevertheless, we would expect that leaders who are perceived as crafting a sense of us (identity entrepreneurship) might be particularly effective in influencing followers' group efficacy beliefs. We were not able to test this assumption statistically in the present research because of the strong intercorrelations between the four sub-scales of the identity leadership inventory. Future research might test it by manipulating the different dimensions of identity leadership and assessing their effectiveness in fostering or changing efficacy beliefs.
Given the pioneering character of our research, we would like to stress that our studies were to a large part exploratory rather than explanatory in nature. As much as the context-independency of all effects via group-efficacy can be explained by the theoretical ideas behind our research, we were not able to predict it a-priory. Moreover, our results do not allow inferring a causal process between identity leadership and group outcomes as we only experimentally manipulated the group context, but not group membership or perception of identity leadership itself, nor can we rule out the possibility of reversed causality. We can also not rule out the possibility of effects produced by method factors (Podsakoff et al., 2012) nor the existence of omitted variables influencing the relationships between identity leadership, ingroup identification, and collective action. To establish the directionality of the hypothesized processes further experimental research or longitudinal designs will be necessary.
One characteristic of our design was that, after being randomly allocated to one of the two larger group contexts, participants were asked to choose the specific leadership subcontext that was most relevant for them. This method had the double advantage of sampling a 31 rather large scope of contexts and at the same time to assure some ecological validity. Our design had, however, the disadvantage that it did not allow to disentangle differences between specific sub-contexts and inter-individual differences within the same specific subcontext. Moreover, our design might have created different demand characteristics, as it is something else to choose between partisan groups one identifies with than selecting a workgroup one is part of. While these are certainly limitations affecting the internal validity of our research, we are encouraged by the robustness of the data pattern across the two studies. Nevertheless, it might be worthwhile in future research to disentangle the two components of variation between participants by a highly powered replication that has a database broad enough to apply multi-level analyses. To avoid the possibility of different demand characteristics when studying multiple contexts, however, experimental research would be necessary that keeps the context more or less constant within conditions. Such research can use artificial group contexts. On the other hand, to increase external validity it can also be conducted in real-life contexts as long as participants can connect to the same specific group context (e.g., sampling from all members of the same union within the same work context). Another possibility to test the overall proposed idea of the current research would be an alternative experimental approach manipulating types of collective action (i.e., fitting versus non-fitting the specific social identity at stake) rather than manipulating group contexts.
Other limitations refer to information about our participants we did not assess in our studies. Although we asked our participants, who were mainly part-time students, about their employment status, we did not assess information about their professions, years of working experiences or the areas in which they work. This information would have provided additional insights, particularly for the workgroup context. Likewise, we did not control whether participants in the partisan contexts were actual members of the political parties or 32 trade unions they selected (Study 1) or whether they were active supporters of the selected NGOs (Study 2). Despite all these limitations, the current research revealed in two studies that the process connecting identity leadership with the mobilisation for collective action depends on the group context with regard to the mediating role of ingroup identification, but not with regard to the mediating role of group efficacy beliefs.
In conclusion, identity leadership is indeed a promising approach to extend our understanding of collective action and social change. It is promising because it conceptualises leadership as being through followers rather than done to followers (Turner et al., 2008, p. 70), it defines leadership as influencing both social stability and social change, and it provides a theoretical framework to study the influence of leaders in different group contexts.
Finally, it is promising because it does not only provide us with accounts why leaders such as Martin Luther King Jr., Mahatma Gandhi, Nelson Rolihlahla Mandela, and Mgcineni Mambush Noku (the man in the green blanket) were able to mobilize the masses to fight for freedom and social justice, but also why leaders such as Omar Al-Bashir or Robert Mugabe were able to successfully mobilize followers to support murder, torture, and persecution of political opponents; and to plunder their own countries. Identity leadership can mobilize collective action effectively, for the better or for the worse, depending on one's political position. Understanding how it works is therefore of utmost importance for everyone who cares about social change or stability. The present research aimed at addressing this question theoretically from the perspective of the model of identity leadership (Haslam, Reacher, & Platow, 2011, Reacher, Haslam, & Platow, 2018, which proposes that leaders influence followers through social identity processes. As social identity, like efficacy beliefs, is key for collective action, we consider the identity leadership approach as most appropriate to address the how question.
Besides examining the social identity processes involved in leaders' influence on followers to act collectively, we argue and provide empirical evidence that leaders' influence through social identity depends on the group context in which the relationship between leader and followers is embedded.
Social identity refers to the psychological process of cognitively grouping oneself as identical (i.e., similar, interchangeable) with ingroup members in contrast to outgroup members (i.e., self-categorisation; Tajfel & Turner, 1979;Turner, Hogg, Oakes, Reicher, & Wetherell, 1987) and to one's knowledge of belonging to a certain psychological group that has "some emotional and value significance" (i.e., social identity; Tajfel, 1974, p. 72; see also Turner, Reynolds, & Subasic, 2008). As people define and identify themselves as group members, the norms, values, and beliefs that define this group are internalized through the process of depersonalisation (Turner, et al., 1987) and influence their attitudes and behavior (Turner et al., 2008;Reinholds, Jones, O'Brien, & Subasic, 2013). Thus, social identity makes group behavior possible; that is, people care about and act collectively to advance group interests and goals because of their shared social identity rather than their differing personal identities (Turner et al., 2008).
Shared social identity also makes mutual influence possible; that is because ingroup members are perceived as similar to oneself they become a source for validating information (Turner et al., 2008). However, some ingroup members are more influential than others in representing and shaping norms, values, and beliefs of the ingroup; in changing attitudes and in mobilizing group members.

Identity leadership
Group members follow influential ingroup members because they embody the group; they define who the group is and should be. Social influence as social identity process, which has been conceptualised by social identity theory (Tajfel & Turner, 1979) and selfcategorisation theory (Turner et al., 1987), builds the point of departure of the model of identity leadership (e.g., Haslam & Platow, 2001;Haslam et al., 2001;Reicher, Haslam, & Hopkins, 2005;Haslam et al., 2011;van Dick et al., 2018, Reicher et al., 2018. This model posits that both leadership and followership are only possible because of a shared sense of "us" and defines leadership as "a recursive, multidimensional process that centers on leaders' capacities to represent, advance, create, and embed a shared sense of social identity for group members" , p. 1002. Thus, different from other leadership models that conceptualise shared social identity (e.g., organisational identification) as an outcome of perceived similarity or fit between leaders' and followers' interests and concerns (e.g., between CEOs ideology and organisational ideology, see Hambrick & Wowak, 2021; or between corporate socio-political activism and key stakeholders' values and brand image, see Bhagwat et al., 2020), the identity leadership model views shared social identity as the starting point, which determines a shared understanding of organisational and social reality; including specific configurations of similarities and differences of leaders' and followers' interests and concerns.
Moreover, leadership does not only emerge from a shared social identity but is an essential part of social identity processes (Haslam et al., 2011;. Findings from previous research suggest that leaders shape social identities because they are one of us (Identity prototypicality), they are doing it for us (Identity advancement), they craft a sense of us (Identity entrepreneurship), and they make us matter (Identity impresarioship) (Haslam et al., 2011;. For instance, the more prototypical an ingroup member is, that is, the more she or he differs from the outgroup and the less from the ingroup, the more she or he is trusted and followed as a leader (van Dick et al., 2018), the more she or he is perceived as an effective leader (Hains, Hogg, & Duck, 1997;Hogg, Hains, & Mason, 1998;van Knippenberg & van Knippenberg, 2005;Giessner & van Knippenberg, 2008;van Knippenberg, 2011), the more she or he is endorsed as leader even if she or he fails to deliver (Giessner, van Knippenberg, & Sleebos, 2009, Ullrich, Christ, & van Dick, 2009, the more followers cooperate (De Cremer, van Dijke, & Mayer, 2010), and the more followers identify and report a personal bond with the leader . Moreover, leaders are influential on followers because they advance group identity through championing group interests and goals, they are entrepreneurs of group identity through defining and influencing core values, norms, and ideals of what the group stands for, they determine who belongs and who does not belong to the group, and they initiate structures and patterns as group reality experienced by both ingroup and outgroup members .

Mobilising for collective action: Identification as mediator
Although the theoretical conceptualization of the model of identity leadership is elaborate, it has not yet been applied to study systematically how leadership takes effect 5 through social identity processes in mobilizing followers to act collectively. Most of the research addressing the impact of identity leadership (with particular reference to the Identity Leadership Inventory; see van Dick et al., 2018) focused on group members' performance (Steffens, Haslam, Kerschreiter, Schuh, & van Dick, 2014;Fransen et al., 2016;Steffens, Yang, Jetten, Haslam, & Lipponen, 2018;van Dick et al., 2018), job satisfaction and well-being To answer the question why members would sign a petition or participate in a strike action because a leader asks them to do so for the interest of their group from the identity leadership perspective, we should first recall the main premises on which this approach is built. First, a leader is always the leader of a particular constituency which can be a political party, a workgroup, a religious group, a state and so on (Reicher et al., 2018). Consequently, leadership refers to the relationship between the leader and followers within a particular social context. Or to put it differently, both leadership and followership are more effective for leaders and followers who do not only share the same group membership but also a social identity. Secondly, the model of identity leadership proposes that leaders who are one of, and act for, the group represent and shape the shared social identity (e.g., by defining "who we are" and "what we stand for") and, thus, influence how followers perceive social reality and whether they get mobilized "to transform the social world so that it comes into line with their norms and values" (Haslam et al., 2011, p. 70). Thus, different from other leadership approaches that view followers as rather indifferent by default and in need to be commanded and controlled, the identity leadership model conceptualises leadership as a group process in which both leaders and followers are viewed as engaged, energised and agentic (Haslam, Steffens, Reicher, & Bentley, 2021). Leadership is therefore about both being and becoming whereby both processes determine each other. To put it differently, as social identity processes affect leadership so does leadership affect social identity processes (Haslam et al., 2011;Reicher et al., 2018).
The involvement of the shared social identities in the leadership process has at least two consequences. First, the shared identity of the group plays a key role in the mobilization process itself. As leadership promotes ingroup identification and ingroup identification has been found to foster collective action, we hypothesise that:

Group context as moderator
Secondly, a leader who exercises identity leadership and calls followers to sign a petition or to participate in a strike action might rely in some groups more on shared social identity in mobilising followers than in other groups; that is, the mediation of identity leadership effects through group identification can be assumed to vary across different social 7 contexts. For instance, there are social contexts that according to Simon et al. (1998), De Weerd andKlandermans (1999), Stürmer, Simon, Leowy and Jörger (2003), and van Stekelenburg and Klandermans (2013) produce politicised social identities, which by definition oblige members to advance group interests and goals through social actions aiming for social change. These politicised social identities correspond with distinct ideologies and orientations that prevail within and predominate the context of partisan groups and that are different from other contexts such as workgroups (Brown et al., 1992). As partisan groups, such as political parties, trade unions or NGOs, only have meaning because they either compete with relevant outgroups such as other political parties or they are set up against powerful adversaries such as employers or governments, they are most likely to define their ingroup and its outcomes with reference to other relevant groups and challenges; that is to say, a relational orientation is most likely to dominate (Brown et al., 1992). In contrast, workgroups can be conceived as task-based that is not necessarily related either objectively or psychologically to other groups. Therefore, workgroups might be better described in their orientation as autonomous as they tend to evaluate their ingroup and its outcomes with reference to some abstract standards such as efficiency and professionalism that typically do not implicate other groups (Brown et al., 1992). Thus, different group orientations and consequently different social identities might determine that the same degree of ingroup identification evokes different group outcomes.
Essentially, we argue that collective action to advance group interests and goals is in some social groups, such as partisan groups, part and parcel of the shared social identity. Or, to put it differently, to act collectively for the interest of the group is more normative in some social groups than in others. It is, however, important to note that we are not implying that only members of politicised groups engage in collective actions aiming at social change. As countless examples teach us, most politicised groups such as trade unions resulted from collective actions of non-politicised group members such as factory workers. We are, however, implying that the importance of social identity processes for the effectiveness of identity leadership to mobilise followers might vary depending upon the social context in which the relationship between leader and followers is embedded.
Consequently, although we assume that social identity influences collective action as one of its key predictors (e.g., social identity theory, Tajfel & Turner, 1979: selfcategorisation theory, Turner et al., 1987; and social identity model of collective actions, van , we propose that the social group context of the leader-followers relationship determines the role that shared social identity plays in the relationship between the perception of identity leadership and the compliance to the request by the leader to act collectively. More specifically, we propose that the relationship between perceived identity leadership and followers' compliance to the leader's request to act collectively should involve the mobilization of identification with the social group in some social contexts more than in others. We expect identification with the group to be more important in social groups for which collective action is normative (e.g., partisan groups) than in social groups for which collective action is less normative (e.g., workgroups). Therefore, we hypothesise: H2. The indirect effect of identity leadership on collective action through ingroup identification is moderated by group context. More specifically, the link between ingroup identification and collective action intentions is stronger in the partisan group context than in the workgroup context (see Figure 1).
We tested these two hypotheses in two studies using a quasi-experimental survey design. We acknowledge the limitations of this approach as it captures the individual subjective perceptions of identity experiences rather than the shared and lived experiences themselves (Fukuyama, 2018). To capture the latter other methods involving direct observations would be necessary as they are often applied in the study of dynamics in large-9 scale movements (e.g., participant observation, Drury & Reicher, 2000; ethnographic designs, Drury & Reicher, 2005;and/or interviews, Stevenson, Reicher, Pandey, Shankar, Tewari, & Hopkins, 2020). Studying such dynamics, however, would go beyond the scope of the current research as we aimed to examine how identity leadership experienced by an individual can motivate her or him to engage in collective action and how the role that social identification plays in this process varies across different group contexts. Such individual-level research, which is common in identity leadership studies in the context of workgroups (e.g., Steffens, Haslam, Kerschreiter et al., 2014) or in the context of sport and exercise (e.g., Fransen, Steffens & Haslam, 2015;Stevens et al., 2020), is a necessary complement to the indispensable observational approaches that are able to capture directly the group-level phenomena and historically contextualised intergroup dynamics.

Study 1 Participants
In total, 293 students from the University of South Africa took part in the study. The University of South Africa is a distance learning university. Thus, most students work fulltime and study part-time. The sample consisted of 231 females and 56 males (six participants did not indicate their gender). The age of the participants ranged from 18 to 65 years with a mean age of 29.6. Of all participants, 129 indicated to be employed and 99, not to be employed (65 did not answer the question).
As our database of Study 1 included 23 missing values on the items of the dependent variable, collective action (0.8%), we analysed whether they were missing completely at random (MCAR). Little's MCAR test was significant, Chi-Square = 369.16, df = 315, p = .019, indicating that these missing values were not missing completely at random. However, removing one item of the ingroup identification scale from the analysis rendered the MCAR test insignificant, Chi-Square = 344.246, df = 305, p. = .060, that means that the nonrandomness was mainly due to the dependency on certain aspects of ingroup identification.
At the same time, these results suggest that it might be reasonable to consider that the less demanding (though not directly testable) assumption that missing values were missing at random was met. Using AMOS 28, we, nevertheless, applied different techniques of dealing with the missing values such as listwise deletion, maximum likelihood estimation, regression imputation (which for each case, regresses the unobserved values on the observed values, assuming that the population means and covariances of all variables are equal to their maximum likelihood estimates) and multiple stochastic regression imputation (which draws at random "from the conditional distribution of the missing values given the observed values, with the unknown model parameters" again "set equal to their maximum likelihood estimates", Arbuckle, 2021, p. 477). The latter technique generated ten data output files with imputed missing values. Apart from negligible differences in the estimations, the overall results were the same in all these analyses (apart from the fact that AMOS does not allow bootstrapping when missings are dealt with by maximum likelihood estimation, see Supplemental Material, Tables S14-16). For analyses including missing values, we therefore report in this paper only the pooled results of the multiple stochastic regression imputation, applying Rubin's rules (Rubin, 1976) and the Degree-of-Freedom approximation suggested by Lipsitz et al. (2002). All datasets including the 10 stochastic regression imputations are available on the OSF repository (osf.io/zeuv6).

Procedure
The study was conducted using the online platform Qualtrics. Participants were invited via email and informed that the study aims at understanding people's perceptions about leaders of various groups and organizations in South Africa. After providing consent, 11 participants were randomly assigned to one of two different social group contexts: partisan (political party: n = 121, trade union: n = 61) and workgroup 1 (n = 111) contexts.
Participants, randomly assigned to the partisan context, were either presented with a list of political parties currently occupying most seats in the South African parliament (i.e., Each participant chose one political party/trade union she or he identified with. Subsequently, participants were presented with a picture of the leader matching the political party/trade union they chose and were asked to briefly think about this leader before they completed the measures assessing identity leadership, ingroup identification, and collective action. Due to the low number of participants in the trade union condition, we decided to merge the political party and trade union conditions as both present patisan contexts. More precisely, we propose that relational orientation (Brown et al., 1992) dominantes in both contexts as they derive meaning from competing with or standing up against relevant outgroups or powerful institutions. Moreover, participants in these two conditions did not differ in their responses with regard to identity leadership, ingroup identification, and collective action intentions (see summary statistics in Supplementary Material, Table S1).

African
Participants, randomly assigned to the workgroup context, were told to think about their current workgroup and workgroup leader (n = 68). In situations where participants were not employed (n = 41), they were asked to think about a past or imagined workgroup and its leader. Participants then proceeded to complete the aforementioned measures.
Identity leadership was assessed through the identity leadership inventory  which measures four dimensions: identity prototypicality (e.g., "This leader embodies what the group stands for"), identity advancement (e.g., "This leader promotes the interests of members of the group"), identity entrepreneurship (e.g., This leader creates a sense of cohesion within the group"), and identity impresarioship (e.g., "This leader creates structures that are useful for the group members").
Ingroup identification was assessed through ten items that were selected from the ingroup identification scale by Leach et al. (2008): "I feel a bond with my group", "I feel committed to my group", "I think that my group has a lot to be proud of", "It is pleasant to be a member of my group", "The fact that I am a member of this group is an important part of Collective action was assessed by participants' attitudes towards (i.e., support items) and intentions to engage in actions (i.e., participation items) for the interest of the group when instructed by the group's leader (adapted from van Zomeren et al., 2010). More precisely, the following instruction was provided. If the leader of the group asks you to take part in various actions for the sake of the group, to what extent would you agree or disagree with the following statements: " I would support future demonstrations of fellow group members", " I would support raising a collective voice as a group", "I would support doing something with fellow group members", "I would support those who participate in some form of action for the group", "I would participate in a future demonstration with fellow group members", "I 13 would participate in raising our collective voice as a group", "I would do something together with fellow group members", and "I would participate in some form of action for the group".
All measures were rated on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). The items of the identity leadership inventory were presented in a fixed order as proposed by ; while the items within the ingroup identification and collective action measures were presented in randomized order to each participant. The order of the measures was the same in both group contexts.

Preliminary Analysis
We conducted a confirmatory factor analysis to assess discriminant validity for the four-dimensional identity leadership inventory and the ingroup identification measure. This analysis was necessary to ascertain that these measures represented distinct constructs.
Accordingly, three competing models were tested. The first model was a one-factor model Again, however, the results of the chi-square difference tests should be interpreted carefully, given that they are based on models that might be misspecified. Although the results suggested that the five-correlated factors model (i.e., Model 3) represents the comparably best fit to the data, the intercorrelations between the four dimensions of identity leadership indicated some potential multicollinearity (rs > .81). According to , p. 1009, strong intercorrelations between the leadership dimensions suggest that participants perceive prototypicality, advancement, entrepreneurship and impresarioship as overlapping in their leaders. As the aim of the present study was to examine the overall role of identity leadership for collective action, we decided to base our further analyses on one underlying latent factor that captured the shared variance of the items of all four dimensions.
However, for the sake of completeness, we will also report in the Supplementary Material the results for a model in which we added four independent factors, each one capturing the specific variance of one of the four leadership dimensions. Table 1 reports the means, standard deviations, and intercorrelations of identity leadership, ingroup identification and collective action separately for the partisan and workgroup contexts as well as the internal consistency for each construct. We further tested Mnot employed = 4.82, SDnot employed = 1.17), t(285) = -1.08, p = .282.
Insert Table 1 and Table 2 around here Insert Figure 1 around here

Hypotheses testing
We tested whether the indirect effect of identity leadership on collective action through ingroup identification (Hypothesis 1) is conditional upon the group context in which the leader-follower relationships are embedded (Hypothesis 2) through multiple group comparison in structural equation analyses using AMOS 27. The model included the direct paths from identity leadership and ingroup identification on collective action, and the indirect path from identity leadership on collective action through ingroup identification (Figure 1).
One measurement weight of each latent construct was set to be equal to one in order to define the scale of the construct. For all parameter estimates we applied bootstrapping with 2000 iterations calculating 95% percentile confidence intervals.
In a first step, we examined the context-dependency of the relationships between the theoretical constructs by comparing the data of the partisan and the workgroup contexts through model specification (see Table 2). We first compared the totally unconstrained model that allowed between-group differences in all estimated parameters (Model 1), with the limited intercept model that assumed measurement weights for all three variables and item intercepts for identity leadership and collective action to be equal between the two group contexts (Model 2). The results of the model comparisons as depicted in Table 2 (see under well the data as the less parsimonious Model 1. Restricting intercepts of the ingroup identification items to be equal between the two contexts would have significantly worsened the model fit. The assumption that the intercepts of the ingroup identification items differ between the two group contexts is reasonable, given that item difficulties in the two contexts might be influenced by different social norms (e.g., in terms of loyalty expressions). Thus, the subsequent models kept the assumptions of Model 2 but accounted for the partial measurement non-equivalency of ingroup identification by allowing the item intercepts to differ. Secondly, we compared a limited structural weight model, which assumes that only the effect of identity leadership on ingroup identification is group context invariant (Model 3), with the totally unconstrained model (Model 1) and the limited intercept model (Model 2).
The most parsimonious Model 3 fitted equally well the data as Model 1 and Model 2. Thus, the path from identity leadership on ingroup identification was assumed to be equivalent between the two contexts in all subsequent models. Thirdly, we compared three additional structural weight models assuming either the path from identity leadership on collective action (Model 4) or the path from ingroup identification on collective action (Model 5) or both paths (Model 6) as group context invariant. Each of these models fitted the data worse than Model 1, Model 2, and Model 3; and Model 6 turned out to be the worst fitting model in comparison to almost all other models (see under Model Comparisons Δ Chi-Square in Table   2). It is, however, important to note that even though the model-fit of Model 3 was satisfying according to conventional criteria, the significant Chi-Square test indicated again substantial misfit of the model to the data (Hayduck, 2014). Comparisons of models that fail their Chi-Square test of fit need to be treated with caution (Yuan & Bentler, 2004). Most of the misspecification in our models was due to the fact that they did not account for pairwise covariation of item-specific residuals, particularly of some items of the identity leadership 17 inventory and the ingroup identification scale. Such specific covariations can exist for various reasons such as similarity in wording or references to similar sub-dimensions of the constructs. In the Supplementary Material (Tables S2 to S10), we provide a documentation of the sources of misfit of the final models of both Study 1 and Study 2. Table 3 reports the direct and indirect paths between identity leadership, ingroup identification and collective action for the limited structural weight Model 3. In the partisan context identity leadership explained 25.6% of the variance of ingroup identification and both identity leadership and ingroup identification explained together between 29.2% and 30.2% (varying across the 10 imputations) of the variance of collective action intentions. The explained variance in the workgroup context ranged between 38.5% and 38.9% for ingroup identification and between 15.1% and 16% for collective action. The path between identity leadership and ingroup identification was statistically significant in both the partisan and workgroup contexts. However, the path between ingroup identification and collective action was only statistically significant in the partisan group context, but not in the workgroup context, whereas the direct path between identity leadership and collective action was only statistically significant in the workgroup context but not in the partisan group context. Moreover, the indirect path between identity leadership and collective action through ingroup identification as predicted in Hypothesis 1 only reached statistical significance in the partisan group context, but not in the workgroup context. Supporting Hypothesis 2, the difference in the indirect effects was statistically significant, Estimate = 0.244, SE = 0.117, 95% CI [0.014; Adding four independent factors to the model, each one capturing the specific variance of one of the four leadership dimensions, did not change the predicted results for the general identity leadership factor. Coefficients of this model, χ 2 (1000) = 1874.31, p < .001, are reported in Table S11 in the Supplementary Material.
In sum, Study 1 presented first empirical evidence that leaders' influence on followers' tendencies to act collectively through social identity processes does indeed depend on the group context in which the leader-followers relationship is embedded (Reicher et al., 2018, see also van Dick et al., 2018). More specifically, we found evidence for Hypothesis 1 only in the partisan context, that is, only in this context was social identity the key intervening variable for followers to act collectively when instructed by their leader they perceive as exercising identity leadership. The same was not the case in the workgroup context, in which identity leadership was correlated with ingroup identification as well, but was directly related to collective action, independent of its relation to ingroup identification.
One possible reason for such a direct path of identity leadership could be that leadership has an impact on group efficacy perceptions. Self-categories, social identity and leadership are not just about "being" but also about "becoming" (Spears, Jetton, & Doosje, 2001;Reicher, Spears, & Haslam, 2010). Unsurprisingly, research on collective action predicted and showed that social identity co-varies with the belief of group members to be capable of transforming a situation through coordinated actions, which makes group efficacy another key predictor of collective action (van Zomeren, Postmes, & Spears, 2008, p. 507).
We acknowledge the unresolved debate of the directional relationship between group identification and group efficacy (see van Zomeren et al., 2010;Thomas, Mavor, & McGarty, 2012). For instance, previous research reasoned that highly identified individuals put more emphasis on collective than individual efficacy (Mummendey, Kessler, Klink, & Mielke, 1999, p. 238), that social identity inspires people's self-definitions as empowered individuals 19 (e.g., through expressing their confidence in the followers) through strengthening followers' collective sense of "us" (Fransen et al., 2015). Others assumed the reversed effect and showed experimentally that group efficacy beliefs influence ingroup identification through collective action tendencies (van Zomeren et al., 2010). Without denying the complexity of the relationship between group efficacy and social identity, we propose for the current research that beliefs in a group's capability to achieve change through coordinated action are key for the mobilisation of social identity and therefore should positively influence group members' ingroup identification. Our proposal is informed by social identity theory's reasoning that general efficacy beliefs as part of a social change belief system render intergroup relations insecure (Tajfel & Turner, 1979), which has been shown to positively affect ingroup identification (Bettencourt, Charlton, Dorr, & Hume, 2001). More specifically, we argue that leaders do not only shape the shared social identity by defining "who we are" and "what we stand for" but also "what can be done" by portraying the current social context as changeable and the group as capabale to execute these changes. Consequently, we extended Hypothesis 1 and further assumed: H1a. The more followers perceive their leader to exercise identity leadership the more they experience group efficacy and identify with their ingroup, which in turn makes them more likely to respond with intentions to act collectively when instructed by the leader (see Figure 2). For the same reasons as for Hypothesis 1, this extended mediation is conditional upon group context (Hypothesis 2).

Study 2 Participants
In total, 338 students from the same university as in Study 1 participated in Study 2 of which 239 indicated to be currently employed (1 participant did not answer the question).
The majority of participants were female (n =298; male: n = 39; one participant did not average age of 35.5. Similar to Study 1, the database of Study 2 included 21 missing values (0.6%), one on an item of the group efficacy scale and 20 on the items of the dependent variable, collective action. Again Little's MCAR test was statistically significant, Chi-Square = 496.13, df = 382, p < .001, but when the ingroup identification items were removed from the analysis it was not significant any more, Chi-Square = 291.27, df = 272, p = .202. Thus, although missingness depended on ingroup identification, the assumption that values are missing at random might be plausible. We applied the same multiple techniques to treat the missing values as in Study 1. Like in Study 1, we report in this paper the pooled estimates of ten multiple stochastic regression imputations (available on the OSF repository: osf.io/zeuv6) as well as the result of alternative missing value treatment techniques in the Supplemental Material (Tables S14-S16). Again the results were robust across these different analyses.

Procedure
The procedure of Study 2 was similar to Study 1. After participants provided consent to participate in the study, they were randomly assigned to one of two group contexts, namely the partisan context (n = 121) and workgroup context (n = 217). Different from Study 1, the partisan context referred to Non-Governmental Organizations (NGOs) rather than to political parties and trade unions. Participants in this context were provided with a list of South African NGOs (i.e., Save South Africa, Treatment Action Campaign, International Federation of Red Cross, Black Management Forum, and Nelson Mandela Children's Fund) and asked to select the NGO they identify mostly with. After selecting the NGO, participants were provided with a picture of the NGO's leader. The instructions for employed (n = 157) and not employed participants (n = 59) allocated to the workgroup context were the same as in Study and collective action. 2

Measures
As in Study 1, all measures were rated on a seven-point Likert scale, the items assessing identity leadership were presented in a fixed order, the items within the group efficacy, ingroup identification and collective action measures were presented in random order, and the order of the measures was the same in the two contexts. Identity leadership and ingroup identification were measured as in Study 1.
Group-efficacy was assessed with three items proposed by van Zomeren et al. (2010): "I think together we are able to change the situation of the group", "I think together as group members we are able to stand up for our rights", and "I think group members are able to influence the decisions of the group".
Collective action assessment used the same instruction as in Study 1 and the four participation items. Different to Study 1, the four items assessing attitudes towards collective action (i.e., support items) were replaced by four items assessing concrete behavioural intentions (Tausch et al., 2011) in order to capture a more proximal predictor of actual behavioural tendencies: "I would participate in discussion meetings with fellow group members", "I would participate in plenary meetings with fellow group members", "I would write flyers on behalf of the group", and "I would sign a complaint with fellow group members".

Preliminary analysis
validity of the four-dimensional identity leadership inventory, the ingroup identification and the group-efficacy measures by comparing three competing models. The first model was a one-factor model (Model 1: identity leadership, group-efficacy and ingroup identification items loading on a single factor), the second model was a three-correlated factors model  Table 4 and Table 5  We again examined in an initial step the context-dependency of the relationships between the theoretical constructs by comparing the data of the partisan and the workgroup contexts through model specification. We first compared the totally unconstrained model that allowed between-group differences in all estimated parameters (Model 1) with a limited intercept model that assumed measurement weights for group efficacy and collective action and item intercepts for collective action to be equal between the two group contexts (Model 2). Model 2 fitted equally well the data as the less parsimonious Model 1. Restricting the measurement weights for identity leadership, ingroup identification or the item intercepts of the identity leadership or ingroup identification or group efficacy items to be equal between the two group contexts would have significantly worsened the model fit. We, therefore, kept in the subsequent models the assumptions of Model 2 and allowed these latter parameters to differ between group contexts. Secondly, we compared the limited structural weight model that corresponded to the best fitting model in Study 1 assuming the effects of identity leadership on ingroup identification, both the direct and the indirect via efficacy, as well as the direct path from efficacy on collective action intentions to be group context invariant (Model 3) with the totally unconstrained model (Model 1) and the limited intercept model (Model 2). The most parsimonious Model 3 fitted equally well the data as Model 1 and additionally the direct path from identity leadership on collective action (Model 4) as group context invariant with the previous models. This most parsimonious Model 4, in which the effect of ingroup identification on collective action was the only structural weight assumed to differ between group contexts, fitted equally well the data as Models 1, 2 and 3. Finally, we compared Model 4 with Model 5 assuming an invariant group context for all structural paths.
Model 5 fitted the data worse than the previous models (see under Model Comparisons Δ Chi-Square in Table 5). Consequently, we report for Study 2 the direct and indirect effects for the limited structural weight Model 4.
In the partisan context identity leadership explained 17.8% of the variance of group efficacy, identity leadership and group efficacy explained between 37.0% and 37.2% of the variance of ingroup identification (varying across the 10 imputations), whereas identity leadership, group efficacy and ingroup identification explained together between 44.5% and 45.4% of the variance of collective action. The explained variance in the workgroup context was 36.4% for group efficacy, 59.9% to 60% for ingroup identification and 29.5% to 30% for collective action. The direct and indirect paths between identity leadership, group efficacy, ingroup identification, and collective action are reported in Table 3. The direct paths between identity leadership and group efficacy, between identity leadership and ingroup identification, between group efficacy and ingroup identification, and between group efficacy and collective action were statistically significant in both contexts. Consistent with Hypothesis 2, and replicating the findings of Study 1, the path between ingroup identification and collective action was only significant in the partisan group context, but not in the workgroup context.
The direct path between identity leadership and collective action was negative and weak, but reached statistical significance in both the partisan context and the workgroup context. and collective action through ingroup identification (predicted by Hypothesis 1 and bypassing efficacy) reached statistical significance only in the partisan group context, but not in the workgroup context. The differences between these two indirect paths was significant, Estimate = 0.12, SE = 0.06, 95% CI [0.000; 0.241]. Moreover, the extended path linking identity leadership with collective action through group efficacy and ingroup identification (Hypothesis 1a) was only significant in the partisan context, but not in the workgroup context. Again, supporting Hypothesis 2, the difference between these two indirect paths reached statistical significance, Estimate = 0.10, SE = 0.05, 95% CI [0.003; 0.193].
Adding the four independent factors into the model, χ 2 (1179) = 2653.26, p < .001, did not change the predicted results for the general identity leadership factor (Supplementary Material, Table S12).
As we found in Study 2, unlike to Study 1, that the participants' age and status of employment were related to identity leadership, ingroup identification and group-efficacy, we estimated the same limited structural weight Model 4 while controlling for age and employment status (see Table S13 in the Supplementary Material). The inclusion of these covariates did neither influence the findings concerning the indirect paths supporting Hypotheses 1 and 1a nor the conditionality of these indirect paths supporting Hypothesis 2.
Overall, the results of Study 2 replicated the findings of Study 1 that the given group context in which the leader-followers relationship is embedded matters for leadership exercised through social identity processes. Moreover, the results of Study 2 imply that identity leadership nurtures followers' beliefs in being capable to change a situation or the destiny of the ingroup , which also makes the ingroup matter for followers. Group efficacy and its implication for collective action was, however, group context invariant.
The overall aim of the present research was to address the question what makes followers respond to a leader's instruction to act collectively. More specifically, we were interested in the group context-dependency of the hypothesized relationships between identity leadership, group efficacy, ingroup identification, and intentions to act collectively when instructed by the leader. In two studies we found that the link between identity leadership and collective action through ingroup identification is indeed conditional upon the group context in which the leader-followers relationship is embedded. More specifically, we found that the statistical indirect effect of identity leadership on collective action through ingroup identification was evident in the partisan group but not in the workgroup context (Study 1 and 2). Moreover, the chain mediation capturing the relationship between identity leadership and collective action intentions through the sequence of group efficacy and ingroup identification was context-dependent as well. As predicted, it was only significant in the partisan group and not in the workgroup context, and the difference between these two effects was statistically significant (Study 2). Interestingly, however, is that this group context-dependency only applied to effects that involved a link between ingroup identification and collective action.
Relationships between identity leadership and ingroup identification (Studies 1 and 2), and between group efficacy and all three other variables (Study 2) did not differ between group contexts, nor did the indirect effect through efficacy that bypassed ingroup identification. In conclusion, our findings imply that the group context in which the leader-follower relationship is embedded matters for collective action evoked by ingroup identification due to identity leadership but not for collective action evoked by group efficacy due to identity leadership.

28
Our results demonstrate that identity leadership is a promising approach not only for leadership in business and work contexts but also for leadership in the contexts of politics and volunteer work (van Dick et al., 2018). More specifically, our results are consistent with the idea that identity leadership does not only influence workplace-specific behaviors such as job satisfaction and work performance but also group behavior that aims at advancing group interests and goals. Accordingly, leaders are able to evoke not only personal and organizational change but also large-scale social change through social identity processes.
Secondly, our results support the notion that the group context in which the leaderfollower relationship is embedded matters for leadership when exercised through social identity processes; although, it does so in a particular way. Whereas group efficacy and ingroup identification were related to perceived identity leadership irrespective of whether participants were allocated to partisan or workgroup contexts, ingroup identification was only directly linked to intentions to act collectively in the partisan context. It indicates that ingroup identification due to identity leadership is more important for collective action in partisan than in workgroup contexts.
Our results suggest that the shared identity plays a more important role for the mobilization of followers in groups, such as partisan groups, in which they seem to be by definition obliged to act collectively in the interest of the group. From a broader point of view, this particular context effect can be explained by a more general principle, namely by the fact that in such groups collective action has a better normative fit to the behavior that is expected from a prototypical member (Turner et al., 1987). According to this principle, our results do not imply that a leader who exercises identity leadership can mobilize followers to act collectively for or against just anything. Even in partisan groups, the collective actions that leaders ask followers to take need to be appropriate according to the norms and values of the social group. For instance, a trade union functionary might ask his or her followers to sign 29 a petition against the introduction of affirmative action in the composition of decision boards.
This request is likely to be appraised as inappropriate if members of this group share feminist and even post-feminist norms and values because they might appraise affirmative action as an effective tool to increase gender equality in the organisation. Consequently, the leader requesting his followers to sign such a petition is rather unlikely to find broader support among his followers. Our reasoning related to group normativity has at least two implications, which should be addressed by future research. First, while social identification in workgroup contexts might be less important for identity leadership effects on collective action as measured in the present research, it might play a stronger role in mobilising followers for work-related collective efforts (e.g., extra shifts of medical teams to treat Covid-19 patience) or within liberal-leaning organisations in which collective action of employees is accepted part of the organisational culture (Gupta & Briscoe, 2020). Secondly, although collective action forms part of the social identity in partisan group contexts, shared social norms and values will have a crucial function in regulating whether and which kind of collective action is appropriate and can be expected in a certain situation (e.g., Haslam, Reicher, & Birney, 2016).
Finally, our results showed that group-efficacy evoked by identity leadership, ingroup identification evoked by identity leadership and group efficacy, and collective action evoked by group efficacy due to identity leadership were group invariant. These results imply that irrespective of the group context, the perception of the leader as being one of us and as doing it for us fosters followers' beliefs that they can act as and achieve something for the group, which is linked to their ingroup identification and their readiness to engage in collective action. Moreover, our results suggest that different to ingroup identification effects, the effects of followers' beliefs about their capabilities to act as a group are not necessarily shaped by social group context and probably less dependent on whether these acts are 30 actually appropriate or inappropriate or fit a certain identity. By nurturing followers' group efficacy, leaders actually exercise identity entrepreneurship by defining and influencing what the group is able to do, which is, however, different from defining and influencing what the group stands for. Nevertheless, we would expect that leaders who are perceived as crafting a sense of us (identity entrepreneurship) might be particularly effective in influencing followers' group efficacy beliefs. We were not able to test this assumption statistically in the present research because of the strong intercorrelations between the four sub-scales of the identity leadership inventory. Future research might test it by manipulating the different dimensions of identity leadership and assessing their effectiveness in fostering or changing efficacy beliefs.
Given the pioneering character of our research, we would like to stress that our studies were to a large part exploratory rather than explanatory in nature. As much as the context-independency of all effects via group-efficacy can be explained by the theoretical ideas behind our research, we were not able to predict it a-priory. Moreover, our results do not allow inferring a causal process between identity leadership and group outcomes as we only experimentally manipulated the group context, but not group membership or perception of identity leadership itself, nor can we rule out the possibility of reversed causality. We can also not rule out the possibility of effects produced by method factors (Podsakoff et al., 2012) nor the existence of omitted variables influencing the relationships between identity leadership, ingroup identification, and collective action. To establish the directionality of the hypothesized processes further experimental research or longitudinal designs will be necessary.
One characteristic of our design was that, after being randomly allocated to one of the two larger group contexts, participants were asked to choose the specific leadership subcontext that was most relevant for them. This method had the double advantage of sampling a 31 rather large scope of contexts and at the same time to assure some ecological validity. Our design had, however, the disadvantage that it did not allow to disentangle differences between specific sub-contexts and inter-individual differences within the same specific subcontext. Moreover, our design might have created different demand characteristics, as it is something else to choose between partisan groups one identifies with than selecting a workgroup one is part of. While these are certainly limitations affecting the internal validity of our research, we are encouraged by the robustness of the data pattern across the two studies. Nevertheless, it might be worthwhile in future research to disentangle the two components of variation between participants by a highly powered replication that has a database broad enough to apply multi-level analyses. To avoid the possibility of different demand characteristics when studying multiple contexts, however, experimental research would be necessary that keeps the context more or less constant within conditions. Such research can use artificial group contexts. On the other hand, to increase external validity it can also be conducted in real-life contexts as long as participants can connect to the same specific group context (e.g., sampling from all members of the same union within the same work context). Another possibility to test the overall proposed idea of the current research would be an alternative experimental approach manipulating types of collective action (i.e., fitting versus non-fitting the specific social identity at stake) rather than manipulating group contexts.
Other limitations refer to information about our participants we did not assess in our studies. Although we asked our participants, who were mainly part-time students, about their employment status, we did not assess information about their professions, years of working experiences or the areas in which they work. This information would have provided additional insights, particularly for the workgroup context. Likewise, we did not control whether participants in the partisan contexts were actual members of the political parties or 32 trade unions they selected (Study 1) or whether they were active supporters of the selected NGOs (Study 2). Despite all these limitations, the current research revealed in two studies that the process connecting identity leadership with the mobilisation for collective action depends on the group context with regard to the mediating role of ingroup identification, but not with regard to the mediating role of group efficacy beliefs.
In conclusion, identity leadership is indeed a promising approach to extend our understanding of collective action and social change. It is promising because it conceptualises leadership as being through followers rather than done to followers (Turner et al., 2008, p. 70), it defines leadership as influencing both social stability and social change, and it provides a theoretical framework to study the influence of leaders in different group contexts.
Finally, it is promising because it does not only provide us with accounts why leaders such as Martin Luther King Jr., Mahatma Gandhi, Nelson Rolihlahla Mandela, and Mgcineni Mambush Noku (the man in the green blanket) were able to mobilize the masses to fight for freedom and social justice, but also why leaders such as Omar Al-Bashir or Robert Mugabe were able to successfully mobilize followers to support murder, torture, and persecution of political opponents; and to plunder their own countries. Identity leadership can mobilize collective action effectively, for the better or for the worse, depending on one's political position. Understanding how it works is therefore of utmost importance for everyone who cares about social change or stability.     Note. ***p < .001, ** p < .01, * p < .05, † p < .10. The correlation coefficients for the partisan context are reported in the upper part of the table, and the correlation coefficients for the workgroup context are reported in the lower part of the table.