Leadership in the Face of a Looming Threat: The Brexit Referendum
Wout de Vries S2393107
W.B.de.Vries@Student.rug.nl
August 2019
Research Master Thesis
Research Master in Economics and Business Faculty of Economics and Business
University of Groningen
Abstract
The current paper aims to assess the leadership response following the referendum vote in the United Kingdom to leave the European Union. I apply the threat-rigidity hypothesis to the field of leadership to make predictions on the response of individual leaders to the macro-level threat of the Brexit. By conducting a multi-macro-level event study, I show that the outcome of the referendum caused an increase in directive leadership behaviours for a large sample of leaders representing a wide range of organizations. My findings indicate that this effect may endure until after the Brexit has occurred. I discuss the implications for policy makers and organizations with respect to the Brexit and future large-scale events. Overall, this paper serves as further evidence on how the context can serve as an antecedent to leadership.
Leadership in the Face of a Looming Threat
On 23 June 2016, the population of the United Kingdom (UK) voted in a referendum on whether the UK should remain a member of or leave the European Union (EU). Contrary to general expectation, this ‘Brexit’ referendum resulted in a narrow majority (51.9 per cent) of votes in favour of leaving—an outcome the UK government ensured would be respected (Stone, 2016a). In the period that followed, organizations postponed their investment decisions, uncertain whether or when the government would formally start the withdrawal procedure and what kind of agreement would be made with the EU (Breinlich, Leromain, Novy, & Sampson, 2019; Cooper, 2016). However, certain was that a wide range of sectors would be seriously impacted by new restrictions on the movement of people and barriers to trade, such as tariffs (Blanchflower, 2018).
Official negotiations for a Brexit deal between the UK and the EU commenced a year later, on 19 June 2017. By December, less than 15 months remained before the official deadline, yet the magnitude of the consequences that Brexit would have for organizations remained unclear (Musaddique, 2017). In fear of a no-deal scenario, UK’s key employers such as Airbus and Siemens threatened to move away from the UK, together putting 30,000 jobs at risk (Bruce, 2018). In fact, many organizations were re-evaluating their decision-making following the shift in context that the referendum outcome induced. In early 2018, leaders of over 150 UK companies had been discussing potential relocation plans with the Dutch government (Stone, 2019). Overall, in the years following the referendum many organizations felt the looming threat of an impending Brexit (Shaheen, Miller, Luyendijk, & Newman, 2018).
current research aims to assess this response, by means of focusing on these organizations’ key representatives: their leaders. These individuals play a large role in the performance of their organization, and their decisions will largely shape how the UK proceeds as we move closer towards the Brexit. Indeed, a multitude of studies report on the strong relationship between organizational performance and both managers’ personal characteristics (Hambrick & Mason, 1984) and behaviours, such as their style of leading (Avolio, Reichard, Hannah, Walumbwa, & Chan, 2009; Bertrand & Schoar, 2003; Waldman, Ramírez, House, & Puranam, 2001). The importance of leadership is further exemplified by it being one of the most-studied topics in the disciplines of industrial-organizational psychology and
organizational behaviour (Dinh et al., 2014; Porter & Schneider, 2014). A focus on leaders should therefore permit valuable insights in how organizations performed in the wake of the referendum. Specifically, the current paper revolves around the following question: How did the outcome of the Brexit referendum impact leadership in the UK?
Leadership and Context
The referendum outcome implied a sudden shift in the context that UK leaders operate in. These leaders are likely to differ in the extent to which they are affected. For instance, the outlook of a future increase in tariffs will have greater implications for a manager working in a sector where trade with the EU is common than for a manager working in education (Los, Chen, McCann, & Ortega-Argilés, 2017; Thomas & McDaniel, 1990). That is, there are relevant contextual factors that may influence the leadership response to the referendum outcome.
Historically, those studying leadership varied in the extent to which they
Antonakis, 2009; Osborn, Hunt, & Jauch, 2002; Pawar & Eastman, 1997; Shamir & Howell, 1999; see also Johns [2006]). Instead, to a large extent the field has focused on the
dispositions and traits of leaders, and their relationship with a leader’s effectiveness
regardless of context (Boal & Hooijberg, 2001; Day & Antonakis, 2012; Lord, Day, Zaccaro, Avolio, & Eagly, 2017; but see Fiedler [1967], House & Aditya [1997], Hunt & Osborn [1981]). As by definition these dispositions and traits are relatively stable, such an approach may fail to appreciate the variations in behaviour that leaders display (Dinh & Lord, 2012; Shamir, 2011).
In recent decades the focus on the context in which leaders operate has increased (Day & Harrison, 2007; Dinh et al., 2014; Osborn, Uhl-Bien, & Milosevic, 2014). A substantial amount of studies have been conducted that assess how the relationship between leadership traits, behaviours and outcomes differ across contexts (Liden & Antonakis, 2009; Oc, 2018). Some scholars claim that this area of ‘contextual leadership’ is the most common new direction of research (Gardner, Lowe, Moss, Mahoney, & Cogliser, 2010; Oc, 2018; see Porter and McLaughlin [2006] for a review). As this subfield gains ground, and the
relationship between leadership and context is further explored, some researchers have noted a lack of consensus on what contextual variables are relevant (Ayman & Lauritsen, 2017; Hackman & Wageman, 2007). Though there does exist consensus that leadership does not occur in a vacuum, the term ‘context’ is broad and relates to leadership through multiple pathways (Johns, 2006; Oc, 2018; Osborn et al., 2002). The most commonly explored pathway is the moderating role of context. For instance, many have studied how the
Context as an Antecedent to Leadership
Frequently calls have been made for more research on the causal impact of context on leadership (Osborn & Marion, 2009; Oc, 2018), or on organizational behaviour more
generally (Johns, 2006; Rousseau & Fried, 2001). That is, as opposed to treating the context as shaping the relationship between leadership behaviours and outcomes, a change in context may instead cause a change in the way leaders lead (e.g., Dierdorff, Rubin, & Morgeson, 2009). In assessing such a role of the context—as an antecedent to leadership—Dinh and Lord (2012) argue for the role of events as a “fundamental level of analysis” (p. 651). Events have the capacity to engender the required shift in context that in turn affects leaders (Johns, 2006; Morgeson, 2005; Shamir, 2011). For my current purpose, I therefore treat the outcome of the Brexit referendum as an event that may have caused a change in leadership.
Applying the taxonomy of event dimensions by Hoffman and Lord (2013), a defining dimension of the referendum is that it is a large-scale event—a change in the macro-context with global consequences. In the domain of events and leadership, the impact of such shifts in the macro-level of context has been largely neglected (Oc, 2018; Osborn et al., 2002; Staw, 2016). There are two prime causes for this. First of all, there exists a gap in levels: Whereas large-scale events transpire on a macro level, leadership is essentially an individual
phenomenon. As a result, one requires specific theories and methods that allow for bridging this micro-macro divide (Aguinis, Boyd, Pierce, & Short, 2011). Second, establishing a causal relation between a change in context and a change in leadership necessitates particular research designs. However, the standard experimental design is not feasible when dealing with macro-level constructs. I turn to these two issues next.
conceptualize of leaders as embedded in organizations, which are in turn embedded in sectors (Kozlowski & Klein, 2000). Contextual leadership researchers argue that one cannot fully understand leadership by studying a single level of analysis (House, Rousseau, & Thomas-Hunt, 1995), and instead one has to explicitly account for these multiple levels in both theory and measurement (Yammarino, Dionne, Chun, & Danseraeu, 2005). However, in doing so, few leadership researchers tend to evaluate processes that transcend the organization level (Batistič, Černe, & Vogel, 2017; Dionne et al., 2014; Dinh et al., 2014; Uhl-Bien & Marion, 2009). Large-scale events such as the referendum outcome encompass a macro-level that exceeds this organization level. As vice versa, those studying the macro-context tend to neglect what occurs inside organizations, this makes for what has been coined as the micro-macro divide (Aguinis et al., 2011). This divide has caused a propensity for both theories and methods to be restricted to either a micro or macro level, with limited linkages between both. As a result, there is a lack of research on the behavioural effects within organizations of large-scale events (Morgeson, Mitchell, & Liu, 2015; Osborn et al., 2002; Staw, 2016).
Causality. A second issue concerns the dynamic relationship between context and leadership. Just as a change in context can shape leadership, leaders in turn determine their context, such as when deciding to relocate their offices (Stone, 2019). As a result, potential endogeneity problems should caution contextual leadership researchers against heedlessly drawing conclusions about causality (Antonakis, Bendahan, Jacquart, & Lalive, 2014; Wooldridge, 2002). An example of this is the correlational study by Cogliser and
of large-scale events makes it impossible to manipulate how and when an event transpires, or randomize who is exposed to it.
Conducting an event study is an alternative approach to drawing conclusions about the causal impact of a shift in the macro-context (Meyer, 1995). Such a design accords well with the calls for a greater focus on events (Dinh & Lord, 2012; Hoffman & Lord, 2013; Rousseau & Fried, 2001), but to infer causality one requires that the event is (1) exogenous and (2) ‘potent’—capable of bringing about a change in leadership. I argue that this is the case for the Brexit referendum outcome. Regarding the exogeneity of the event I point to the unexpected nature of the outcome. Prior to the referendum, confidence in a likely majority vote for the UK remaining in the EU resonated across financial (Cohn, 2016) and betting markets (Shaddick, 2016), as well as several opinion polls (Stone, 2016b). Regarding the potency of the event, I apply event system theory (Morgeson et al., 2015). This theory holds that events are capable of bringing about change to the extent that they are “novel, disruptive, and critical” (p. 515). Given that it is unprecedented that a member state will withdraw from the EU, and given the widespread risks the Brexit poses (Los et al., 2017), I believe it is safe to conclude that this event is likely to be a “dominant contextual condition” capable of causing a change in leadership (Osborn et al., 2014, p. 11). Consequently, conducting an event study will allow me to assess how the referendum outcome impacted leadership in the UK.
individual level. Moreover, they found that this response depended on the magnitude of the threat to the managers’ organizations—the context in which these leaders operate. In the current study I build forward on the theory and methods they introduced.
The Present Study
The goal of the current research is to investigate the leadership style of managers in the UK in the period surrounding the Brexit referendum. To this end, I conduct an event study and analyze an extension of the data used in Stoker et al. (2019), as collected by a global consultancy agency. This data allows for a multi-level approach, so that I account for the context in which these leaders are embedded.
The remainder of this paper is structured as follows. In the following section I present my theoretical framework, where I reiterate the theoretical foundation established by Stoker et al. (2019) and apply it to the Brexit to develop my hypotheses. Subsequently I outline the data collection procedure and describe in detail the sample of UK managers I study, as well as the measures I use. In the following section I estimate a series of multi-level models to assess the impact of the referendum outcome on the managers’ style of leading. The section afterwards includes a critical discussion of my results, where I return to my theoretical framework and the assumptions I made throughout the paper. The final section concludes.
Theoretical Framework The Threat-Rigidity Hypothesis
environment of an organization are categorized in this manner, and an organization’s response depends largely on how its managers interpret the situation (Dutton & Jackson, 1987; Latham & Braun, 2009; Patel & Cooper, 2014; Thomas, Clark, & Gioia, 1993). When under threat, the threat-rigidity hypothesis predicts “a general tendency for individuals, groups, and organizations to behave rigidly” (p. 502). Overall, this rigidity takes the form of a restriction in information processing and a constriction of control. Individuals increase their reliance on previously established information in their decision-making, and consequently respond to threatening situations in ways familiar to them (i.e., react with a well-learned response). On the organization level exists a similar reliance on previous knowledge, resulting in a reduction in the number of alternatives considered. This combines with a centralization of authority and a formalization of procedures. Given that a threat to an individual can originate at a macro-level, the threat-rigidity hypothesis allows for bridging the micro-macro divide (Aguinis et al., 2011). Furthermore, the multi-level nature of the theory permits considering managers in the organizational context in which they lead. The organizational response to a threat will affect its individual managers. An example of this is the finding by Stoker et al. (2019) that the threat of the financial crisis was larger for
individuals working in the manufacturing sector—a sector with a relatively large increase in unemployment (Groot, Möhlmann, Garretsen, & de Groot, 2011).
The threat-rigidity hypothesis is not the only theory that makes predictions on the nature of a response under situations of threat. Prospect theory (Kahneman & Tversky, 1979), a hallmark theory in economics, predicts an increase in risk-taking when individuals face “impending negative consequences” (Staw et al., 1981, p. 502) and perceive a situation in a loss frame (Tversky & Kahneman, 1981). This runs counter to the threat-rigidity hypothesis, where the predicted rigid response is more in line with a decrease in risk-taking.
opposing views. They apply the argument by Ocasio (1995), who claims that “the
experimental results of prospect theory deal with the consideration of objectively risky, but well-specified alternatives, while threat-rigidity deals with the failure to consider alternative responses that are not well understood, whose outcome is highly ambiguous, and for which a probability distribution of outcomes is not well-defined” (p. 297). Chattopadhyay et al. (2001) argue that one should differentiate between the relevant dimensions of a threat to ascertain which of the two theories applies in a given situation. In situations of
uncontrollability and ambiguity we should turn to the threat-rigidity hypothesis, whereas in situations of likely loss prospect theory applies. Stoker et al. (2019), in applying this framework, argued that the threat of the financial crisis corresponded to the former
description, and is therefore best described as a “control-reducing threat” (Chattopadhyay et al., 2001, p. 939). I maintain that a similar argument holds for the Brexit. First of all, the preponderance of negative consequences predicted to follow from the Brexit warrants classifying it as a threat. Second, the fact that it is unprecedented that a member state will withdraw from the EU makes for a unique situation for which no probability distribution of outcomes exists. Consequently, I will classify the Brexit as a control-reducing threat and the outcome of the referendum as the event that introduced this threat. Therefore, I will apply the threat-rigidity hypothesis in developing my hypotheses.
The Leadership Response to a Threat
attempts to regain control in the face of a control-reducing threat. In line with this, previous research has found directive leadership behaviours to be more common in threatening
situations (Kamphuis, Gaillard, & Vogelaar, 2011; Hannah, Uhl-Bien, Avolio, & Cavarretta, 2009; see also Yun, Faraj, and Sims [2005]). Potential mechanisms underlying these findings are that aspects of threatening situations restrict leaders in the range of behaviours they can perform (Kerr & Jermier, 1978), or that leaders adapt to follower perceptions of prototypical effective leadership in these situations (Foti & Lord, 1987; Lord, Brown, Harvey, & Hall, 2001). In line with this, Stoker et al. (2019) found an increase in directive leadership following the financial crisis. Based on this I formulate the following hypothesis:
H1: The Brexit referendum outcome led to an increase in directive leadership in the UK.
I aim to test this hypothesis by comparing UK managers’ directive leadership
behaviours before and after the referendum outcome. In an additional analysis I will explore their leadership response over time. In testing the persistence of the increase in directive leadership, Stoker et al. (2019) found the effect to have disappeared one year after the crisis. They explained this by arguing that the main threat that the crisis posed had vanished after one year, both in terms of its macro-economic impact and the fear of an economic depression. The same does not apply to the Brexit, as a year after the referendum official negotiations between the EU and the UK had only just commenced, and as of now the Brexit itself has not occurred yet. Exploring how directive leadership develops over time will allow me to
disentangle whether it is the case that leaders remain more directive for as long as the threat persists, or if leaders instead can only maintain a change in their style of leading for a given period before regressing to old habits.
leaders display a change in participative leadership following the referendum. This style of leading is characterized by behaviours such as delegating authority and involving
subordinates in decision-making (Somech, 2006). Such behaviours run counter to the predictions of the threat-rigidity hypothesis, and my classification is strengthened if I do not find an increase in participative leadership. Second, I will analyse additional samples of French and German managers. Whereas the Brexit is likely to pose a threat to these managers, the degree of this threat is undoubtedly smaller than for managers in the UK. Therefore, my results are stronger if following the referendum, I find an increase in directive leadership for managers in the UK that is larger than for French and German managers.
The magnitude of the threat. The degree to which the Brexit poses a threat is likely to differ across UK managers. For instance, the damage of future barriers to trade varies for managers working across different sectors (Los et al., 2017). Chattopadhyay et al. (2001) argue that “the effect of threats and opportunities on organizational actions may be moderated by contextual characteristics” (p. 940). In line with this, Shimizu (2007) found the degree organizational risk-seeking in the face of a threat to crucially depend on features of the organization, such as the size of its sub-units (see also Caballero & Hammour, 2001). Stoker et al. (2019) demonstrated that a similar argument holds for leadership, in finding a relatively large increase in directive leadership for managers active in the manufacturing sector. This is in line with the study by Thomas and McDaniel (1990), who found organization-level
I propose that the magnitude of the threat that the Brexit poses varies for different types of organizations (Porter & McLaughlin, 2006). Specifically, I argue that a relatively strong threat exists for (1) organizations in sectors with a large amount of exports to the EU, (2) multinational organizations, and (3) organizations with foreign headquarters. Due to their strong connections to outside the UK, the Brexit will cause a relatively large disruption for these types of organizations (Los et al., 2017; Musaddique, 2017; Super, 2018). I expect this to be reflected in the leadership style of UK managers. Specifically, I hypothesize the following:
H2: The increase in directive leadership is stronger for a) organizations in sectors with large exports to the EU, b) multinational organizations c) organizations with a foreign headquarters.
Method
The hypotheses were tested using a large dataset as collected by a global consultancy agency. This dataset was an extension of the data used by Stoker et al. (2019), to now include additional time periods, and was merged with a set of additionally collected organization- and sector-level variables.
Population and Sample
companies. On average, they work for large firms that Korn Ferry serves in multiple countries.
Organizations make an appointment for a single training of their managers. They select themselves which managers are eligible for training, as well as a set of subordinates to rate the managers. The location of the trainings is oftentimes the headquarters of the
organization, but sometimes a public place such as a hotel is decided on.
For my current purpose I primarily considered a subset of the Korn Ferry data and focused on participating managers in training programs held in the UK around the time of the Brexit referendum. This subset included 5,048 managers working for 281 organizations, measured between the 30th of May 2015—a year before the referendum—and the 23rd of August 2018. This latter date was the final date for which I had access to the Korn Ferry data. Measures
Directive leadership. The main dependent variable of the study was managers’ scores on directive leadership. For this construct, Korn Ferry employed both subordinate ratings and managers’ self-ratings. Previous research has indicated frequent lack of agreement between such self- and other-ratings (Harris & Schaubroeck, 1988), as social desirability leads to a tendency for individuals to inflate self-ratings (Mabe & West, 1982; Podsakoff & Organ, 1986). In a review article, Fleenor, Smither, Atwater, Braddy and Sturm (2010) advocate the use of subordinate ratings of leadership, provided that one can establish a sufficient degree of agreement between the various raters. As a consequence, the current study utilizes only subordinate ratings in the analyses. Overall, the data included a total of 25,983 subordinate ratings, making for an average of 4.8 subordinates per manager rated.
to make most of their own decisions’ and ‘makes most decisions for the people in the team’ (see Appendix A). Subordinates indicated to what extent these statements applied to the manager they rated. For the analysis the reversed scored items were recoded so that higher scores imply a leader is rated as more directive (M = 3.05, SD = 0.62; α = .74). These items were a slightly revised version of those used by Stoker et al. (2019). Eighteen subordinates that failed to respond to more than two of the items were excluded from the analysis. An exploratory factor analysis had a one-dimension factor solution, with the six item loadings ranging between 0.57 and 0.78, indicating the 6-item scale is unidimensional. This finding was not sensitive to imputing the missing values. Regarding the inter-rater reliability of the various subordinates, an ICC(1) of 0.31 and an ICC(2) of 0.70 was found. These values are well above the mean of previous estimates reported in the literature for aggregating a group-level construct (Castro, 2002; Woehr, Loignon, Schmidt, Loughry, & Ohland, 2015). Lastly, an rwg of 81 per cent indicated on average “strong agreement” between the various
subordinates (LeBreton & Senter, 2008, p. 836). Taken together, this was deemed as sufficient evidence to aggregate the subordinate ratings to the manager level, making for a single directive leadership score to be used in subsequent analyses.
Participative leadership. Similar to directive leadership, for participative leadership subordinates indicated on a 6-point Likert scale their agreement with five sets of two
statements (see Appendix B; M = 4.53, SD = 0.57; α = .75), which closely resembled the items used in Stoker et al. (2019). Exploratory factor analysis revealed a single dimension, onto which the item loadings ranged between 0.60 and 0.82. Eight subordinates that failed to respond to more than two of the items were excluded from the analysis. On average,
subordinates displayed lower, but sufficient agreement on the participative leadership items [ICC(1) = 0.21, ICC(2) = 0.57, rwg = 0.81], which were subsequently aggregated to the
Even though consensus in the literature is that directive leadership is conceptually distinct from participative leadership (Somech, 2006; Yammarino et al., 2005), some scholars have conceptualized of the two styles as two ends of the same continuum (e.g., Lewis, Welsh, Dehler, & Green, 2002; Sagie, Zaidman, Amichai-Hambuger, Te'eni, & Schwartz, 2002). For this reason, an exploratory factor analysis on the combined set of 11 directive and
participative leadership items was performed to assess the construct validity of the two scales. As table 1 shows, the factor solution identified the two constructs as separate dimensions, onto which the respective items loaded sufficiently for the most part. Only the directive leadership item ‘Makes most decisions for the people in the team’ erroneously had a larger loading onto the participative leadership factor. However, the results are robust to the exclusion of this item.
Table 1
Results of an Exploratory Factor Analysis on the Directive & Participative Leadership Items. Factor 1 Factor 2 Directive leadership (α = .74)
o To ensure instructions are followed exactly, this person requires people to provide detailed updates.
0.15 0.78 o Expects people to carry out instructions immediately. 0.02 0.65 o When team members deviate from this person’s directions, they are
quickly corrected.
-0.05 0.77 o To ensure compliance, this person monitors what people are doing
very closely.
-0.08 0.54 o Makes most decisions for the people in the team. -0.62 0.32 o To ensure there are no surprises, this person pays very close
attention to what team members are doing.
-0.17 0.64
o Frequently encourages the team to make decisions for themselves. 0.67 -0.16 o Strongly prefers that decisions be made through consensus. 0.63 0.11 o Keeps everyone in the team involved and well-informed about
organizational issues that may affect them.
0.70 0.15 o Encourages people to participate in most decision-making. 0.81 0.04
o Regularly adopts new ideas from the team. 0.75 0.00
Eigenvalue 3.62 1.94
Proportion of variance explained 0.28 0.23
Note. For purpose of exposition, only those items for which higher scores reflect more directive/participative leadership are displayed. Estimation was performed in the R language (R Core Team, 2019), using package PSYCH (Revelle, 2019).
Brexit. A dummy variable ‘Brexit’ was created to assess the impact of the result of the referendum. This dummy takes the value 1 for the 69 per cent of managers in the sample that were trained after the 23rd of June 2016 and 0 otherwise. To explore whether there are effects over time, I further considered the period after the referendum, which I divided into two sub-periods. In deciding on what sub-periods to assess I reasoned as follows: On 19 June 2017, a year after the referendum, official negotiations between the UK and the EU
commenced. This could be a potential turning point, as here it became clear that there would truly be a Brexit. Moreover, in the study by Stoker et al. (2019), the effect of the financial crisis on directive leadership disappeared after one year. However, there, the actual threat of the crisis was largely diminished after a year, whereas the same does not apply for the Brexit a year after the referendum. For these reasons, in evaluating the effect of the Brexit
measured between the 23rd of June 2016 and the 23rd of June 2017. ‘Brexit – Year 2’ takes
the value 1 for managers measured between the 23rd of June 2017 and the 23rd of August 2018.
Control variables. Only a small proportion of organizations opted for a second training of the same managers within the scope of a few years. Specifically, out of the 5,048 managers 386 were measured a second time and 21 a third time, so that only 7.5 per cent of the observations was a repeat measurement. As a consequence, in assessing the effect of the referendum outcome on leadership style, essentially two groups of different managers are compared. In order to draw conclusions about causality, it is of key importance that the group of managers trained before the referendum is comparable to the group trained afterwards. If a priori these two groups would differ in their leadership style, any conclusion involving a change in leadership style following the referendum would be invalid. To minimize this risk, four individual-level control variables were included in the analyses, as recorded by Korn Ferry during the training. These four variables were included in Stoker et al. (2019), allowing for a similar analysis.
The four control variables are a manager’s age, gender, tenure and nativeness. In addition, Korn Ferry recorded a fifth variable of potential interest: the trainees’ level of management. Though not included in Stoker et al. (2019), there are good reasons to include this variable in the analyses, which will be explained below. To the extent that these five variables have been shown to be related to directive or participative leadership in previous research, including them as controls should serve to reduce bias. Furthermore, this
Age. The consensus in the literature is that there are “minimal differences between
generations” in both leadership style as well as preferences for specific styles (Rudolph, Rauvola, & Zacher, 2018, p. 55). In line with this, Walter and Scheibe (2013) found no relation between age and directive leadership in their literature review. In contrast with this stands the study by Oshagbemi (2008), who analysed data from over 400 UK managers and suggested that “age may positively impact the use of directive leadership” (p. 1906). Overall, despite this limited evidence I will follow the approach by Stoker et al. (2019) and include age as a control variable in the analyses.
During their trainings, Korn Ferry had managers indicate which of six age categories they belonged to, ranging from ’19 or younger’ to ’60 or older’. In contrast with the findings by Oshagbemi (2008), older managers were rated as significantly less directive (rτ = -.05, p <
0.01). Age was not related to participative leadership (rτ = .01, p = 0.70).
Gender. Differences in leadership style between male and female leaders are a
that female leaders were rated as significantly more directive (rpb = .03, p < 0.05), as well as
more participative (rpb = .06, p < 0.01).
Tenure. A manager’s tenure is defined as the number of years she has assumed her
function as a leader. Research on the link between tenure and leadership style is scarce, but some studies do suggest a link between the two constructs. For instance, Korac-Kakabadse, Korac-Kakabadse and Myers (1998) argue that managers with long tenure rely increasingly on their previous experience and consequently may be less flexible in their style of leading. In addition, their experience may allow them to more easily decide on the appropriate style of leading in a given situation (Oshagbemi, 2008). This is in line with upper echelon theory (Hambrick & Mason, 1984), which posits that one can effectively use such background characteristics as tenure to predict managers’ values, which in turn shape their style of leading. However, these studies primarily suggest a relationship between tenure and the variance of a given leadership style, rather than its level.
Five categories were used to describe how long a trainee had been in a manager position, ranging from ‘no formal experience’ to ‘more than 10 years’. Manager’s tenure was negatively related with directive leadership (rτ = -.07, p < 0.01), and unrelated with
participative leadership (rτ = -.01, p = 0.31).
Nativeness. Discrepant cultural values may result in differences in leadership style
between native and non-native managers (House, Javidan, Hanges, & Dorfman, 2002). Within multicultural organizations, such differences are indeed a common finding (e.g., Gabrielsson, Darling, & Seristö, 2009). In a study comparing expatriate and national managers in the United Arab Emirates, Bealer and Bhanugopan (2014) found strong
Korn Ferry recorded the country of birth of those trained. Managers born in the UK were coded as ‘native’, which constituted 85 per cent of the sample. ‘Nativeness’ correlated negatively with directive leadership (rpb = -.05, p < 0.01), and positively with participative
leadership (rpb = .06, p < 0.01).
Level. The relationship between level of management within an organization’s
hierarchy and leadership style has frequently been a topic of research (Oshagbemi & Gill, 2004). In a meta-analysis, Lowe, Kroeck, and Sivasubramaniam (1996) found that higher level managers tend to use all types of leadership behaviours less, compared to lower level managers. In line with this, in a more recent study van Emmerik et al. (2010) found that managers at higher levels use less directive leadership. Jago and Vroom (1977) found more participative and less directive behaviour at higher levels of a large R&D organization. Oshagbemi (2008), in his study on managers in the UK, reported negative relationships between management level and both participative and directive leadership.
These findings can be explained in multiple ways. A common explanation is that lower level managers are more malleable in their leadership style, whereas higher level managers tend to stick to those styles that have been effective in their previous experiences. Whereas this explanation is difficult to disentangle from an explanation based on tenure (note the relatively strong negative correlation between tenure and directive leadership), for
directive leadership one can also consider managers’ job description at different levels: Higher level managers “tend to give only broad outlines, opinions and suggestions rather than directives to their lower level managers. On the other hand, supervisors or foremen often need to give specific directives to facilitate operatives in doing exactly what is expected of them and when” (Oshagbemi, 2008, p. 1906).
examined leadership style across two levels of management and found a manager’s leadership style to affect her chances during selection. As a result, including ‘Level’ as a control variable introduces potential endogeneity issues (Antonakis et al., 2014). For my current purposes I will therefore only include this variable in an auxiliary set of analyses.
Korn Ferry employed six categories to distinguish between various levels of management, ranging from ‘Early-level individual contributors’ to ‘Senior management’. ‘Level’ was negatively related with directive leadership (rτ = -.09, p < 0.01), as well as with
participative leadership (rτ = -.08, p < 0.01). Note that the magnitude of these correlations is
relatively large.
Organization- and sector-level variables. The testing of hypotheses 2a, 2b and 2c necessitated data that was only partly collected by Korn Ferry. The required additional information was gathered by means of triangulation of a wide range of data sources, including data from Bureau van Dijk (https://orbis.bvdinfo.com), Compustat
(https://www.compustat.com), Bloomberg (https://www.bloomberg.com), as well as the annual reports and corporate websites of the individual organizations in the sample. For 41 managers of 10 organizations no reliable information was found, so that these are excluded from the core sample. The reported results are not sensitive to including these observations in the analyses.
Export risk indicator (ERI). Korn Ferry recorded the sectors in which managers were
active for 227 of the 281 organizations represented in the sample. For this they employed a classification system they designed themselves. Using the aforementioned sources, the remaining missing classifications were completed, and some obvious errors were corrected. Furthermore, the sectors ‘Consumer Products (excluding Food & Beverage)’ and
In order to quantify the degree of the threat that the Brexit poses to organizations in these various sectors, I was able to use data from the 2016-release of the World Input-Ouput database (WIOD; Timmer, Los, Stehrer, & de Vries, 2016). The WIOD holds information on the volume of world trade across 56 industries (See Appendix D), up until 2014. Importantly, this includes imports of services and goods at any stage of the production, so that industries trading in intermediaries within a global value chain are appropriately accounted for (Los, Timmer, & de Vries, 2016; see Ijtsma, Levell, Los, and Timmer [2018] for the position of UK companies in these global value chains).
Industries that currently add value in a value chain that crosses the UK-EU border are exposed to Brexit. Based on the WIOD data, Los et al. (2017) developed a set of indicators that convey how much of this ‘value added’ is at risk as a result of the Brexit (see Chen et al. [2018] for the techniques involved). For industries in the UK, these export risk indicators (ERIs) capture an industry’s degree of exposure to Brexit by means of its dependency on trade with the EU. Specifically, the ERIs encompass the percentage of sectoral value added that would be lost if the UK could not export anything to EU countries. That is, the ERIs do not include the risk of losing access to intermediate goods from outside the UK. Neither do these ERIs capture the potential domestic loss of value resulting from the relocation of an organization to outside the UK. Overall this allowed Los et al. (2017) to quantify the risk that the Brexit would pose for 54 of the 56 WIOD industries.
Appendix E displays the ERIs for the 54 WIOD industries, in addition to the
industries. The ERIs for the Korn Ferry sectors to be used in the analyses are presented in Appendix F.
Part of a multinational. Using the previously listed method, for each of the 281
organizations in the sample it was established whether they had operations outside of the UK. Companies were labelled as a multinational organization if they owned or produced a
substantial amount of goods or services abroad during the period surrounding the Brexit referendum. Specifically, a dummy variable ‘Part of a multinational’ was created to identify managers in these organizations. Given that Korn Ferry often serves organizations in multiple countries simultaneously, their broader dataset on managers trained outside of the UK proved to be useful in the classifications. That is, a multinational status could be assigned with greater confidence if in the data a given organization had a presence in multiple countries.
Foreign headquarters. Similar to assessing multinational organizations, the location
of headquarters was determined for each of the 281 UK organizations. This was performed based off the ultimate owner of the organization in terms of owning a majority of its shares. Organizations with headquarters outside of the UK were labelled using a ‘foreign
headquarters’ dummy.
Results Descriptive Statistics
variables on the organization- and sector-level, only 1 per cent of managers was excluded from the core sample due to a lack of data availability. The ERIs range between 0 and 27 per cent of sectoral value at risk, with a mean of 12 per cent across all managers (see Appendix F). Lastly, out of the 5,455 trainings given, 76% were given to a manager of a multinational organization and 49% to a manager of an organization with a foreign headquarters. Note that all organizations with a foreign headquarters are multinationals, so that 27% of managers work for a multinational with headquarters in the UK.
In testing my hypotheses, my prime focus will be on comparing the 69 per cent of managers trained after the Brexit referendum with the 31 per cent trained before. In all
analyses I include the dummy variables ‘Female’ and ‘Nativeness’, which take the value 1 for female and native leaders, and 0 for male and non-native leaders, respectively. For the
categorical control variables I include dummies for each level, with as reference category those managers that are youngest or have the shortest tenure. Though there are no outliers present in the data, for many variables there does exist a substantial amount of missing data. Respondents with missing data on any of the relevant variables are excluded from the analyses, following the approach of Stoker et al. (2019).
Table 2
Descriptive Statistics.
N Mean St. Dev. Min. Max.
Directive leadership 5,453 3.05 0.62 1 6
Participative leadership 5,454 4.53 0.57 1 6
Brexit 5,453 0.69 0.46 0 1
Brexit – Year 1 5,453 0.38 0.48 0 1
19 or younger 4,116 0.00 0.01 0 1 20 – 29 4,116 0.03 0.17 0 1 Age 30 – 39 4,116 0.28 0.45 0 1 40 – 49 4,116 0.44 0.50 0 1 50 – 59 4,116 0.24 0.43 0 1 60 or older 4,116 0.01 0.10 0 1 Female 5,103 0.35 0.48 0 1
No formal management role 4,123 0.03 0.16 0 1
Less than 3 years 4,123 0.12 0.33 0 1
Tenure 3 to 5 years 4,123 0.12 0.33 0 1
5 to 10 years 4,123 0.21 0.41 0 1
More than 10 years 4,123 0.52 0.50 0 1
Nativeness 4,906 0.85 0.353 0 1
Early-level indiv. contributor 5,070 0.01 0.10 0 1 Mid-level indiv. contributor 5,070 0.04 0.20 0 1
Level Senior indiv. contributor 5,070 0.06 0.23 0 1
First level manager 5,070 0.23 0.42 0 1
Mid-level manager 5,070 0.30 0.46 0 1
Senior management 5,070 0.37 0.48 0 1
Core sample 5,413 0.99 0.09 0 1
Export risk indicator 5,413 0.12 0.10 0 0.27
Part of a multinational 5,413 0.76 0.43 0 1
Foreign headquarters 5,413 0.49 0.50 0 1
Note. Mean for categorical variables refers to the proportions in each respective category and sample size N for refers to all non-missing data across all subcategories combined; Indiv. = Individual.
Correlations
Important are the low correlations between ‘Brexit’ and all other variables, as they attest to the exogeneity of the event (the referendum outcome). Further notable are the large
correlations between the ERIs and both (1) whether a manager works for a multinational organization and (2) whether the manager’s organization has a foreign headquarters. This indicates that these types of organizations tend to be active in sectors where the Brexit poses a relatively large risk to exports. Remarkably, these three variables also all correlate
positively and significantly with the Brexit dummy (ERI: rpb = .10, p < 0.01; Part of a
multinational: rpb = .13, p < 0.01; Foreign headquarters: rpb = .12, p < 0.01; these correlations
are similar for the larger sample of 5,413 managers). This indicates that those organizations for which the Brexit is hypothesized to be a relatively large threat are overrepresented in the sub-sample measured after the referendum as compared to the sub-sample before the
referendum. Given that these three variables are time-invariant, this points to the possibility of a selection effect, where the relatively threatened organizations are more likely to decide on a training for their managers than the organizations less threatened. I will return to this point in the discussion.
11. 60 or older -0.023 0.022 -0.019 -0.023 0.004 -0.002 -0.02 -0.07 -0.098 -0.063 12. Female 0.046 0.097 -0.02 0.009 -0.03 0.02 0.079 0.074 -0.034 -0.069 13. No formal manag. role -0.011 0.027 0.007 -0.006 0.013 -0.003 0.143 0.058 -0.049 -0.058 14. Less than 3 years 0.073 0.015 -0.002 0.017 -0.019 0.043 0.227 0.168 -0.103 -0.142 15. 3 to 5 years 0.031 -0.006 -0.022 -0.007 -0.016 -0.006 0.081 0.224 -0.109 -0.137 16. 5 to 10 years 0.023 -0.03 -0.022 -0.006 -0.016 -0.008 -0.075 0.195 0.009 -0.176 17. More than 10 years -0.083 0.01 0.032 0 0.032 -0.016 -0.186 -0.436 0.148 0.345 18. Nativeness -0.074 0.056 -0.016 -0.03 0.015 0.007 0.006 -0.063 0.011 0.04 19. Early-lvl indiv. contr. 0.017 0.007 0.006 0.002 0.004 -0.001 0.135 0.002 -0.029 -0.021 20. Mid-lvl indiv. contr. -0.004 0.055 0.022 -0.018 0.041 -0.003 0.055 0.057 -0.034 -0.048 21. Senior indiv. contr. 0.003 0.033 0.014 -0.015 0.029 -0.004 0.011 0.029 -0.001 -0.034 22. First level manager 0.108 0.028 0.031 -0.035 0.068 0.037 0.12 0.092 -0.062 -0.069 23. Mid-level manager 0.009 -0.006 -0.006 0.016 -0.023 -0.01 -0.005 0.101 -0.025 -0.068 24. Senior management -0.089 -0.052 -0.033 0.024 -0.059 -0.014 -0.13 -0.195 0.086 0.15 25. Export risk indicator 0.023 -0.051 0.097 0.087 0.008 -0.017 0.033 0.065 -0.044 -0.022 26. Part of multinational -0.037 -0.077 0.13 0.055 0.075 -0.024 -0.01 0.038 0.008 -0.032 27. Foreign headquarters -0.013 -0.018 0.12 0.101 0.016 -0.012 0.006 0.014 -0.002 -0.006 28. Brexit*ERI 0.015 -0.026 0.555 0.327 0.223 -0.013 0.049 0.045 -0.052 0.001 29. Brexit*PoaM -0.017 -0.045 0.697 0.354 0.339 -0.016 0.014 0.001 -0.009 0.011 30. Brexit*Foreign HQ -0.001 -0.006 0.438 0.271 0.162 -0.01 0.029 -0.002 -0.006 0.004 11 12 13 14 15 16 17 18 19 20 11. 60 or older 1 12. Female -0.013 1 13. No frml manag. role -0.018 0.076 1
14. Less than 3 years -0.041 0.076 -0.061 1
15. 3 to 5 years -0.021 0.078 -0.062 -0.14 1
16. 5 to 10 years -0.035 0.012 -0.085 -0.191 -0.195 1
17. More than 10 years 0.076 -0.134 -0.17 -0.384 -0.391 -0.536 1
18. Nativeness 0.042 -0.013 -0.033 -0.039 -0.062 -0.035 0.105 1
19. Early-lvl indiv. contr. -0.009 0.032 0.252 0.021 0.01 -0.032 -0.074 -0.002 1
20. Mid-level indiv. contr. 0.017 0.087 0.311 0.039 0.01 -0.017 -0.117 0.019 -0.015 1 21. Senior indiv. contr. 0.001 0.071 0.14 0.056 0.019 -0.021 -0.076 -0.041 -0.019 -0.047 22. First level manager -0.024 0.068 -0.011 0.316 0.112 -0.01 -0.269 0.029 -0.034 -0.081 23. Mid-level manager -0.027 0.019 -0.089 -0.012 0.093 0.112 -0.116 -0.009 -0.05 -0.12 24. Senior management 0.037 -0.137 -0.129 -0.265 -0.182 -0.074 0.394 -0.001 -0.07 -0.168 25. Export risk indicator -0.032 -0.135 0.008 -0.014 0.038 0.009 -0.026 -0.112 0.001 -0.002 26. Part of multinational -0.047 -0.201 -0.01 -0.049 0.01 -0.011 0.037 -0.127 -0.006 -0.032 27. Foreign headquarters -0.031 -0.121 -0.011 0.026 0.016 0.002 -0.026 -0.093 0.026 -0.03 28. Brexit*ERI -0.025 -0.075 0.008 -0.004 0.022 0.003 -0.017 -0.099 0.015 -0.006 29. Brexit*PoaM -0.03 -0.11 -0.003 -0.041 0.002 -0.015 0.039 -0.092 0.013 -0.021 30. Brexit*Foreign HQ -0.025 -0.067 0.002 0.013 0.009 0.004 -0.018 -0.089 0.03 -0.018 21 22 23 24 25 26 27 28 29 30
21. Senior indiv. contr. 1
23. Mid-level manager -0.161 -0.281 1
24. Senior management -0.225 -0.391 -0.583 1
25. Export risk indicator -0.054 0.048 0.022 -0.03 1
26. Part of multinational -0.058 -0.048 0.03 0.047 0.568 1
27. Foreign headquarters -0.043 0.04 0.023 -0.023 0.443 0.505 1
28. Brexit*ERI -0.033 0.042 0.021 -0.035 0.755 0.435 0.375 1
29. Brexit*PoaM -0.034 -0.014 0.022 0.012 0.404 0.652 0.382 0.719 1
30. Brexit*Foreign HQ -0.031 0.02 0.021 -0.017 0.379 0.408 0.803 0.575 0.627 1
Note. Significance levels not reported, but available on request. Correlations reported between leadership style and control
variables reported here differ slightly from those reported earlier, as this table only includes complete cases; Indiv. = Individual; Contr. = Contributor; Lvl = Level; ERI = Export risk indicator; PoaM = Part of a multinational; HQ = headquarters. Table was generated in the R language (R Core Team, 2019), using package CARET (Kuhn et al., 2019).
Analysis
In order to test my hypotheses I estimate a series of hierarchical linear (multi-level) models (HLMs; Castro, 2002; Schriesheim, 1995). These models hold two levels: the individual and the organization level. In doing so, I account for the fact that managers are nested within organizations (Kozlowski & Klein, 2000). Across the various organizations, differences in practices may make directive leadership more common for some organizations prior to the referendum. For instance, directive leadership behaviours may be more frequently observed for organizations in the construction sector as compared to the education sector. To this end I include in the HLMs a random intercept that varies across organizations. In the line of thinking of hypotheses 2a, 2b and 2c, I also assume that managers’ leadership response to the Brexit referendum may differ across organizations. This is captured by the inclusion of a random slope for the Brexit dummy, again varying across organizations.
Main results. Results of the estimation of a first set of HLMs are presented in table 4. Column (1) displays an intercept-only model that serves to assess how much of the variation in directive leadership can be explained by the knowledge that managers are nested in
between the individual managers. This is comparable to the variance explained by the
organization level reported in Stoker et al. (2019) and justifies the multi-level approach taken (see also Emmerik et al., 2010).
Columns (2) to (7) display my main findings. All specifications presented include the four control variables ‘Age’, ‘Gender’, ‘Tenure’ and ‘Nativeness’. In line with hypothesis 1, column (2) shows a significant positive effect of the Brexit referendum on directive
leadership (β = 0.053, p < .05). The magnitude of this coefficient is larger than the effect of the financial crisis on directive leadership (0.043) reported by Stoker et al. (2019) for a global sample. The authors also report an average yearly change in directive leadership of 0.014 for the period 2005-2014. An Equivalence test (Lakens, 2017) with a Cohen’s d of 0.8
(indicating a large effect) rejects the null hypothesis that the increase in directive leadership following the referendum is no different than this average yearly variation in directive leadership. Overall, this finding lends support to the threat-rigidity hypothesis (Staw et al., 1981), and the classification of the Brexit as a control-reducing threat (Chattopadhyay et al., 2001). The specifications in columns (3) and (4) allow for assessing in what period following the referendum the increase in directive leadership has taken place. Column (3) demonstrates that both a dummy for the year after the referendum (β = 0.054, p < .1) and a dummy for the subsequent year (β = 0.048, p < .1) are marginally significant and of similar magnitude. This indicates that the effect of the referendum is not driven by either of these periods in
particular, but instead has remained stable over time. The regression in column (4) confirms this by demonstrating no significant increase or decrease in directive leadership on top of the original effect of the referendum (β = 0.003, n.s.).
Columns (5) to (7) add interaction effects between the Brexit dummy and
hypotheses 2a, 2b, and 2c. Specifically, the effect of the referendum on directive leadership is not found to be significantly different for organizations with large exports to the EU (β = -0.323, n.s.), for multinational organizations (β = -0.013, n.s.), and for organizations with a foreign headquarters (β = 0.027, n.s.). The main effect of the Brexit dummy is now
interpreted as the increase in directive leadership for sectors with an ERI of 0 per cent (β = -0.078, p < .01), non-multinationals (β = 0.060, n.s.) and organizations with headquarters in the UK (β = 0.039, n.s.), respectively. Whilst there are no organizations in the sample with an ERI of 0 per cent, some are not far off (see Appendix F). For instance, for the 267 managers from 17 organizations in the Health sector (ERI = 0.23%), the estimated model in column (5) predicts an increase in directive leadership of 0.077 following the referendum. This indicates a larger increase in directive leadership for managers whose organization’s exports are at a relatively low risk due to the Brexit, and runs counter to hypothesis 2a.
Table 4
Results of Hierarchical Linear Model Estimations for Directive Leadership. Dependent variable: Directive Leadership
(0.593) (0.592) (0.593) (0.592) (0.593) (0.592)
60 or older -0.505 -0.508 -0.506 -0.513 -0.507 -0.498
(0.598) (0.598) (0.598) (0.598) (0.598) (0.598)
Female 0.045** 0.044** 0.045** 0.039* 0.036* 0.038*
(0.020) (0.020) (0.020) (0.021) (0.021) (0.021)
Less than 3 years 0.151** 0.150** 0.151** 0.135** 0.135** 0.137**
(0.064) (0.064) (0.064) (0.065) (0.065) (0.065)
3 to 5 years 0.078 0.077 0.077 0.060 0.061 0.062
(0.065) (0.065) (0.065) (0.065) (0.065) (0.065)
5 to 10 years 0.087 0.087 0.087 0.071 0.072 0.072
(0.063) (0.063) (0.063) (0.063) (0.063) (0.063)
More than 10 years 0.043 0.043 0.043 0.027 0.028 0.029
(0.063) (0.063) (0.063) (0.063) (0.063) (0.063) Nativeness -0.098*** -0.099*** -0.098*** -0.098*** -0.100*** -0.097*** (0.027) (0.027) (0.027) (0.027) (0.027) (0.027) Brexit*ERI -0.323 (0.303) Brexit*PoaM -0.013 (0.052) Brexit*Foreign HQ 0.027 (0.049) Constant 3.073*** 3.456*** 3.460*** 3.457*** 3.472*** 3.503*** 3.475*** (0.018) (0.596) (0.596) (0.596) (0.596) (0.596) (0.596) Observations 5,455 4,061 4,061 4,061 4,024 4,024 4,024 Log Likelihood -5,050.44 -3,690.51 -3,690.1 -3,690.47 -3,653.50 -3,653.45 -3,653.90 Akaike Inf. Crit. 10,106.880 7,415.014 7,422.207 7,422.942 7,344.997 7,344.904 7,345.806 Bayesian Inf. Crit. 10,126.690 7,522.271 7,554.700 7,555.435 7,464.697 7,464.604 7,465.507
Note: Standard errors in parentheses; ERI = Export risk indicator; PoaM = Part of a multinational; HQ =
headquarters; *p<0.1; **p<0.05; ***p<0.01. All estimations were performed in the R language (R Core
Team, 2019), using packages NLME (Pinheiro et al., 2019), LME4 (Bates et al., 2019), and STARGAZER (Hlavac, 2019).
Table 5
Results for Random Intercept Specifications.
Dependent variable: Directive Leadership
(1) (2) (3) (4) (5) (6) (7) Brexit 0.048** 0.046* 0.074** 0.060 0.038 (0.024) (0.026) (0.036) (0.043) (0.030) Brexit - Year 1 0.046* (0.026) Brexit - Year 2 0.049* 0.003 (0.028) (0.026) ERI 0.147 (0.303) PoaM -0.033 (0.052) Foreign HQ -0.018 (0.050) 20 - 29 -0.334 -0.334 -0.334 -0.356 -0.350 -0.341 (0.595) (0.595) (0.595) (0.594) (0.595) (0.594) 30 - 39 -0.401 -0.401 -0.401 -0.415 -0.409 -0.400 (0.593) (0.593) (0.593) (0.592) (0.592) (0.592) 40 - 49 -0.419 -0.419 -0.419 -0.433 -0.427 -0.418 (0.593) (0.593) (0.593) (0.592) (0.593) (0.592) 50 - 59 -0.406 -0.405 -0.405 -0.422 -0.418 -0.407 (0.593) (0.593) (0.593) (0.593) (0.593) (0.593) 60 or older -0.495 -0.495 -0.495 -0.505 -0.502 -0.491 (0.599) (0.599) (0.599) (0.599) (0.599) (0.599) Female 0.044** 0.044** 0.044** 0.039* 0.036* 0.038* (0.020) (0.020) (0.020) (0.021) (0.021) (0.021)
Less than 3 years 0.154** 0.154** 0.154** 0.137** 0.136** 0.139**
(0.064) (0.064) (0.064) (0.065) (0.065) (0.065)
3 to 5 years 0.077 0.077 0.077 0.060 0.061 0.061
(0.065) (0.065) (0.065) (0.065) (0.065) (0.065)
5 to 10 years 0.086 0.086 0.086 0.070 0.071 0.071
(0.063) (0.063) (0.063) (0.063) (0.063) (0.063)
More than 10 years 0.042 0.042 0.042 0.026 0.027 0.027
Note: Standard errors in parentheses; ERI = Export risk indicator; PoaM = Part of a multinational; HQ = Headquarters; *p<0.1; **p<0.05; ***p<0.01. All estimations were performed in the R language (R Core
Team, 2019), using packages NLME (Pinheiro et al., 2019), LME4 (Bates et al., 2019), and STARGAZER (Hlavac, 2019).
Participative leadership. Thus far, the results presented are in line with
characterizing the Brexit as a control-reducing threat (Chattopadhyay et al., 2001). Such a portrayal would be challenged if in addition to an increase in directive leadership we would also observe an increase in participative leadership. That is, the threat-rigidity hypothesis predicts managers to constrict control following a control-reducing threat, which runs counter to participative leadership behaviours (see Appendix B). Moreover, the correlations in table 3 show a strong negative relationship between directive and participative leadership.
Table 6 reproduces the specifications in table 4, changing only the dependent variable to participative leadership. Out of the total variation in participative leadership, 7.4% can be explained by the organization level, so that the residual 92.6% of variance is between the individual managers. Across all specifications, no change in participative leadership
following the referendum is observed. This reinforces the characterization of the Brexit as a control-reducing threat.
Table 6
Results of Hierarchical Linear Model Estimations for Participative Leadership. Dependent variable: Participative Leadership
(0.045) Foreign HQ 0.008 (0.044) 20 - 29 -0.906 -0.837 -0.842 -0.869 -0.835 -0.827 (0.561) (0.558) (0.558) (0.559) (0.557) (0.557) 30 - 39 -0.825 -0.767 -0.771 -0.800 -0.777 -0.769 (0.559) (0.556) (0.556) (0.557) (0.555) (0.555) 40 - 49 -0.824 -0.773 -0.777 -0.803 -0.786 -0.778 (0.559) (0.556) (0.556) (0.557) (0.555) (0.555) 50 - 59 -0.831 -0.776 -0.781 -0.808 -0.789 -0.780 (0.559) (0.557) (0.557) (0.557) (0.555) (0.555) 60 or older -0.794 -0.736 -0.740 -0.718 -0.694 -0.683 (0.564) (0.562) (0.562) (0.563) (0.561) (0.561) Female 0.103*** 0.095*** 0.095*** 0.098*** 0.092*** 0.095*** (0.019) (0.019) (0.019) (0.019) (0.019) (0.019)
Less than 3 years -0.080 -0.074 -0.076 -0.078 -0.075 -0.073
(0.060) (0.060) (0.060) (0.061) (0.060) (0.060)
3 to 5 years -0.092 -0.078 -0.080 -0.092 -0.080 -0.081
(0.061) (0.061) (0.061) (0.061) (0.061) (0.061)
5 to 10 years -0.118** -0.102* -0.104* -0.116* -0.102* -0.103*
(0.059) (0.059) (0.059) (0.059) (0.059) (0.059)
More than 10 years -0.070 -0.049 -0.050 -0.064 -0.043 -0.044
(0.059) (0.059) (0.059) (0.059) (0.059) (0.059) Nativeness 0.072*** 0.069*** 0.068*** 0.066*** 0.062** 0.066*** (0.025) (0.025) (0.025) (0.025) (0.025) (0.025) Brexit*ERI 0.062 (0.313) Brexit*PoaM -0.040 (0.052) Brexit*Foreign HQ 0.011 (0.049) Constant 4.489*** 5.306*** 5.227*** 5.236*** 5.317*** 5.267*** 5.236*** (0.016) (0.562) (0.560) (0.559) (0.560) (0.559) (0.558) Observations 5,455 4,061 4,061 4,061 4,024 4,024 4,024 Log Likelihood -4,573.07 -3,437.90 -3,424.55 -3,424.33 -3,389.75 -3,378.99 -3,379.94 Akaike Inf. Crit. 9,152.147 6,909.797 6,891.102 6,890.656 6,817.494 6,795.973 6,797.887 Bayesian Inf. Crit. 9,171.960 7,017.053 7,023.595 7,023.149 6,937.194 6,915.674 6,917.588
Note: Standard errors in parentheses; ERI = Export risk indicator; PoaM = Part of a multinational; HQ =
Headquarters; *p<0.1; **p<0.05; ***p<0.01. All estimations were performed in the R language (R
Core Team, 2019), using packages NLME (Pinheiro et al., 2019), LME4 (Bates et al., 2019), and STARGAZER (Hlavac, 2019).
0.71, rwg = 0.81). Overall, the threat that the Brexit poses to these managers should not be
larger than it is to managers in the UK. Consequently, my findings are strengthened in the case that I do not observe an increase in directive leadership of a magnitude similar to the UK.
Table 7 displays the directive leadership response to the French and German managers. Columns (2) and (6) show no change in directive leadership following the referendum for these managers. However, the magnitude for the coefficient for French managers is relatively large. Column (3) indicates that there was no change in directive leadership directly following the referendum, but that this effect is driven by the period that started a year after the referendum (presumably in response to the 2017 elections). Overall, these findings indicate that there was no change in directive leadership for French or German managers following the referendum.
Table 7
Results for French and German Managers.
Dependent variable: Directive Leadership
France Germany (1) (2) (3) (4) (5) (6) (7) (8) Brexit -0.052 -0.045 -0.016 0.005 (0.049) (0.047) (0.039) (0.043) Brexit - Year 1 -0.056 0.025 (0.046) (0.049) Brexit - Year 2 -0.097** -0.047 -0.031 -0.063 (0.049) (0.042) (0.043) (0.043) 30 - 39 -0.271 -0.296 -0.288 0.175* 0.167 0.160 (0.186) (0.185) (0.185) (0.106) (0.106) (0.106) 40 - 49 -0.348* -0.371** -0.362* 0.151 0.145 0.140 (0.187) (0.186) (0.187) (0.108) (0.108) (0.108) 50 - 59 -0.474** -0.49*** -0.483** 0.143 0.133 0.127 (0.190) (0.189) (0.190) (0.111) (0.111) (0.111) 60 or older -0.343 -0.351* -0.347 0.171 0.170 0.168 (0.213) (0.212) (0.213) (0.171) (0.170) (0.171) Female 0.071* 0.075* 0.074* 0.044 0.044 0.046 (0.042) (0.042) (0.042) (0.035) (0.035) (0.035)
Less than 3 years -0.042 -0.057 -0.048 -0.066 -0.068 -0.071
3 to 5 years -0.134 -0.146 -0.139 -0.066 -0.071 -0.070
(0.143) (0.143) (0.143) (0.084) (0.084) (0.084)
5 to 10 years 0.031 0.023 0.029 -0.028 -0.030 -0.030
(0.136) (0.136) (0.136) (0.081) (0.080) (0.081)
More than 10 years 0.058 0.049 0.056 -0.002 -0.006 -0.006
(0.135) (0.134) (0.135) (0.080) (0.080) (0.080) Nativeness 0.072 0.058 0.062 -0.16*** -0.17*** -0.16*** (0.065) (0.065) (0.065) (0.039) (0.039) (0.039) Constant 3.267*** 3.550*** 3.605*** 3.584*** 3.092*** 3.107*** 3.113*** 3.131*** (0.028) (0.229) (0.226) (0.227) (0.032) (0.123) (0.122) (0.124) Observations 1,420 1,271 1,271 1,271 1,981 1,565 1,565 1,565 Log Likelihood -1,347.6 -1,206.2 -1,205.1 -1,204.7 -1,662.7 -1,299.2 -1,295.8 -1,296.4 Akaike Inf. Crit. 2,701.21 2,444.42 2,450.22 2,449.31 3,331.39 2,630.36 2,631.51 2,632.77 Bayesian Inf. Crit. 2,717.0 2,526.7 2,553.2 2,552.3 3,348.2 2,716 2,738.6 2,739.9
Note: Standard errors in parentheses. The reference category for ‘Age’ is 20 – 29, as there were no managers of 19 or younger present in the French or German sample. *p<0.1; **p<0.05; ***p<0.01. All estimations
were performed in the R language (R Core Team, 2019), using packages NLME (Pinheiro et al., 2019), LME4 (Bates et al., 2019) and STARGAZER (Hlavac, 2019).
Level of Management. As stated, a wealth of literature exists on the relationship between level of management and leadership style. Furthermore, in the UK data a strong negative correlation between level of management and directive leadership was found (rτ =
-.09, p < 0.01). For this reason, I explore the possibility that the earlier found increase in directive leadership is primarily driven by managers of a certain level. In table 8, column (2), I add ‘Level’ as a control variable to the earlier specification. In this specification the
predicted increase in directive leadership is 20 per cent lower and is only marginally
category (N = 1,511), and (3) a group of high-level managers forming ‘Senior management’ (N = 1,852), which is the sixth category in table 2. The analysis for the low-, middle- and high-level manager samples are displayed in columns (3), (4) and (5), respectively. These columns show that the increase in directive leadership following the referendum is driven by the sub-sample of low-level managers (β = 0.114, p < .01). Remarkably, the coefficient for this subgroup is 215 per cent the size of the coefficient for the pooled sample.
Table 8
Including Level of Management into the Analysis.
Dependent variable: Directive Leadership
Low-level Mid-level High-level
(1) (2) (3) (4) (5) Brexit 0.043* 0.114** 0.010 -0.011 (0.024) (0.051) (0.044) (0.036) 20 - 29 -0.313 -0.528 (0.592) (0.571) 30 - 39 -0.365 -0.621 -0.071 -0.140 (0.590) (0.568) (0.109) (0.185) 40 - 49 -0.385 -0.632 -0.118 -0.177 (0.590) (0.568) (0.112) (0.186) 50 - 59 -0.375 -0.607 -0.093 -0.174 (0.590) (0.570) (0.119) (0.187) 60 or older -0.463 -0.957 0.015 -0.253 (0.596) (0.595) (0.234) (0.214) Female 0.039* -0.072* 0.062 0.103*** (0.021) (0.038) (0.038) (0.031)
Less than 3 years 0.148** 0.133* 0.041 -0.023
(0.069) (0.069) (0.277) (0.258)
3 to 5 years 0.109 0.097 0.041 -0.099
(0.070) (0.074) (0.275) (0.250)
5 to 10 years 0.134* 0.172** 0.010 -0.082
(0.069) (0.075) (0.274) (0.245)
More than 10 years 0.109 0.126* 0.003 -0.103
(0.069) (0.076) (0.274) (0.244)
Nativeness -0.103*** -0.143*** -0.105** -0.087**
(0.027) (0.052) (0.050) (0.040)
Mid-lvl indiv. contr. -0.190
(0.132)
Senior indiv. contr. -0.226*
(0.129)
First level manager -0.116
Mid-level manager -0.220* (0.127) Senior management -0.280** (0.127) Constant 3.073*** 3.613*** 3.707*** 3.209*** 3.322*** (0.018) (0.604) (0.574) (0.289) (0.307) Observations 5,455 4,004 1,026 1,177 1,801 Log Likelihood -5,050.437 -3,620.034 -904.244 -1,088.277 -1,632.990
Akaike Inf. Crit. 10,106.880 7,284.068 1,842.487 2,208.554 3,297.980
Bayesian Inf. Crit. 10,126.690 7,422.559 1,926.355 2,289.686 3,385.918
Note: Low-level = Early-level individual contributor, Mid-level individual contributor, Senior individual contributor, & First level manager. Mid-level = Mid-level manager. High-level = Senior
management. Standard errors in parentheses. *p<0.1; **p<0.05; ***p<0.01. All estimations were
performed in the R language (R Core Team, 2019), using packages NLME (Pinheiro et al., 2019), LME4 (Bates et al., 2019) and STARGAZER (Hlavac, 2019).
Figure 1
Change in Directive Leadership following the Brexit Referendum by Level of Management.
Note:Figure was constructed in the R language (R Core Team, 2019), using packages DPLYR (Wickham, François, Henry, & Müller, 2019), and GGPLOT2 (Wickham et al., 2019).
Discussion
The current study demonstrates the leadership consequences of the UK vote to leave the EU. My results show that the referendum outcome has caused an increase in directive leadership for a large sample of managers covering a wide range of sectors. From this I conclude that these managers perceive the Brexit as a control-reducing threat (Chattopadhyay et al., 2001). This conclusion is strengthened by the fact that the increase in directive
supports the assertion that the Brexit primarily poses a threat the managers in the UK. Overall, my findings are in line with predictions from the threat-rigidity hypothesis (Staw et al., 1981) and serve as further evidence for the link between the framework by
Chattopadhyay et al. (2001) and the field of leadership (Kamphuis et al., 2011).
My contributions are as follows. First of all, I provide unique insights in the response inside organizations following the Brexit referendum outcome. In doing so, I take a further step in demonstrating how the context can serve as an antecedent to leadership, and provide additional support for using events as a level of analysis that allows for capturing the
dynamics in leadership behaviour (Dinh & Lord, 2012; Hoffman & Lord, 2013; Johns, 2006; Shamir, 2011). In addition, I strengthen the case for considering leaders in the multi-level context in which they lead, including the macro-context. As such, I answer the call for more leadership research on events external to an organization (Morgeson et al., 2015; Osborn et al., 2002; Staw, 2016) and assist in bridging the micro-macro divide (Aguinis et al., 2011). Theoretical implications
Timing and duration of the threat. A key difference between the current study and the study by Stoker et al. (2019) revolves around the timing and the duration of the threat under consideration. Whereas the threat from the financial crisis emanated from its
immediate negative impact, the Brexit referendum outcome marked the start of a long-lasting period of impending negative consequences. A crucial difference is therefore that, in contrast to the crisis, the threat from the Brexit results primarily from the anticipation of these
