Synthesis Paper

A method for measuring destructive leadership and identifying types of destructive leaders in organizations

James B. Shaw a,⁎, Anthony Erickson a, Michael Harvey a,b

a Bond University, Queensland 4229, Australia b University of Mississippi, University, MS, USA

a r t i c l e i n f o a b s t r a c t

Available online 25 May 2011 This study describes the development of a measure of the nature of destructive leadership in organizations. We then use scales developed from that measure in a cluster analysis to empirically derive a behavior-based taxonomy of destructive leaders. Data were obtained through a web-based survey that generated 707 respondents. Based on follower perceptions, the results identified seven types of destructive leaders using behavior-focused scales. An interesting discovery was that most of the types of destructive leaders identified were not “all destructive” but rather perceived as extreme on just one or two characteristics.

© 2011 Elsevier Inc. All rights reserved.

Keywords: Dysfunctional leadership Destructive leader typology Destructive leadership

A Method for Identifying the Prevalence and Types of Destructive Leadership in Organizations

“Leaders are not always interested in effecting change for the purpose of benefiting the organization and its members as a whole: rather, the leader may be more interested in personal outcomes.” (O'Conner, Mumford, Clifton, Gessner, & Connelly, 1995)

The history of leadership research appears to have been dominated largely by an attempt to understand “good” or “effective” leadership. Trait approaches to leadership (for reviews see Bass, 1990; Judge, Bono, Ilies, & Gerhardt, 2002; Kirkpatrick & Locke, 1996; Lord, DeVader, & Alliger, 1986; Stogdill, 1948.), Behavioral approaches (Fleishman, 1953; Hemphill & Coons, 1957; Katz, Maccoby, & Morse, 1950), Contingency theories (Fiedler, 1972; Hersey & Blanchard, 1969; House, 1971; Kerr & Jermier, 1978; Vroom & Jago, 1988; Vroom & Yetton, 1973), dyadic theories (Dansereau, Graen, & Haga, 1975; Dienesch & Liden, 1986; Graen & Cashman, 1975, Graen & Uhl-Bien, 1995), Neo-Charismatic theories (Bass, 1985; Bennis & Nanus 1985; Burns, 1978; Conger & Kanungo, 1987; House, 1977; Sashkin, 1988; Shamir, House, & Arthur, 1993; Trice & Beyer, 1986), social network theory (Balkundi & Kilduff, 2005) and complexity theory (Marion & Uhl-Bien, 2001) all examine the various requirements for effective leadership.

However, the occurrence of destructive leadership (i.e., the “systematic and repeated behavior by a leader, supervisor, or manager that violates the legitimate interest of the organization by undermining and/or sabotaging the organization's goals, tasks, resources, and effectiveness and/or the motivation, well-being or job satisfaction of subordinates“ as defined by Einarsen, Aasland, & Skogstad, 2007, p. 20) in complex organizations would seem to be an important consideration for top management (Einarsen & Raknes, 1997). Destructive leadership can have detrimental effects on productivity (Keelan, 2000), the financial bottom-line (Field, 2003), and employee morale (Olafsson & Johannsdottir, 2004). Yet, destructive leadership is often misdiagnosed and/or mismanaged once identified in organizations. Many victims of poor leadership suffer from a form of social stress that is similar in nature to post-traumatic stress syndrome that can have a debilitating impact on the individual (Leymann & Gustafsson, 1996; Wilson, 1991). Thus, the follower of a dysfunctional leader can suffer social, psychological, and psychosomatic effects, which can

The Leadership Quarterly 22 (2011) 575–590

⁎ Corresponding author at: Faculty of Business, Technology & Sustainable Development, Bond University, Robina, Queensland 4229, Australia. E-mail address: [email protected] (J.B. Shaw).

1048-9843/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.leaqua.2011.05.001

Contents lists available at ScienceDirect

The Leadership Quarterly

journal homepage: www.elsevier.com/locate/leaqua

manifest themselves in a negative impact on the individual's self-efficacy and ability to perform his/her job (Einarsen, 1999; Einarsen & Raknes, 1997).

As a result, there has been a growing interest by leadership scholars in examining the “darker” side of leadership (Hogan & Hogan, 2001). A variety of research streams have examined such topics as bullying, (Ferris, Zinko, Brouer, Buckley, & Harvey, 2007; Harvey, Treadway, & Heames, 2007), toxic leadership (Lipman-Blumen, 2006), abusive supervision (Tepper, 2000), bad leadership (Erickson, Shaw, & Agabe, 2007; Kellerman, 2004), narcissistic leadership (Paunonen, Lönnqvist, Verkasalo, Leikas, & Nissinen, 2006) and destructive leadership (Einarsen et al., 2007). Such approaches to the examination of the dark side of leadership have been labelled collectively as “destructive” leadership.

In combination, the above approaches would seem to represent a taxonomy of destructive leaders. However, there has been no attempt to date to empirically substantiate the differences between leadership variously described as destructive, toxic, bad, bullying, or abusive. Kellerman (2004) identified seven types of bad (i.e., destructive) leaders: (1) Incompetent; (2) Rigid; (3) Intemperate; (4) Callous; (5) Corrupt; (6) Insular, and (7) Evil. However, this typology is based on case studies and anecdotal information. While Einarsen et al. (2007) developed a conceptual typology that identified supportive-disloyal leadership, derailed leadership and tyrannical leadership, this categorization scheme also remains empirically untested.

One of the problems associated with developing an effective typology of destructive leadership is the lack of a broad ranging measure of destructive leadership in organizations. While researchers have described and assessed a number of negative leader characteristics and behavior including abusive supervision (Tepper, 2000), bullying (Notelars, Einarsen, De Witte, & Vermunt, 2006), narcissistic leadership (Rosenthal & Pittinsky 2006), toxic leadership (Lubit, 2004; and Padilla, Hogan, & Kaiser 2007), and destructive leadership (Schaubroeck, Walumbwa, Ganster, & Kepes 2007; and Sommers, Schell, & Vodanovich 2002) it is likely that “destructive leadership” may possess a number of traits associated with each of these distinct types of negative leadership.

The following study attempts to address the measurement and taxonomic deficits outlined above by building on the work of Erickson et al. (2007). Our examination of such leadership is consistent with the definition provided by Einarsen et al. (2007), that is, that destructive leadership is defined as a systematic and repeated set of behaviours by a leader that have a significant negative (i.e., destructive) impact on organizational and/or employee outcomes. However, it is also important to note that our approach is one that defines destructive leadership specifically in terms of the perception of subordinates. This is important since subordinate perception of leader behavior may differ from managers higher-up in the organization. We are taking a cognitive categorization approach to leadership. We are attempting to define the prototypical attributes of a particular cognitive schema (category)—that of the “Destructive Leader”—and then seeking to understand what distinctions (sub-categorizations) followers make among such leaders (see Crocker, Fiske, & Taylor, 1984 for a discussion of the basic processes underlying cognitive categorization). This cognitive categorization approach to leadership is consistent with earlier work by Lord, Foti, and Phillips (1982), Lord (1985), and Cronshaw and Lord (1987). Lord et al. (1982) stated that “perceiving someone as a leader involves a relatively simple categorization (leader/non-leader or leader/follower) of the stimulus person into already existing categories.” (p. 104). In our current study, the categorization issue is whether a subordinate labels his/her superior as a “good leader or destructive leader.” As noted by Shaw (1990, p. 628):

“… the labelling of an individual as a leader is important because someone recognized as a leader can gain social power and influence. These people are more likely to be viewed as responsible for causing positive group outcomes. If individuals are labelled as non-leaders, they may have difficulty in influencing the behavior of subordinates and may not receive credit for positive group outcomes for which they are actually responsible.”

If we were to substitute the label “good leader” for “leader” and “destructive leader” for “non-leader” in the previous sentence, the implications are identical. Individuals categorized as destructive leaders may be unable to gain social power and influence over subordinates, are more likely to be linked to negative individual and group outcomes and may not be given credit for positive outcomes for which they are actually responsible. As mentioned earlier, previous research on the “dark side” of leadership also shows that destructive leadership may cause a number of direct negative consequences on employees and ultimately on the organization.

Using qualitative data collection methods, Erickson et al. (2007) identified a variety of personal characteristics and behaviors which were associated with the perception by subordinates of their superiors as “bad leaders.” In the current study, we have utilized these results along with information gleaned from other research on the “dark side” of leadership to develop a survey instrument to identify prototypical attributes of destructive leaders and to also examine the extent to which these attributes can further be used to identify sub-categories of destructive leaders.

1. Method

The data for this study came from a convenience sample and were collected using a web-based survey. While there are a number of limitations in using such a sample, which will be discussed at length later in this paper, we felt that using a web-based sample would gain us access to a large number of respondents, industries and types of leaders. In particular, since one of our goals was the investigation of the factor structure of a survey of destructive leadership with over 100 items, we needed a large number of respondents. A “beta” version of the Destructive Leadership Questionnaire (DLQ) contained items that asked respondents to rate their current leader (i.e., their current immediate supervisor, regardless of whether their leader was a good, destructive or average leader) as to the extent the leader had characteristics or engaged in behaviors associated with destructive leadership. Another set

576 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

of questions asked respondents to indicate the age, gender, nationality, and position title of their current leader. A final set of questions asked respondents to indicate their own age, gender, nationality, and education level. The survey was made available on the website for approximately 3 months with 707 individual respondents completing all or part of the questionnaire.

2. Survey construction

The construction of the DLQ represented a continuation of a qualitative study by Erickson et al. (2007). As indicated earlier, the focus of this paper is on the development of a measure of destructive leadership as viewed from a subordinate “cognitive schema” perspective. With this perspective and using a qualitative, web-based survey, Erickson et al. (2007) collected behavioral examples of destructive leadership from 240 respondents. Seven hundred and sixty-seven specific examples were collected of behaviors and other characteristics which respondents indicated had caused them to classify their leader as a “bad leader.” These 767 behavioral examples were then classified into eleven major categories: (1) Autocratic Behavior; (2) Poor Communication; (3) Unable to Effectively Deal with Subordinates; (4) Poor Ethics/Integrity; (5) Inability to Use Technology; (6) Inconsistent/Erratic Behavior; (7) Poor Interpersonal Behavior; (8) Micromanagement; (9) Poor Personal Behavior; (10) Excessive Political Behavior and (11) Lack of Strategic Skills.

In addition to the results of this qualitative study, work by Kellerman (2004) and research on leader bullying, narcissistic leadership, toxic leadership, and destructive leadership were examined as sources of potential items for the DLQ. The intent of item construction was to yield a set of items which encompassed a very broad range of personal and behavioral aspects of destructive leadership. Draft questions for the web-based survey were constructed by the authors and then reviewed by academic colleagues. This process yielded one hundred twenty-seven questionnaire items.

The web survey was constructed so that responses to the 127 items were made by ticking a response category with six responses ranging from Strongly Disagree to Strongly Agree. A response of “don't know” was also provided on each item. Four of the 127 items asked respondents to indicate an overall judgment of their current supervisor,(e.g., My boss is a terrible boss to work for; Of all the bosses I have known, my boss is one of the very best {reverse scored}). Nineteen items related to broader personal characteristics of the leader, (e.g., my boss is compulsive; my boss is arrogant; my boss is self-centred; or my boss is lazy). The remaining 104 items focused on fairly specific behaviors in which a leader might engage, (e.g., my leader often makes knee jerk reactions to problems or; my leader often takes credit for the work that others have done).

For all 127 items, responses were reverse scored where necessary so that a high rating indicated a high level of destructive leadership behavior or characteristic. One additional item asked respondents to rate their overall perception of their current supervisor by indicating a number from 1 to 100, where “1=the absolute WORST leader you could possibly imagine working for, and 100=the absolute BEST leader you could possibly imagine working for” (from now on this item will be referred to as the Worst-Best Leader item). Two other items asked respondents to provide demographic information on their current supervisor and themselves using a “type in the box” response format.

A graduate assistant organized the placement of the questions onto the host website. A pilot test of the questionnaire was then conducted using a convenience sample of academic colleagues, administrative staff, post-graduate, and undergraduate students to ensure that the website worked effectively and that the wording of the items was clear and understandable. Some adjustments to item wording were made prior to the official opening of the survey website to the public.

3. Elicitation of survey respondents

A variety of means were used to encourage participation in the study. Announcements of the study were posted on several international professional list server websites (e.g., Emonet, HRNet). Announcements were also sent to destructive leadership- related websites (e.g., Badbossology.com). A description of the study and an invitation to participate was sent to local community newsletters and to the alumni newsletter of our university. In addition, “snowball sampling” was used. Staff and students at our university were asked to announce the study to their colleagues, family, and friends who in turn were asked to make the presence of the survey known to their contacts. This process was quite successful in increasing the number and breadth of our respondents. Press releases about the study were sent out by our university's public relations department to major news organisations as well. Announcements on local radio stations were used and one of the authors was interviewed by a local radio station as well as by some radio stations in the USA. Thanks to publicity surrounding the results of the Erickson et al. (2007) study—Reuters Newsagency picked up the story and it appeared in over 600 websites and newspapers around the world—we had made contact with a number of HR managers, HR consultants and others interested in the topic. We contacted these individuals and many of them agreed to spread the word of our study to their colleagues and clients.

4. Sample

Seven hundred and seven respondents completed at least some of the questions in the survey. Except for a single item (n=382), sample size for individual items ranged from 501 to 691. Five hundred and fifty respondents provided information about their age with the age of the current leader indicated in 553 cases. The average respondent was 43.8 years old (Std. Dev.=11.6) with a range of 19 to 76 years. Average age of the current leaders described by respondents was 49.1 (Std. Dev.=8.9) with a range of 24 to 74 years. Data on the gender of current leaders and respondents were available for 556 and 555 cases respectively. While the majority of the current leaders described were male (54.9%), the majority of respondents to the survey were female (63.6%). Nationality data were available

577J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

for 525 leaders and 523 respondents. Most respondents were from Australia (53.9%) or the United States (22.9%), as were their leaders (Australian=50.1%, US=20.4%). The remaining respondents represented 28 nationalities while the remaining current leaders represented 41 different nationalities.

Data on education level and professional work experience were available for 536 and 547 respondents respectively. Respondents were highly educated, with 92.3% having university undergraduate or postgraduate degrees. Respondent professional work experience averaged 20.1 years (Std. Dev.=11.2) and ranged from 1 to 46 years. While we cannot claim that the sample is representative of any particular population, the current leaders rated by respondents represented a wide variety of fairly senior positions. Data on the position title of the current leader were available for 514 cases. Of these leaders, 63.6% were in positions with titles that could be categorized as CEO/Managing Director, Vice President, Director, Executive Officer, Senior Partner, General Manager, Functional Manager, Unit Manager, Department Manager, Policy Advisor or Senior Consultant. Academic managers such as President, Vice Chancellor, Dean, or Department Head represented 17.4% of the current leaders. Lower level managers such as supervisors or team leaders comprised only 4.3% of the current leader sample with the remaining leaders representing a range of generally mid-level managerial positions. Respondents who were the immediate subordinate of these leaders had a wide variety of position titles.

5. Analyses and results

As noted earlier in the paper, our primary goals in this study were to (1) identify the prototypical attributes underlying the categorization of a leader as a “destructive leader” by subordinates, (2) develop scale items to reliably measure these attribute prototypes, and (3) to use the scale measures to identify a typology of destructive leader sub-categories within our sample. All items in the study were first recoded so that a high score on the item represented a high level of destructive leadership behavior or characteristics. One exception to this was the Worst-Best Leader item with a scale ranging from 1 to 100. This item was not recoded—thus, a high score indicates good leadership. Basic descriptive analyses, factor analyses, reliability analyses, and cluster analysis techniques were used in the pursuit of our study objectives.

6. Descriptive statistics

Means and standard deviations were calculated for each of the items in the DLQ. Given the large number of items involved, only summary data will be provided in this report. For all items except the single Worst-Best Leader item, responses ranged from one to six, with item means ranging from 2.12 (Item 14, My boss is lazy.) to 3.99 (Item 5, When my boss wants something, he/she obsesses about it). Item standard deviations ranged from 1.28 (Item119, You can rarely predict how my boss is likely to behave.) to 1.93 (Item 102, “My boss is not a very good boss”). Both the average and median item standard deviation were 1.63. For the Worst-Best Leader item, the mean was 60.59 (Std. Dev.=29.39). Note that in two cases, respondents ignored our rating instructions on this item and gave their current leader a negative rating, (e.g., −20). We arbitrarily assigned a rating of zero for those two cases, given that the respondent felt so strongly negative about their leader that they were not willing to even rate them as a 1.0. This basic descriptive data seems to indicate that respondents used the full range of response options across all items with sufficient variance in their ratings for subsequent analyses.

7. Item intercorrelations

As with our basic descriptive statistics, given the large number of items involved, only summary data will be provided in this report. Due to the large sample size in our study, individual correlations between items were, with only a very few exceptions, statistically significant. Item intercorrelations ranged from moderately negative, (e.g., -.12) to highly positive (e.g., .84). A very large portion of the item intercorrelations were positive and in the magnitude range of .40 to .70. This indicated both the possibility of significant common method variance as well as the possibility of a “halo effect” embedded within our respondents’ rating. This high level of item collinearity presented a significant challenge for conducting our further analyses. However, in a later section of this paper on Factor Analyses Based on Standardized Residual Item Ratings, we describe an approach that we believe has allowed us to identify the underlying structure of these items and accomplish the primary goals of our study.

8. Factor analyses

The primary goal of our study was to identify the underlying attributes of destructive leadership as perceived by subordinates. Although each of the items on the DLQ might be considered a behavioral or personal “attribute” of a destructive leader, from a practical perspective this large number of potential attributes (items) was not desirable. We needed to identify the “essential attribute factors” underlying subordinate categorizations of their leaders as destructive leaders. To examine the factor structure of our items we chose to use SPSS principal components analysis (PCA) with oblimin (oblique) rotation. We chose principal components analysis due to the exploratory nature of our study. We had no underlying theoretical model that would specify a particular factor structure which we could then test using confirmatory factor analysis techniques such as those found in structural equation modelling. Further, our primary purpose at this stage in the development of the DLQ was data reduction—a purpose for which principal components analysis is well suited. We used oblique rotation since we had no reason to believe that the

578 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

underlying factors within our data should be orthogonal to one another. In fact, there was every reason to believe that factors would be non-orthogonal. As noted by Costello and Osborne,

“Conventional wisdom advises researchers to use orthogonal rotation because it produces more easily interpretable results, but this is a flawed argument. In the social sciences we generally expect some correlation among factors, since behavior is rarely partitioned into neatly packaged units that function independently of one another. Therefore, using orthogonal rotation results in a loss of valuable information if the factors are correlated, and oblique rotation should theoretically render a more accurate, and perhaps more reproducible, solution” (2005, p. 3).

Pair-wise deletion of missing data was used in all of the factor analyses. As noted earlier, on each of the items in the DLQ, a “don't know” option was available to respondents. Although this response was not used extensively by any one respondent or on any one item (except Item 126), respondents did use this option to some extent and the response was widely scattered throughout the items. A “don't know” response was treated as missing data, since the response indicated that the respondent did not legitimately feel comfortable in making a rating on a particular aspect of the leader's characteristics or behavior. Analysis of missing and “don't know” responses indicated that a list-wise deletion of data would yield a sample size of less than 160. Given the large number of items in the DLQ, this was not feasible for the purposes of our analyses. With pair-wise deletion of data in the correlation matrix used in the factor analyses, each item intercorrelation was based on generally between 530 and 550 cases. Only one set of item intercorrelations had nb500. This exception was Item 126 (When my boss has to punish someone, the punishment is always appropriate for the offense, reverse scored) for which data were available for only 382 cases. Most of the data missing for Item 126 was due to a “don't know” response. Factor analyses were conducted on the data set and are described below. For all analyses, four items assessing the overall judgment of respondents about their current leader as well as the Worst-Best Leader Item were deleted from the factor analyses.

9. Factor analyses based on Raw item ratings

Two principal components analyses (PCA) with oblimin rotation were conducted using the raw item ratings for 104 “behavior- focused items” and then for the 19 “personality items” of the DLQ. We conducted the analyses separately on the behavior-focused and personality-oriented items after an initial PCA with all items indicated a rather confused and difficult to interpret factor structure. This made sense in that the personality-oriented items tended to represent a potentially broad range of behaviors found in the behavior-focused items. For example, Item 39: “My boss could best be described as mean” could involve a variety of behaviors found in the behavior-focused items, e.g., Item 126: “The punishment my boss gives is often inappropriate for the offense,” or Item 26: “My boss wants to dominate/control everything,” or perhaps Item 75: “My boss blames others for his/her own mistakes.” Having both personality and behavior items in the same PCA was, in reality, introducing a potentially significant amount of collinearity among these items that was not helpful in identifying the underlying factor structure. This difficulty with using both personality and behavior items in a single PCA was evidenced in initial analyses using raw items scores as well as with residual scores (as described in the next section of this paper).

Given the pattern of intercorrelations among raw item ratings noted above, we expected and found that for both sets of items in the PCA analysis results were dominated by a single general factor. In the case of the personality items, three factors with eigenvalues greater than 1.0 were extracted that accounted for a total of 66.2% of the variance. Of the total variance accounted for, 50.3% was due to the first factor extracted. In a similar fashion, the results using the 104 behavior-focused items yielded 10 factors with eigenvalues greater than 1.0 accounting for a total of 72.3% of the variance. Of this total, 55% of the variance was explained by the first factor extracted. In both sets of analyses, the remaining factors appeared somewhat conceptually fragmented in nature and appeared to add little to our understanding of the underlying dimensions of destructive leadership. The identification of such large general factors is consistent with our categorization view of destructive leadership. Cognitive categorization theory suggests that once an individual perceives a person exhibit a small number of attributes prototypical of a particular category, they are then likely to assume that the individual exhibiting those few attributes possesses all attributes associated with the category (Crocker et al., 1984). While consistent with a cognitive categorization view of destructive leadership, in order to identify the attributes underlying the overall classification of the leader as a destructive leader, an alternative analysis approach was needed.

10. Factor analyses based on standardized residual item ratings

In an effort to tease out the underlying structure of the items of the DLQ, it was decided to remove from each item the variance predicted by the respondent's overall “good-destructive leader” judgment of their current leader. To do this, we regressed the Worst-Best Leader rating provided by each respondent onto their ratings for each of the 104 behavior-based and 19 personality- oriented DLQ items. A standardized residual score was created for each of the 123 items. This residual score represented each item rating with the variance attributable to the respondent's overall judgment (categorization) of their leader removed. These standardized residual scores were then used as the basis for two principal components factor analyses with oblimin rotation—one using the behavior-focused items and the second using the personality-oriented items. The uses of separate PCAs for the behavior and personality items was done here for the same reasons we discussed in the previous section on raw item analyses.

The use of these residual scores as the basis for our PCA procedures is certainly not without its problems. The removal of the “overall liking/disliking” of their leader from the respondents' ratings is, without doubt, removing a potentially important variable

579J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

from our study. From a cognitive categorization perspective, feelings of liking/disliking may well contribute to the overall categorization of a leader as either a “good leader” or a “destructive leader.” As noted earlier in the paper, this good-destructive categorization may have a substantial impact on the leader's ability to effectively lead his or her subordinates. However, from a practical perspective, building a measure of destructive leadership around a categorization process based on a generalized feeling of liking/disliking is of little use to an organization. Such a measure would be of little use in directing organizations as to what training or perhaps even selection criteria would be useful in reducing the level of destructive leadership in an organization. Thus, some process of identifying the underlying cause of those general feelings of liking/disliking is necessary. Our residual score approach is one such method.

Quite appropriately, these residual scores might be seen as “black boxes,” the contents of which cannot be clearly observed and understood. However, in discussing the use of the residual score approach with colleagues, we concluded that the value of these scores would be determined by the nature of the factors which resulted from our PCA procedures. If the residual scores truly reflected “unknowables” representing a mixture of concepts within each score, then one would expect the PCA analyses to yield a corresponding illogical jumble of factors. The results of our two PCA analyses are shown in Table 1. As seen in this table, this was not the case. Factors based on residual items scores “made sense.” There is also a level of convergent and discriminant validity among the resulting attribute scales scores as will be discussed later in this paper. For the 104 behavior- focused items, the factor analysis extracted 22 factors with eigenvalues greater that 1.0 accounting for 64.3% of total variance. For the 19 personality-oriented items, the factor analysis extracted 4 factors with eigenvalues greater that 1.0 accounting for 52.5% of total variance. In Table 1 are the pattern matrix loadings for the items within each of the factors along with the name of that factor.

One of the issues we wanted to explore was whether the factor structure of items extracted using the entire data set, which included data from respondents who rated good, destructive, and average leaders, would be different if only data relating to destructive leaders was included in the factor analysis. To examine this issue, we trichotomized the sample into three groups using the average item rating for the four items in the DLQ that measured the respondent's overall judgment as to how good or destructive a leader their current leader was (coefficient alpha for this Overall Good-Destructive scale was .96). The average Overall Good-Destructive scale score for leaders in this “destructive leader” group was 5.44 (Std. Dev.=.54) and their average rating on the Worst-Best Leader item was 28.2 (Std. Dev.=19.3). Given that scores on the Overall Good-Destructive Leader scale could range from 1.0 to 6.0 with 6.0 representing a really destructive leader and that the rating on the Worst-Best Leader items went from 1 (worst leader I could possibly work for) to 100 (best leader I could possibly work for), the leaders in this subsample were very destructive leaders as viewed by their subordinate.

Unfortunately, there were only 232 “destructive leaders” who fell into the worst 1/3 of leaders in the sample. With 104 behavioral-focused items, and thus a very small sample size to variable ratio, the results of such an analysis would be highly unstable. We did, however, conduct a factor analysis on “destructive leader” data using the 19 personality-oriented items. As with the total sample results, four factors were extracted accounting for 53.6% of total variance, i.e., almost identical to that found in the total sample. In addition, the pattern coefficient matrix showed a factor structure almost identical to that found in the total sample. Thus, while we were unable to examine the factor structure of the 104 behavior-focused items, there is evidence based on the personality-oriented items that the factor structure identified in the total sample may have external generalizability.

Constructing destructive leadership attribute scales

The PCA analyses provided the basis for constructing scales using DLQ items to measure different prototypical attributes of destructive leadership as perceived by subordinates. Due to the widely scattered nature of missing data and the “don't know” response, the use of factor scores as measures of the destructive leadership attributes was impractical. Due to missing data on individual items as we discussed earlier, factor scores could be calculated for less than 160 cases. Thus, the DLQ scales created were based simply on the average of the individual items that had loaded on a factor. In fact, two sets of DLQ scale scores were calculated for each attribute factor. The first scale score was based on the raw item ratings while the second scale score used the standardized residual scores for items within the factor. In the cases of Factors 11 and 15 which contained only a single item, these factors were dropped from all subsequent analyses in this study. In future research, we plan to write additional items for these two dimensions with the hope of adding reliable multi-item measures of these two dimensions. In addition, Item 51 in Factor 16 and Item 47 in Factor 19 were dropped from scale calculations since both items did not seem to fit conceptually with other items in the factor and their loadings on the factors were in the opposite direction of all other items within the factor.

The SPSS scale reliability program was used to calculate coefficient alpha internal consistency reliability for all scales. The results for both the raw item and residual scales are found in Table 1 in parenthesis next to the factor name. The first number represents the coefficient alpha (decimal points omitted) for the raw items and the second that for the residual scores. Reliabilities for all scales using raw item ratings were consistently high. This was not unexpected given the potential common method variance and “cognitive categorization effect” that we believed were incorporated within these ratings. For scales calculated using the standardized residual scores, coefficient alphas were acceptable in most cases, though reliabilities for Factors 13, 14 and 4P were below acceptable limits. In the case of Factors 14 and 4P, however, there were only two items that loaded significantly within those factors.

Scale intercorrelations were also computed for both the raw item and standardized residual scales. These are presented in Table 2. Correlations in bold font above the diagonal are related to the raw item scales, while those below the diagonal in normal

580 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Table 1 Total variance explained and item pattern coefficients resulting from factor analyses of 104 Behavior-Focused and 19 Personality-Oriented Items.

Initial Eigenvalues Extraction sums of squared loadings Rotation sums of squared loadings

Factor Total % of Var Cum % Total % of Var Cum % Total

Factor analysis with OBLIMIN rotation of 104 Behavior-Focused Items 1 22.19 21.34 21.34 22.19 21.34 21.34 8.34 2 9.66 9.29 30.63 9.66 9.29 30.63 6.12 3 4.49 4.32 34.94 4.49 4.32 34.94 6.63 4 3.19 3.06 38.01 3.19 3.06 38.01 5.58 5 2.71 2.61 40.62 2.71 2.61 40.62 5.01 6 2.31 2.22 42.84 2.31 2.22 42.84 5.55 7 2.09 2.01 44.85 2.09 2.01 44.85 7.06 8 1.92 1.85 46.70 1.92 1.85 46.70 4.20 9 1.81 1.74 48.43 1.81 1.74 48.43 7.22 10 1.61 1.55 49.99 1.61 1.55 49.99 7.55 11 1.56 1.50 51.49 1.56 1.50 51.49 3.89 12 1.47 1.42 52.90 1.47 1.42 52.90 6.38 13 1.37 1.32 54.22 1.37 1.32 54.22 5.59 14 1.33 1.28 55.50 1.33 1.28 55.50 2.72 15 1.23 1.19 56.69 1.23 1.19 56.69 3.22 16 1.22 1.17 57.86 1.22 1.17 57.86 2.13 17 1.20 1.15 59.01 1.20 1.15 59.01 4.66 18 1.18 1.13 60.14 1.18 1.13 60.14 4.53 19 1.13 1.09 61.23 1.13 1.09 61.23 3.42 20 1.10 1.06 62.29 1.10 1.06 62.29 9.77 21 1.06 1.02 63.31 1.06 1.02 63.31 8.22 22 1.04 1.00 64.31 1.04 1.00 64.31 5.15

Factor analysis with OBLIMIN rotation of 19 Personality-Oriented Items 1 5.02 26.44 26.44 5.02 26.44 26.44 4.08 2 2.32 12.20 38.64 2.32 12.20 38.64 2.27 3 1.48 7.77 46.41 1.48 7.77 46.41 3.71 4 1.16 6.09 52.49 1.16 6.09 52.49 1.38

Pattern matrix coefficients for 104 Behavior-Focused Items (Coefs. N or=.30) Pattern coefficients

FACTOR 1: making decisions based on inadequate information (α=92/76) 78 My boss has his/her head in the sand 0.73 40 My boss does NOT have a clue what is going on in our business unit 0.73 83 My boss is ignorant of things are not part of the immediate environment 0.41 31 My boss often makes knee jerk reactions 0.34 59 My boss does NOT pay enough attention to what really matters 0.32

FACTOR 2: acting in a brutal bullying manner (α=94/86) 114 My boss places brutal pressure on subordinates 0.62 77 My boss enjoys making people suffer 0.59 76 Anyone who challenges my boss is dealt with brutally 0.53 112 I have often seen my boss bully another employee 0.52 38 My boss holds grudges 0.47 20 My boss rarely shows a high level of respect for others 0.34 30 My boss sees every negotiation issue as a win/lose conflict 0.33

FACTOR 3: lying and other unethical behavior (α=96/86) 111 My boss lies a lot 0.72 37 My boss often acts in an unethical manner 0.66 74 My boss rarely acts with a high level of integrity 0.63 36 My boss often takes credit for the work that others have done 0.59 75 My boss blames others for his/her own mistakes 0.52 118 My boss spends too much time promoting him/herself 0.49 91 My boss often says one thing while doing exactly the opposite 0.48

FACTOR 4: micro-managing and over-controlling (α=92/83) 62 My boss is a micro-manager 0.82 101 My boss attempts to exert total control over everyone 0.63 63 My boss is autocratic 0.56 96 My boss does NOT trust others to do tasks properly 0.52 26 My boss wants to dominate/control everything 0.51 58 My boss does not show trust in subordinates by assigning them important tasks 0.39 25 My boss does not share power with the people with whom he or she works 0.34

(continued on next page)

581J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Pattern matrix coefficients for 104 Behavior-Focused Items (Coefs. N or=.30) Pattern coefficients

FACTOR 5: not making expectations clear to subordinates (α=88/69) 19 I rarely know what my boss expects of me 0.63 95 I often have to guess what my boss really expects of me 0.56 50 I rarely know what my boss thinks of my work 0.49 42 My boss spends more time politicking than working 0.44 52 My boss does not provide an appropriate level of supervision and oversight 0.38

FACTOR 6: ineffectual at negotiation and persuasion (α=90/76) 53 My boss is very ineffective in persuading others −0.65 105 When negotiating with others my boss is usually a total failure −0.58 68 My boss is a poor negotiator −0.55 18 My boss is not very good at inspiring others −0.36

FACTOR 7: inability to deal with new technology and other changes (α=89/86) 16 My boss avoids having to use new technology −0.93 54 My boss seems not to enjoys new technology −0.92 92 Sometimes I think my boss is frightened by new technology −0.91 109 My boss has a difficult time dealing with change −0.48

FACTOR 8: inability to deal with interpersonal conflict or similar situations (α=91/74) 126 The punishment my boss gives is often inappropriate for the offense 0.56 49 My boss manages interpersonal conflict poorly 0.45 125 My boss is very poor at dealing with personal or interpersonal issues 0.42 113 My boss has often committed a serious breach of trust 0.38

FACTOR 9: lack of credibility within the organization (α=92/75) 46 Very few people see my boss as a credible manager 0.76 122 My boss has a very poor reputation in our organization 0.76 84 My boss has lost credibility with stakeholders 0.57

FACTOR 10: playing favorites and other divisive behavior (α=91/79) 120 My boss has personal favourites −0.80 82 My boss tends to show excessive favouritism −0.80 98 My boss tends to act in ways that divide employees against one another −0.36

FACTOR 11: telling people what they want to hear (dropped) 80 My boss will tell superiors what they want to hear −0.79

FACTOR 12: ineffective in coordination and management of issues (α=88/64) 89 My boss is an ineffective coordinator −0.65 93 If my boss makes a mistake someone else usually has to fix the problem −0.58 88 My boss often fails to provide subordinates with information and resources −0.40 17 When my boss makes a mistake he or she rarely corrects it −0.38

FACTOR 13: not seeking information from others (α=84/51) 24 My boss rarely seeks opinions from a wide variety of people 0.79 61 My boss does not seek out or pay attention to the opinions and wishes of

subordinates 0.48

55 If my boss screws something up it stays screwed up forever 0.33 23 My boss tends to alienate certain groups 0.33

FACTOR 14: Acting in an insular manner relative to other groups in the organization (α=75/57) 121 My boss does not care about things happening in other units 0.74 45 My boss demonstrates no concern for anyone outside his/her own unit 0.43

FACTOR 15: ineffective communication (dropped) 32 My boss is an ineffective communicator −0.45

FACTOR 16: not having the skills to match the job (α=91/68) 51 My boss treats both good and destructive performers the same way (This item was

dropped from the Factor 16 Scale) 0.62

110 The skills of my boss do not match his/her job very well −0.40 73 My boss lacks the skills and/or experience needed to function job effectively −0.37 104 My boss makes poor decisions under pressure or difficult conditions −0.31

FACTOR 17: inability to prioritize and delegate (α=85/63) 21 My boss is unable to prioritize very well 0.60 34 My boss is unable to effectively manage change 0.46

Table 1 (continued)

582 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Pattern matrix coefficients for 104 Behavior-Focused Items (Coefs. N or=.30) Pattern coefficients

100 My boss is unable to delegate properly 0.32 86 My boss is invulnerable to reason 0.31

FACTOR 18: exhibiting inconsistent, erratic behavior (α=72/65) 119 You can rarely predict how my boss is likely to behave −0.78 81 You never know from day to day how my boss will behave −0.65 43 My boss will act one way and then later acts in the exact opposite manner −0.45

FACTOR 19: unwillingness to change mind and listen to others (α=87/63) 99 My boss is very poor at listening to what others are saying 0.54 124 Once my boss has made up his/her mind there is no changing it 0.49 116 My boss does NOT know what subordinates are thinking 0.37 47 My boss is unable to take a stand and stick to it (This item was dropped from the

Factor 19 Scale) −0.34

FACTOR 20: inability to understand and act on a long term view (α=96/89) 79 My boss can only talk about issues that are very short-term −0.65 28 My boss does not understand the big picture well −0.64 117 My boss does NOT know what the goal of our unit is or should be −0.59 72 My boss does not adapt well to new and changing circumstances −0.58 103 My boss often ignores the big picture −0.57 66 My boss has poor strategic planning skills −0.57 41 My boss is poor at developing a vision for our business unit −0.54 97 My boss is unable to focus very well on the most important issues −0.48 67 My boss deals very ineffectively with complex situations −0.47

FACTOR 21: inability to develop and motivate subordinates (α=93/82) 71 My boss is ineffective at educating and developing subordinates −0.69 108 My boss is NOT very good at developing the skills of subordinates −0.67 33 My boss does not systematically develop the skills of his or her subordinates −0.54 87 My boss does not understand the needs strengths, weaknesses and

responsibilities of subordinates −0.44

56 My boss has no idea what it takes to motivate subordinates −0.42

FACTOR 22: inability to make clear, appropriate decisions (α=85/69) 70 My boss is very poor at getting to the point quickly and clearly −0.60 107 I have trouble understanding what my boss means or wants −0.48 106 My boss is very poor at solving problems and making decisions −0.40 29 In an ambiguous situation my boss has great difficulty making a decision −0.33

Pattern matrix coefficients for 19 Personality-Focused Items (Coefs. N or = .30) Pattern coefficients

FACTOR 1P: an inconsiderate tyrant (α=94/81) 39 My boss could best be described as mean 0.74 115 My boss is a tyrant 0.72 6 My boss is an inconsiderate person 0.69 8 My boss is arrogant 0.66 48 My boss is pig headed i.e. extremely stubborn 0.64 9 My boss is self-centred 0.59

FACTOR 2P: lazy and incompetent (α=81/67) 14 My boss is lazy 0.75 35 My boss is incompetent 0.68 4 My boss lacks drive and energy 0.67 11 My boss is not very smart 0.61

FACTOR 3P: overly emotional with negative psychological characteristics (α=85/77) 12 My boss often gets emotional 0.76 10 My boss lacks self-control 0.71 15 My boss seems to have huge mood swings 0.65 7 My boss seems extremely paranoid about many things 0.62 3 My boss is indiscrete 0.56 1 My boss is compulsive 0.47 5 When my boss wants something he/she obsesses about it 0.38

FACTOR 4P: careless when dealing with people in various situations (α=74/36) 2 My boss lacks emotional intelligence 0.65 13 My boss is often careless when dealing with situations 0.55

Table 1 (continued)

FACTOR 17: inability to prioritize and delegate (α=85/63)

583J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

font are related to the standardized residual scales. All correlations for the raw item scales were statistically significant at pb.05. For the residual item scales, all correlations above .084 in magnitude were significant at pb .05. As can readily be seen in Table 2, there is a high consistent level of intercorrelation among scales based on the raw item ratings. For scales based on standardized residual scores, there is a moderate level of intercorrelation between some scales, and a low level of intercorrelation for others. This level of intercorrelation among the residual-based scales is not unexpected given the nature of the behaviors involved and in fact indicates a level of convergent and discriminant validity within the scale scores. For example, Factors 1 and 17 are moderately intercorrelated. This makes sense given that Factor 1 measures the extent to which the leader makes decisions based on inadequate information while Factor 17 measures the extent to which the leader is unable to prioritize and delegate. These two aspects of leader behavior would logically have some relationship to one another. On the other hand, making decisions based on inadequate information (Factor 1) would likely be less directly related to micro-managing and over-control (Factor 4). The intercorrelation of these two scales was 0.18—low as one would expect. Thus, the correlation data in Table 2 indicate some reasonable level of both convergent and discriminant validity among the various residual item-based scales. This is less evident when looking at the raw item-based scale intercorrelations.

As another way to examine the extent to which the high reliabilities found for the raw items scales were due to common method variance and/or a cognitive categorization effect, we arbitrarily created five 4-item scales from the behavior-focused items, and one 4-item scale from the personality-oriented items. “Arbitrary” scale 1 contained the four items that individually loaded highest on Factors 1, 2, 3, and 4. Arbitrary scale 2 contained the four items that individually loaded highest on Factors 5, 6, 7, and 8, and so on for the other arbitrary scales (excluding items from Factors 11 and 15). Coefficient alpha was calculated for each of these arbitrary scales. For these raw item arbitrary scales, coefficient alphas were .74, .71, .82, .64, .84, and .66 indicating that common method variance and/or a categorization effect comprised a significant portion of the internal consistency among items in the various factor scales.

As with the raw item scales, we created another set of “arbitrary” scales, but in this case used the standardized residual item scores. Coefficient alphas were calculated on each of the six arbitrary residual scales. For these residual item-based arbitrary scales, coefficient alphas were .30, .37, .36, .36, .51, and .21. This gave some indication that the attribute dimensions being tapped within the residual-based factors represented meaningful conceptions of prototypical destructive leadership behaviors and personal characteristics. As can also be seen in Table 1, while some of the attribute dimensions involve recognized aspects of destructive leadership such as bullying, abuse, or narcissism, the range of attribute dimensions identified is quite extensive and wide ranging in nature.

Identifying a typology of sub-categories of destructive leadership

Our third major objective in this study was to use scales created for the DLQ to identify sub-categories of destructive leaders. Kellerman (2004) identified seven types of destructive leaders and there are many other typologies of destructive leaders found in various websites and books focused on the issue. However, most if not all of these typologies are—as Kellerman readily admits—

Table 2 Intercorrelations for raw item-based and standardized residual item-based scales. a

Factors 1 2 3 4 5 6 7 8 9 10 12 13 14 16 17 18 19 20 21 22 P1 P2 P3 P4

1 * .74 .79 .68 .78 .77 .55 .77 .83 .76 .81 .78 .72 .84 .81 .61 .81 .84 .81 .72 .77 .74 .61 .79 2 .24 * .85 .82 .71 .61 .41 .79 .71 .81 .73 .78 .65 .71 .67 .68 .77 .62 .69 .55 .91 .56 .75 .71 3 .38 .61 * .74 .75 .63 .44 .78 .76 .81 .81 .78 .67 .80 .73 .66 .77 .69 .71 .63 .87 .67 .69 .73 4 .18 .58 .34 .70 .62 .42 .74 .71 .73 .69 .76 .60 .71 .70 .64 .77 .62 .70 .60 .83 .49 .72 .66 5 .42 .28 .33 .32 * .68 .43 .77 .75 .74 .76 .76 .64 .77 .75 .64 .78 .71 .82 .68 .72 .63 .56 .73 6 .45 .08 .09 .16 .30 * .50 .71 .79 .63 .69 .70 .60 .80 .73 .51 .72 .81 .77 .70 .61 .72 .47 .68 7 .34 .08 .16 .13 .18 .32 * .45 .52 .45 .50 .46 .41 .54 .58 .38 .48 .63 .50 .55 .41 .56 .38 .43 8 .36 .47 .40 .41 .41 .32 .15 * .75 .77 .78 .78 .63 .76 .72 .68 .80 .71 .77 .66 .80 .60 .66 .78 9 .54 .21 .30 .28 .35 .52 .31 .35 * .73 .77 .77 .63 .85 .79 .58 .78 .80 .79 .73 .73 .72 .59 .73 10 .38 .55 .53 .40 .37 .15 .16 .45 .32 * .76 .76 .62 .72 .68 .68 .77 .66 .74 .64 .80 .58 .66 .69 12 .51 .33 .43 .30 .46 .31 .27 .42 .39 .39 * .76 .64 .82 .77 .65 .80 .75 .79 .71 .76 .67 .60 .77 13 .39 .43 .38 .45 .41 .29 .15 .42 .39 .41 .44 * .66 .76 .72 .61 .80 .71 .78 .62 .81 .60 .60 .75 14 .42 .26 .31 .23 .24 .28 .14 .23 .30 .26 .25 .30 * .65 .63 .49 .66 .64 .66 .53 .69 .57 .49 .62 16 .53 .15 .30 .23 .33 .50 .33 .31 .50 .26 .41 .30 .26 * .81 .57 .76 .85 .80 .81 .74 .74 .59 .76 17 .53 .23 .34 .38 .43 .44 .44 .33 .50 .28 .49 .38 .29 .47 * .62 .78 .80 .78 .75 .70 .69 .61 .74 18 .25 .45 .38 .40 .36 .13 .12 .42 .20 .46 .38 .31 .18 .14 .32 * .63 .53 .56 .52 .67 .45 .67 .62 19 .46 .41 .32 .47 .43 .28 .19 .50 .38 .44 .44 .45 .30 .29 .49 .33 * .74 .82 .72 .81 .60 .65 .78 20 .60 .02 .19 .12 .31 .58 .50 .23 .51 .18 .40 .25 .29 .61 .58 .08 .31 * .78 .76 .65 .76 .51 .71 21 .53 .22 .22 .30 .58 .44 .27 .47 .46 .37 .49 .45 .35 .43 .52 .21 .53 .45 * .72 .73 .66 .56 .78 22 .44 .07 .20 .22 .40 .52 .41 .29 .44 .21 .42 .25 .18 .53 .51 .21 .31 .61 .45 * .57 .68 .51 .65 P1 .28 .80 .59 .58 .29 .04 .04 .43 .21 .48 .33 .47 .32 .15 .25 .40 .45 .05 .27 .07 * .58 .76 .75 P2 .44 .04 .23 -.06 .23 .43 .39 .07 .38 .10 .26 .12 .23 .38 .38 .02 .06 .50 .26 .41 .02 * .42 .64 P3 .19 .58 .39 .51 .17 .06 .16 .37 .19 .39 .24 .25 .09 .17 .31 .51 .34 .09 .15 .20 .55 -.03 * .62 P4 .44 .27 .28 .19 .34 .25 .14 .47 .32 .29 .45 .35 .23 .31 .40 .32 .43 .28 .45 .28 .34 .21 .30 *

a Correlations above diagonal=raw item scales; Correlations below diagonal=standardized residual item scales.

584 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

based on anecdotal rather than empirical data. In an attempt to produce a typology of empirically based sub-categories of destructive leaders, we conducted a series of cluster analyses with our sub-sample of destructive leaders using both the behavior- focused and personality-oriented raw item-based and standardized residual item-based scales. Unlike our earlier PCA analyses, our intent here was not to reduce the number of items in the DLQ and identify the underlying behavioral or personality dimensions of destructive leadership. Our intention, rather, was to use the attribute scales derived from our PCA analyses to group together leaders with similar patterns of scores across the various attribute scales. Thus, we must refer to our clusters as representing a taxonomy of destructive “leaders” rather than of destructive “leadership.” The numbers of both behaviour-focused and personality-oriented scales were sufficiently small that our sample to variable ratios for the cluster analyses were reasonable and allowed us some sense of confidence in the potential stability of our clustering results.

However, as we suspected, the clusters formed using the raw item scores were not particularly illuminating and were, instead, dominated by the overall judgment of how destructive the leader was. Raw item clusters showed relatively little pattern differences across scale measures and instead simply distinguished between really destructive leaders and slightly less destructive leaders. Additionally, the cluster results using the four personality-based scales yielded clusters that were overlapping in nature and did not provide a clear delineation among personality-based categories of destructive leadership. For these reasons only results from cluster analyses using the standardized residual item scores on behavior-based scales are reported below.

For all cluster analyses we used Ward's hierarchical cluster analysis method found within SPSS. In Ward's method, at each stage of the clustering process, the combination of all cluster pairs is considered. The two clusters joined at each step represent those clusters whose joining results in the smallest increase in terms of an error sum-of-squares criterion. One common problem with any cluster analysis method is determining the exact number of clusters that should be extracted. The “elbow criterion” says that if you graph the percentage of variance explained by the clusters against the number of clusters, the number of clusters chosen should be the point at which increasing the number of clusters does not add markedly to the variance explained. Graphically this appears as a noticeable bend in the graph (the elbow). Unfortunately, this elbow is not always easy to unambiguously identify.1

Kellerman (2004) identified seven types of “bad leaders” in her book, and popular lists of destructive leaders such as those found on websites often have five to 10 types of destructive leaders mentioned. In our study, the cluster analyses were conducted using data from respondents who were describing leaders who fell into the bottom third of leaders in our sample.

As noted previously, we had trichotomized our sample into good, average and destructive leaders using scores from the 4-item Overall Good-Destructive Leader scale (final n=203 for the cluster analyses). Ward's hierarchical cluster analysis procedure was conducted on the sample of destructive leaders using the standardized residual scale scores derived from the behavioral-focused items. A criterion was set so that the clustering algorithm identified cluster membership for cases in the sample for cluster solutions ranging from 5 to 15 clusters. Given the sample size involved, however, results of these initial analyses indicated that cluster solutions with greater than eight clusters yielded clusters with membership of only two or three individuals. For practical reasons, therefore, our examination of clusters was limited to cluster solutions ranging from five to seven.

Multivariate analysis of variance (MANOVA) procedures were conducted using cluster membership as the fixed variable and the behavior-focused scale scores as the multiple dependent variables. For all three of the cluster solutions (5, 6 and 7 clusters) the MANOVAs indicated that there were significant main effects for cluster membership. Univariate comparisons indicated that within the sets of clusters, there was a clear pattern of cluster differences across all scale measures. For many of the variables, each cluster within the set was significantly different from all other clusters.

Further intuitive analysis of the nature of the clusters along with plotting of the cluster mean for each of the dependent variables indicated that a 7-cluster solution had clusters that each contained a sufficient number of cases to be meaningfully interpreted. In addition, the pattern of cluster mean scores across the dependent variables allowed us to identify reasonably interesting and different types of “destructive leaders.” A graph for the 7-cluster solution showing mean scale scores for the behavior-based scales is found in Fig. 1.

Categories of destructive leaders derived from cluster analyses of scores on behavior-focused attribute scales

To aid in our interpretation of the different clusters, the pattern of scale scores within each cluster were graphed as was the average of each scale score for the top third “good leaders” in our sample. Thus, we could compare the destructive-leader cluster pattern to that of the average good leader on each variable. In addition, since the scales in the cluster analyses were based on standardized residual item ratings, the average scale scores across all types of leaders in our sample was between −.01 and .01 and thus, could serve as a comparison standard. Remember that when examining Fig. 1 that a high score represents destructive leader behavior. Below is a summary description for each of the clusters identified using the behavior-focused scales shown in Fig. 1.

Cluster 1 (n=40)—this type of leader scored worse than the average good leader on all behavioral scales. Factors on which this type of leader scored particularly badly included Making Decisions Based on Inadequate Information; Lying and Other Unethical Behavior; Inability to Deal with New Technology and Other Changes; Inability to Prioritize and Delegate; and Inability to Make Clear Appropriate Decisions. This type of leader was not much different however, from the average of all leaders in terms of their propensity to micro-manage and over-control or in terms of their willingness to change their minds and listen to others.

1 Found at http://en.wikipedia.org/wiki/Data_clustering

585J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Cluster 2 (n=19)—this type of leader was better than the average good leader on some factors but much worse than the average good leader on others. In particular, he/she lacked some common skills that a normal leader might be expected to have. This type of leader was described as Ineffectual at Negotiation and Persuasion; Not Having the Skills to Match the Job; having an Inability to Understand and Act on a Long Term View; and having an Inability to Develop and Motivate Subordinates. This leader was considerably better than the average good leader, however, with lower levels of Acting in a Brutal Bullying Manner; Lying and Other Unethical Behavior; and Micro-Managing and Over-Controlling.

Cluster 3 (n=36)—this leader was actually a pretty good leader, with better scores than an average good leader on 12 of the 20 factors. In fact the only really negative trait evidenced by this leader—relative to both the average good leader and the average of all leaders in the sample—was that he/she tended to engage in higher levels of Micro-Managing and Over- Controlling. It is interesting to note that this type of leader tended to collect an adequate amount of information before making decisions (consistent with a high level of micro-management), was a fairly effective negotiator and persuader, and kept in touch not only with what was happening in his/her own group but also that of others (again, consistent with a tendency to micro-manage and over-control).

Cluster 4 (n=17)—this type of leader showed high scores on three factors. These were Inability to Deal with Interpersonal Conflict or Similar Situations; Playing Favourites and Other Divisive Behavior; and Exhibiting Inconsistent, Erratic Behavior. This type of leader, however, was not viewed as completely bad as they were seen as having the skills necessary to do their job and could both understand and act on strategic or long terms issues affecting their group.

Cluster 5 (n=40)—this type leader could be referred to as the “not all that bad but not all that good” leader. In many ways this type of leader was not that different from the average of all leaders within our sample, although they did tend to be worse than the average good leader on many of the behavioral factors. Relative to other types of bad leaders, this type of leader's worst characteristics were only moderately bad, with their worst score being .36 on Not Seeking Information from Others. Other modestly negative aspects of this type of leader were that they were seen as Ineffective in Coordination and Management of Issues; showed an Unwillingness to Change Mind and Listen to Others; and (perhaps as a result of these other factors) showed an Inability to Develop and Motivate Subordinates. After all, if you do not communicate and listen to your subordinates how you can motivate and develop them?

Cluster 6 (n=32)—the predominant feature of this leader's behavior is their tendency for Acting in an Insular Manner Relative to Other Groups in the Organization. Their scale scores also indicate a moderate tendency for Acting in a Brutal Bullying Manner; Micro-Managing and Over-Controlling; and an Inability to Deal with New Technology and Other Changes.

Cluster 7 (n=19)—anyone unfortunate enough to have one of these leaders undoubtedly works in a situation of significant misery and despair. This leader's two highest scores (and they were very high) were for Acting in a Brutal Bullying Manner as well as Lying and Other Unethical Behavior. In addition, this leader had very high scores for Micro-Managing and Over-Controlling; Inability to Deal with Interpersonal Conflict or Similar Situations; Not Having the Skills to Match the Job; and Unwillingness to Change Mind and Listen to Others.

As a final means of understanding the nature of the clusters, job titles for leaders falling within each of the clusters were identified. A list of job titles by cluster is presented in Table 3. There were no obvious differences in the types of jobs that fell within each cluster, although it did seem that in some clusters there were substantially more job titles associated with university/ academic leader positions than in other clusters.

To see if there were any systematic differences among the clusters in terms of academic versus non-academic leader prevalence, we assigned codes to each of the job titles provided for leaders in the worst-leaders sample. We then conducted a chi-

-1.50

-1.00

-0.50

0.00

0.50

1.00

1.50

1 2 3 4 5 6 7 8 9 10 12 13 14 16 17 18 19 20 21 22

Z S

c a

le S

c o

re

Behavioral Factors

Cluster 1

Cluster 2

Cluster 3

Cluster 4

Cluster 5

Cluster 6

Cluster 7

Good Avg

Fig. 1. Seven clusters using behavior-focused scales on a sample of the worst third of leaders in the sample.

586 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Table 3 List of jobs of current boss in each of seven destructive leader clusters. a

Cluster 1: N=40 (Academic=20.5%)

• Assistant Professor • Director of Nursing • Manager of Training and Development

• Central HR Director • Director of the Library • Managing Director

• Commanding Officer • Executive Director • Officer in Charge (Military)

• Committee Management Officer • General Manager×3 • Principal

• Dean • Head of School×2 • Provost×2

• Department Superintendent • Head of the Department • Regional Librarian

• Director×4 • HR Manager • Section Head

• Director (Section Head) • Information Centre Manager • Section Manager×2

• Director Human Resources • IT Coordinator • Team Leader in Investment Bank

• Director of Marketing and Student Recruitment and Admissions

• Manager (unspecified

Cluster 2: N=19 (Academic=52.6%)

• Business Manager×2 • General Manager • Principal

• Dean (unspecified×2) • Head of Department • Professor

• Dean of Faculty (Head of School) • Interim Dean • Program Control Manager

• Department Chair (Academic) • Library Manager • Team leader

• Director×2 • PhD • Vice President of Academic Affairs (Provost)

• Workshop Manager

Cluster 3: N=36 (Academic=30.6%)

• 2nd in Command (Military) • General Manager×2 • Principal

• Academic Department Head×3 • Head of Agency • Principal Director

• Associate Director • Head of Service • Professor

• CEO • Library Director • Registrar

• Commander • Major General • Senior Manager

• Director×3 • Manager×4 • Senior Officer Human Resources, Faculty State Manager

• Director Enterprise Portfolio Management • Owner of a Restaurant • Superintendent of School System

• Executive Manager • Partner • Superintendant of Schools

• University Rector • President

Original Cluster 4: N=17 (Academic=56.3%)

• Branch Manager • Owner/Managing Director • Senior Financial Accountant

• Department Chair • Principal • State Manager

• General Manager, International • Professor×3 • Vice Chancellor for Academic Affairs

• Head of School×2 • Registrar

• Manager (just below the CEO) • Senior Consultant

Original Cluster 5: N=40 (Academic=39.4%)

• Area Supervisor • Department Chair or Head×5 • Pro-Vice Chancellor

• Assistant Secretary • Director×2 • Sales Manager

• Associate Director, Human Resources • General Manager • Senior VP of Human Resources

• CEO • Head of School (academic) • Team Leader×2

• Chief • Head of Service • Vice President

• Corporate Quality Manager • Manager×3 • VP Business Mgmt

• Dean×6 • Principal Investigator

Original Cluster 6: N=32 (Academic=19.4%)

• Acting Manager • Director×3 • Marketing Manager

• Ambulance Service Operations Manager • Director of Operations • Operations Manager HR, Training, Staffing

• President • Primary School Principal×2 • Retail Manager

• Associate Dean and Interim Head of Department • Executive Director • Senior Vice President

• Associate Professor • General Manager×2 • Vice President

• Campus Boss • HR Director • Vice President of Human Resources

• Chief Facilitator - Coordinator • Manager×4

• Dean • Manager, Library and Record Services

• Department Chair • Managing Director

(continued on next page)

587J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

square test to see if there were significant differences in the relative proportion of academic versus non-academic leaders in each cluster. Indeed, a chi-square of 14.12 was significant at pb .03 (df=6). Table 3 also indicates the percentage of academic job titles represented in each cluster, and it can be seen that academic positions seemed to be particularly prominent in clusters 2 and 4. For those readers familiar with the world of academia and the behavior of their academic leaders, these findings might well appear to enhance the face validity of our clusters!

We also examined the leader's gender to see whether our clusters might reflect differing groups of leaders based on whether the leader was male or female, i.e., did our clusters reflect gender differences rather than more generic aspects of destructive leadership. There were no significant differences in the relative proportion of male and female leaders within the clusters. However, when we examined whether the match between the leader's and subordinate's gender was reflected in our cluster structure, our chi-square tests did show significant differences in leader–subordinate gender match across the clusters (chi- square=14.57, pb .03, df=6). Leaders whose gender did not match the gender of their subordinate were found in clusters 5 (60%) and 7 (58.8%), while leaders whose gender did match that of their subordinate were predominant in clusters 1 (60.0%), 2 (73.7%), 3 (75.0%) and 6 (65.6%). These findings are not unexpected as the clusters are based on data reflecting the perception of the leader by his/her subordinate. Thus the gender match between the subordinate and leader may well influence the perception by the subordinate of any given behavior exhibited by their leader. Of course, it could be that in gender-mismatch situations, leaders actually engage in different types of behavior than leaders in gender-match dyads. Further research would be required to answer this interesting question.

Discussion

There has been increased interest by leadership scholars recently in exploring the nature of leadership variously described as destructive, toxic, bullying, or abusive. This study is the first to attempt to empirically examine the differences between leaders who are collectively labelled as “destructive.” It should be emphasized again that our approach to the study of such leadership is based on follower perceptions and resulting categorization of leadership behavior. Given the relative infancy of research into destructive leadership and considering that leadership is often viewed as socially constructed between leaders and followers, then our “cognitive schema “ approach represents a valid point from which to examine this phenomenon.

As with most research, our study and interpretation of results is subject to several limitations. Firstly, as noted earlier in this paper, there are a number of limitations with convenience sampling. Convenience sampling involves non-random sampling which can lead to biased results. In the case of our sample there are, indeed, a greater proportion of respondents and leaders from academic institutions than would be expected in the general population. The sample is certainly more educated than the general population. However, the average respondent was 43.8 years old (Std. Dev.=11.6) meaning that the majority of our sample were aged 32 to 54, which would be typical of working professionals. The majority of current leaders described were male (54.9%), which would not be unexpected in the countries from which most of our respondents came. When we examined the nature of ratings for male and female leaders in terms of their overall rating of their leaders, 46.8% of female leaders were rated between 1.0 and 3.0 on the Overall Good-Destructive Leader scale while 49.3% of male leaders were rated within this range. The majority of respondents to the survey were female (63.6%) which could bias the nature of responses provided. However, when we examined the responses from male and females in terms of their overall rating of their leaders, 49.6% of females rated their leader between 1.0 and 3.0 on the Overall Good-Destructive Leader scale while 46.5% of males did so.

A second potential bias in our sample could have been that respondents who were willing to complete the entire survey may have been those particularly affected by the behavior of destructive leaders. While this could occur, respondents in our sample rated leaders who ranged from exceptionally good to extremely destructive. In fact, the median of all respondent ratings on the Overall Good-Destructive Leader scale was 3.0 out of 6.0. We felt that the advantages of using a web-based sample to gain a large number of respondents, industries and types of leaders outweighed the potential biases that might occur. Given the data above, we believe that while biases certainly exist within our sample, they may not be sufficient to invalidate our conclusions. Obviously, however, replication of both our DLQ factor structure and the nature of our destructive leader taxonomy are necessary with more carefully selected samples representative of particular leader populations.

A third potential limitation of the current study is the use of residual scores as the basis for our PCA procedures. As noted previously in this paper, the removal of the “overall liking/disliking” of their leader from the respondents' ratings may remove a potentially important variable. However, as noted previously, the nature of the factors identified using this approach “make sense” and is consistent in nature with previous research on the characteristics of destructive leaders. In addition, inter-item correlations among the residual score-based factors show evidence of both convergent and discriminant validity among the various factors. We

Original Cluster 7: N=19 (Academic=31.9%)

• Assistant Dean External Relations • CEO • Manager×3

• Assistant Secretary/Branch Head • Division Director • Manager Research Services

• Associate Professor • General Manager • Operations Director

• Centre Manager • Head of Teaching and Learning • Professor

• Supervisor • Maintenance Manager • Works Manager

a Some respondents did not provide job title information.

Table 3 (continued)

588 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

would certainly encourage researchers to devise and use other methods for constructing factors, but we believe that the current method is a legitimate one.

The behavioral factors identified in our study can readily be construed within the overall current label of destructive leadership. However, much of previous research on the dark side of leadership has tended to focus on fairly specific leadership characteristics or behaviors as the cause of that destruction. That is, we have looked at the negative impact of abusive or bullying leaders and have examined the effects that narcissistic leaders have on employee and organizational outcomes. The attribute factors identified in our study, however, are broader in nature, ranging from typical destructive leaders behaviors, e.g., bullying, to less severe forms of destructive leadership, e.g., being a failure at negotiations. The implications of these milder aspects of destructive leadership become more apparent when we examine the sub-categories of destructive leaders identified by our cluster analysis procedure. While previous taxonomies of destructive leaders have been presented based on anecdotes, case studies and conceptual notions; our study used cluster analysis to empirically identify seven types of destructive leaders based on scores of the behavior-focused attribute scales. As previously mentioned, we had trichotomized our sample into good, average and destructive leaders using scores from the 4-item Overall Good-Destructive Leader scale. Our cluster analyses were conducted using data from respondents who were describing leaders who fell into the worst third of leaders in our sample.

Interestingly, despite the fact that our analysis was based on the very worst of the leaders represented in our study, most types of leaders identified via the cluster analysis could not be described as truly destructive. Rather, they tended to be destructive based on high scores on just a few negative behaviors. It seems that there are very specific behaviors that followers identify in their leaders that cause them to perceive their leaders as destructive. Such leaders may also have positive characteristics. However, if they rate highly on just one or two negative behaviors, they are likely to be categorized as “destructive leaders” by their subordinates. This then can affect the leader's ability to obtain social power and influence and gain credit for positive aspects of their behavior. Having been classified as a “destructive leader” such individuals would also likely be blamed for negative outcomes which occur to subordinates and their work teams. One reason for this finding may be that subordinates define “destructiveness” in terms of their own, fairly limited personal experiences and outcomes. This is consistent with the common definition of destructive leadership which focuses on behaviours by leaders that violate “the legitimate interest of the organization” and/or “undermining and/or sabotaging the motivation, well-being, or job satisfaction of subordinates” (Einarsen et al. 2007, p.20).

The results of this study contribute to the developing research stream that examines destructive leadership and have both theoretical and practical implications. At the theoretical level, they indicate that it is possible to empirically identify different types of destructive leaders, each of which may provide a unique opportunity for further research. At the practical level, perhaps the most important implication is that top managers need to effectively monitor both the extent of destructive leadership that is present in their organization as well as the particular types of destructive leaders that predominate. If many of the leaders within the organization are destructive, then significant improvements in leader selection might be warranted. Such data would also have implications for the training of future leaders in terms of behaviors that may lead to negative perceptions by subordinates, and also for possible training interventions to deal with leaders whose followers already view them in a negative light.

References

Balkundi, P., & Kilduff, M. (2005). The ties that lead: A social network approach to leadership. Leadership Quarterly, 16(6), 941–961. Bass, B. M. (1985). Leadership and performance beyond expectations. New York: Free Press. Bass, B. M. (1990). Bass & Stogdill's handbook of leadership: Theory, research and managerial applications (3rd ed.). New York: Free Press. Bennis, W. G., & Nanus, B. (1985). Leadership: The strategies for taking charge. New York: Harper and Row. Burns, J. M. (1978). Leadership. New York: Harper & Row. Conger, J. A., & Kanungo, R. N. (1987). Toward a behavioral theory of charismatic leadership in organizational settings. Academy of Management Review, 12(4),

637–647. Costello, A. B., & Osborne, J. W. (2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical

Assessment, Research & Evaluation, 10(7), 1–9, electronic journal found at: http://pareonline.net/pdf/v10n7.pdf Crocker, J., Fiske, S. T., & Taylor, S. E. (1984). Schematic bases of belief change. In J. R. Eiser (Ed.), Attitudinal judgment (pp. 197–226). New York: Springer-Verlad. Cronshaw, S. F., & Lord, R. G. (1987). Effects of categorization, attribution, and encoding processes on leadership perceptions. Journal of Applied Psychology, 72(1),

97–106. Dansereau, F., Graen, G., & Haga, W. J. (1975). A vertical dyad linkage approach to leadership within formal organizations. Organizational Behavior and Human

Performance, 13(1), 46–78. Dienesch, R. M., & Liden, R. C. (1986). Leader member exchange model of leadership: A critique and further development. Academy of Management Review, 11(3),

618–634. Einarsen, S. (1999). The nature and causes of bullying at work. International Journal of Manpower, 20(1/2), 16–27. Einarsen, S., Aasland, M. S., & Skogstad, A. (2007). Destructive leadership behavior: A definition and conceptual model. Leadership Quarterly, 18(3), 207–216. Einarsen, S., & Raknes, B. (1997). Harassment at work and the victimization of men. Violence and Victims, 12, 247–263. Erickson, A., Shaw, J., & Agabe, Z. (2007). An empirical investigation of the antecedents, behaviors, and outcomes of bad leadership. Journal of Leadership Studies, 1

(3), 26–43. Ferris, G., Zinko, R., Brouer, R., Buckley, M. R., & Harvey, M. (2007). Strategic bullying as a supplementary, balanced perspective on destructive leadership.

Leadership Quarterly, 18(3), 195–206. Fiedler, F. E. (1972). The effectiveness of leadership training and experience: A contingency model interpretation. Administrative Science Quarterly, 17(4), 455–463. Field, T. (2003). Workplace bullying. British Medical Journal, 326, 776–777. Fleishman, E. A. (1953). The description of supervisory behavior. Personnel Psychology, 37(1), 1–6. Graen, G., & Cashman, J. F. (1975). A role-making model of leadership in formal organizations: A developmental approach. In J. G. Hunt & L. L. Larson (Eds.),

Leadership frontiers (pp. 143–165). Kent, OH: Kent State University Press. Graen, G. B., & Uhl-Bien, M. (1995). Relationships- based approach to leadership: Development of Leader–Member Exchange (LMX) theory of leadership over

25 years: Applying a multi-level domain approach. Leadership Quarterly, 6(2), 219–247. Harvey, M., Treadway, D., & Heames, J. (2007). The occurrence of bullying in global organizations: A model and issues associated with social/emotional contagion.

Journal of Applied Social Psychology., 37(11), 2576–2599.

589J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590

Hemphill, J. K., & Coons, A. E. (1957). Development of the leader behavior description questionnaire. In R. M. Stogdill & A. E. Coons (Eds.), Leader behavior: Its description and measurement (pp. 6–38). Columbus, OH: Bureau of Business Research, Ohio State University.

Hersey, P., & Blanchard, K. H. (1969). Life cycle theory of leadership. Training and Development Journal, 23(2), 26–34. Hogan, R., & Hogan, J. (2001). Assessing leadership: A view from the dark side. International Journal of Selection and Assessment, 9(1–2), 40–51. House, R. J. (1971). A path-goal theory of leader effectiveness. Administrative Science Quarterly, 16(3), 321–339. House, R. J. (1977). A 1976 theory of charismatic leadership. In J. G. Hunt & L. L. Larson (Eds.), Leadership: The cutting edge (pp. 189–207). Carbondale, IL: Southern

Illinois University Press. Judge, T. A., Bono, J. E., Ilies, R., & Gerhardt, M. W. (2002). Personality and leadership: A qualitative and quantitative review. Journal of Applied Psychology, 87(4),

765–780. Katz, D., Maccoby, N., & Morse, N. (1950). Productivity, supervision, and morale in an office situation. Ann Arbor, MI: Institute for Social Research. Keelan, E. (2000). Bully for you. Accountancy, 125(1281), 56. Kellerman, B. (2004). Destructive leadership: What it is, how it happens, why it matters. Cambridge, Mass: Harvard Business School Press. Kerr, S., & Jermier, J. M. (1978). Substitutes for leadership: Their meaning and measurement. Organizational Behavior and Human Performance, 22(3), 375–403. Kirkpatrick, S. A., & Locke, E. A. (1996). Direct and indirect effects of three core charismatic leadership components on performance and attitudes. Journal of Applied

Psychology, 81(1), 36–51. Leymann, H., & Gustafsson, A. (1996). Mobbing at work and the development of post-traumatic stress disorders. European Journal of Work and Organizational

Psychology, 5(2), 251–276. Lipman-Blumen, J. (2006). The allure of toxic leadership: Why we follow destructive bosses and corrupt politicians—and how we can survive them. New York: Oxford

University Press. Lord, R. G. (1985). An information processing approach to social perceptions, leadership and behavioral measurement. In L. L. Cummings & B. M. Staw (Eds.),

Research in organizational behavior, Vol. 7. (pp. 87–128)Greenwich, Conn: JAI Press. Lord, R. G., DeVader, C. L., & Alliger, G. M. (1986). A meta-analysis of the relationship between personality traits and leadership: An application of validity

generalization procedures. Journal of Applied Psychology, 71(3), 402–410. Lord, R. G., Foti, R., & Phillips, J. S. (1982). A theory of leadership categorization. In J. G. Hunt, U. Sekaran, & C. Schriesheim (Eds.), Leadership: Beyond establishment

views (104–121). Carbondale, Ill: Southern Illinois University Press. Lubit, R. (2004). The tyranny of toxic managers: Applying emotional intelligence to deal with difficult personalities. Ivey Management Services: Improving the practice

of management, March/April, Retrieved July 29, 2009 from http://home.mycybernet.net/~taylors/Publish/Tyrany%20of%20toxic%20Managers.pdf Marion, R., & Uhl-Bien, M. (2001). Leadership in complex organizations. Leadership Quarterly, 12(4), 389–418. Notelars, G., Einarsen, S., De Witte, H., & Vermunt, J. K. (2006). Measuring exposure to bullying at work: The validity and advantages of the latent class cluster

approach. Work & Stress, 20(4), 289–302. O'Conner, J., Mumford, M., Clifton, T., Gessner, T., & Connelly, M. (1995). Charismatic leadership and destructiveness: A historiometric study. Leadership Quarterly, 6

(4), 529–555. Olafsson, R. F., & Johannsdottir, H. L. (2004). Coping with bullying in the workplace: The effect of gender, age and type of bullying. British Journal of Guidance &

Counselling, 32(3), 319–333. Padilla, A., Hogan, R., & Kaiser, R. B. (2007). The toxic triangle: Destructive leadership, susceptible followers, and conducive environments. The Leadership Quarterly,

18(3), 176–194. Paunonen, S. V., Lönnqvist, J. E., Verkasalo, M., Leikas, S., & Nissinen, V. (2006). Narcissism and emergent leadership in military cadets. Leadership Quarterly, 17(5),

475–486. Rosenthal, S. A., & Pittinsky, T. L. (2006). Narcissistic leadership. The Leadership Quarterly, 17(6), 617–633. Sashkin, M. (1988). The visionary leader. In J. A. Conger & R. A. Kanungo (Eds.), Charismatic leadership: The elusive factor in organizational effectiveness

(pp. 122–160). San Francisco, CA: Jossey-Bass. Schaubroeck, J., Walumbwa, F. O., Ganster, D. C., & Kepes, S. (2007). Destructive leader traits and the neutralizing influence of an “enriched” job. The Leadership

Quarterly, 18(3), 236–251. Shamir, B., House, R. J., & Arthur, M. B. (1993). The motivational effects of charismatic leadership: A self-concept based theory. Organization Science, 4(4), 577–594. Shaw, J. B. (1990). A cognitive categorization model for the study of intercultural management. Academy of Management Review, 15(4), 626–645. Sommers, J. A., Schell, T. L., & Vodanovich, S. J. (2002). Developing a measure of individual differences in organizational revenge. Journal of Business & Psychology, 17

(2), 207–222. Stogdill, R. M. (1948). Personal factors associated with leadership: A survey of the literature. Journal of Psychology, 25(1), 35–71. Tepper, B. J. (2000). Consequences of abusive supervision. Academy of Management Journal, 43(2), 178–190. Trice, H. M., & Beyer, J. M. (1986). Charisma and its routinization in two social movement organizations. In B. M. Staw & L. L. Cummings (Eds.), Research in

organizational behavior (pp. 113–164). Greenwich, CT: JAI Press. Vroom, V. H., & Jago, A. G. (1988). The new leadership. Managing participation in organizations. Englewood Cliffs, NJ: Prentice Hall. Vroom, V. H., & Yetton, P. W. (1973). Leadership and decision-making. Pittsburgh: University of Pittsburgh Press. Wilson, C. (1991). US businesses suffer from workplace trauma. July: Personnel Journal 47–50.

590 J.B. Shaw et al. / The Leadership Quarterly 22 (2011) 575–590