cu diversity study

Upload: matt-sebastian

Post on 07-Feb-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/21/2019 CU diversity study

    1/65

    1

    David R. Hekman, Ph.DAssistant Professor of Management and Entrepreneurship

    Leeds School of BusinessUniversity of Colorado

    [email protected]

    Wei YangPh.D. Student

    McCombs School of BusinessUniversity of Texas at Austin

    Maw Der Foo

    Associate Professor of Management and EntrepreneurshipLeeds School of Business

    University of Colorado

    mailto:[email protected]:[email protected]:[email protected]
  • 7/21/2019 CU diversity study

    2/65

    2

    AN EXAMINATION OF WHETHER AND HOW DIVERSITY-VALUING BEHAVIOR

    RESULTS IN WORSE PERFORMANCE RATINGS FOR MINORITY AND FEMALE

    LEADERS THAN WHITE AND MALE LEADERS

    ABSTRACT

    We seek to solve the puzzle of why the glass ceiling persists despite the presence of

    ethnic minorities and women in organizational leadership positions. We suggest that the glass

    ceiling persists partly because ethnic minority and women leaders are discouraged from engaging

    in diversity-valuing behaviors. Specifically, we hypothesize and test in both field and laboratory

    samples that ethnic minority or female leaders who engage in diversity-valuing behavior are

    penalized with worse performance ratings than their equally-diversity-valuing white or male

    counterparts. We generally find that this divergent effect partly results from traditional negative

    racial and gender stereotypes (i.e. lower competence and warmth judgments, respectively) placed

    upon diversity-valuing ethnic minority and women leaders. We discuss how our findings extend

    and enrich the vast literatures on the glass ceiling, tokenism and workplace discrimination and

    imply that researchers might benefit by shifting from a focus on diversity-valuing behavior

    (which only appears to be viewed negatively when performed by ethnic minorities and women)

    to demographic-unselfishness behavior (which is likely more universally admirable).

  • 7/21/2019 CU diversity study

    3/65

    3

    I dont think that I would consider myself a feminist.First female CEO of Yahoo, Marissa

    Mayer

    People get bent out of shape about the factthat when I was a kid, you could not drink out of

    certain water fountains. Well, the water was the same.African-American U.S. Supreme Court

    Justice Clarence Thomas

    For decades management researchers have been interested in understanding the degree to

    which ethnic minorities and women face a glassceilingin the highest corporate echelons

    (Finkelstein, Hambrick, and Cannella, 2009). Indeed, most research attention has focused on

    documenting that the glass ceiling exists, and little has been done to understand why it exists and

    how it might be dismantled. As Park and Westphal (2013: 543) conclude littleattention has

    been devoted to identifying possible sources of social discrimination against minorities who have

    managed to acquire high-status positions.However, some researchers are beginning to try to

    solve this puzzle and have argued that the glass ceiling is held in place, in part, because bosses

    hold stereotypical views of ethnic minority and women leaders (Desai, Chugh & Brief, 2014;

    Hoobler, Wayne & Lemmon, 2009; Joshi, 2014; Westphal & Stern, 2007), and therefore hold

    such leaders to higher performance standards (Kulich, Trojanowski, Ryan, Haslam, &

    Renneboog, 2011).

    The purpose of our research is to build on these findings and to provide deeper insight

    into the puzzle of why nearly all large U.S. organizations are controlled by white men (Hoobler

    et al., 2009; Hymowitz & Schellhardt, 1986; Lyness & Heilman, 2006; Ridgeway, 2011).

    Especially now that nonwhites and women outnumber white men in the U.S. workplace by a

    margin of 2 to 1 (Burns, Barton & Kerby, 2012), it is surprising that white men continue to hold

  • 7/21/2019 CU diversity study

    4/65

    4

    85 percent of the CEO and corporate board positions in Fortune 500 companies (Catalyst, 2012;

    Hillman, Shropshire & Cannella, 2007). Moreover, white men have been found to receive lower

    performance ratings than their nonwhite or female workplace counterparts in some contexts

    (Rosette & Tost, 2010; Zenger & Folkman, 2012). Taken together, this body of evidence

    suggests that despite nonwhites and women outnumbering and sometimes outperforming their

    white male counterparts, only rarely are they given the reigns of the most powerful organizations

    in society. Economists are perhaps most disturbed by this phenomenon because orthodox

    economic theory would predict that it is suboptimal for society to select its top leaders from only

    34 percent of the population (i.e. the white men; Economist, 2008).

    1

    Organizational diversity programs are thought to empower ethnic minority and female

    leaders to individually place tiny cracks in the glass ceiling by championing the cause of

    organizational diversity wherever such leaders are located (Ely 1994; Ibarra 1995; Ragins &

    Scandura 1999). However, there is a growing body of work indicating that ethnic minority and

    women leaders might avoid hiring and promoting fellow ethnic minorities and women because

    such leaders feel threatened by fellow members of their own demographic groups (Duguid, 2011;

    Duguid, Loyd, & Tolbert, 2012). We take this idea one step further and suggest that engaging in

    workplace diversity-valuing behavior is not universally beneficial, and that ethnic minority and

    women leaders may actually be penalized in the form of lower performance ratings from

    diversity-valuing behavior whereas white men may not. In fact, we go as far as to argue that this

    divergent effect of diversity-valuing behavior on performance ratings results from supervisors

    1Indeed, the glass ceiling is a major reason why nonwhites and women are considered minorities, even thoughtogether they comprise a numerical majority. The term minority does not refer to a smaller number of people

    compared to the dominant group, but rather refers to a group that holds few positions of social power (Schaefer,1996). Affirmative action programs and corporate diversity offices have been put in place with the purported goal ofhelping minorities break through this glass ceiling and achieving greater organizational power and influence(Harrison, Kravitz, Mayer, Leslie & Lev-Arey, 2006; Levi & Fried, 2008). However, despite the increasingemphasis on promoting diversity, ethnic minorities and women are still under-represented at the highestorganizational levels and over-represented at the lower organizational levels (Leslie, Mayer & Kravitz, 2013).

  • 7/21/2019 CU diversity study

    5/65

    5

    tendency to negatively stereotype diversity-valuing ethnic minority leaders as incompetent, and

    their tendency to negatively stereotype diversity-valuing female leaders as interpersonally cold.

    Moreover, we contend that ethnic minority and women leaders tend to avoid being negatively

    stereotyped and tend to be viewed as strong performers to the extent they engage in low levels of

    diversity-valuing behavior. We suggest that when nonwhite or women leaders engage in

    diversity-valuing behaviors, they violate the expectation that as minorities, they will play a

    supporting, rather than a leading, role in society (Sheppard & Aquino, 2013).

    Our theoretical rationale challenges the conventional wisdom that ethnic minority and

    women leaders will be motivated to shatter the glass ceiling (Ely 1994; Ibarra 1995; Ragins &

    Scandura, 1999), and helps begin to solve the puzzle of why the glass ceiling persists despite its

    societal costs. Specifically, the glass ceiling may persist in part because nonwhite and women

    leaders who engage in diversity-increasing behaviors in the highest organizational ranks are

    systematically penalized with lower performance ratings for doing so. Our findings suggest that

    nonwhite and women leaders may increase their own chances of advancing up the corporate

    ladder by actually engaging in a very low level of diversity-valuing behavior. This idea is

    consistent with the anecdotal evidence portrayed in our introductory quotes, where a powerful

    woman distances herself from her gender, and a powerful African-American distances himself

    from his ethnicity. By downplaying their race and gender, these leaders may be viewed as more

    warm and competent, and thus be viewed as worthy of being promoted into the highest

    organizational echelons. Our results also help explain why there are so few leaders willing to

    publicly advocate for nonwhite or women leaders to be promoted (Ibarra, Carter & Silva, 2010),

    and why ethnic minorities and women might feel threatened at the prospect of hiring a fellow

    member of their demographic group (Duguid, 2011; Hansen, Ibarra & Peyer, 2010).

  • 7/21/2019 CU diversity study

    6/65

    6

    Our theoretical rationale also contributes to the literature on tokenism which has

    struggled to explain why minorities and women may impede the advancement of their fellow

    women (queenbee syndrome,Staines, Tavris, & Jayaratne, 1973: 55) and nonwhite (crab

    mentality,Mendoza, 2002, p. 57) coworkers. Indeed the tokenism literature suggests that token

    nonwhites and women are placed in positions of power by powerful white or male executives

    with the express purpose of keeping other nonwhites and women out of the upper echelons of

    organizations (Kanter, 1977). Not only does the idea that white men are collectively conspiring

    against their competent nonwhite or women coworkers defy common sense, but also no

    empirical link between tokenism and nonwhite or female career outcomes, such as performance

    ratings has been observed (Derks, Ellemers, van Laar, & de Groot, 2011; Sheppard & Aquino,

    2013). However, our findings explain that rather than being a conscious decision to promote

    white men instead of nonwhites or women, tokenism could emerge from minority and female

    leaders' awareness that engaging in diversity-valuing behavior is personally costly, and avoiding

    such behavior is personally beneficial.

    Our approach of building and extending organizational theory in order to understand and

    solve organizational problems aligns withAMJsstrategic vision of bringingorganizational

    problems to the forefront,which helps make organizational research more relevant to

    managers, and more interesting for our readers(George, 2014: 2). Our findings also highlight

    the potential importance of researchers like ourselves in shifting from a focus on studying

    diversity to a focus on studying demographic unselfishness. As our results suggest, diversity-

    valuing minorities and women leaders justifiably may be viewed as promoting race- and gender-

    based selfishness because "valuing diversity" typically refers to greater inclusion of demographic

    minorities and women (Richard, 2000, p.164). Just as it might be unfair to ask white men how

  • 7/21/2019 CU diversity study

    7/65

    7

    much they value homogeneity, our findings suggest that it may also be unfair to ask minorities

    and women how much they value diversity. An approach that might be fairer to members of all

    demographic groups would be to simply study the degree to which people value hiring and

    promoting individuals who are demographically dissimilar from themselves. By leading the

    charge in this respect, perhaps we researchers can have a better chance of rectifying the

    demographic imbalances that characterize today's organizations.

    Our research also has several methodological strengths that help bolster confidence in our

    findings. Our moderated mediation model (Muller, Judd, & Yzerbyt, 2005) linking the

    interaction of leader diversity-valuing behavior and demographics to performance ratings

    through the mechanism of stereotypical leader attribute ratings was confirmed using both the

    Sobel test (MacKinnon, Warsi, & Dwyer, 1995) as well as with Preacher and colleague's

    bootstrapping methodology (Preacher, Rucker & Hayes, 2007). Moreover, our full theoretical

    model was partially confirmed in a large field sample of 362 high-level executives (CEOs, other

    C-level leaders, vice presidents, directors, and board-level professionals), and fully confirmed in

    a laboratory setting using trained actors where we were able to manipulate our predictor

    variables (diversity-valuing behavior and leader demographics). This approach is consistent with

    the idea of full-cycle research (Chatman & Flynn, 2005), wherein a naturally occurring

    phenomenon is observed in the field, and then brought into a carefully controlled laboratory

    setting to verify the causal process and intervening mechanisms. A final strength is that our field

    study data were derived from multiple respondents (peer and supervisor ratings) which helps

    minimize concerns regarding common-method bias (Podsakoff, MacKenzie, Lee, & Podsakoff,

    2003).

    Diversity-valuing behavior, leader demographics and performance ratings

  • 7/21/2019 CU diversity study

    8/65

    8

    Glass ceiling background

    Although at a macro-level there is a great deal of evidence that a glass ceiling exists for

    ethnic minority and women leaders (i.e. minorities and women are under-represented at the

    highest organizational levels; Hitt & Barr, 1989; Hillman, Cannella & Harris, 2002; Zweigenhaft

    & Domhoff, 2006; Hillman, Shropshire & Cannella, 2007), at a micro level, there is little

    evidence of a main effect of race or gender on leader performance evaluations (Eagly, Makhijani

    & Klonsky, 1992; Joshi & Roh, 2009; Ng, Eby, Sorensen & Feldman 2005; Rosette, Leonardelli

    & Phillips 2008). In fact, the most recent meta-analytic findings regarding leader gender show

    that women tend to receive slightly better average performance ratings than their male

    counterparts (Roth, Purvis, & Bobko, 2012). Certainly, ethnic minorities and women have made

    some progress in recent years in attaining leadership positions as the percentage of Fortune 500

    ethnic minority or women CEOs has doubled from 4 to 8 percent over the last decade (Cook &

    Glass, 2013; Zweigenhaft & Domhoff, 2011) and 79 percent of working men report having

    worked for a female boss at some point in their careers (Elsesser & Lever, 2011). Indeed, recent

    evidence suggests that ethnic minority and women leaders tend to be viewed as belonging to the

    "in-group" of white male top executives to the extent they are similar to the white male

    incumbents in other ways (i.e. similar age, leadership experience, or functional background;

    Urban and Miller, 1998; Gaertner and Dovidio, 2000; Zhu, Shen & Hillman, 2014:1), or

    ingratiate themselves to these incumbents (Westphal & Stern, 2006; Westphal & Stern, 2007).

    Building on these ideas, we suggest that only when minority or women leaders behave in a way

    that highlights their atypical race or gender attributes will these demographic differences

    negatively influence such leaders' performance evaluations.

  • 7/21/2019 CU diversity study

    9/65

    9

    When evaluating others, people tend to gather information that helps them determine

    whether an individual is a potential threat or an opportunity (Wojciszke, Bazinska, & Jaworski,

    1998). Because threatening information (i.e. a perceived dissimilarity) is generally viewed to be

    more important than information indicating that a person presents an opportunity (i.e. a

    perceived similarity; Baumeister, Bratslavsky, Finkenauer & Vohs, 2001), perceived

    dissimilarities loom larger than perceived similarities (Bodenhausen, Kang, & Peery, 2012;

    Fiske, Cuddy & Glick, 2007). Race and gender have been found to be the two of the most

    important demographic markers leaders use to determine the degree to which fellow leaders are

    different from themselves (i.e. more important than age, education, functional background, or

    leadership experience; Zhu et al., 2014). Individuals tend to believe that their race- and gender-

    based demographic groups are engaged in a zero-sum competition with other demographic

    groups, such that if other demographic groups gain status, their own demographic groups lose

    status (Sidanius, Pratto, & Mitchell, 1994). In the United States at least, most people consider

    white men to be members of a high-status social group and ethnic minorities and women to

    belong to low-status social groups (see Ridgeway [1991] for a review).

    Warmth and competence stereotypes

    In general, members of high-status demographic groups tend to be viewed as more

    interpersonally warm and competent than members of low-status demographic groups (Aquino

    & Bommer, 2003; Giannopoulos, Conway, & Mendelson, 2005; Fiske et al., 2007). Individuals

    who are judged as warm or competent are evaluated much more favorably than those who are not

    (Wojciszke et al., 1998). Indeed, warmth and competence are at the heart of demographic

    stereotypes, and together they account for 82% of the variance in perceptions of everyday social

    behaviors (Abele, Cuddy, Judd, & Yzerbyt, 2008; Fiske et al., 2008). Women tend to be held to

  • 7/21/2019 CU diversity study

    10/65

    10

    higher standards of warmth than men, and ethnic minorities tend to be held to higher standards of

    competence than whites (Cuddy, Fiske & Glick, 2007). Warmth perceptions involve judgments

    of others' morality and sociability, and competence perceptions involve judgments of others'

    efficiencyand capability (Abele et al., 2008; Cuddy, Fiske, & Glick, 2008; Leach, Ellemers, &

    Barreto, 2007).

    To the extent ethnic minority or women leaders behave in a way that emphasizes their

    low-status race or gender characteristics, their perceived low-status demographics (and higher

    warmth and competence standards) become instantly salient, triggering these leaders to be

    viewed as social outsiders (Gaertner, Dovidio, Rust, Nier, Banker, Ward, Mottola & Houlette,

    1999; Dovidio & Gaertner, 2000; Park & Westphal, 2013). Indeed, when minority or women

    leaders engage in behaviors that are unusual or atypical, they highlight their demographic

    dissimilarity from the white male incumbents and tend to be undervalued and discounted by their

    workplace peers (Eagly and Karau, 1991; Eagly et al., 1992), especially in contexts where such

    demographic minorities also represent a numerical minority (e.g. a high-level leadership context;

    Ridgeway & Smith-Lovin, 1999; Carli, 2010). The particular behavior we focus on is diversity-

    valuing behavior, which we argue leads observers to pay attention to the race and gender of the

    leader engaging in that behavior, and thus enables race and gender biases to affect how those

    leaders are perceived and evaluated. As noted above, "valuing diversity" is generally viewed to

    be a euphemism for hiring and promoting demographic minorities and women (Richard, 2000:

    164). Minority and women leaders' engagement in diversity-valuing behavior may be viewed as

    selfishly advancing the social standing of their own low-status demographic groups.

    Women leaders, warmth, and diversity-valuing behavior

  • 7/21/2019 CU diversity study

    11/65

    11

    Starting with our rationale regarding gender, we suggest that female leaders who engage

    in diversity-valuing behavior will tend to be viewed as less warm and receive lower performance

    ratings than their equally-diversity-valuing male leader counterparts. Women are explicitly and

    implicitly expected by society and even themselves to be interpersonally warmer than men

    (Rudman, Greenwald & McGhee, 2001). Traditionally women are expected to serve and support

    men (Wurtzell, 1998), and to the extent they are viewed as desiring to selfishly increase their

    own social standing by engaging in diversity-valuing behavior we expect that they will be

    viewed by their bosses as cold and scheming to subvert the existing social order. When leaders

    display selfishness, they lose status and are viewed as less leader-like (Hardy & Van Vugt, 2006;

    Melwani, Mueller, & Overbeck, 2012). Indeed as an article from the popular press notes, a range

    of powerful women leaders have been derogated by the media as being interpersonally cold:

    "England's Margaret Thatcher, was called "Attila the Hen." Golda Meir, Israel's first female

    prime minister, was "the only man in the Cabinet." Richard Nixon called Indira Gandhi, India's

    first female prime minister, "the old witch." And Angela Merkel, the current chancellor of

    Germany, has been dubbed "The Iron Frau.""(Vedantam, 2007). Certainly, women leaders who

    threaten the status quo tend to be viewed negatively by observers (Rudman, Moss-Racusin,

    Phelan & Nauts, 2012). We suggest that the cold woman stereotype will be placed upon

    diversity-valuing female leaders, delegitimizing their authority and preventing such women from

    gaining power in organizations, thereby perpetuating the glass ceiling.

    Simply put, women leaders who engage in diversity-valuing behavior will be viewed as

    selfishly promoting women, which will lead their bosses to stereotype them as cold, and this

    judgment will result in lower performance ratings. We may even find evidence of factors

    perpetuating tokenism in the form of the queen bee syndrome (Chattopadhyay, Tluchowska, and

  • 7/21/2019 CU diversity study

    12/65

    12

    George, 2004; Derks, Van Laar, Ellemers, & de Groot, 2011; Staines et al., 1973), such that

    women leaders who perform low levels of diversity-valuing behavior are rewarded with higher

    warmth and performance ratings and thereby increased chances of advancing up the corporate

    ladder.

    Hypothesis 1. Leader sex will moderate the influence of leader diversity-valuing behavior

    on performance ratings, and this moderating effect will be mediated by perceived

    warmth. That is, women leaders who engage in diversity-valuing behavior will be rated

    as less warm than their equally-diversity-valuing male leader counterparts, and these

    lower warmth ratings will be associated with lower performance ratings.

    Nonwhite leaders, competence, and diversity-valuing behavior

    Our argument linking leader diversity-valuing behavior and race to performance ratings

    parallels our argument for leader diversity-valuing behavior and sex. Specifically, we suggest

    that nonwhite leaders who engage in diversity-valuing behavior will tend to be viewed as

    selfishly advancing nonwhites, leading to negative stereotypical judgments of such leaders'

    competence and ultimately lower performance ratings. In contrast, we expect white leaders who

    engage in diversity-valuing behavior to avoid being viewed negatively because their diversity-

    valuing behavior is unlikely to be viewed as selfishly promoting their own demographic group.

    Demographic group status largely determines stereotypical judgments of group competence

    (Cuddy et al., 2008). In general, nonwhites tend to be stereotyped as incompetent and thus

    society tends to demand that racial minorities prove a higher level of competence, work harder

    and attain more education in order to be judged as equally competent to their white counterparts

    (Biernat & Kobrynowicz, 1997; Ng, Eby, Sorensen & Feldman, 2005; Yarkin, Town, &

  • 7/21/2019 CU diversity study

    13/65

    13

    Wallston, 1982). Indeed, 65% of people have been found to stereotype nonwhites as

    "unintelligent" and 75% have been found to stereotype nonwhites as lazy(Devine, 1989, p. 8).

    Thus we suggest that the incompetent nonwhite stereotype is a subtle social mechanism that

    applies to diversity-valuing nonwhite leaders, delegitimizing their authority, and thereby

    strengthening the glass ceiling. To be clear, we are not arguing that there will be a main effect of

    leader race on competence and performance ratings, but rather that leader race will influence

    competence and performance ratings to the extent leaders make their race salient by engaging in

    diversity-valuing behavior.

    Simply put, nonwhite leaders who engage in diversity-valuing behavior will be viewed as

    selfishly promoting nonwhites, which will lead their bosses to stereotype them as incompetent,

    and this judgment will result in lower performance ratings than their equally-diversity-valuing

    white leader counterparts. We may even find evidence of factors facilitating racial tokenism (i.e.

    Mendoza, 2002), such that nonwhite leaders who engage in an extremely low level of diversity-

    valuing behavior are rewarded with higher competence and performance ratings and thereby

    increased chances of advancing up the corporate ladder.

    Hypothesis 2. Leader race will moderate the influence of leader diversity-valuing

    behavior on performance ratings, and this moderating effect will be mediated by

    perceived competence. That is, nonwhite leaders who engage in diversity-valuing

    behavior will be rated as less competent than their equally-diversity-valuing white leader

    counterparts, and these lower competence ratings will be associated with lower

    performance ratings.

    Method

  • 7/21/2019 CU diversity study

    14/65

    14

    We conducted two studies to test our hypotheses. Following the full-cycle research

    approach (Chatman & Flynn, 2005), we first tested our conceptual model in the field, and then

    sought to replicate our findings in a highly controlled laboratory context.

    Study 1

    Sample

    Our sample consisted of an entire calendar year cohort of 362 executives working in the

    United States who went through an executive development program. These executives were

    rated by their bosses and peers via a confidential online survey in which these peers and bosses

    reported their ratings of the executives diversity-valuing behavior, warmth, competence and

    performance two weeks prior to the start of the executive development program (100% response

    rate). As far as demographics are concerned, 13.8% of the executives were nonwhite and 30.7%

    were female. As for organizational rank, 88.7% were executives (vice presidents, directors, and

    board level professionals), and 11.3% were one rank above executives (i.e. top managers such

    as CEOs or other C-level leaders). The majority of ratees (89%) had abachelorsdegree or

    graduate degree. Each ratee was rated by a single boss, and an average of 3.61 peers (median =

    3). Peers and bosses were demographically similar to executive ratees, as 31% of the peers were

    women and 16% were nonwhite. Thirteen percent of the executivesbosses were women and

    8.8% were nonwhite.

    Measures

    Leader performance. We used an adaptation of Sadri, Weber and Gentrys(2011) three-

    item measure of leader performance by using bosses ratings on a five-point Likert scale (1 =

    among the worst, 5 = among the best): (1) How would you rate this person's performance in

    his/her present job?, (2) How effectively would this person handle being promoted in the same

  • 7/21/2019 CU diversity study

    15/65

    15

    function or division (moving a level up)?, and (3) Where would you place this person as an

    executive relative to other executives inside and outside your organization? Coefficient alpha for

    this measure was .88.

    Leader sex. We coded male leaders as 0 and females as 1. One-hundred eleven leaders

    were female (30.7%), which is consistent with the national average of female leaders nationwide

    (Lyness & Judiesch, 2008). Only eight leaders (2.2%) were both nonwhite and female.

    Leader race. Each leader reported their own race and each boss reported the race of the

    leader they were rating. There were no disagreements between boss and leader reports of leader

    race and we coded whites as 0 and nonwhites as 1. Of the 362 leaders in the sample, 312 were

    Caucasians (86.2%) and 50 were nonwhite (13.8%). The percentage of nonwhite leaders is fairly

    consistent with the national average of nonwhite leaders nationwide (Lyness & Judiesch, 2008).

    Leader diversity-valuing behavior. We used an adaptation of Kossek & Zonias (1993)

    measure to capture the degree to which peers perceived the leader as engaging in racial and

    gender diversity-valuing behaviors. Other-reports of behavior are thought to be more accurate

    than self-reports (Morgeson, Campion, Dipboye, Hollenbeck, Murphy, & Schmitt, 2007), which

    is why many researchers have argued for the use of observer reports (Connelly & Ones, 2010;

    Oh, Wang, & Mount, 2011; Zimmerman, del Carmen Triana, & Barrick, 2010), and which is

    why we used peer-reports in our analyses. On average, each leader was rated by 3.61 peers

    regarding the peers perceptions of the leaders diversity-valuing behaviors, and we averaged the

    peer ratings to create the mentoring score for each leader. The following three items were rated

    on a five-point Likert scale (1 = not at all, 5 = to a very great extent): (1) Understands and

    respects cultural, religious, gender and racial differences, (2) Values working with a diverse

    group of people, and (3) Is comfortable managing people from different racial or cultural

  • 7/21/2019 CU diversity study

    16/65

    16

    backgrounds. Coefficient alpha for the average peer ratings of each leader of this measure was

    .87. The peer-ratings of each leaders diversity-valuing behavior had an acceptable level of

    agreement as indicated by a significantFstatistic for ANOVA (F = 2.11, df = 361, p< .01) and

    rwgand ICC(1,2) statistics above acceptable cutoffs (rwg= .92; ICC(1) = .27; ICC(2) = .52), and

    thus were aggregated. Aggregating data is appropriate when theFstatistic for ANOVA is

    significant, rwgis higher than .70 (James, Demaree, & Wolf, 1984), and ICC(1) is non-zero

    (Bliese, 2000). Certainly, ICC(2) values should be higher than .70 (Kozlowski & Klein, 2000),

    but a low ICC(2) value simply indicates lower power in detecting relationships involving Level

    2 variables (and thus a more conservative test of the hypotheses), and thus does not prevent

    aggregation (Bliese, 2000).

    Leader warmth. We used an adaptation of Cuddy, Fiske and Glicks (2007) warmth

    scale to capture the degree to which leaders were perceived as being interpersonally warm by

    their bosses. The core of perceived warmth is perceived morality (Leach et al., 2007), and in fact

    this construct was initially labeled morality (Wojciszke et al., 1998: 1251) but the label

    evolved to morality (warmth) (Fiskeet al., 2007: 77) and currently is often referred to as

    simply warmth (Cuddy, Glick & Beninger, 2011: 73). Although researchers may disagree on

    the label for this construct, they broadly agree on its content as including the primary sub-

    dimension of morality and the secondary sub-dimension of sociability (Abele & Wojciszke,

    2007; Abele, Cuddy, Judd & Yzerbyt, 2008; Brambilla, Rusconi, Sacchi & Cherubini, 2011).

    This measure was rated by leadersbosses in order to reduce social desirability bias (Nederhof,

    1985), and to further enhance the measures validity (Zimmerman et al., 2010). Bosses rated the

    frequency that they believed the leader exhibited the following four characteristics on a six-point

    Likert scale (1 = never, 6 = always): (1) Deceptive - Conceals the truth for selfish reasons

  • 7/21/2019 CU diversity study

    17/65

    17

    (reverse-coded) (2) Credible - Worthy of trust, believable (3) Ethical - Lives within societys

    standards of right and wrong (4) Scheming - Develops sly and devious plans (reverse-coded).

    Coefficient alpha was .83.

    Leader competence.We used an adaptation of Cuddy, Fiske and Glicks (2007)

    competence scale to capture the degree to which leaders were perceived as being competent by

    their supervisors. Bosses rated the frequency that they believed the leader exhibited the following

    three characteristics on a six-point Likert scale (1 = never, 6 = always): (1) Ambitious - Highly

    motivated; determined to make progress, (2) Productive - Gets a lot done, (3) Driven - Has a

    burning, over-whelming passion to succeed. Coefficient alpha for our measure was .72.

    Controls. Following Spector and Brannicks(2011) advice to avoid over-controlling

    variance, we only included those control variables that we expected to affect our hypothesized

    relationships. Industry. Because different industries have different norms regarding the

    appropriate ranges of job performance ratings (Brutus, Fleenor & London, 1998), we created 26

    industry dummy variables representing each of the 26 industries in our sample (e.g. Aerospace,

    Automotive, Banking, Chemicals, Computer Hardware, Computer Software, Food and beverage,

    Government, Manufacturing, Non-profit, Pharmaceuticals, Retail, and Utilities).Job function.

    Because different job functions have different norms regarding the appropriate ranges of job

    performance ratings (Goodhue & Thompson, 1995), we created 20 job function dummy variables

    representing each of the 20 job functions in our sample. Leaders were distributed fairly equally

    across these functions: Accounting, Administration, Advertising/Public Relations,

    Credit/Finance, Education, Engineering, Human Resources/Training, Information Systems/Data

    Processing, Law, Manufacturing, Marketing, Materials Management/Purchasing, Medicine,

    Operations, Other, Product Development, Research and Development, Research/Analysis, Sales,

  • 7/21/2019 CU diversity study

    18/65

    18

    and Top Management.Pay. Higher paid leaders likely have more influence in the organization,

    thereby decreasing the likelihood of negative stereotypes and increasing their performance

    ratings. The bosses were asked to report how much the leaders were paid annually. On average

    the leaders were paid $173,824 per year. Organization Size. Diversity initiatives are likely more

    common in large organizations, arguably making diversity valuing less notable in these types of

    workplaces. Bosses were asked to report the approximate number of employees working for the

    organization. The leaders managed fairly large organizations, as the organizations in our sample

    employed an average of 4,717 workers. Leader education level. Because more educated leaders

    may be more savvy about avoiding being stereotyped negatively, the leaders were asked to report

    their highest educational degree (1 = high school; 2 = associates; 3 =bachelors; 4 = Masters;5

    = Doctorate/Professional). The majority of leader ratees (89%) had abachelorsdegree or

    graduate degree, with the mean education being 3.36). Leader organizational level. Because it

    becomes more difficult to differentiate between leader quality at higher levels (Avolio,

    Walumbwa & Weber, 2009), the leaders and their bosses were asked to determine each leaders

    organization level according to six choices (1 = hourly employee, machine operators,

    clerical/secretarial and support staff, technicians; 2 = first level forepersons, crew chiefs, section

    supervisors; 3 = middle, office managers, professional staff, mid-level administrators; 4 = upper

    middle, department leaders, plant managers, senior professional staff; 5 = executives, vice

    presidents, directors, board-level professionals; and 6 = Top managers, chief executives, or

    operating officers, presidents). All individuals in our sample were classified by themselves and

    their bosses as being in the top two categories. The average leader organization level was 5.11.

    Boss familiarity with leader. Because bosses who are more familiar with their followers tend to

    rate these individuals as better performers (Kingstrom & Mainstone, 1985), we controlled for

  • 7/21/2019 CU diversity study

    19/65

    19

    boss familiarity with the leader ratee using a single-item measure evaluated with a four-point

    Likert scale. This item was, How well do you know the ratee? (1 = I hardly know this person;

    2 = I do not know this person well; 3 = I know this person moderately well; 4 = I know this

    person extremely well). On average bosses rated themselves as knowing the leaders in their

    charge as quite well (M= 3.51).Boss organizational level. To account for the tendency for top

    managers to be more lenient in their ratings (Avolio, Walumbwa & Weber, 2009), we controlled

    for each bosss organizational level. The bosses classified themselves according to the same

    scale as the leaders. The mean boss organizational ranking was 5.27, somewhat higher than the

    leaders they rated. In addition, because individuals tend to have a demographic similarity bias

    (Turban & Jones, 1988), we also controlled for the interaction of boss race X leader raceas well

    as the interaction of boss sex X leader sex. We measured and coded bosses race (0 = white, 1 =

    nonwhite). We measured and coded boss sex (0 = male, 1 = female). There were 314 male

    bosses (86.7 percent) and 48 female bosses (13.3 %). As with the leaders, only 8 of the leaders

    bosses (2.2%) were both nonwhite and female.

    Results

    Table 1 reports the means, standard deviations, and correlation coefficients between the

    criterion, predictor, and control variables. We also ran two ANOVAs testing whether there is a

    race and gender difference in reported diversity-valuing behavior and found that women are

    significantly more likely to be reported by their peers as engaging in diversity-valuing behavior

    than men (mean difference is .17,F= 9.81, df =1,p< .01), and nonwhites are significantly more

    likely to be reported by their peers as engaging in diversity-valuing behavior than whites (mean

    difference is .36,F= 24.26, df =1,p< .001). Hierarchical moderated regression models were

    used to examine the hypothesized interaction effects. Following Aiken and West (1991), all

  • 7/21/2019 CU diversity study

    20/65

    20

    variables involved in the interaction terms were mean-centered. Table 2 presents the results of

    the analysis for the influence of leader demographics and diversity-valuing behavior on the

    mediators, and Table 3 presents the results of the analysis for the influence of the main effects

    and proposed mediators on leader performance.

    ------------------------------------------------Insert Tables 1 through 3 About Here

    ------------------------------------------------

    Figure 1 shows that we hypothesized two models of moderated mediation, which means

    that we predicted that the joint influence of a main effect (diversity-valuing behavior) and a

    moderator (leader race/sex) on a criterion variable will operate through a mediator (leader

    warmth/competence; Muller et al., 2005). Thus, in Models 1 and 4 of Table 2, all the control

    variables and main effects for predicting the mediators are included. In Models 2 and 5 (Table 2),

    all the control variables, main effects and the hypothesized interactions of our mediators are

    included.

    In Model 1 of Table 3, all the control variables for predicting our criterion variable are

    included. In Model 2 (Table 3) all the control variables and main effects for predicting our

    criterion variable are included. In Model 3 (Table 3), all the control variables, main effects and

    the interaction of leader sex X diversity-valuing behavior are included for predicting our

    criterion variable. In Model 4 (Table 3), all the control variables, main effects, the interaction of

    leader sex X diversity-valuing behavior and the proposed mediator of leader warmth are included

    for predicting our criterion variable. In Model 5 (Table 3) all the control variables, main effects

    and the interaction of leader race X diversity-valuing behavior are included for predicting our

    criterion variable. In Model 6 (Table 3), all the control variables, main effects, the interaction of

    leader race X diversity-valuing behavior and the proposed mediator of leader competence are

  • 7/21/2019 CU diversity study

    21/65

    21

    included for predicting our criterion variable. In Model 7 (Table 3), all the control variables,

    main effects and both interactions are included for predicting our criterion variable. In Model 8

    (Table 3), all the control variables, main effects, both interactions and both proposed mediators

    are included to demonstrate that our mediation effects hold even when both mediators are

    included in the model.

    In hypothesis 1 we argued that the gap in performance ratings between male and female

    leaders resulting from diversity-valuing behavior would be mediated by perceived warmth. That

    is, women leaders who valued diversity would be rated by their bosses as less warm than their

    male leader counterparts who valued diversity and this lower warmth rating would be associated

    with a lower performance rating. As shown in Table 2, the coefficient for the interaction term

    involving leader sex and diversity-valuing behavior was significant for leader warmth (= -.12,

    p < .05). The Aiken and West (1991) methodology demonstrated leader diversity-valuing

    behavior was only negatively associated with boss ratings of leader warmth for women, = -.22,

    p< .05, not men,= .28,p< .05.(See Figure 2). Likewise, Table 3 shows that the coefficient

    for the interaction term involving leader sex and diversity-valuing behavior was significant for

    leader performance (= -.16,p < .01). Looking at the plots in Figure 3, diversity-valuing

    behavior was only negatively associated with boss ratings of leader performance for women, =

    -.33,p< .01, not men, = .25,p< .01. Moreover, when leader warmth was entered into the

    model predicting leader performance rating, the strength and significance of the coefficient for

    the interaction of leader sex and diversity-valuing behavior decreased, which suggests that leader

    warmth partially mediated the joint influence of leader diversity-valuing behavior and sex on

    leader performance ratings.

  • 7/21/2019 CU diversity study

    22/65

    22

    To further test for mediation, we ran Sobel (MacKinnon et al., 1995) and bootstrap

    (Lockwood, Hoffman, West & Sheets, 2002) analyses and both revealed evidence of mediation.

    Specifically the Sobel mediation test statistic was significant (t= 2.17,p< .05), indicating leader

    warmth mediated between the interaction of leader sex X diversity-valuing behavior and leader

    performance. For our bootstrapped mediation analysis (MacKinnon, Lockwood & Williams,

    2004; Schneider, Ehrhart, Mayer, Saltz & Niles-Jolly, 2005), we created 1,000 bootstrap samples

    and thus 1,000 estimates of the mediated effect. This analysis tested for the joint effect of leader

    sex and diversity-valuing behavior on leader performance rating through the mediator of leader

    warmth. The overall indirect effect was -.06 (the 95% confidence interval for the mediated effect

    ranged from -.13 to -.02 and did not straddle zero), indicating that the mediating effect was

    significant (p< .05). Taken together, the evidence suggests that hypothesis 1 is supported.

    In hypothesis 2, we argued that the gap in performance ratings between white and

    nonwhite leaders resulting from diversity-valuing behavior would be mediated by perceived

    competence. That is, nonwhite leaders who engaged in diversity-valuing behavior would be rated

    by their bosses as having less competence than their white, diversity-valuing leader counterparts,

    and this lower competence rating would be associated with a lower performance rating. As

    shown in Table 2, the coefficient for the interaction term involving leader race and diversity-

    valuing behavior was significant for leader competence (= -.16,p < .05). The Aiken and West

    (1991) methodology demonstrated leader diversity-valuing behavior was only negatively

    associated with boss ratings of leader competence for nonwhite leaders, = -.29,p< .05, not

    white leaders, = -.04, n.s.(See Figure 4). However, Table 3 shows the coefficient for the

    interaction term involving leader race and diversity-valuing behavior was not significant for

    leader performance (= -.03, n.s.).

  • 7/21/2019 CU diversity study

    23/65

    23

    Although we tested for mediation using the Sobel (Baron & Kenny, 1986) and

    bootstrapped (Lockwood, Hoffman, West & Sheets, 2002) analyses, neither revealed evidence of

    mediation. Certainly, we found evidence that the interaction of leader diversity-valuing behavior

    and leader race predicted boss ratings of leader competence; however we found no evidence that

    this interactive effect influenced performance ratings through competence ratings. Taken

    together, the evidence suggests that hypothesis 2 is not fully supported.

    ------------------------------------------------Insert Figures 1 through 4 about here

    ------------------------------------------------

    Discussion, Study 1

    Our first study explored whether leaders are judged differently for diversity-valuing

    behavior depending on their demographic characteristics. We found that diversity-valuing

    behavior was only negatively related to evaluations of leaders who were nonwhite or female

    leaders who are thought to have the greatest potential to dismantle the glass ceiling. This finding

    suggests that minorities and women might be able to advance their own careers by acting as

    tokens and engaging in a low level of diversity-valuing behavior.

    The relationships we observed indicate that biases against diversity-valuing minority

    employees may creep into performance evaluations. Certainly, our first study has several

    methodological strengths including the realism of a field study, leaders drawn from a range of

    industries, organizations, and functions, as well as multiple respondents from two sources (peers

    and supervisors). However, it also has some notable weaknesses including the fact that it is

    cross-sectional, it only includes a small percentage of nonwhites (13.8% of the surveyed leaders),

    the predictor variable of diversity-valuing behavior is measured subjectively (albeit with fairly

    objective peer ratings), and we cannot be sure that some unobserved variables such as leader

  • 7/21/2019 CU diversity study

    24/65

    24

    speech accents, communication styles, nonwhite-sounding leader names, or objective

    performance are driving or obscuring the effects that we observed. Therefore we conducted a

    second study in a highly controlled laboratory context where we manipulated our predictor

    variables of diversity-valuing behavior as well as leader demographics to address the limitations

    of our first study, and to provide a more rigorous test of our conceptual model.

    Study 2

    The goal of this study was to show that the specific behavior of promoting and advancing

    nonwhites and women into management, rather than simply observersperceptions of these

    behaviors, caused women and nonwhite leaders to be negatively stereotyped and receive lower

    performance evaluations. To that end, we designed an experiment wherein student participants

    were asked to observe a presentation given by leaders (trained actors hired from the acting

    department of a university) regarding a hiring decision for a vacant project manager position. In

    the experiment, all job candidates promoted by leaders were equally qualified and only differed

    based on their demographics. This aspect of the study design allowed us to reduce variability of

    the context under which diversity is promoted thereby providing a better test of whether the same

    behavior of advocating for nonwhite or female job candidates would produce different ratings of

    leadership performance depending on the leadersdemographics. Second, we assessed how

    student participants not only evaluated the leader, but also evaluated the job candidate.

    Sample

    A total of 395 university students from a major western U.S. public university watched a

    presentation where a leader from human resources (a trained actor from the University's Acting

    and Theater Department) gave a presentation advocating for one of four potential project

    manager job candidates. Participants were asked to take the role of human resource department

  • 7/21/2019 CU diversity study

    25/65

    25

    supervisors and evaluate the warmth, competence and job performance of both the leader (the

    actor) as well as the job candidate (fictional candidate information displayed on a PowerPoint

    slide). We randomly assigned 92 participants to watch presentations given by nonwhite male

    actors, 57 participants to watch presentations given by white male actors, 132 to watch

    presentations given by white female actors, and 114 to watch presentations given by nonwhite

    female actors. Because of difficulties scheduling actors and because of participant no-shows, we

    had unequal cell sizes. However, we ensured that our cell sizes satisfied the 20 cases per cell

    rule-of-thumb (VanVoorhis & Morgan, 2007). Overall, 15.8% of our participants were nonwhite

    and 41.1% were female, which is not too different from the overall U.S. population.

    Design

    To reduce participant awareness that they were participating in a race and gender related

    study, we used a between subjects design such that each participant observed one actor

    advocating for one candidate. To reduce concerns regarding single-stimulus bias and increase the

    generalizability of our experimental results (Highhouse, 2009), we hired two white male actors,

    two white female actors, two nonwhite male actors (American-born Latinos), and two nonwhite

    female actors (American-born Chinese). In their scripted speeches, each actor told the audience

    that the four finalists for the position were equally qualified and proceeded to advocate for one of

    the four candidate finalists during the presentation, with the facilitation of PowerPoint slides.

    The four candidate finalists included one white male, one white female, one nonwhite male

    (Latino), and one nonwhite female (Asian). To provide a conservative test of our theory, we used

    Asian and Latino American actors because Asian Americans are generally regarded as the model

    minority (Hurh & Kim, 1989; Lee, 1994), whereas Latino Americans are typically associated

    with more negative stereotypes (Burns & Gimpel, 2000; Dixon & Rosenbaum, 2004), although

  • 7/21/2019 CU diversity study

    26/65

    26

    less negative than those associated with African Americans (Devine & Elliot, 1995; Krueger,

    1996). We provided one slide for each of the four candidates, with information on the left half of

    the slide detailing each candidate's intelligence score, emotional intelligence score, interview

    performance rating, reference letter quality, personality test scores, work experience quality,

    ability test scores, and overall rating. The ratings for the sub-dimensions for all candidates were

    positive and equivalent and overall ratings for each candidate were 5 out of 5. On the right half

    of the slide we provided a high-resolution headshot photo purported to be of the candidate that

    was taken in a professional setting. The candidate information was fictional in order to minimize

    the risk that participants knew the candidates. After the presentation, the leader remained in the

    front of the room and participants were asked to anonymously rate the leader and candidate's

    perceived warmth, competence and performance, and to report their own demographic

    information.

    Measures

    Leader performance. Our measure of leader performance was an average score of the

    same three items used to measure leader performance in Study 1. Participants were asked to rate

    on a five-point Likert scale (1 = among the worst, 5 = among the best): (1) How would you rate

    this person's performance in his/her present job? (2) Where would you place this person as a

    leader relative to other leaders? (3) Where would you place this person as an executive relative to

    other executives inside and outside your organization? Coefficient alpha for this measure was

    .93.

    Leader warmth. We measured the perceived warmth of the actor/actress using the same

    measure in Study 1. Participants were asked to rate on a five-point Likert scale (1=strongly

    disagree, 5=strongly agree) the degree to which the actor is: (1) Deceptive - Conceals the truth

  • 7/21/2019 CU diversity study

    27/65

    27

    for selfish reasons (reverse-coded), (2) Credible - Worthy of trust, believable, (3) Ethical - Lives

    within societysstandards of right and wrong, and (4) Scheming - Develops sly and devious

    plans (reverse-coded). Coefficient alpha for this measure was .71.

    To further establish the validity of our warmth scale, we also measured warmth using

    Costa & McRaes (1992) NEO-PI-R 10-item altruism measure. Sample items of this measure

    include, This person makes people feel welcome, This person loves to help others, and This

    person has a good word for everyone. Coefficient alpha for this 10-item measure was .86.

    Across our entire sample, we found that the correlation between the two measures of warmth was

    .68 (p< .001). Moreover, when we ran an exploratory factor analysis including all items for both

    scales, all items loaded on one factor that explained 41% of the variance (Eigenvalue of 5.8).

    Across all 14 items, the average factor loading on this single factor was .64 (lowest value was

    .51). We viewed this result as strongly supporting the validity of our warmth items.

    Leader competence. Perceived competence of the actor/actress was measured with the

    same scale we used in Study 1. Participants were asked to rate on a five-point Likert scale

    (1=strongly disagree, 5=strongly agree) the degree to which the actor is: (1) Ambitious - Highly

    motivated; determined to make progress, (2) Productive - Gets a lot done, and (3) Driven - Has a

    burning, over-whelming passion to succeed. Coefficient alpha for this measure was .88.

    To further establish the validity of our competence scale, we also measured competence

    using Costa & McRaes (1992) NEO-PI-R nine-item achievement-striving measure. Sample

    items of this measure include, This person works hard, This person demands quality, and

    This person sets high standards for self and others. Coefficient alpha for this nine-item

    measure was .89. Across our entire sample, we found that the correlation between the two

    measures of competence was .83 (p< .001). Moreover, when we ran an exploratory factor

  • 7/21/2019 CU diversity study

    28/65

    28

    analysis including all items for both scales, all items loaded on one factor that explained 59% of

    the variance (Eigenvalue of 7.1). Across all 12 items, the average factor loading on this single

    factor was .76 (lowest value was .61). We viewed this result as strong evidence of the validity of

    our competence items.

    Condition. We had three conditionsone for sex, one for race and one that considered

    sex and race at the same time. The sex condition included sessions performed by white male or

    white female leaders (1 = the sessions in which presentations were given by white female actors,

    0 = the sessions in which presentations were given by white male actors). The race condition

    included sessions performed by white male or nonwhite male leaders (1 = the sessions in which

    presentations were given by nonwhite male actors, 0 = the sessions in which presentations were

    given by white male actors). The combined sex and race condition included sessions performed

    by white male or nonwhite female leaders (1 = the sessions in which presentations were given by

    nonwhite female actors, 0 = the sessions in which presentations were given by white male

    actors).

    Diversity-valuing behavior. Diversity-valuing behavior was measured by whether or not

    the leaders advocated for white male candidates. In the sex condition, diversity-valuing behavior

    was 1 when the leaders advocated for the white female candidate, and was 0 when the leaders

    advocated for the white male candidate. In the race condition, diversity-valuing behavior was 1

    when the leaders advocated for the nonwhite male candidate, and was 0 when the leaders

    advocated for the white male candidate. In the combined sex and race condition, diversity-

    valuing behavior was 1 when the leaders advocated for the nonwhite female candidate, and was 0

    when the leaders advocated for the white male candidate. The scripts for the high and low

    diversity-valuing speeches are below:

  • 7/21/2019 CU diversity study

    29/65

    29

    High diversity-valuing behavior script. "Imreally impressed with all four candidates, but

    the one that really stands out to me is [name of minority or woman candidate]. S/he will

    be a fantastic choice and I think s/he will be a great leader. On a personal note, I think we

    should really focus on the demographics of the employees. Diversity is really important

    to me, and I think it should be important to our organization. All of the candidates are

    equally qualified, so letsdo the right thing and hire a minority/woman."

    Low diversity-valuing behavior script. "Imreally impressed with all four candidates, but

    the one that really stands out to me is [name of white male candidate]. He will be a

    fantastic choice and I think he will be a great leader. On a personal note, he looks like a

    leader and looks like somebody who the team will respect. Whoever we hire will need to

    get up to speed really quickly and get a lot done in short amount of time. After

    interviewing all the candidates and looking at all their materials, Imconvinced he is the

    best person for the job."

    Controls. Even though all participants were told that the human resource committee gave

    an overall rating of 5 out of 5 to all job candidates, and thus all candidates were equally

    qualified, we controlled for the perceivedperformance of the candidateusing the same scale as

    leader performance. Participants were asked to rate the candidates on the same performance

    measure items used to evaluate the leader. Coefficient alpha for this measure was .89. Because

    politically liberal individuals tend to have a more favorable evaluation of diversity-valuing

    behavior (Harrison et al., 2006), we controlled forparticipant political affiliation(1 = self-

    identified member of the Democratic or Green parties, 0 = self-identified member of the

    Republican or Libertarian parties). Participant ages ranged from 18 to 29 years, and because

  • 7/21/2019 CU diversity study

    30/65

    30

    older participants may have more experience working under workplace diversity programs, we

    controlled forparticipant age. Because foreign-born individuals may have a unique

    understanding of U.S.-based demographic stereotypes, we controlled forparticipant foreign-

    born status(1 = born outside the U.S., 0 = born in the U.S.). Because being a non-native

    English-speaker could lead to participant misunderstanding the speeches, we controlled for

    participantyears speaking English. In addition, because individuals tend to have a demographic

    similarity bias (Turban & Jones, 1988), we also controlled for the interaction ofparticipant race

    X leader racein the race condition, the interaction ofparticipant sex X leader sexin the sex

    condition, and both of these interactions in the combined race and sex condition.

    Results

    We regressed leader performance on our controls, predictors, and interaction to determine

    the degree to whichparticipantsratings of actor performance reflected race and gender bias.

    Table 4 presents the means, standard deviations and correlations between the study variables and

    Tables 5 and 6 present the regression models we used to test our hypotheses. Consistent with

    Study 1 findings, Table 4 shows that across all conditions, diversity-valuing behavior is slightly

    negatively correlated with performance ratings (r= -.24,p< .05), indicating that on average

    participants did not have extremely positive (or negative) views of diversity-valuing behavior.

    ---------------------------------------------Insert Tables 4 through 6 about here---------------------------------------------

    In hypothesis 1, we argued that the gap in performance ratings between male and female

    leaders resulting from diversity-valuing behavior would be mediated by perceived warmth. That

    is, female leaders who engage in diversity-valuing behavior would be rated by participants as

    less warm than their male leader counterparts who engage in this same behavior, and this lower

    warmth rating would be associated with a lower performance rating. As shown in Tables 5 and 6

  • 7/21/2019 CU diversity study

    31/65

    31

    the coefficient for the interaction term involving leader sex and diversity-valuing behavior was

    significant for leader warmth (= -0.26,p< 0.01) and leader performance (= -0.32,p< 0.001).

    The increase of R-squared for the interaction term was 5% (p< 0.01) for leader warmth and 8%

    for leader performance (p< .001). Figure 5 and simple slope analysis (Aiken & West, 1991)

    demonstrate that the simple slope of diversity-valuing behavior on leader warmth is significant

    and negative for female leaders (= -.33,p< .05) while the simple slope of diversity-valuing

    behavior on leader warmth is slightly positive for male leaders (= .05, n.s.). Figure 6 and

    simple slope analysis (Aiken & West, 1991) demonstrate that the simple slope of diversity-

    valuing behavior on leader performance is significant and negative for female leaders (= -.44,p

    < .05) while the simple slope of diversity-valuing behavior on leader warmth is slightly positive

    for male leaders (= .08, n.s.). Table 6 reflects that when warmth was entered into the model

    predicting leader performance, the strength and significance of the coefficient for the interaction

    of sex and diversity-valuing behavior decreased, which suggests that warmth partially mediated

    the joint influence of leader diversity-valuing behavior and sex on leader performance ratings

    (Aiken & West, 1991). We conducted the Sobel test (MacKinnon et al., 1995) which further

    supports our hypothesized mediated effect (t= 2.57,p< .05).

    To further test for mediation, we ran the bootstrapping test for mediation which is

    thought to be superior to the Sobel test because it imposes no distributional assumptions on the

    sample (Preacher & Hayes, 2008). Across 1,000 bootstrapped samples, the overall indirect

    mediated effect for our full hypothesized model was calculated to be -.19 (95% confidence

    interval does not include zero and ranged between -.03 and -.39). The bootstrapped results show

    that the indirect effects of diversity-valuing behavior on performance ratings through the

  • 7/21/2019 CU diversity study

    32/65

    32

    mediator of warmth are significant and opposite for male and female leaders, providing further

    confirmation of our conceptual model. Thus, hypothesis 1 is supported.

    In hypothesis 2 we argued that the gap in performance ratings between white and

    nonwhite leaders resulting from diversity-valuing behavior would be mediated by perceived

    competence. That is, nonwhite leaders who engage in diversity-valuing behavior would be rated

    by participants as being less competent than their white counterparts who engage in this same

    behavior, and these lower competence ratings would be associated with lower performance

    ratings. As shown in Tables 5 and 6 the coefficients for the interaction terms involving actor race

    and diversity-valuing behavior are significant for leader competence (= -.34,p< 0.001) and

    performance (= -.37,p< 0.001). Figure 7 along with simple slope analysis demonstrate that

    diversity-valuing behavior is negatively associated with competence for nonwhite leaders (= -

    .42,p< 0.05) and positively associated with performance for white leaders (= .18,p< 0.05).

    Figure 8 along with simple slope analysis demonstrate that diversity-valuing behavior is

    negatively associated with performance for nonwhite leaders (= -.48,p< 0.05) and slightly

    positively associated with performance for white leaders (= .07, n.s.). The increase of R-

    squared for the interaction term is 12% (p< 0.001) for both leader competence and performance.

    Moreover, in Table 6, when leader competence was entered into the model predicting leader

    performance, the strength and significance of the coefficient for the interaction of leader race and

    diversity-valuing behavior decreased, which suggests that leader competence partially mediated

    the joint influence of leader diversity-valuing behavior and race on performance ratings. Results

    from the Sobel test further support our hypothesized mediated effect (t= 3.66,p< .001). Results

    from the bootstrapping analysis also supported this hypothesis (Preacher & Hayes, 2008). Across

    1,000 bootstrapped samples, the overall indirect effect was calculated to be -.70 (95% confidence

  • 7/21/2019 CU diversity study

    33/65

    33

    interval did not include zero and ranged between -1.15 and -.33). In this way, hypothesis 2 is

    supported.

    We also wanted to examine whether nonwhite female leaders might be in double

    jeopardy because of their race and sex. In order to test the joint interactive effect of race and sex

    with diversity-valuing behavior, we used the experimental design in which nonwhite female

    leaders were compared with white male leaders. Tables 5 and 6 indicate that the diversity-

    valuing behavior X nonwhite female leader interaction significantly influenced leader warmth (

    = -.26,p< .05), leader competence (= -.41, p < 0.001) and leader performance (= -.44,p