FP Koenig Eagly 2011

Are Leader Stereotypes Masculine?A Meta-Analysis of Three Research Paradigms

Anne M. KoenigUniversity of San Diego

Alice H. EaglyNorthwestern University

Abigail A. MitchellNebraska Wesleyan University

Tiina RistikariUniversity of Tampere

This meta-analysis examined the extent to which stereotypes of leaders are culturally masculine. Theprimary studies fit into 1 of 3 paradigms: (a) In Scheins (1973) think managerthink male paradigm, 40studies with 51 effect sizes compared the similarity of male and leader stereotypes and the similarity offemale and leader stereotypes; (b) in Powell and Butterfields (1979) agencycommunion paradigm, 22studies with 47 effect sizes compared stereotypes of leaders agency and communion; and (c) in Shinars(1975) masculinityfemininity paradigm, 7 studies with 101 effect sizes represented stereotypes ofleadership-related occupations on a single masculinityfemininity dimension. Analyses implementedappropriate random and mixed effects models. All 3 paradigms demonstrated overall masculinity ofleader stereotypes: (a) In the think managerthink male paradigm, intraclass correlation ! .25 forthe womenleaders similarity and intraclass correlation ! .62 for the menleaders similarity; (b) in theagencycommunion paradigm, g ! 1.55, indicating greater agency than communion; and (c) in themasculinityfemininity paradigm, g ! 0.92, indicating greater masculinity than the androgynous scalemidpoint. Subgroup and meta-regression analyses indicated that this masculine construal of leadershiphas decreased over time and was greater for male than female research participants. In addition,stereotypes portrayed leaders as less masculine in educational organizations than in other domains andin moderate- than in high-status leader roles. This article considers the relation of these findings to Eaglyand Karaus (2002) role congruity theory, which proposed contextual influences on the incongruitybetween stereotypes of women and leaders. The implications for prejudice against women leaders arealso considered.

Keywords: leadership, management, gender stereotypes, gender roles, meta-analysis

The characteristics that people commonly ascribe to women,men, and leaders contribute to the challenges that women face inobtaining leadership roles and performing well in them.1 Culturalstereotypes can make it seem that women do not have what it takesfor important leadership roles. This cultural mismatch, or roleincongruity, between women and the perceived demands of lead-ership underlies biased evaluations of women as leaders (Eagly &Karau, 2002). Fueling this mismatch is an inconsistency betweenthe predominantly communal qualities (e.g., nice, compassionate)that people associate with women and the predominantly agentic

qualities (e.g., assertive, competitive) that they believe are requiredfor success as a leader (Eagly & Carli, 2007). Given that agenticqualities are ascribed more to men than women (e.g., Gallup NewsService, 2001; Spence & Buckner, 2000), leadership is generallyassociated with masculinity. This article reports a meta-analysis ofresearch that has assessed the cultural masculinity of leader ste-reotypes and explores the conditions under which this masculinityis more or less pronounced.

The Importance of Stereotypes to Womens Accessto Leadership

Stereotypes often are a potent barrier to womens advancementto positions of leadership. This assertion is the consensus view not

1 In this article, we use the terms leader and manager interchangeably.Although a distinction between leadership and management is useful insome contexts (e.g., Bennis, 1989), the research that we review does notallow leader and managerial roles or functions to be separated. Also, weuse the terms sex and sexes to denote the grouping of people into femaleand male categories. The term gender refers to the meanings that societiesand individuals ascribe to these female and male categories. We do notintend to use these terms to give priority to any class of causes that mayunderlie sex and gender effects (see Wood & Eagly, 2010).

This article was published Online First May 30, 2011.Anne M. Koenig, Department of Psychological Sciences, University of

San Diego; Alice H. Eagly, Department of Psychology, NorthwesternUniversity; Abigail A. Mitchell, Department of Psychology, NebraskaWesleyan University; Tiina Ristikari, Department of Social Research,University of Tampere, Tampere, Finland.

Anne M. Koenig was supported in part by the National Institutes ofHealth under a Ruth L. Kirschstein National Research Service Award(1 F31 MH074251).

Correspondence concerning this article should be addressed to Anne M.Koenig, Department of Psychological Sciences, 5998 Alcala Park, Univer-sity of San Diego, San Diego, CA 92110. E-mail: [email protected]

Psychological Bulletin 2011 American Psychological Association2011, Vol. 137, No. 4, 616642 0033-2909/11/$12.00 DOI: 10.1037/a0023557

616

Ursula Liebhart

Ursula Liebhart

Ursula Liebhart

Ursula Liebhartjhdfhhdfkjfdn

only of social and organizational psychologists (e.g., Glick &Fiske, 2007; Heilman & Parks-Stamm, 2007; S. K. Johnson, Mur-phy, Zewdie, & Reichard, 2008), but also of women who havesubstantial experience as leaders. For example, a survey of 705women at the vice president level and above in Fortune 1,000corporations found that 72% agreed or strongly agreed that ste-reotypes about womens roles and abilities are a barrier to wom-ens advancement to the highest levels (Wellington, Kropf, &Gerkovich, 2003).

The damaging effects of stereotypes for women as leaders donot stem from beliefs about women that are mainly negative. Onthe contrary, consistent with the women-are-wonderful effect (Ea-gly & Mladinic, 1994; Langford & MacKinnon, 2000), women areregarded as the nicer, kinder sex and thus have a cultural stereo-type that is in general more positive than that of men. Neverthe-less, women often experience workplace discrimination (Heilman& Eagly, 2008). Theorists have resolved this paradox by reasoningthat it is not the evaluative content of the stereotype of women butits mismatch with many desirable work roles that underlies biasedevaluations in many employment settings (e.g., Eagly & Karau,2002; Heilman, 2001; Hogue & Lord, 2007; Lyness & Heilman,2006).

In Eagly and Karaus (2002) role congruity model, the mismatchthat produces biased evaluation is between stereotypes of women(e.g., Spence & Buckner, 2000) and stereotypes of leaders (e.g.,Epitropaki & Martin, 2004; Lord & Maher, 1993; Offermann,Kennedy, & Wirtz, 1994). In this model, stereotype content isframed in terms of agency and communion, with communion themore important theme in the female stereotype and agency themore important theme in both leader and male stereotypes. Menare therefore seen as more similar to the leader stereotype thanwomen are, producing disadvantage for women. In Heilmans(1983, 2001) broader lack-of-fit model, to the extent that a work-place role is inconsistent with the attributes ascribed to an indi-vidual, she or he would suffer from perceived lack of fit, producingincreased expectations of failure and decreased expectations ofsuccess. The incongruity between construals of women and leadersis thus one type of lack of fit.

These communal and agentic meanings ascribed to women andmen form a constant backdrop to social interaction, coloring thejudgments made about people encountered in organizations andother contexts (Wood & Eagly, 2010). In both the role congruity(Eagly & Karau, 2002) and the lack-of-fit models (Heilman, 2001),these beliefs comprise two kinds of expectations or norms: de-scriptive beliefs (or stereotypes), which are consensual expecta-tions about what members of a social group actually do, andinjunctive (or prescriptive) beliefs, which are consensual expecta-tions about what group members ought to do or ideally would do(Cialdini & Trost, 1998). Therefore, women are the targets of twoforms of prejudice against them as leaders: a deficit in the ascrip-tion of leadership ability to them and, compared to that of men, aless favorable evaluation of their agentic leadership behavior. Inother words, descriptively, women seem less usual or natural inmost leadership roles; and prescriptively, women often seem in-appropriate or presumptuous when they display the agentic behav-ior often required by these roles (see also Burgess & Borgida,1999).

Because individuals are commonly assimilated to group stereo-types (e.g., von Hippel, Sekaquaptewa, & Vargas, 1995), an eval-

uative penalty is exacted from women leaders or potential leadersregardless of whether they possess the qualities appropriate forleadership roles. This penalty consists of unfavorable performanceexpectations, which in turn enable biased judgments and lessfavorable evaluations (Eagly, Makhijani, & Klonsky, 1992; Lyness& Heilman, 2006). Discrimination is the behavioral outcome ofthese processes (Eagly & Diekman, 2005).

How strong is the evidence for the masculinity of leader stereo-types? Partial reviews of research on this question have substan-tiated the claim about leaders cultural masculinity (e.g., Duehr &Bono, 2006; Heilman, Block, Martell, & Simon, 1989; Powell,Butterfield, & Parent, 2002; Schein, 2001). However, these re-views have not identified all of the relevant research paradigms oraccessed all of the available studies within each paradigm orestimated the magnitude of the effects. Our integration of theresearch literature remedies these omissions and tests the propo-sitions of role congruity theory concerning the conditions underwhich incongruity between the leader and female gender stereo-type is stronger or weaker (Eagly & Karau, 2002).

Three Paradigms for Examining the Masculinity ofLeader Roles

Research in three separate paradigms has addressed the culturalmasculinity of leader stereotypes. Best known is the thinkmanagerthink male paradigm, which was created by Schein(1973). This method provides a direct test of the similarity ofleader stereotypes to male and female stereotypes. In these studies,separate groups of participants rated a leader category (e.g., suc-cessful middle managers), women, or men on a large number ofgender-stereotypical traits. The researchers correlated the meanratings of managers or leaders with the mean ratings of men and,separately, with the mean ratings of women. These correlationsrepresent the similarity of stereotypes of men and women tocultural concepts of leadership. The think managerthink maleeffect occurs when men and leaders are similar and women andleaders are not similar. Although these studies typically provideonly these correlations and not information on the content of thestereotypes of men, women, or leaders, the method yields a directtest of the central assumptions of Eagly and Karaus (2002) rolecongruity model and Heilmans (1983, 2001) lack-of-fit model asapplied to leader roles.

A second method, the agencycommunion paradigm, consistsof studies that examined the gender-stereotypical content of theleader stereotype (Powell & Butterfield, 1979). In these studies,participants rated a leader category (e.g., good manager) on sep-arate masculine (i.e., agentic) and feminine (i.e., communal) gen-der stereotyping scales. Comparison of participants mean ratingson the two scales determined whether the stereotype of leaders wasmore masculine than feminine. By directly addressing the contentof leader roles, this method complements but differs from thesimilarity comparisons yielded by the think managerthink maleparadigm.

A third method, the masculinityfemininity paradigm, consistsof studies that appeared in the research literature as a test of themasculine versus feminine content of occupational stereotypes(Shinar, 1975). A substantial minority of the occupations chosenfor these studies were leader roles (e.g., university president,mayor). Participants rated each of these leader roles, among other

617MASCULINITY OF LEADER STEREOTYPES

occupational groups, on a single bipolar masculine versus femininerating scale. This paradigm became less popular subsequent tocritiques pointing out that bipolar masculinityfemininity scales donot allow masculinity and femininity to vary independently butforce them to function as opposites (e.g., Constantinople, 1973).Nonetheless, these data are informative in part because researchersinvestigated various specific types of leaders rather than merelyleaders (or managers) in general. If the mean ratings of mostcategories of leaders were on the masculine side of a masculinityfemininity scale, the studies would provide a conceptual replica-tion of the basic tendency for leader roles to be stereotyped asmasculine.

Our project separately meta-analyzed studies in these threeparadigms because their distinctively different study designs andmeasuring instruments precluded combination across the para-digms (see Borenstein, Hedges, Higgins, & Rothstein, 2009, pp.359363). The think managerthink male studies correlate (a)ratings of men with ratings of managers and (b) ratings of womenwith ratings of managers. In this think managerthink male para-digm, participants are randomly assigned to rate one of the threegroups (women, men, or leaders). In contrast, in the much simpleragencycommunion and masculinityfemininity paradigms, stud-ies obtain ratings only of leaders in general or specific leader roles(and not women or men) on gender-stereotypical traits or scales.Although the agency communion and masculinityfemininitystudies are similar in presenting only a leader group (or groups) toparticipants, these two paradigms feature different types of mea-suring instruments. The masculinityfemininity studies use a sin-gle bipolar masculine versus feminine rating scale, whereas theagencycommunion studies use two separate scales, one assessingagency (or masculinity) and the other assessing communion (orfemininity). Therefore, in the masculinityfemininity paradigm,the effect sizes are based on a point estimate (in relation to thescale midpoint). In the agencycommunion paradigm, the effectsizes are based on a comparison between ratings on the two scales.For these reasons, the data are noncomparable across the threeparadigms and analyzed separately. Yet all three types of studiestest the correspondence between gender and leader stereotypes,and the studies are methodologically quite homogenous withineach paradigm.

Variation in Stereotypes About Men, Women, andLeaders

Eagly and Karau (2002) hypothesized that the incongruity be-tween leader stereotypes and the female gender stereotype is notfixed but varies with change in either stereotype. This meta-analysis examines several factors hypothesized to influence thisincongruity.

Change in Stereotypes Over Time

Cultural change over historical time is one of these factors,given that stereotypes may have changed in a manner that reduceswomens role incongruity in relation to leadership. Although somehints of the decreasing masculinity of leadership have appeared inprior reviews (e.g., Duehr & Bono, 2006; Eagly & Sczesny, 2009;Powell et al., 2002; Schein, 2001), generalizations have remaineduncertain. Temporal comparisons of relevant studies based on their

dates of publication, which extend back to 1973, allow our meta-analysis to clarify whether the cultural representation of leadershiphas changed. A shift in an androgynous direction would easewomens role incongruity problem in relation to leader roles.

Why might role incongruity have lessened? Organizational ex-perts have often argued that definitions of good managerial prac-tices have changed in response to features of the contemporaryorganizational environment, such as fast social and technologicalchange and unprecedented complexity of organizations missionsand contexts (e.g., Avolio, 1999; Kanter, 1997; Lipman-Blumen,2000). According to such analyses, these changed conditions com-promise the efficacy of top-down command-and-control leadershipand foster democratic relationships, participatory decision-making,delegation, and team-based leadership skills (e.g., Gergen, 2005;Kanter, 1997; Lipman-Blumen, 2000; McCauley, 2004). Suchdescriptions are manifestly less masculine than many traditionalmodels of good leadership.

Another possibility is that the increase of women leaders mightproduce less masculine and more androgynous beliefs about lead-ership. Evidence of the increase of women leaders abounds, in-cluding growth over time in womens emergence as leaders in fieldand laboratory studies of leader emergence in initially leaderlessgroups (Eagly & Karau, 1991). In the United States, women nowconstitute 25% of chief executives when all organizations areconsidered and 43% of all employees in management, business,and financial operations occupations (vs. 31% in 1983; U.S. De-partment of Labor, 2007, 2010b). Women have also become morenumerous in highly visible political leader roles, now constituting17% of the U.S. Congress (vs. 2% in 1950; Center for AmericanWomen and Politics, 2011) and 12% of governors (vs. 0% in 1950;Center for American Women and Politics, 2010). Women haveincreased in leadership roles in many other nations as well (seeEuropean Commission, 2010; Inter-Parliamentary Union, 2010).

Evidence that the mere presence of more women leaders canchange perceptions of leader roles emerged in research on wom-ens occupancy of the chief village councilor role in West Bengal(Beaman, Chattopadhyay, Duflo, Pande, & Topalova, 2009). Peo-ple in the villages that were mandated (vs. not mandated) by thegovernment to elect a woman for this leader role not only per-ceived their current women leaders and women leaders in generalas more effective but also reduced their implicit bias towardassociating men with leadership and elected more women to lead-ership positions in the next election. As additional evidence thatthe presence of women leaders changes perceptions of leadership,female college students with more women professors as rolemodels reduced their implicit associations of leadership qualitieswith men and communal qualities with women (Dasgupta &Asgari, 2004). Empirical research thus has demonstrated that anincrease in the number of women leaders can produce a moreandrogynous concept of leadership and thereby reduce bias towardcurrent and potential women leaders.

What about change in gender stereotypes? If gender stereotypesreflect the differing placements of men and women into socialroles (Wood & Eagly, 2010), womens increase in labor forceparticipation (to 61% vs. 33% in 1950; U.S. Department of Labor,2010a) and in leader roles might predict change in the femalestereotype. However, women still perform the majority of domes-tic work (e.g., Bianchi, Robinson, & Milkie, 2006), and the ma-jority of employed women have remained concentrated in tradi-

618 KOENIG, EAGLY, MITCHELL, AND RISTIKARI

tional occupations. The six most common in the United States aresecretary and administrative assistant; registered nurse; elementaryand middle school teacher; cashier; retail salespersons; and nurs-ing, psychiatric, and home health aides (U.S. Department of Labor,2011). Therefore, it is not surprising that the partial reviews ofgender stereotyping conducted so far have not yielded evidence ofdecreased stereotyping over time (e.g., Lueptow, Garovich-Szabo,& Lueptow, 2001). Nonetheless, the masculinity of leader rolescould be changing without much change in stereotypes about menand women in general.

In sum, leader stereotypes may have become less masculineover time. If the change in leader stereotypes is related to increasesof women in leadership roles, then the number of women leadersshould be related to the masculinity of leadership. Because stereo-type change presumably reflects the updating of impressions basedon new observations (e.g., Weber & Crocker, 1983), leader ste-reotypes at any one time point should correspond best to contem-poraneous observations of women in leader roles.

Other Influences on the Masculinity of Leader RolesA priori moderators. As Eagly and Karau (2002) argued,

men may have a more masculine leader stereotype than women do.There is some existing evidence that men, more than women,believe that good leaders have masculine qualities (e.g., Atwater,Brett, Waldman, DiMare, & Hayden, 2004; Schein, 2001) andmanifest prejudice against female leaders (Eagly et al., 1992).Such effects are understandable, given that men are less likely thanwomen to have experience with female managers (McTague,Stainback, & Tomaskovic-Devey, 2009; Stainback &Tomaskovic-Devey, 2009) and that mens group interest favorsretaining these roles for men. As a result, women leaders, partic-ularly as newcomers entering male-dominated roles, can encounterresistance, especially from men (Eagly & Carli, 2007). Our meta-analysis offers quantitative tests of the extent to which the incon-gruity between women and leaders is greater in men than women.

Role incongruity should also reflect organizational contexts andthe level of leader roles in organizational hierarchies (Eagly &Karau, 2002). Leadership may be less masculine, for example, infemale-dominated fields such as elementary education, nursing, orlibrarianship (U.S. Department of Labor, 2010b). Because thesefields are thought to require traditionally feminine skills, such aswarmth, compassion, and caring for others (Cejka & Eagly, 1999;Glick, 1991), the characteristics that people associate with leader-ship roles in these occupations are likely to incorporate morecommunal attributes. Also, consistent with the preponderance ofmen in executive positions in many organizations (e.g., 84% ofcorporate officers in Fortune 500 companies; Catalyst, 2010),construals of leadership are likely to be more masculine for higherstatus leader positions, thereby increasing role incongruity forwomen. Our inclusion of studies examining various types of lead-ership positions allows comparisons of leader stereotypes betweendifferent organizational contexts and levels of leadership withinorganizations.

Comparisons of leader stereotypes across nations are also im-portant, as Eagly and Karau (2002) also argued. Narrative review-ers have suggested that leader roles are less masculine in theUnited States than in other nations (e.g., Schein, 2001), but therehas been no quantitative analysis across nations. Variation in the

percentages of women (vs. men) in leader roles could underlie anynational differences as well as temporal differences. The partici-pation of women in leader roles and the overall status of womenare considerably greater in Western than Eastern nations (Haus-mann, Tyson, & Zahidi, 2009). Thus, there is reason to believe thatrole incongruity for women leaders is greater in Eastern thanWestern nations.

In summary, consistent with role congruity theory, we predictedthat this meta-analysis would show a robust tendency for leaderroles to be perceived as masculine. However, based on this theoryand the other evidence outlined above, we expected that therelationship between perceptions of leadership and masculinitywould be moderated by several factors. Masculinity of leadershipshould decrease over time. In addition, men should have a moremasculine construal of leadership than women. Also, leadershipmay be more masculine in domains with few women managers andhigher status leader roles. People in different nations may alsodiverge in their construals of leadership, with the presence of fewerwomen in leader roles associated with more masculine stereotypes.These a priori moderators of the gender typing of leadership areexamined within this meta-analysis.

Exploratory Moderators. Several other variables were alsoinvestigated as potential moderators of the masculinity of leader-ship on an exploratory basis. For example, because employmentgenerally entails some experience with women managers, olderparticipants might have a less masculine concept of leadership.However, older peoples more traditional attitudes about gender(e.g., Howell & Day, 2000) might foster a more masculine conceptof leadership. Therefore, we could not predict whether age wouldbe positively or negatively associated with the masculinity ofleadership but address the issue with an exploratory analysis.

Also, in research on gender, authors of one sex have occasion-ally obtained different findings than authors of the other sex (e.g.,Eagly & Carli, 1981; Leaper, Anderson, & Sanders, 1998). Inaddition, originators of research paradigms sometimes obtainedstronger data than other researchers (e.g., B. T. Johnson & Eagly,1989; Wood, Lundgren, Ouellette, Busceme, & Blackstone, 1994).Because originators stronger findings might partially be due totheir choice of stereotype measure in the case of the masculinity ofleadership, we also explored whether the differing measures usedin each paradigm were associated with different effects. Althoughthese variables have influenced other meta-analyses, we had nospecific reason to expect they would also influence the masculinityof leadership, so we tested their effects as exploratory moderators.

Method

Sample of Studies and Criteria for Inclusion andExclusion

Three paradigms of research. The search located studiesthat fit into the three different paradigms, whose defining charac-teristics are the following:

1. In the Schein (1973) think managerthink male paradigm,participants rated a leader category, men (typically men in gen-eral), or women (typically women in general) on various traitsin a between-subjects design (with only two studies having awithin-subjects design). In most of these studies, the ratings werecompleted using the Schein Descriptive Index, which consists of


92 traits, including many agentic or communal personality traits(e.g., adventurous, submissive, aggressive, intuitive, ambitious,modest, kind). Studies collected for this meta-analysis reportedinternal consistency coefficients ranging from .71 to .92 for theSchein Descriptive Index. Researchers reported intraclass correla-tions, computed across the traits, for relating the mean ratings ofleaders and men and of leaders and women.2 Studies were ac-cepted even if they reported only a menleader or womenleadercorrelation, but the great majority of studies reported both corre-lations.

2. In the Powell and Butterfield (1979) agencycommunionparadigm, participants rated leaders (or managers) on two multi-item scales, typically the masculine and feminine scales of theBem Sex Role Inventory (Bem, 1974). One scale contains stereo-typically masculine (i.e., agentic) personality traits, such as asser-tive, forceful, dominant, and competitive; the other scale containsstereotypically feminine (i.e., communal) personality traits, suchas affectionate, compassionate, warm, and gentle. Studies col-lected for this meta-analysis reported internal consistency coeffi-cients for the Bem Sex Role Inventory from .74 to .90 for theagency scale and .79 to .90 for the communion scale. Researchersusually reported means and standard deviations on the two scalesbut sometimes presented item-level data that included a represen-tation of agentic and communal traits or classified respondentsleader ratings into the four quadrants of a two-dimensionalagencycommunion space. The researchers other than Powell andButterfield whose studies fit this paradigm usually did not cite theprecedent of the Powell and Butterfield study, but nonethelessreported participants ratings of a leader category on agency andcommunion.

3. In the Shinar (1975) masculinityfemininity paradigm, par-ticipants rated one or more leader categories on a single bipolar7-point scale that ranged from very masculine to very feminine.Researchers reported means and standard deviations of these rat-ings and compared means to the midscale value, which representsthe gender-neutral or androgynous point on the scale. The meta-analysis includes only those occupations from each study that havea clear leadership focus (most commonly managerial roles, such asoffice manager).

The studies selected for all three of these paradigms presentedparticipants with a general leader category, such as managers orexecutives, or with occupations or job titles denoting managerialauthority, such as personnel directors, head librarians, academicadministrators, military officers, or political office holders. For allparadigms, data were excluded if they provided ratings of leaderbehaviors rather than personality traits (e.g., Gutek & Stevens,1979), of specific people in a leadership position (e.g., Petty &Miles, 1976), or of leader groups identified by sex (e.g., success-ful female managers; Dodge, Gilroy, & Fenzel, 1995). Studies inthe agencycommunion paradigm were eliminated if the majorityof the items in their two scales did not pertain to agentic orcommunal personality traits or if the two types of items were veryunequally represented (e.g., Buttner & Rosen, 1988).

The abstract of each identified document was evaluated by atleast one of the authors; if the study potentially fit the inclusioncriteria, the full document was obtained. In total, 78 documentswere rejected based on their abstracts and 134 documents wererejected after reading the document. In addition, 11 documents metthe inclusion criteria but did not report the appropriate statistics to

calculate an effect size, and when contacted, the authors could notsupply the needed information (e.g., Couch & Sigler, 2001; Ger-stner & Day, 1994).

Studies from any participant population or nation were includedif they fit the above criteria. When documents included data fromdifferent nations or different participant samples within a nation,they were treated as separate studies if the results were reportedseparately. Some documents included data for more than one typeof leader category (e.g., company president, head librarian). Ifseparate groups of participants rated each category, the leadercategories were treated as separate studies. If the same participantsrated more than one leader category, these effect sizes were aver-aged prior to calculating study-level mean effect sizes but retainedas separate effect sizes for moderator analyses. In two documentsin the think managerthink male paradigm (Dodge et al., 1995;Karau, 2005), separate groups of participants rated different typesof leaders. These ratings were correlated with the same ratings ofwomen and men, provided by two other groups of participants. Inour analyses, we treated these semi-independent leader conditionsas separate studies.

Some data qualified in more than one paradigm. For example,we extracted only the leader condition of think managerthinkmale studies for use in the agencycommunion paradigm if thereport contained item-level data that allowed us to separate agenticand communal items into subscales (i.e., van Engen, 2006). Also,parts of the same study that were published separately sometimesqualified for different paradigms. For example, some authors pre-sented intraclass correlation coefficients in one publication butgrouped the data into agentic and communal scales in another(Fullagar, Sverke, Sverke, Sumer, & Slick, 2003; Sumer, 2006) orpresented the intraclass correlation coefficients in one publicationbut the mean for a specific masculinefeminine bipolar scale inanother (Koch, Luft, & Kruse, 2005; Luft, 2003).

Search for studies. Computer-based information searcheswere conducted in the following databases: ABI/INFORM, Aca-demic Press/Ideal, Business Source Elite, Proquest Digital Disser-tations, Educational Resources Information Center (ERIC), Emer-ald Full Text, PAIS International, Proquest Business Databases,PsycINFO, ScienceDirect, Sociological Abstracts, Web of Sci-ence, WilsonWeb, and Worldwide Political Science Abstracts.

In these searches, the keywords leader! (-s, -ship), manager!(-s, -ial), educator, executive, candidate, public office, politicaloffice, principal, or occupation were combined with (a) stereotyp!(-e, -es, -ical), traits, characterization, attribute inventory, image,

2 The design of the studies in this paradigm appears similar to that of astudy by Broverman, Broverman, Clarkson, Rosenkrantz, and Vogel(1970), whose participants rated a mature, healthy, socially competentman, woman, or adult person. Broverman et al. found greater similaritybetween an adult person and men than between an adult person andwomen, but this finding has been criticized as artifactual (see Kelley &Blashfield, 2008; Widiger & Settle, 1987, for details). However, thesimilarity of the Broverman et al. study to the think managerthink malestudies is only superficial because Scheins (1973) method offers superioritem selection and statistical analysis. Specifically, (a) the typical thinkmanagerthink male items (in the Schein Descriptive Index) are moder-ately balanced between agentic and communal qualities (see Duehr &Bono, 2006), and (b) the statistical analysis with an intraclass correlation ismore appropriate to the data.


expectation, and perce! (-ption, -tive) or (b) gender terms, such asman, men, woman, women, male, female, masculinity, femininity, sex,sex role, sex-typing, gender, gender role, and androgy! (-ny, -nous).Additional searches paired the two most common dependent mea-sures (the Schein Descriptive Index and the Bem Sex Role Inven-tory) with manage! or leader!, think managerthink male, orrequisite management characteristics. Web of Science citationsearches were also performed for the seminal articles in eachparadigm (Powell & Butterfield, 1979; Schein, 1973; Shinar,1975). All obtained studies reference sections were also searchedfor relevant studies.

Several foreign psychological and academic databases weresearched with the limited keywords leader! (-s, -ship) or manager!(-s, -ial) paired with (a) stereotyp! (-e, -es, -ical), traits, or simi-larities or (b) gender terms, such as masculinity, femininity, gen-der, and sex (translated, as appropriate). The foreign databaseswere from Finland (University of Joensuu database, Forum ofScience database, Finnish Social Science Achieve, University ofRovaniemi database, University of Tampere database); Germany(Datenbank, PSYNDEX, PSYNDEXalert, PSYTKOM); GreatBritain and Ireland (Index to Theses); Norway (National Library ofNorway, BIBSYS Library); Spain (Psicodoc); and Sweden (Chal-mers University of Technology database, the Center for ResearchLibraries). Documents in languages other than English were trans-lated by appropriately skilled individuals who assisted the twocoders.

To locate unpublished studies, messages asking for relevant datawere sent to the e-mail distribution lists of several organizations:European Association of Social Psychology, European Associa-tion of Work and Organizational Psychology, Interamerican Soci-ety of Psychology, Society for the Psychology of Women, SocialIssues in Management Division of the Academy of Management,Society for Personality and Social Psychology, and Society for thePsychological Study of Social Issues. Data were also sought fromthe originators of each of the paradigms, but they had no additionaldata to offer. Finally, when the authors of this article gave severaltalks on the meta-analysis in progress, they asked the audience tocontribute their own data or to provide leads about relevant sourcesof data. Both unpublished and published studies meeting the in-clusion criteria were included in the meta-analysis.

The initial search extended from the beginning of each paradigmthrough the end of 2002. The search was then updated in June 2007by consulting the databases that had yielded studies in 2002, withthe addition of Google and Google Scholar, and updated oncemore in March 2009 with PsycINFO, Google, and Google Scholar.

Variables Representing Each Study in All Paradigms

The coded variables represent the a priori moderators as well asother variables that we investigated on an exploratory basis. Theincluded studies, their codings, and effect sizes appear in Tables 1, 2,and 3 for the think managerthink male, agencycommunion, andmasculinityfemininity paradigms, respectively.

A priori moderators. Year of publication (or year of datacollection for unpublished data) was recorded as well as thepercentage of male participants. Effect sizes were also calculatedseparately by participant sex when possible. If data were notreported separately for male and female participants, we contactedthe authors and requested this information.

For each paradigm, we recorded the exact description of theleader group and coded its domain: (a) In the think managerthinkmale paradigm, the leader domains were nearly all designated asmanagerial, and those few not so designated were heterogeneous(e.g., leader, educational administrator), so no moderator variablewas tested; (b) in the agencycommunion paradigm, the domainswere managerial (e.g., manager, middle-level manager), political(e.g., mayor, state senator), or educational (e.g., elementary schoolprincipal, school superintendent); and (c) in the masculinityfemininity paradigm, the domains, or economic sectors, of theoccupations were business (e.g., company president, marketingmanager), education (e.g., educational administrator, school prin-cipal), politics (mayor, politician), judicial (e.g., federal judge,Supreme Court justice), arts (orchestra conductor, theatrical direc-tor), or other (boat captain, farm manager, park manager). For allthree paradigms, leader status was coded as high (defined asprestigious political roles such as president or governor or orga-nizational positions higher than middle manager such as upper-level managers or executives) or moderate (e.g., managers,middle-level managers, all other leader occupations).

Participant nationality was coded by the nation where the studywas conducted and classified as Western (United States, GreatBritain, Germany, Canada, New Zealand, Australia, Sweden,South Africa, Netherlands, Denmark, and Portugal) or Eastern(China, Japan, Hong Kong, Egypt, Saudi Arabia, and Turkey).Other exploratory classifications (e.g., North American vs. otherWestern vs. Eastern) did not improve prediction of the effect sizes.

To interpret cross-temporal and cross-national comparisons of stud-ies, we imported data on the percentage of female managers, definedas the percentage of managers who are women, based on the likelytime of data collection for the nations in which the studies wereconducted, defined as 2 years prior to studies dates of publication.When available, these data came from the Human DevelopmentReports of the United Nations Development Programme (e.g., 2006)and otherwise from the Yearbooks of Labour Statistics of the Inter-national Labour Organization (e.g., 1985). Aggregate indexes of thestatus of women available from the United Nations DevelopmentProgramme or other sources were not appropriate for this meta-analysis because they are available only for more recent years.

Exploratory moderators. Average participant age was eitherprovided in each research report or estimated from sample char-acteristics. The distributions of the effect sizes across the types ofparticipant populations were as follows: (a) for the think managerthink male paradigm, 55% undergraduate students, 18% managers,8% MBA students, 6% nonmanagerial employees, and 14% otheror mixed; (b) for the agencycommunion paradigm, 62% under-graduate students, 13% nonmanagerial employees, 9% managers,9% MBA students, and 9% other or mixed; and (c) for themasculinityfemininity paradigm, 86% undergraduate studentsand 14% other or mixed.

The percentage of male authors was coded. Research group wascoded as originators of paradigm (Schein, Powell and Butterfield, orShinar) or other. Stereotype measure was coded based on the use ofthe originators versus other measures: (a) In the think managerthinkmale paradigm, Schein Descriptive Index or other (e.g., task vs.person-orientation scales); (b) in the agencycommunion paradigm,Bem Sex Role Inventory or other (e.g., initiating structure vs. con-sideration scales); and (c) in the masculinityfemininity paradigm,always a masculinityfemininity 7-point scale.


Tabl

e1

AllI

nclu

ded

Stud

ies

(k!

51)W

ithM

oder

ator

Valu

esand

Effec

tSize

sfor

the

Thin

kM

anag

erT

hink

Mal

ePa

radi

gm

Rep

ort

Sam

plea

Sep.

byse

xb

Pub.

sourc

ec%

mal

eau

thor

sSt

ereo

type

mea

sure

dN

atio

ne%

fem

ale

mgr

.M

age

Lead

erro

lef

Lead

erst

atus

gn

%m

ale

part.

Item

nLe

ader

sim.h

ICCi

Var

ianc

e

Boo

ysen

&N

kom

o(20

06)

1Y

10

1ZA

3032

SMM

139

169

92W

.46

0.01

404

6992

M.70

0.01

Boy

ce&

Her

d(20

03)

1Y

10

1U

S46

20SM

O1

508

8292

W"

.03

0.01

501

8292

M.39

0.01

Bre

nner

etal

.(19

89)

1Y

167

1U

S35

42SM

M1

407

7292

W.15

0.01

395

6992

M.68

0.01

Byl

er(20

00)

1Y

20

1U

S44

44EA

D1

8382

92W

.45

0.01

8578

92M

.23

0.01

Cohe

n-K

aner

(1992

)1

Y2

01

US

3736

SMM

111

939

92W

.26

0.01

119

3992

M.53

0.01

Dio

n&

Schu

ller(

1990

)1

Y1

502

US/

CA36

32SM

M1

5345

51M

.58

0.02

Dio

n&

Schu

ller(

1991

)1

Y1

502

US/

CA37

33SM

M1

4348

22M

.66

0.05

Dod

geet

al.(

1995

)1

Y1

331

US

4031

SMM

148

5992

W.15

0.01

4859

92M

.50

0.25

SUM

248

5992

W.02

0.01

4859

92M

.59

0.01

Due

hr&

Bon

o(20

03)

1Y

20

1U

S42

30SM

M1

7639

92W

.54

0.01

7431

92M

.56

0.01

Due

hr&

Bon

o(20

06)

2Y

10

1U

S42

47SM

M1

178

6092

W.67

0.01

169

5792

M.56

0.01

US

4220

SMM

119

730

92W

.28

0.01

203

3392

M.43

0.01

Fern

ande

s&

Cabr

al-C

ardo

so1

Y1

502

PT32

21M

119

946

42W

".01

0.03

(2003

)19

946

42M

.40

0.03

Fost

er(19

94)

1Y

10

1G

B32

42SM

M1

160

5092

W.11

0.01

160

5092

M.46

0.01

Fulla

gere

tal

.(20

03)

2Y

150

1TR

821

SMM

153

658

92W

.21

0.01

536

5892

M.58

0.01

SE30

21SM

M1

182

2792

W.15

0.01

182

2792

M.68

0.01

Hei

lman

etal

.(19

89)

1M

150

1U

S35

41SM

M1

7810

092

W"

.24

0.01

7810

092

M.54

0.01

Kar

au(20

05)

1Y

210

01

US

4222

SMM

111

259

92W

.01

0.01

104

6392

M.39

0.01

CEO

211

358

92W

.19

0.01

105

6392

M.93

0.01

Kar

au&

Elsa

id(20

05)

2Y

210

01

US

4222

SMM

121

758

92W

.31

0.01

220

5992

M.40

0.01

EG9

21SM

M1

371

7592

W"

.24

0.01

357

7392

M.43

0.01

Kar

au&

Han

sen

(2005

)2

Y2

100

1SE

3121

SMM

147

5592

W.44

0.01

4757

92M

.33

0.01

US

4221

SMM

161

5192

W.28

0.01

6156

92M

.39

0.01

Ken

t(19

84)

1Y

20

1U

S28

47SS

A1

240

4792

W.58

0.01

247

5292

M.59

0.01


Tabl

e1

(contin

ued)

Rep

ort

Sam

plea

Sep.

byse

xb

Pub.

sourc

ec%

mal

eau

thor

sSt

ereo

type

mea

sure

dN

atio

ne%

fem

ale

mgr

.M

age

Lead

erro

lef

Lead

erst

atus

gn

%m

ale

part.

Item

nLe

ader

sim.h

ICCi

Var

ianc

e

Kun

kele

tal

.(20

03)

1Y

133

1U

S46

21CE

O2

4146

92W

.18

0.01

4246

92M

.80

0.01

Luft

(2003

)1

Y2

02

DE

2728

L1

202

4725

W"

.27

0.04

202

4725

M.68

0.04

Mar

men

out(

2008

)1

Y2

01

SA31

21SM

M1

144

4092

W.18

0.01

171

3792

M.61

0.01

Mas

seng

ill&

DiM

arco

1Y

110

01

US

2244

SM1

8960

92W

.15

0.01

(1979

)83

5792

M.64

0.01

Moo

k(20

05)

1Y

20

1N

L26

39SM

185

4892

W.24

0.01

9645

92M

.58

0.01

Nee

rgaa

rdet

al.(

2007

)1

Y2

331

DK

2543

MSC

111

985

92W

.44

0.01

113

8292

M.72

0.01

Nor

ris&

Wyl

ie(19

95)

1Y

10

2U

S/CA

4021

SMM

161

348

24W

.44

0.05

605

4824

M.81

0.05

Ors

er(19

94)j

2Y

10

1CA

4121

SM1

198

092

W.31

0.01

198

092

M.46

0.01

CA41

21SM

119

850

92W

.18

0.01

198

5092

M.49

0.01

Para

dine

etal

.(19

95)

1Y

266

2U

S40

29SM

M1

204

5540

W.15

0.03

213

5440

M.76

0.03

Rya

net

al.(

inpr

ess)

2Y

225

1A

U37

22M

SC1

3538

92W

.36

0.01

3651

92M

.72

0.01

X27

MSC

191

2292

W.46

0.01

7525

92M

.79

0.01

Saue

rset

al.(

2002

)1

Y1

671

NZ

3821

SMM

121

248

92W

.43

0.01

212

4892

M.71

0.01

Sche

in(19

73)

1M

10

1U

S18

43SM

M1

193

100

92W

.06

0.01

212

100

92M

.62

0.01

Sche

in(19

75)

1F

10

1U

S17

44SM

M1

113

092

W.30

0.01

115

092

M.54

0.01

Sche

in&

Dav

idso

n2

Y1

01

GB

3221

SMM

115

254

92W

.19

0.01

(1993

)k15

254

92M

.52

0.01

GB

3221

SMM

110

148

92W

.17

0.01

101

4892

M.64

0.01

Sche

in&

Mue

ller

2Y

150

1G

B32

21SM

M1

101

4892

W.17

0.01

(1992

)10

148

92M

.64

0.01

DE

1924

SMM

129

763

92W

.05

0.01

297

6392

M.71

0.01

Sche

inet

al.(

1989

)1

Y1

331

US

3521

SMM

115

264

92W

.23

0.01

152

6492

M.64

0.01

Sche

inet

al.(

1996

)2

Y1

501

JP9

21SM

M1

195

6692

W"

.06

0.01

210

6492

M.60

0.01

CN12

21SM

M1

186

5692

W.10

0.01

183

5392

M.91

0.01

Scze

sny

(2003

a)1

Y1

02

DE

2724

L1

7358

40W

.65

0.03

7358

40M

.76

0.03

(tabl

eco

ntin

ues)


Lastly, the source of publication was coded as published orunpublished (including dissertations, masters theses, and otherunpublished documents) to provide one estimate of possible pub-lication bias (Sutton, 2009; see the subsection Publication Bias).

Two of the authors, who had successfully completed a course onmeta-analysis, independently coded the studies, with high agree-ment across the variables. The mean kappas for intercoder agree-ment were .90, .90, and .97 for the think managerthink male,agencycommunion, and masculinityfemininity paradigms, re-spectively. Disagreements were resolved by discussion.

Computation of Effect Sizes and Data AnalysisEffect sizes were calculated with a hand calculator or DSTAT

software and then entered into Comprehensive Meta-Analysis(Version 2.2.050) and Statistical Package for the Social Sciences(SPSS). Subgroup analyses were conducted using ComprehensiveMeta-Analysis, and meta-regressions were conducted with SPSSmacros provided by D. Wilson (see http://mason.gmu.edu/#dwilsonb/ma.html). For all analyses, p values less than .05 wereconsidered significant and values between .05 and .10 were de-fined as marginal.

Calculation of effect sizes. In the think managerthink maleparadigm, researchers reported intraclass correlation coefficients(ICCs) for the menleaders and womenleaders relationships. Inall cases in which ICCs were not reported, the researchers providedadditional information that allowed us to produce intraclass cor-relations. The ICCs used in the primary data reports were com-puted using a one-way, single-rater, random effects model, whichassesses absolute agreement among measurements (Case 1 inMcGraw & Wong, 1996). However, the more appropriate ICCwould have been the 2-way ICC(A,1), which is also a measure ofabsolute agreement but which takes the fixed column factor (lead-ers vs. women; leaders vs. men) into account. Thus, the ICCprovided by researchers in the primary studies would be biaseddownward somewhat but quite close to the calculations forICC(A,1) (K. O. McGraw, personal communication, January 8,2003). We used the one-way, single-rater, random effects ICCbecause it was available for all data sets, providing a commonmetric for the studies. The menleaders and womenleaders sim-ilarities, or ICCs, were analyzed separately as Fishers Z, using theconversion .5 * log({1 $ [(k " 1) * r]}/(1 " r)), where k is thenumber of observations made on each object of measurement, andgiven an inverse variance within-study weight of 1/{k/[2 * (n "2)(k " 1)]}, where n is the number of items (see formulas inAppendix B of McGraw & Wong, 1996). The Zs were transformedback to ICCs for presentation of the results.3 In the random-effectsmodels, the variances in these study weighting terms consisted ofthe sum of the within-study variance and the between-studiesvariance (see Borenstein et al., 2009, p. 72).

In the agencycommunion paradigm, researchers reported meansand standard deviations separately on the agentic and communalscales, allowing the computation of a d effect size comparing the

3 Because the womenleaders and menleaders correlations are depen-dent in that they share the data for leaders, we did not compare the twoeffects statistically. These types of comparisons would require informationabout the correlation between male and female targets, which is notavailable in the primary studies.Ta

ble

1(co

ntin

ued)

Rep

ort

Sam

plea

Sep.

byse

xb

Pub.

sourc

ec%

mal

eau

thor

sSt

ereo

type

mea

sure

dN

atio

ne%

fem

ale

mgr

.M

age

Lead

erro

lef

Lead

erst

atus

gn

%m

ale

part.

Item

nLe

ader

sim.h

ICCi

Var

ianc

e

Scze

sny

(2003

b)1

Y1

02

DE

2745

L1

6058

40W

.87

0.03

6470

40M

.87

0.03

Sylv

an(19

83)

1Y

20

1U

S27

38SM

160

5392

W.11

0.01

6354

92M

.48

0.01

van

Enge

n(20

06)

1Y

20

1N

L26

22SM

M1

7023

92W

.29

0.01

310

92M

.38

0.01

Yim

&B

ond

(2002

)1

Y1

502

HK

2519

SMM

114

249

32W

.44

0.03

142

4932

M.72

0.03

Note

.In

each

sam

ple,

sepa

rate

grou

psofp

artic

ipan

tsra

ted

ale

ader

cate

gory

,wom

en,

or

men

on

gend

er-s

tere

otyp

ical

trai

ts.Th

eef

fect

sizes

are

the

intra

clas

sco

rrel

atio

nsbe

twee

nth

em

ean

ratin

gsofm

anag

ers

(orle

ader

s)an

d(a)

the

mea

nra

tings

ofm

enor

(b)th

em

ean

ratin

gsofw

om

enac

ross

allt

hetr

aits.

Sep.!

sepa

rate

d;pu

b.!

publ

icat

ion;

mgr

.!m

anag

ers;

part.!

parti

cipa

nts;

sim.!

simila

rity;

ICC!

intra

clas

sco

rrel

atio

nco

effic

ient

.a

The

num

bero

fdiff

eren

tsam

ples

(e.g.,

from

diffe

rent

popu

latio

nsor

countr

ies)

repo

rted

with

inth

ear

ticle

.b

Sepa

rate

dsa

mpl

esby

sex:

N!

no;Y

!ye

s;M!

sam

ple

was

allm

ale;

F!

sam

ple

was

allf

emal

e.c

Publ

icat

ion

sourc

e:1!

publ

ished

;2!

unpu

blish

ed.

d1!

Sche

inD

escr

iptiv

eIn

dex;

2!

oth

er.

eA

U!

Aus

tralia

;CA!

Cana

da;C

N!

Chin

a;D

K!

Den

mar

k;EG!

Egyp

t;D

E!

Ger

man

y;G

B!

Gre

atB

ritai

n;H

K!

Hon

gK

ong;

JP!

Japa

n;N

L!

Net

herla

nds;

NZ!

New

Zeal

and;

PT!

Portu

gal;

SA!

Saud

iAra

bia;

ZA!

Sout

hA

frica

;SE!

Swed

en;T

R!

Turk

ey;

US!

Uni

ted

Stat

es;X!

incl

uded

data

from

sever

alco

untr

ies.

fEA

D!

effe

ctiv

eat

hlet

icdi

rect

ors;

L!

lead

ersh

ip;M

!m

anag

ers;

MSC!

man

ager

sofs

ucc

essf

ulco

mpa

nies

;CEO

!su

cces

sful

CEO

s;SM

!su

cces

sful

man

ager

s;SM

M!

succ

essf

ulm

iddl

eman

ager

s;SM

O!

succ

essf

ulm

ilita

ryoffi

cers

;SSA!

succ

essf

ulsc

hool

adm

inist

rato

rs;S

UM!

succ

essf

ulupp

erle

velm

anag

ers.

g1!

mode

rate

;2!

high

.h

M!

men

le

ader

ssim

ilarit

y;W!

wom

enle

ader

ssim

ilarit

y.iEf

fect

size

(ICC)

com

pute

dusin

ga

one-

way

singl

era

ter

rando

mef

fect

sm

ode

l,re

late

dth

em

ean

ratin

gsof

the

lead

erro

leto

the

ratin

gsof

men

or

wom

en.

Thes

eco

rrel

atio

nsre

pres

ent

the

simila

rity

of

ster

eoty

pes

of

men

or

wom

ento

ster

eoty

pes

of

lead

ers;

high

erco

rrel

atio

nsin

dica

tea

grea

ter

simila

rity.

j The

sam

ple

with

50%

mal

epa

rtici

pant

swas

com

pose

dofb

usin

esss

tude

nts,

and

the

sam

ple

with

0%m

ale

parti

cipa

ntsw

asco

mpo

sed

ofn

onbu

sines

sstu

dent

s.k

The

sam

ple

with

54%

mal

epa

rtici

pant

sw

asfro

mN

orth

Gre

atB

ritai

n,an

dth

esa

mpl

ew

ith48

%m

ale

parti

cipa

nts

was

from

Sout

hG

reat

Brit

ain.


Tabl

e2

AllI

nclu

ded

Stud

ies

(k!

48)W

ithM

oder

ator

Valu

esand

Effec

tSize

sfor

the

Agen

cyC

omm

unio

nPa

radi

gm

Rep

ort/l

eade

rrole

Sam

plea

Sep.

byse

xb

Pub.

sourc

ec%

mal

eau

thor

sSt

ereo

type

mea

sure

dN

atio

ne%

fem

ale

mgr

.M

age

Lead

erdo

mai

nfLe

ader

stat

usg

n%

mal

epa

rt.gh

Var

ianc

e

But

terfi

eld

&Po

wel

l(19

81)

1N

110

01

US

25G

ood

pres

iden

t20

22

128

592.

360.

03Ca

nn&

Sieg

fried

(1987

)1

Y1

100

2U

S33

Man

ager

213

111

053

0.79

0.02

Hud

dy&

Terk

ildse

n(19

93)

1Y

10

2U

S37

21G

ood

Cong

ress

mem

ber

22

133

540.

480.

02G

ood

loca

lcounci

lmem

ber

21

149

500.

200.

01G

ood

may

or2

215

351

0.75

0.01

Goo

dpr

esid

ent

22

139

561.

000.

02A

vera

geofr

ole

s0.

610.

02In

derli

ed&

Pow

ell(

1979

)4

Y1

501

US

22G

ood

man

ager

433

141

632.

650.

09G

ood

man

ager

373

151

02.

140.

06G

ood

man

ager

283

115

469

1.41

0.02

Goo

dm

anag

er22

31

259

581.

220.

01M

aier

(1993

)1

Y1

100

1U

S37

20Ty

pica

lman

ager

31

6050

0.72

0.04

Ave

rage

ofr

ole

s3

159

501.

790.

05Id

ealm

anag

er1.

250.

04M

aron

giu

&Ek

eham

mar

(2000

)1

Y1

501

SE27

Man

ager

413

187

480.

650.

02Po

wel

l(19

92)

2Y

110

01

US

3721

Goo

dm

anag

er3

187

471.

460.

03Ty

pica

lAm

eric

anm

anag

er3

179

441.

630.

03Po

wel

l&B

utte

rfiel

d(19

79)

2Y

110

01

US

22G

ood

man

ager

203

111

082

2.35

0.20

Goo

dm

anag

er27

31

574

701.

980.

05Po

wel

l&B

utte

rfiel

d(19

84)

1N

110

01

US

28G

ood

man

ager

203

162

762

2.47

0.01

Pow

ell&

But

terfi

eld

(1987

)2

Y1

100

1U

S33

20G

ood

pres

iden

toft

heU

.S.

22

5046

1.53

0.05

Goo

dvic

e-pr

esid

ent

22

4446

0.81

0.05

Pow

ell&

But

terfi

eld

(1989

)2

Y1

100

1U

S35

Goo

dm

anag

er20

31

199

432.

560.

15G

ood

man

ager

273

112

660

1.97

0.23

Pow

elle

tal

.(20

02)

2Y

167

1U

S45

Goo

dm

anag

er32

31

123

651.

490.

13G

ood

man

ager

213

120

657

1.08

0.04

Pow

ell&

Kid

o(19

94)

2Y

110

1JP

8G

ood

man

ager

213

188

77"

1.18

0.03

Typi

calJ

apan

ese

man

ager

213

186

230.

270.

02R

osen

was

ser

&D

ean

(1989

)9

N1

01

US

3521

City

counci

lmem

ber

21

1454

2.44

0.25

Gov

erno

r2

214

542.

660.

27M

ayor

22

1454

2.31

0.24

(tabl

eco

ntin

ues)


Tabl

e2

(contin

ued)

Rep

ort/l

eade

rrole

Sam

plea

Sep.

byse

xb

Pub.

sourc

ec%

mal

eau

thor

sSt

ereo

type

mea

sure

dN

atio

ne%

fem

ale

mgr

.M

age

Lead

erdo

mai

nfLe

ader

stat

usg

n%

mal

epa

rt.gh

Var

ianc

e

Pres

iden

t2

214

541.

400.

18Sc

hool

boar

dm

embe

r1

114

541.

010.

16St

ate

repr

esen

tativ

e2

114

542.

400.

25St

ate

senat

or2

114

541.

590.

19U

.S.r

epre

sent

ativ

e2

214

542.

550.

26U

.S.s

enat

or2

214

541.

980.

21R

uste

mey

er&

Thrie

n(19

89)

2Y

10

1D

E19

Goo

dm

anag

er23

31

109

643.

110.

25G

ood

man

ager

453

154

100

1.84

0.23

Stok

er(20

07)

1Y

10

1N

L26

Idea

lman

ager

363

132

2967

1.07

0.00

1Su

mer

(2006

)1

Y1

02

TR8

Succ

essf

ulm

iddl

em

anag

er21

31

369

551.

290.

01van

Enge

n(20

06)

1Y

20

2N

L26

Succ

essf

ulm

iddl

em

anag

er22

31

4028

1.00

0.06

Vin

nico

mbe

&Ca

mes

(1998

)1

Y1

02

LU9

Succ

essf

ulm

anag

erat

thei

rba

nk43

31

6650

1.02

0.03

Vin

nico

mbe

&Si

ngh

(2002

)1

Y1

02

GB

33Su

cces

sful

man

ager

sw

hore

ache

dth

eto

pte

am43

32

363

331.

710.

01

Will

emse

n(20

02)

1Y

10

2N

L27

Succ

essf

ulm

anag

erat

thei

rba

nk22

31

139

521.

190.

02

Will

iam

s(19

89)

1N

20

1U

S35

50A

cade

mic

dean

ofs

tude

nts

11

5443

1.00

0.04

Ass

istan

tprin

cipa

l1

154

430.

760.

04El

emen

tary

scho

olpr

inci

pal

11

5443

0.28

0.04

Scho

olsu

perin

tend

ent

12

5443

0.34

0.04

Seco

ndar

ysc

hool

prin

cipa

l1

154

431.

200.

04A

vera

geofr

ole

s0.

710.

04

Note

.In

each

sam

ple,

parti

cipa

nts

rate

da

lead

erro

leon

sepa

rate

mas

culin

e(i.

e.,ag

entic

)and

fem

inin

e(i.

e.,co

mm

unal

)gen

ders

tere

otyp

ing

scal

es.C

om

paris

onofp

artic

ipan

tsm

ean

ratin

gson

the

two

scal

esde

term

ined

whe

ther

the

ster

eoty

peofl

eade

rsw

asm

ore

mas

culin

eth

anfe

min

ine.

The

effe

ctsiz

efo

rthe

aver

age

oft

hero

les

ispr

esen

ted

fors

tudi

esusin

ga

with

in-s

ubjec

tsde

sign.

Sep.!

sepa

rate

d;pu

b.!

publ

icat

ion;

mgr

.!m

anag

ers;

part.!

parti

cipa

nts.

aTh

enum

ber

of

diffe

rent

sam

ples

(e.g.,

from

diffe

rent

popu

latio

nsor

countr

ies)

repo

rted

with

inth

ear

ticle

.So

me

stud

ies

with

multi

ple

sam

ples

repo

rted

parti

cipa

ntch

arac

teris

tics

acro

ssal

lsa

mpl

es.

bSe

para

ted

sam

ples

byse

x:

N!

no;Y!

yes.

cPu

blic

atio

nso

urc

e:1!

publ

ished

;2!

unpu

blish

ed.

d1!

Bem

Sex

Rol

eInv

ento

ry;2!

oth

er.

eD

E!

Ger

man

y;G

B!

Gre

atB

ritai

n;JP!

Japa

n;LU

!Lu

xem

bour

g;N

L!

Net

herla

nds;

SE!

Swed

en;T

R!

Turk

ey;U

S!

Uni

ted

Stat

es.

f1!

educ

atio

nal;

2!

polit

ical

;3!

man

ager

ial.

g1!

mode

rate

;2!

high

.h

Effe

ctsiz

e(g)

com

pute

dby

subt

ract

ing

the

mea

nra

ting

on

the

fem

inin

esc

ale

from

the

mea

nra

ting

on

the

mas

culin

esc

ale

and

divi

ding

byth

epo

oled

stan

dard

devi

atio

n,

adjus

tedfo

rsm

alls

ampl

ebi

as.I

fonly

frequ

enci

esor

perc

enta

gesi

na

2%

2ta

blec

reat

edby

split

ting

the

scal

esat

the

med

ians

wer

ere

porte

d,ag

ency

and

com

munio

nw

ere

trea

ted

asdi

chot

omou

s,an

dg

was

estim

ated

from

d Cox

.La

rger

gsin

dica

tea

more

mas

culin

eco

nce

ptofl

eade

rshi

p.


Tabl

e3

AllI

nclu

ded

Stud

ies

(k!

101)

With

Mod

erat

orVa

lues

and

Effec

tSize

sfor

the

Mas

culin

ityF

emin

inity

Para

digm

Rep

ort/l

eade

rrole

Sam

plea

Sep.

byse

xb

%m

ale

auth

ors

Nat

ionc

%fe

mal

em

gr.

Mag

eLe

ader

dom

aind

Lead

erst

atus

en

%m

ale

part.

gfV

aria

nce

Beg

gs&

Doo

little

(1993

)1

Y0

US

3721

Boa

tcap

tain

61

139

502.

440.

007

Com

pany

pres

iden

t2

213

849

1.22

0.00

7D

ean/

educ

atio

nala

dmin

istra

tor

11

140

500.

610.

007

Dire

ctor

,lab

orre

latio

ns2

114

150

0.88

0.00

7D

irect

or,o

pera

tions

21

141

501.

110.

007

Farm

man

ager

61

140

501.

640.

007

Fede

ralju

dge

51

141

501.

460.

007

Hea

dlib

raria

n1

114

150

"1.

660.

007

Man

ager

,cre

ditu

nio

n2

114

150

0.87

0.00

7M

anag

er,p

rocu

rem

ents

ervic

es2

114

150

0.72

0.00

7M

anag

ing

edito

r6

113

950

0.85

0.00

7M

ayor

41

141

501.

400.

007

Mot

elm

anag

er2

114

150

0.45

0.00

7O

ffice

man

ager

21

141

500.

260.

007

Orc

hestr

aco

ndu

ctor

31

140

501.

350.

007

Park

man

ager

61

141

501.

100.

007

Pers

onne

ldire

ctor

21

139

500.

300.

007

Polit

icia

n4

114

051

1.40

0.00

7Pu

blic

rela

tions

dire

ctor

21

141

500.

060.

007

Sale

sm

anag

er2

114

050

0.60

0.00

7Sa

les

pres

iden

t2

214

050

1.35

0.00

7Sc

hool

prin

cipa

l1

114

150

0.92

0.00

7Th

eatri

cald

irect

or3

114

150

0.01

0.00

7U

nive

rsity

pres

iden

t1

214

050

1.44

0.00

7U

.S.S

upre

me

Cour

tjusti

ce5

113

950

1.51

0.00

7A

vera

geofr

ole

s0.

890.

007

Gat

ton

etal

.(19

99)

1Y

33U

S44

21A

ccou

ntin

gm

anag

er2

117

547

0.84

0.00

6Co

mpu

tero

pera

tions

man

ager

21

175

470.

670.

006

Dat

apr

oces

sing

man

ager

21

175

47"

0.16

0.00

6En

gine

erin

gm

anag

er2

117

547

1.46

0.00

6Fi

nanc

ialm

anag

er2

117

547

0.76

0.00

6H

uman

reso

urc

esm

anag

er2

117

547

"0.

310.

006

Mar

ketin

gm

anag

er2

117

547

0.20

0.00

6O

ffice

man

ager

21

175

470.

270.

006

Plan

tman

ager

21

175

471.

770.

006

Quali

tyco

ntr

olm

anag

er2

117

547

0.84

0.00

6R

&D

dire

ctor

21

175

470.

450.

006

Sale

sm

anag

er2

117

547

0.38

0.00

6A

vera

geofr

ole

s0.

600.

006

Koc

het

al.(

2005

)1

Y0

DE

2728

Lead

ersh

ip2

120

247

0.63

0.00

5(ta

ble

contin

ues)


Tabl

e3

(contin

ued)

Rep

ort/l

eade

rrole

Sam

plea

Sep.

byse

xb

%m

ale

auth

ors

Nat

ionc

%fe

mal

em

gr.

Mag

eLe

ader

dom

aind

Lead

erst

atus

en

%m

ale

part.

gfV

aria

nce

Mah

oney

&B

lake

(1987

)1

Y10

0U

S33

26Fa

rmm

anag

er6

129

521.

520.

036

Fina

ncia

lman

ager

21

3450

0.48

0.03

0Sa

les

man

ager

21

2952

0.58

0.03

5A

vera

geofr

ole

s0.

860.

033

Mun

oz

Sastr

eet

al.(

2000

)2

N20

FR9

21B

oatc

apta

in6

123

445

2.35

0.00

4Co

mpa

nypr

esid

ent

22

234

450.

980.

004

Dea

n/ed

ucat

iona

ladm

inist

rato

r1

123

445

0.92

0.00

4Fa

rmm

anag

er6

123

445

1.97

0.00

4Fe

dera

ljudg

e5

123

445

0.53

0.00

4H

ead

libra

rian

11

234

45"

1.03

0.00

4M

anag

ing

edito

r2

123

445

0.52

0.00

4M

ayor

41

234

450.

780.

004

Mot

elm

anag

er2

123

445

0.80

0.00

4O

rche

stra

condu

ctor

31

234

451.

220.

004

Park

man

ager

61

234

450.

930.

004

Pers

onne

ldire

ctor

21

234

450.

490.

004

Polit

icia

n4

123

445

1.15

0.00

4Pu

blic

rela

tions

dire

ctor

21

234

450.

280.

004

Sale

sm

anag

er2

123

445

0.30

0.00

4Sa

les

pres

iden

t2

223

445

0.72

0.00

4Sc

hool

prin

cipa

l1

123

445

0.07

0.00

4Su

prem

eCo

urtju

stice

51

234

450.

600.

004

Thea

trica

ldire

ctor

31

234

450.

530.

004

Uni

vers

itypr

esid

ent

12

234

450.

870.

004

Ave

rage

ofr

ole

s0.

750.

004

ES32

21B

oatc

apta

in6

112

641

2.03

0.00

8Co

mpa

nypr

esid

ent

22

126

410.

750.

008

Dea

n/ed

ucat

iona

ladm

inist

rato

r1

112

641

0.74

0.00

8Fa

rmm

anag

er6

112

641

1.25

0.00

8Fe

dera

ljudg

e5

112

641

0.59

0.00

8H

ead

libra

rian

11

126

41"

0.09

0.00

8M

anag

ing

edito

r2

112

641

0.74

0.00

8M

ayor

41

126

410.

610.

008

Mot

elm

anag

er2

112

641

0.68

0.00

8O

rche

stra

condu

ctor

31

126

411.

300.

008

Park

man

ager

61

126

410.

780.

008

Pers

onne

ldire

ctor

21

126

410.

700.

008

Polit

icia

n4

112

641

0.62

0.00

8Pu

blic

rela

tions

dire

ctor

21

126

410.

070.

008

Sale

sm

anag

er2

112

641

0.51

0.00

8Sa

les

pres

iden

t2

212

641

0.45

0.00

8Sc

hool

prin

cipa

l1

112

641

0.40

0.00

8Su

prem

eCo

urtju

stice

51

126

410.

880.

008


Tabl

e3

(contin

ued)

Rep

ort/l

eade

rrole

Sam

plea

Sep.

byse

xb

%m

ale

auth

ors

Nat

ionc

%fe

mal

em

gr.

Mag

eLe

ader

dom

aind

Lead

erst

atus

en

%m

ale

part.

gfV

aria

nce

Thea

trica

ldire

ctor

31

126

410.

320.

008

Uni

vers

itypr

esid

ent

12

126

410.

570.

008

Ave

rage

ofr

ole

s0.

700.

008

Shin

ar(19

75)

1N

0U

S18

21B

oatc

apta

in6

112

050

3.91

0.00

9Co

mpa

nypr

esid

ent

22

120

509.

060.

011

Dea

n/ed

ucat

iona

ladm

inist

rato

r1

112

050

0.99

0.00

8Fa

rmm

anag

er6

112

050

2.09

0.00

8Fe

dera

ljudg

e5

112

050

1.76

0.00

8H

ead

libra

rian

11

120

50"

1.59

0.00

8M

anag

ing

edito

r2

112

050

0.68

0.00

8M

ayor

41

120

501.

730.

008

Mot

elm

anag

er2

112

050

1.08

0.00

8O

rche

stra

condu

ctor

31

120

502.

730.

009

Park

man

ager

61

120

503.

030.

009

Pers

onne

ldire

ctor

21

120

500.

500.

008

Polit

icia

n4

112

050

0.96

0.00

8Pu

blic

rela

tions

dire

ctor

21

120

500.

680.

008

Sale

sm

anag

er2

112

050

1.13

0.00

8Sa

les

pres

iden

t2

212

050

1.62

0.00

8Sc

hool

prin

cipa

l1

112

050

0.67

0.00

8Th

eatri

cald

irect

or3

112

050

1.09

0.00

8U

nive

rsity

pres

iden

t1

212

050

3.94

0.00

9U

.S.S

upre

me

Cour

tjusti

ce5

112

050

7.08

0.01

0A

vera

geofr

ole

s2.

160.

009

Note

.In

each

sam

ple,

parti

cipa

nts

rate

dle

ader

role

son

asin

gle

bipo

larm

ascu

linity

fem

inin

itysc

ale.

Com

paris

onoft

hese

ratin

gsto

the

mid

poin

toft

hesc

ale

dete

rmin

edw

heth

erth

est

ereo

type

of

lead

ers

was

mas

culin

eor

fem

inin

e.Th

eef

fect

size

fort

heav

erag

eoft

hero

les

ispr

esen

ted

fors

tudi

esusin

ga

with

in-s

ubjec

tsde

sign.

All

stud

ies

wer

epu

blish

edan

duse

da

17

mas

culin

efe

min

ine

ratin

gsc

ale.

Sep.!

sepa

rate

d;m

gr.!

man

ager

s;pa

rt.!

parti

cipa

nts.

aTh

enum

bero

fdiff

eren

tsam

ples

(e.g.,

from

diffe

rent

popu

latio

nsor

countr

ies)

repo

rted

with

inth

ear

ticle

.b

Sepa

rate

dsa

mpl

esby

sex:

N!

no;Y!

yes.

cFR!

Fran

ce;D

E!

Ger

man

y;ES!

Spai

n;U

S!

Uni

ted

Stat

es.

d1!

educ

atio

n;2!

busin

ess;

3!

arts

;4!

polit

ical

;5!

judici

al;6!

oth

er.

e1!

mode

rate

;2!

high

.fEf

fect

size

(g)co

mpu

ted

bysu

btra

ctin

gth

em

idpo

into

fth

esc

ale

from

the

lead

erra

ting,

divi

ded

byth

est

anda

rdde

viat

ion

and

corr

ecte

dfo

rsm

alls

ampl

ebi

as.


ratings on the two scales: (M1 " M2)/sp. The effect sizes wereconverted to g with the correction for small sample bias: 1" [3/(4N" 9)] (Borenstein et al., 2009). Some authors split their sample atthe median on both scales and reported the frequencies or percentagesin each quadrant of the resulting 2 % 2 table. If only this report wasavailable, agency and communion were treated as dichotomous, andg was estimated from dCox, which is a logistic transformation of theodds-ratio (Sanchez-Meca, Marn-Martnez, & Chancon-Moscoso,2003, Formula 18). The within-study weighting term was the con-ventional inverse variance for standardized comparisons of means(Lipsey & Wilson, 2001, p. 72) or dCox (Sanchez-Meca et al., 2003,Formula 19), with the random-effects models also incorporating thebetween-studies variances in the study weight.

In the masculinityfemininity paradigm, authors reported meansand standard deviations on a 7-point masculinityfemininity scale,which yielded a d as a comparison with the scales midpoint [(M"midpoint)/SD]. The effect sizes were then corrected for smallsample bias with the formula 1 " [3/(4N " 5)] to create a g(Borenstein et al., 2009). The within-study weighting term forthese effect sizes was (1/n) $ {(d * d)/[2n(n " 1)]} (B. Becker,personal communication, June 19, 2008), and the random-effectsmodels included the between-studies variance in the study weight.

Mean and distribution of effect sizes. Within each para-digm, the presentation first considers the mean weighted effect size(with within-subjects effect sizes combined prior to averaging),calculated by a random-effects model because the studies effectsizes were not assumed to be consistent with a single underlyingmean value. We tested for outliers, as defined by more than 1.5times the interquartile range beyond the upper quartile (i.e., theupper inner fence, Tukey, 1977). We used Cohens (1988) bench-marks for d and r to describe the magnitude of the effect sizes gand ICC, under which a g of .20 or an ICC of .10 is consideredsmall, a g of .50 or an ICC of .30 is considered medium, and a gof .80 or an ICC of .50 is considered large. Along with the overallmean, we present several distributional statistics recommended byBorenstein et al. (2009): (a) T (or tau, the estimated standarddeviation of the true effect sizes); (b) 95% confidence interval (CI;a measure of the accuracy of the mean; 95% of mean effect sizeswould fall inside this interval); (c) 90% prediction interval (PI; ameasure of the dispersion of effect sizes; 90% of true effects innew studies with a sample selected at random would fall inside thisinterval); (d) Q (a measure of uncertainty, or whether heterogene-ity is genuine); and (e) I2 (a measure of the magnitude of hetero-geneity, defined as the proportion of the observed dispersion thatis real rather than spurious on a 1 to 100% scale). An I2 near zeroindicates that almost all of the observed variance in the effects isspurious and that there is no variance to explain, whereas a largeI2 indicates that investigation of the reasons for this variance iswarranted.

Publication bias. Analyses of the distribution of effect sizeschecked for potential biases in publication or our retrieval ofstudies (see Borenstein et al., 2009). We first examined the funnelplot of the effect sizes plotted by the standard error and assessedwhether Eggers test of the plots asymmetry was significant. Wenext implemented the trim-and-fill procedure, which estimates thenumber of studies that should be removed to create a more sym-metric

FP Koenig Eagly 2011

Documents

Transcript of FP Koenig Eagly 2011