J8M2Nh30

1Contents lists available at ScienceDirect
2The Leadership Quarterly
3journal homepage: www.elsevier.com/locate/leaqua
4The queen bee: A myth? The effect of top-level female leadership on
5subordinate females
6Paulo Roberto Arvate a, ⁎ , Gisele Walczak Galilea b , Isabela Todescat c
7a São Paulo School of Business Administration and Center of Studies on Microeconomic Applied, Getulio Vargas Foundation, Rua Itapeva, 474, 01332-010 São Paulo, SP,
8Brazil
9b São Paulo School of Business Administration, Getulio Vargas Foundation and Institute of Education and Research, INSPER, Brazil
10c São Paulo School of Business Administration, Getulio Vargas Foundation, Brazil
11A B S T R A C T
12We investigate the effect of female leadership on gender differences in public and private organizations. Female
13leadership was constructed using a quasi-experiment involving mayoral elections, and our research used a
14sample of 8.3 million organizations distributed over 5600 Brazilian municipalities. Our main results show that
15when municipalities in which a woman was elected leader (treatment group) are compared with municipalities
16in which a male was elected leader (control group) there was an increase in the number of top and middle
17managers in public organizations. Two aspects contribute to the results: time and command/role model. The
18time effect is important because our results are obtained with reelected women – in their second term – and the
19command/role model (the queen bee phenomenon is either small, or non-existent) is important because of the
20institutional characteristics of public organizations: female leaders (mayor) have much asymmetrical power and
21decision-making discretion, i.e., she chooses the top managers. These top managers then choose middle man-
22agers influenced by female leadership (a role model). We obtained no results for private organizations. Our work
23contributes to the literature on leadership by addressing some specific issues: an empirical investigation with a
24causal effect between the variables (regression-discontinuity design – a non-parametric estimation), the im-
25portance of role models, and how the observed effects are time-dependent. Insofar as public organizations are
26concerned, the evidence from our large-scale study suggests that the queen bee phenomenon may be a myth;
27instead, of keeping subordinate women at bay, our results show that women leaders who are afforded much
28managerial discretion behave in a benevolent manner toward subordinate women. The term “Regal Leader”
29instead of “Queen Bee” is thus a more appropriate characterization of women in top positions of power.
30Introduction
31“There is a special place in hell for women who don't help each
32other!” These words, which were spoken by Former Secretary of State
33Madeleine Albright urging other women to support Hillary's candidacy
34in the last USA presidential election, had a great repercussion in the
35world's press (including the New York Times, The Guardian, and TIME
36magazine). Once the election had been set and following this pre-
37monition, would a portion of American women have their place in hell
38guaranteed (and another portion take their place in heaven)?
39Thankfully, between the heaven and hell of the declarations, there is an
40empirical purgatory trying to understand if and under what conditions
41women support each other in different areas of society (such as in
42politics, business, government).
43Our work is an empirical investigation that seeks to shed some light
44on what is apparently a well-established effect, the QUEEN BEE phe-
45nomenon – QBP (Derks, Ellemers, Van Laar, & De Groot, 2011; Derks,
46Laar, Ellemers, & Raghoe, 2015; Derks, Van Laar, Ellemers, & De Groot,
472011; Faniko, Ellemers, & Derks, 2016). Our investigation focuses on
48women in leadership; with our empirical strategy, we have strong
49control over the environment for estimating the causal effect of a
50woman in power on other females. Up to this point, the literature on
51leadership has not decisively addressed the issue of endogeneity bias
52(Antonakis, Bendahan, Jacquart, & Lalive, 2010). In the presence of this
53bias, which bedevils much of the observational and correlational re-
54search on which the validity of the QBP phenomenon rests, it is im-
55possible to know what the causal relation is between a woman in a
56position of power and gender-oriented outcomes.
57To identify the effect of female leadership independent of the en-
58dogeneity bias due to reverse causality and omitted variables, we use
59https://doi.org/10.1016/j.leaqua.2018.03.002
60Received 28 July 2017; Received in revised form 8 March 2018; Accepted 13 March 2018
61⁎ Corresponding author.
62E-mail address: paulo.arvate@fgv.br (P.R. Arvate).
63The Leadership Quarterly xxx (xxxx) xxx–xxx
641048-9843/ © 2018 Elsevier Inc. All rights reserved.
65Please cite this article as: Arvate, P.R., The Leadership Quarterly (2018), https://doi.org/10.1016/j.leaqua.2018.03.002
66the procedure established by Lee (2001); Lee and Card (2008); and Lee,
67Moretti, and Butler (2004). Basically, we study the effect of a female
68mayor chosen in a gender race–where a man is in first place and a
69woman in second place, or vice-versa–by a very small margin of votes.
70If this margin is close to zero, this type of election mimics an experi-
71ment because the final result under these conditions is almost random.
72Mayors are visible and uncontestable leaders with much asymmetrical
73power (Rucker, Dubois, & Galinsky, 2010; Sturm & Antonakis, 2015)
74and decision-making discretion (Finkelstein & Hambrick, 1990;
75Finkelstein & Peteraf, 2007; Hambrick & Finkelstein, 1987); that is,
76they have the means to asymmetrically enforce their will (preferences)
77over others, and mayors can significantly shape an organization. In
78adopting this causal identification procedure, we can compare the dif-
79ference in outcomes in municipalities that have a female leader (i.e.,
80treatment group) vis-à-vis municipalities with a male leader (i.e., con-
81trol group).
82A female leader, such as a mayor, permits us to observe gender
83differences in heterogeneous environments on municipalities because
84she may both impose her choice by command and influence on pre-
85ferences lower down the ranks in public organizations; and her influ-
86ence on other women in private organizations.
87However, as the existing literature on top-level female leadership
88suggests, women heading up organizations may provoke the so-called
89QBP. The QBP is a situation in which women who succeed in male-
90dominated settings play a negative role in the advancement of their
91female subordinates (Derks et al., 2011).
92In contrast to what the QBP may suggest, we add to the leadership
93literature the importance of influence through “the role model (RM)
94effect”. Hoyt (2005) and Hoyt and Blascovich (2007) report that
95women may react to the tendency of thinking that only men are suitable
96for management roles by demonstrating greater confidence and per-
97forming better. There must be a factor, however, that provokes this
98reaction. In line with social cognitive theory (Bandura, 1977, 1986,
991992, 1999, 2005), we believe that the existence of female leaders may
100increase women's self-esteem and encourage them to enter historically
101male-dominated environments. Thus, an elected female leader may
102influence and work as a RM who triggers a positive dynamic within
103public and private organizations, which reduces gender-related differ-
104ences. There is a tradition in political science and economics literature
105showing that the RM effect has an influence on other women (Atkeson,
1062003; Beaman, Chattopadhyay, Duflo, Pande, & Topalova, 2008;
107Carroll, 1994; Hansen, 1997; Schlozman, Burns, & Verba, 1994). The
108results of our investigation allow us to infer what occurs in organiza-
109tions when female leaders are quasi-randomly appointed.
110We conduct our research in Brazil because, to our knowledge, there
111are no other empirical studies with a database as large as the one we
112use; also, despite the effort involved in conducting our study, it is en-
113tirely replicable. Brazil has approximately 5600 municipalities aver-
114aging 20,000 inhabitants each, which ensures an investigation having
115sufficient data points and statistical power to detect any effects. Mayors
116hold an important political position in Brazil (Miguel, 2003). To verify
117the changes that occur in organizations following the appointment of a
118female leader, we consult a database containing the individual in-
119formation of workers in approximately 8.3 million registered firms
120(private and public organizations) at the municipal level. The propor-
121tion of women in the labor market is higher (59%) in Brazil than in
122other developed countries, such as France (52%) and the United
123Kingdom (57%).
124The purpose of our study is to observe changes in gender results at
125different positions in public and private organizations: top managers,
126middle managers, and lower positions. Public organizations have
127command: in other words, a mayor may choose the top-managers.
128Furthermore, as leader she can influence women in lower positions in
129her own organization for leadership, such as, for instance, for the po-
130sition of middle manager. We also study the effect of female leadership
131in terms of influence in private organizations, because the female leader
132does not have direct authority over private organizations.
133It would be ideal to observe what occurs within organizations in
134“each position” (e.g., CEO, middle managers) from the highest to the
135lowest levels, but we do not have this type of information. As it is
136reasonable to believe that an investigation into different earnings levels
137reflects the organizational hierarchies, we use this fact to establish what
138the top, middle managers, and lower positions are. Higher salaries
139mean a top management position, middle salaries mean a middle
140management position, and lower salaries mean lower positions in the
141hierarchy.
142Briefly, our results suggest that the QBP may be a myth. We find that
143there is a pro-female causal effect of female leadership in public orga-
144nizations: in other words, we find a larger number of women than men
145in both top and middle management positions. We find no robust evi-
146dence to show that this result extends to lower positions. We interpret
147the first result (at the top level) as command and the second result as the
148RM effect reflecting a female leader's influence. Thus, QBP is either non-
149existent, or less than the command and RM effect.
150In public organizations, the top manager can be chosen directly by
151the mayor and indirectly by the same mayor by way of political
152agreement with different levels of government (state and federal gov-
153ernment) if the public organization is in a municipality but is not owned
154by the municipal government (e.g., patronage). Middle managers de-
155pend on the internal dynamic of organizations: top managers choose
156middle managers and are “influenced” by the mayor as leader (via RM)
157in their choice for these positions. Therefore, we expand our under-
158standing of the process of change by investigating different pathways to
159gender-related outcomes (Fischer, Dietz, & Antonakis, 2017).
160The most robust effect favorable to women occurs when the same
161woman is reelected, that is, she serves two consecutive terms in office.
162The time effect as to how long it takes for leaders to assert their choices
163is also an issue that is not well investigated in leadership literature (see
164Antonakis, Day, & Schyns, 2012; Fischer et al., 2017). Delayed effects
165exist because the choice of leaders, the implementation of new pre-
166ference proposals and the change in women's preferences in organiza-
167tions caused by “the RM effect” (mainly, the results at the intermediate
168level in public organizations) all take time. In fact, as indicated by the
169eponymous title of the article by Shamir (2011) “Leadership takes
170time.”
171Contrary to what we find in public organizations, our results show
172that there is no observed improvement for women in private organi-
173zations. The choice of top managers in private organizations is different
174from those in public organizations. Our non-result for private organi-
175zations can be related to the work by Bertrand, Black, Jensen, and
176Lleras-Muney (2014). In Bertrand's work, the change proposed by
177Norwegian legislation (2003) relating to newly-appointed female board
178members is the only change observed. We cannot correlate the private
179result with QBP because we do not observe the emergence of female
180leaders in organizations. Thus, there is no positive effect in private
181firms, which makes sense, at least in the short to medium term, given
182that majors do not have much command and there is no RM effect on
183gender-related outcomes in private organizations.
184Our work is organized as follows. We present a review of the related
185literature on both QBP, RM, and our main hypothesis in Section 0. In
186Section 0, we present the institutional background, dataset and em-
187pirical strategy of our study. In Section 0, we report our results. Finally,
188in Section 0, we summarize our findings and discuss their implications.
189Theoretical overview and hypothesis
190The Queen Bee Phenomenon in Business
191Much of the research on female leadership is based on assumptions
192of sisterhood and solidarity between women (see Mavin, 2006; Mavin,
1932008). Women consider other women to be their natural allies. How-
194ever, the expectation that women will align themselves with other
195P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
1962
197women may not materialize and in these cases the “ally” label is re-
198placed by the “queen bee” label. Some even suggest that women are
199more like “evil stepmothers” than “fairy godmothers” and therefore will
200be eternally punished for not supporting other women. For example,
201Margaret Thatcher, the UK's first female prime minister, received from
202the world's press (see BBC News's Reality Check team, 2018; Moreton,
2032015) the “queen bee label” for not promoting the careers of other
204women in her cabinet. However, many specificities may have explained
205Thatcher's choices at that time; moreover, that was just one case, which
206obviously may not be generalized.
207The queen bee label is given to women who distance themselves
208from other women in organizations where the majority of leadership
209positions are held by men. Such women apparently seek individual
210success by adjusting to the predominantly masculine culture in the
211organization (see Kanter, 1977 and 1987; Staines, Tavris, & Jayaratne,
2121974). Negative relations between women in organizations–that is, not
213building alliances–have been highlighted in the literature for several
214decades. Women in male-dominated organizations, aid and abet the
215status quo by turning against other women, ignoring derogatory re-
216marks about them and contributing to the derogation of these other
217women by being disloyal to them (Nieva & Gutek, 1981).
218In a general sense, female leaders are expected to be more under-
219standing, more nurturing, more giving and more forgiving than men
220(O'Leary & Ryan, 1994). However, when displaying such character-
221istics, they fail to meet the perceived requirements of the managerial
222role, which mainly calls for masculine characteristics (Mavin, 2006).
223This stereotyped view of women–that is, the perception of women as
224being “less adequate” for a position–might lead to detachment from
225their identity reference group, that is, other women. It also leads to
226double bind due to descriptive, but also prescriptive, stereotyping be-
227cause women cannot violate social role expectations (Eagly & Karau,
2282002). To reach a senior position woman need to prove that they are
229different from the other women in that environment; thus, in addition
230to alienation they also describe themselves as possessing characteristics
231considered to be masculine (e.g., assertive, competitive, risk-taking),
232and this, in addition to having a stereotyped view of other women
233(Ellemers, Heuvel, Gilder, Maass, & Bonvini, 2004).
234An important argument in the literature about the QBP is that it is
235not men or women who suppress better job outcomes for women; rather
236it is the presence in the organization of gender stereotypes that hinders
237the success of women in their careers and compels top-level female
238leaders to behave in a hostile way toward subordinate females.
239According to the QBP, women in the work environment are responsible
240for “undermining” the careers of other women, but this phenomenon is
241a response to gender inequality. Apparently, the QBP is present among
242women who have low gender identification with their reference group
243(i.e., other women). This low gender identification is manifested in
244environments where women have experienced gender inequalities in
245their careers. After being encouraged to remember the gender in-
246equalities and bias they experienced in their careers, women then take
247on stereotypically masculine characteristics, which involves empha-
248sizing being different from other women and minimizing the presence
249of gender inequality (Derks et al., 2011; Derks et al., 2011; Faniko et al.,
2502016). The QBP is, therefore, thought to be a response to gender in-
251equality, a consequence of gender discrimination experienced in the
252workplace, and not a cause of it (Derks et al., 2015; Derks, Van Laar, &
253Ellemers, 2016). Of course, once in place, the QBP would add to a si-
254multaneity effect by increasing anti-female bias.
255The QBP is, however, a questionable phenomenon because it is
256difficult to establish a causal relationship between female behavior and
257the low participation of women in top management positions. Deloitte's
258study (see Women in the boardroom: A global perspective), for example,
259makes us wonder if QBP “may have lost its sting”. There is evidence in
260this study that when organizations have women in top leadership po-
261sitions the number of board seats held by women is almost double that
262of organizations with men at the top. Moreover, the effect of the
263presence of women in top management positions would extend “beyond
264the walls of any single corporation": women leaders are role models and
265mentors to other women and can break down stereotypes, encourage
266young women to pursue careers in business, and break down the wage
267gap between men and women. Still, such observational studies, and the
268many others that claim to have documented a QBP have not been un-
269dertaken using appropriate designs to determine the causal effect of
270top-level female appointments on female-related work outcomes.
271Female Leaders and Role Models
272Organizational and government leaders are responsible for making
273far-reaching decisions that can influence many aspects of society.
274Individuals in key leadership positions may seize the opportunity to
275change things for the better because leaders have the opportunity to be
276proactive (Parker, Williams, & Turner, 2006). Despite the importance of
277leadership for institutions and for ensuring the most capable are ap-
278pointed to power, the existence of female leaders in a labor market is
279very unequal when we compare it with the male situation. Women are
280severely under-represented among business and political leaders, and
281whereas they are generally increasing their share in the labor market in
282different countries, female leaders are not particularly well represented
283in the upper echelons; only 19% of firms have female top managers and
284only 23% of the seats in national parliaments are held by women
285(Miller, 2017).
286Moreover, one can assume that male and female leaders behave
287differently and favor those of the same sex in the group; given that
288males dominate in many consequential settings suggests that it is
289women who will pay the price in the long run as the male-oriented
290hierarchical structures are reinforced. Another important argument,
291therefore, is that if top leaders can reshape the allocation of resources
292and influence the outcomes of large numbers of other people, current
293gender imbalances in leadership may create additional distortions in
294the overall distribution of wealth, power and wellbeing (Miller, 2017).
295An explanation for this dearth of female leaders is the belief that
296only men are suitable for managerial roles. According to Hoyt (2005)
297and Hoyt and Blascovich (2007), top management positions require
298achievement-oriented aggressiveness and emotional toughness. An in-
299congruity exists between women's qualities and the leadership role,
300which makes it more difficult for women to attain top leadership po-
301sitions, and more difficult for these women to be viewed as effective in
302these roles. Women are also negatively stereotyped if they violate
303prescriptive expectations (Eagly & Karau, 2002).
304Although the incongruity exists, many authors suggest that its pre-
305valence depends on women's responses to this situation: The reaction of
306women to the socially accepted ideas that men are more suitable for
307managerial roles and that the qualities associated with women are in-
308compatible with the qualities necessary for leadership roles (i.e., such
309roles require more agent-like qualities). Redressing the situation can
310occur by influence from one institution (e.g., observing more female
311leaders in public institutions and politics) to other institutions (e.g.,
312private), or encourage more women to enter public service and thereby
313reduce descriptive stereotyping and what is considered normal. This
314means that the more female individuals are seen in positions of power,
315the less prevalent is the male-oriented stereotype of leadership in those
316institutions and the more the cognitive structures chip away at the
317“think manager think male” stereotype (the “think manager-think
318male” view is a global phenomenon, especially among males according
319to Schein, Mueller, Lituchy, & Liu, 1996).
320The mechanism by which these effects may occur include (a) raising
321self-efficacy, that is, belief about one's competence for addressing spe-
322cific tasks; and (b) counteracting negative stereotypes of women from
323observing female gender role models (Hoyt, 2005; Hoyt & Blascovich,
3242007; Hoyt & Simon, 2011). Within this framework, research needs to
325advance toward understanding how these positive pro-female influ-
326ences (role model) can be triggered.
327P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
3283
329With regard to the role model argument in social cognitive theory
330(Bandura, 1977, 1986), there is an element that can activate women's
331responses to female leaders as role models. Although there are several
332ways of developing leadership (Antonakis et al., 2012), action by
333learning vicariously from observing a female leader, this is also a pos-
334sibility. Other women observe and remember the sequence of events
335that legitimized this leader and use this information to guide them in
336their subsequent behaviors. It is possible, therefore, to establish a link
337between a person's perceived self-efficacy stemming from learning vi-
338cariously, and future actions taken to emulate the role model.
339Individual action can be replicated in the aggregate behavior of
340individuals and, as a consequence, of an organization. Exposure to fe-
341male leaders challenges the incongruity that exists between female
342leaders and the male stereotype of a leader and can help lessen the
343gender difference in organizations. Theoretically, this increases the
344sense of female self-efficacy and thereby improves their aspirations and
345may even encourage them to enter historically male-dominated en-
346vironments. Evidence of this development pattern is found in political
347science and economics literature (Beaman et al., 2008; Carroll, 1994). A
348great number of studies from different settings confirm that women in
349important political positions become positive role models for other
350women and young girls due to an increase in the intention to change.
351We will show some work on this below.
352Carroll (1994) showed that the presence of women in political
353leadership positions transformed beliefs about the appropriateness of
354politics for women, thereby increasing interest in political issues. Ad-
355ditionally, the research carried out by Koch (1997), Sapiro and
356Johnston Conover (1997), Hansen (1997), and Atkeson (2003) sug-
357gested that the presence of female candidates under certain con-
358ditions–legislative elections–has a positive effect on women's political
359engagement, resulting in broader political discussion. Likewise, Fox
360(1997) showed that political campaigns that have female candidates
361tend to have a greater impact among families, generating further dis-
362cussion and greater interest from young women in political matters.
363Moreover, Campbell and Wolbrecht (2006) showed that the presence of
364women in government can raise the level of involvement of young
365women in politics, which then positively affects their likelihood of
366political participation.
367In a similar vein, Koch (1997), like Schlozman et al. (1994), stated
368that women report greater knowledge of campaign details (e.g., can-
369didate names and campaign schedules) when female candidates run for
370leadership positions. However, Koch (1997), Sapiro and Johnston
371Conover (1997) and Hansen (1997) believe the “role model effect” is
372limited to the degree of visibility of the candidate. Atkeson (2003)
373analyzed senatorial and gubernatorial campaigns and found that the
374presence of female candidates affected the engagement of other women
375only in campaigns that were highly competitive.
376The role model effect can stem from other channels too. For in-
377stance, Beaman et al. (2008) showed that increased public exposure to
378female leaders tended to improve the perception of the effectiveness of
379female leaders, thereby weakening stereotypes about gender roles in
380both public and private spheres. They also observed that there was a
381greater tendency for voters to vote for a woman if those same voters
382had been previously exposed to a government with a female leader.
383This work showed that the effect of female leaders on other women led
384to an effect that was more widespread than merely affecting women.
385Research has also explored the position of women as an unusual
386phenomenon (Campbell & Wolbrecht, 2006). They showed that it is
387“unusual” for young women to become candidates for office and that
388female candidates evoke a greater impact when the number of women
389in politics is low. As more women participate in politics, however, their
390involvement becomes less unusual and, therefore, there is a perceived
391reduction in public interest. Thus, the increasing presence of women in
392politics may have less of an impact on the aspirations of younger
393women over time.
394Gender is a relevant factor when the candidacy of a woman is
395considered unique or unusual, such as when gender issues are central to
396the campaign agenda (Campbell & Wolbrecht, 2006; Wolbrecht &
397Campbell, 2007). This type of visibility suggests that the candidacy of a
398woman is sufficiently important to potentially result in victory. This
399effect may shape the agendas of future election campaigns and have a
400large impact on the political socialization of young women.
401In short, we believe that women in power are likely to provide a
402nurturing ground for other women to succeed and that these effects
403stem from the direct effects that women in power exert in their spheres,
404or under the influence of the RM effect. In considering our development
405and these studies, the main hypothesis we examine is the following:
406H1. Female leaders reduce gender differences in organizations.
407Institutional Background, Dataset and Empirical Strategy
408Institutional Background
409Brazilian Political System
410Brazil is a federalist country with three levels of government,
411namely, federal, with 27 states and 5565 municipalities. There are three
412electoral district sizes in the country. For local elections, the district is
413the municipality, where elections for mayor and councilors are held.
414For national and state elections, the electoral district is the state, where
415federal, district, and state deputies, governors, and senators are elected.
416The electoral district of the country as a whole exists only for pre-
417sidential elections. Elections are held every two years in Brazil, with
418local elections occurring every four years and elections for president,
419governors, senators and federal, district and state deputies occurring
420mid-term to local elections. Except for senators, who are elected for
421eight-year terms, all office-holders in the Executive and Legislative
422branches have four-year fixed terms. The constitutional amendment of
4231997 established that Executive officials, including mayors, governors
424and the president, can only be reelected once, which is the term limit
425rule. Federal, District, and State Deputies and Councilors are elected by
426way of an open-list proportional representation system; voters can
427order the list of either candidates or parties. Although the parties in-
428itially order candidates, like countries such as Belgium, Austria, the
429Netherlands, Switzerland, and Luxembourg do, voters can still alter the
430order of candidates by what is known as preference voting. There is no
431term limit for legislative members.
432Since the 1988 Constitution, mayors have been chosen in one-round
433elections using a majoritarian system in municipalities with fewer than
434200,000 registered voters. Mayors are chosen in run-off elections in
435municipalities with over 200,000 registered voters if no candidate
436achieves a majority of valid votes in the first round (50% plus one of the
437valid votes). We only use data for municipalities below the 200,000-
438voter threshold. In doing so, we exclude approximately 120 of the
439largest municipalities from the initial country sample. The reason for
440this decision is to avoid strategic voting behavior, when voters do not
441necessarily reveal their preferences in the first round (Fujiwara, 2011).
442In the first-round election, voters choose the party or candidate of their
443preference. In the second-round election, some voters do not have a
444party of their preference. Given that the vote is compulsory, these vo-
445ters can vote strategically. They can choose a party or candidate that is
446not as bad for them in line with their initial preferences in the first-
447round. Thus, voters reveal their preference for parties and candidates
448only in the first-round.
449A. Electoral Gender Quota
450Law 9100 of 1995 established an electoral gender quota in Brazilian
451elections. In accordance with the law, all parties and coalitions had to
452“to reserve” 20% of their candidacies for women. Law 9504 of 1997
453changed the percentage of “reserve” candidacies for women from 20%
454to 30%. Law 12,034 of 2009 subsequently changed the word “reserve”
455P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
4564
457to “filled by” 30% female candidates, with a maximum of 70%. This
458change was important because parties and coalitions did not under-
459stand the reserve requirement as a “commitment”. In the judgment of
460the Superior Electoral Court (Resolution No. 23,373, 2011), the court
461established that if female candidates do not represent 30% of the can-
462didates of the parties or coalitions, all the candidates of these parties or
463coalitions will be contested. Therefore, institutionally the electoral
464gender quota was valid only after 2011. Before, there was an estab-
465lished idea that the quota may or may not be complied with by the
466party or coalitions. The electoral quota was technically non-mandatory
467for coalitions or parties in the period of our investigation.
468Formal Labor Market
469The Brazilian labor market has both formal and informal elements.
470The formal market is regulated and comprises natural and legal persons
471both registered according to the Brazilian legal system, whereas the
472informal market has neither registered individuals nor employment ties
473for professionals, nor legal obligations for companies.
474As in other countries, Brazilian firms are either public or private.
475The rules and conditions for employment in the public sector differ
476from those in the private sector. Formal public employment includes
477special, statutory, and contractual employment and employment by
478appointment. Employment by appointment refers to employees ap-
479pointed by government officials. Special employment refers to workers
480hired under exceptional circumstances (e.g., cases in which there are
481urgent needs or that require extraordinary skills) and who have a pre-
482determined term of employment specified in their contracts. Statutory
483employment is governed by a set of special rules, including lifetime
484contracts from which employees cannot be discharged and pension
485schemes that are far better than those available to formal employees in
486the private sector. Finally, contractual employees are subject to the
487same rules as employees in the private sector, this type of employment
488being governed by the Consolidation of the Labor Laws (Consolidação
489das Leis do Trabalho), which was enacted in 1943 to consolidate all
490Brazilian labor legislation.
491The 1988 Constitution established a minimum monthly salary for
492formal workers in the public and private sectors. The minimum salary is
493fixed for full-time work of 8h per day, or 44h a week. Employees who
494work part-time–a maximum of 25 h per week–can receive less than one
495minimum salary. The minimum salary is the lowest salary that an
496employer can legally pay their employees for a full-time job, and the
497lowest price for which a person can legally sell their labor. The Nominal
498Minimum Wage was BRL 937.00 (BRL 31.23 per day) on January 1st
4992016. This amount is approximately US$292.82 per month (exchange
500rate: US$ 1/BRL 3.20).
501Command in Public Organizations
502The influence of mayors on the management of public-sector orga-
503nizations can be observed in the appointments made. Appointments in
504these Brazilian state-owned enterprises are direct in municipal-owned
505enterprises and indirect at the federal and state levels. Indirect ap-
506pointments are fundamentally associated with the distribution of public
507resources and the partisan-electoral arena, in other words following the
508practice of political “patronage” (Barbosa & Ferreira, 2017). Essentially,
509this concept is an exchange of favors between federal/state politicians
510and of public resources between municipal politicians–jobs in state-
511owned enterprises, allowances or tariff protection for particular in-
512dustries, construction projects in particular districts–by way of votes or
513other forms of political support (see Gordin, 2002Graziano, 1976). The
514literature justifies this exchange of public resources by way of votes or
515other forms of political support: Voters are heterogeneous in their af-
516finity to parties. They also care about private benefits, and this interest
517tempers their basic party loyalties. Consequently, voter willingness to
518compromise their party affinities in response to offers of private ben-
519efits gives rise to this exchange (see Cox & McCubbins, 1986; Dixit &
520Londregan, 1996; Lindbeck & Weibull, 1993).
521State ownership of large-scale enterprises in Brazil occurs at the
522federal, state and municipal levels. Federal state-owned enterprises may
523be found in a great number of Brazilian municipalities. The Empresa
524Brasileira de Correios e Telégrafos, for example, is set to expand the postal
525service through its business units to all municipalities with a population
526of> 500 residents by 2018. 1 State-owned enterprises at the state level,
527such as the Companhia de Processamento de Dados do Estado de São Paulo
528(PRODESP) 2 [data processing] and the São Paulo Desenvolvimento Ro-
529doviário S.A. (DERSA) [highways], both owned by São Paulo, which is
530richest state in Brazil, are located in several municipalities in that state.
531Whereas mayors cannot occupy an office that involves a formal
532position of command in a state enterprise (e.g., CEO), they delegate
533power to trusted collaborators 3 by making these collaborators the di-
534rectors of state enterprises with power to appoint scores of other sub-
535ordinates. According to Schneider (1991), appointments create a dif-
536ferent informal control hierarchy: A trusted collaborator occupying a
537top position gives the appointer a type of control, to the extent that they
538are appointing someone who will exhibit predictable behavior. The
539great appeal comes with the appointer's ability to gain control over
540uncertainties and elicit the desired performance. The prior personal
541bond between appointer and appointee establishes the subsequent
542working relationship between government and enterprise and gives the
543appointer control over management of the firm.
544Geddes's (1994) approach is probably the most comprehensive re-
545ference to the topic of Brazilian appointment strategies. According to
546Geddes, politicians who appoint able administrators committed to
547achieving political goals have a greater probability of carrying out
548successful policies. She argued that appointment strategies can be
549predicted on the assumption that politicians want to continue to ex-
550ercise political power in the future.
551To conduct her tests, Geddes (1994) developed a quantitative
552measure, the Appointment Strategy Index, of the extent to which po-
553liticians used competence and personal loyalty as the basis for selecting
554administrative personnel in Latin American countries in the 1945–1993
555period; these countries included Argentina, Chile, Brazil, Colombia,
556Venezuela and Peru. The Appointment Strategy Index is based on an-
557swers to eight questions by the forty-four constitutional governments
558(e.g., Was the criterion for choosing administrative personnel primarily
559partisan rather than competence-driven? Was there a concentration of
560appointments among party or coalition members? Were the public-
561sector jobs of party members protected?). Scores on this index depend
562on the number of negative answers to the questions for each adminis-
563tration.
564The Brazilian Appointment Strategy Index indicates that the poli-
565tical selection of top administrative personnel by politicians emphasizes
566competence and loyalty in some periods and remaining within the
567politician's party or coalition in other periods; note, loyalty and holding
568the coalition together take precedence over competence.
569Data
570We collect microdata from three different sources for our key hy-
571pothesis: the Supreme Electoral Court (Tribunal Supremo Eleitoral - TSE),
572the Annual Report of Social Information (Relação Anual de Informações
573Sociais - RAIS), and the Brazilian Institute of Geography and Statistics
574(Instituto Brasileiro de Geografia e Estatística - IBGE).
575The TSE began publishing electoral data electronically only after
5761996, following introduction of the electronic ballot by Law 9100/95.
577The 1996 municipal elections were the first to have electronic ballots in
5781 According to Portaria, MC N° 6206 of 13/11/2015 there are 4261 municipalities.
5792 For example, developing the “Poupatempo Program”, a program for reducing the
580state's service bureaucracy.
5813 As stated by Schneider (1991), trusted collaborators are people who have school or
582professional ties and other personal relationships with the politicians and thus are ap-
583pointed to “positions of trust”.
584P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
5855
586the vast majority of the contests. Candidate information (mainly their
587gender) is incomplete for the 1996 election. Our quasi-experiment fo-
588cuses on the first municipal election in 2000; the immediate term after
589the election started in 2001 (2001–2004); the next election that oc-
590curred in 2004; and the subsequent term, which began in 2005
591(2005–2008). By choosing data for municipalities below the 200,000-
592voter threshold, we exclude approximately 120 of the largest munici-
593palities from the initial country sample.
594The RAIS database contains information about workers employed in
595formally-constituted firms in Brazil. Formal workers are employees who
596have a formal labor contract, which in Brazil is defined as having a
597booklet (i.e., the carteira de trabalho) that registers a worker's entire
598employment history in the formal sector. Formal firms are those re-
599gistered with the tax authorities, which means they possess the tax
600identification number required for Brazilian firms (i.e., the Cadastro
601Nacional de Pessoa Juridica – CNPJ). Our database contains no informal
602firms; 40% of the economically active workforce in Brazil is involved in
603informal employment (Census, 2010).
604We know the municipality, wage information, and the gender of
605employees. To carry out our investigation, we assume that the salary
606information reflects the hierarchical position of workers in firms. For
607example, individuals receive low salaries at the bottom of the hierarchy
608(e.g., cleaning workers), intermediate salaries are middle managers,
609and individuals with high salaries at the top managers (e.g., CEO). We
610build the position of individuals and show results considering three
611levels, observing the ratio between the groups (women and men): low
612position (earning up to 1.5 minimum salaries), middle (earning be-
613tween 10 and 20 minimum salaries), and top (earning over 20
614minimum salaries). As over 20 minimum salaries is the highest-level
615piece of information we obtained, “over 20” is regarded as the salaries
616of top managers. Lower than this, it is salary of intermediate position:
617middle managers. However, what is the empirical salary difference
618between low and intermediate positions given that the separation point
619between ranges is not so obvious? We would have found it difficult to
620separate and classify these positions if we had had a large number of
621significant intermediate results. This would have been particularly
622difficult if we had had significant results close to 10: between 9.51 and
62310, for example. This did not occur. The non-significant result of the
624intermediate ranges, between 1.51 and close to 10, helped us establish
625the difference in positions between low and middle level careers.
626We aggregate the salary information by gender at the municipal
627level annually. Then we build the average results for each municipality
628in the three terms: 1997–2000, 2001–2004, and 2005–2008. The
6291997–2000 term is important for the internal validity of the quasi-ex-
630periment (Eggers, Fowler, Hainmueller, Hall, & Snyder, 2015; Imbens &
631Lemieux, 2008). The 2001–2004 term is the result immediately after
632the 2000 election, our quasi-experiment election. The 2005–2008 term
633serves to evaluate the gender difference of workers in firms in the fol-
634lowing term when the mayor in the electoral quasi-experiment is re-
635elected. By establishing different moments in time, we expect to con-
636tribute to the “time” dimension that is lacking in literature, as indicated
637by Fischer et al. (2017). Finally, we extract municipal data from the
638census that was held in 2000, which was produced by IBGE.
639Table 1 shows the variables used in our empirical investigation,
640their creation, and sources.
641Tables 2A and 2B show the descriptive statistics of the variables of
642three different groups of margins of victory (30%, 10%, and 5%) and
643the significant difference between them when the margin of victory is
6445%. The statistical difference between the variables is small.
645There is variation in the number of observations between the vari-
646ables that reflect gender differences (e.g., the ratio of female to male
647workers with different earnings). This variation reflects the fact that
648there are no organizations paying certain salary levels in some muni-
649cipalities. For example, if a municipality has no large organizations
650then it is unlikely to have organizations paying their employees>20
651minimum salaries. In contrast, there are municipalities with no workers
652earning < 0.5 minimum salaries. One last factor to note is that public
653organizations are found in fewer municipalities than are private orga-
654nizations.
655Empirical Strategy
656Our empirical strategy is designed to investigate whether a female
657leader, compared to a male one, improves the position of female
658workers in organizations over which she has “command/influence”
659(i.e., as an elected mayor), or for which she only has influence, a role
660model (i.e., in private organizations). This test is no easy task, given
661that the non-observable characteristics that define a leader can interfere
662in this relationship. In line with previous literature, for instance, per-
663sonality traits (Fleeson, 2001; Fleeson & Gallagher, 2009) and biolo-
664gical differences (Ashton, 2007) may explain leadership, and voters
665may be influenced by these differences in their choices. Moreover, it is
666possible to assume reverse causality between the variables. The direc-
667tion of our investigation is x (modelled independent variable) → y
668(dependent variable) and not y →x. This reverse causality may occur
669because improvements in the situation of female workers compared to
670male workers in a municipality may explain the emergence of a female
671leader, that is, because of a pro-women environment. An investigation
672that does not consider this possibility, therefore, is likely to contain
673bias.
674To solve this problem, we adopt a sharp regression-discontinuity
675approach, as proposed by Lee et al. (2004) and Lee and Card (2008).
676Municipalities where a woman wins by a large margin are likely to be
677different from those where a woman wins by a small margin. However,
678when we narrow our focus to those municipalities with close-run
679elections, it becomes more plausible to believe that election outcomes
680are determined by idiosyncratic factors. The female leader is elected
681without the influence of observable and non-observable variables, and
682we can observe the influence of the leader on our variables of interest
683(x→y).
684Following the above arguments, we focus our investigation on
685municipalities with close-run mayoral races (i.e., a victory margin close
686to zero) in which the two candidates receiving the most votes are a
687woman and a man. In other words, we are modeling a gender race to
688observe what occurs when the winner is “randomly” appointed.
689Our treatment variable D it is a dummy variable that equals one
690when a woman defeats a male opponent in municipality i in year t. The
691control group (D it = 0) is formed by the municipalities that elect a man.
692The running variable is the margin of victory (Margin it ), which is de-
693fined as the percentage of votes between female and male candidates
694for mayor. Thus, the relationship between D it and Margin it can be
695written as follows:
696= ⎧
697⎨
698⎩
699>
700D
701if Margin
702otherwise
7031 0
7040
705it
706it
707(1)
708The cut-off point at which the margin of victory equals zero corre-
709sponds to a single criterion for determining the candidate's gender. The
710impact of a local female leader on Y it+Ï„ , the dependent variable that
711represents the gender ratio for each municipality–the main variable in
712our investigation–is defined by parameter β, which is an average
713treatment effect near the cut-off point. This effect can be written as
714follows:
715= −
716↓
717+
718↑
719+
720β E Y Margin E Y Margin lim
721(
722|
723)
724lim
725(
726|
727)
728Margin
729it Ï„
730it
731Margin
732it Ï„
733it
7340 0 (2)
735Our causal identification strategy, therefore, is to detect the “jump”
736in the dependent variable at the discontinuity point that can be at-
737tributed to the effect of crossing the discontinuity point. The treatment
738effect is causally interpreted because for a woman winning in munici-
739pality i there should be no systematic differences between the ob-
740servable characteristics of municipalities, the electorate, and so forth,
741and the non-observable characteristics of the leader observed by voters.
742P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
7436
744Under certain conditions, municipalities where women barely lose can
745serve as a reasonable counterfactual for municipalities where they
746barely win; in this way, we have a quasi-experiment, wherein we solve
747the endogeneity problem and can make relatively strong causal claims
748(Antonakis et al., 2010).
749There is the possibility of estimate β with different methods (Calonico,
750Cattaneo, & Titiunik, 2014; Imbens & Lemieux, 2008). Without a great
751technical difference between them, we opt to use the estimator suggested
752by Calonico et al. (2014). This approach shows that the non-parametric
753estimation of (2) by local linear regression typically leads to bandwidth
754choices that are too “large”, which means that there will be a large
755asymptotic bias term; a triangular kernel function, in which the weight of
756each observation decays with the distance from the cut-off.
757In addition to addressing the issue of endogeneity, we can in-
758vestigate the influence of the leader on the variables over time (t+n).
759Insights about how to study the impact of time in leadership literature
760have been discussed by several authors (e.g., Antonakis et al., 2012;
761Fischer et al., 2017; Shamir, 2011); however, as we mentioned before,
762taking time into account in empirical investigations is something that is
763sorely lacking.
764Results
765Validity of the research design
766We validate our quasi-experiment using internal and external tests
767following the procedures established by Imbens and Lemieux (2008)
768and Eggers et al. (2015). First, we determine whether the electoral
769process was manipulated. Electoral manipulation for any candidate
770(i.e., a woman or man) invalidates the quasi-experiment. We use
771McCrary's (2008) test to identify electoral manipulation.
772A visual inspection of the histogram of the density of the margin of
773victory (Fig. 1 –left side) may show electoral manipulation of the cut-off
774point (the percentage margin of votes equals zero) with different class
775intervals or “bins” (2 percentage points-pp, 1pp, and 0.5pp – left side
776of figure). Electoral manipulation occurs when more candidates win the
777Table 1
778Definition of variables, how they are constructed, and their sources.
779Variable Construction of variable Source
780Margin of victory
781Difference in the percentage of
782votes between female and male
783candidates for mayor, considering
784the candidates in first and second
785places in the first round of
786elections.
787We extract the number of votes for
788each candidate. We then build the
789margin of victory.
790Superior Electoral Court
791(TSE) for the election of
7922000 (municipal election
793for mayors and
794councilors).
795(www.tse.gov.br)
796Education
797Mayors who completed primary
798education but not high school;
799mayors who completed high school
800but not higher education; and
801mayors who completed higher
802education.
803Dummy variables with values equal
804to 1 when the definition is met and
805zero otherwise.
806PT (Worker´s Party),
807PSDB
808(Brazilian Social
809Democratic Party), and
810PFL (Liberal Front
811Party) 1
812Dummy variables with values
813equal to 1 when the definition is
814met and zero otherwise. We
815include two parties with left-wing
816ideology (PSDB and PT) and one
817party with right-wing ideology.
818PT (Workers Party), PSDB
819(Brazilian Social Democratic Party),
820and PFL (Liberal Front Party)
821Percentage of population
822with access to mains
823water supply
824Percentage of houses with access to
825mains water supply in the
826municipality
827IBGE - Brazilian
828Institute of Geography
829and Statistics – 2000
830Census
831Population
832Population, percentage of
833municipal population under 1 year
834old; percentage of municipal
835population 1 to 4 years old;
836percentage of municipal population
83749 to 59 years old; percentage of
838municipal population 60 to 69
839years old; and percentage of
840municipal population 70 to 79
841years old.
842Population in thousands
843Percentage of women as
844a proportion of total
845population
846Percentage of women in total
847municipal population
848Gender ratio
849Ratio of female to male workers in
850public and private organizations in
851specific variables: Hours worked;
852age of workers; average number of
853minimum salaries; number of
854workers earning up to 0.5
855minimum salaries, between 0.51
856and 1 minimum salary, between
8571.01 and 1,5 minimum salaries,
858between 10.01 and 15 minimum
859salaries, between 15.01 and 20
860minimum salaries, and over 20
861minimum salaries.
862We extract the information for all
863firms existing in each municipality
864over several years (1996/2008).
865Then, we build the different
866variables annually. The lagged
867variable contains the average
868between the years 1996 and 2000 in
869the municipality. The moment
870immediately after the election
871contains the average between the
872years 2001 and 2004. The next term
873and the mayor being reelected
874contains the average between the
875years 2005 and 2008.
876RAIS – Annual Social
877Information on workers
878in firms in the formal
879sector - produced by the
880Ministry of Employment
881and Labor. The data are
882from 1996 and 2008.
883The income values are
884deflated by IGP-DI
885(2000). The information
886is provided annually by
887firms (information
888centralized in
889December).
890Note: Ideology of Latin American Parties are classified as established by Coppedge (1997).
891P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
8927
893Table 2A
894Descriptive statistics.
895Variable Margin of victory
896Between
897-30% and 30%
898Between
899-10% and 10%
900Between
901-5% and 5%
902Average Std. Dev. Obs. Average Std. Dev. Obs. Average Std. Dev. Obs.
903Covariates
904Mayors who completed primary education but not high school 0.25 0.43 555 0.23 0.42 246 0.26 0.44 134
905Mayors who completed high school but not higher education 0.32 0.46 555 0.36 0.48 246 0.33 0.47 134
906Mayors who completed higher education 0.41 0.49 555 0.39 0.48 246 0.38 0.48 134
907Worker´s Party (PT) 0.01 0.13 555 0.01 0.12 246 0.01 0.12 134
908Brazilian Social Democratic Party (PSDB) 0.17 0.38 555 0.15 0.36 246 0.12 0.33 134
909Front Liberal Party (PFL) 0.23 0.42 555 0.24 0.43 246 0.24 0.43 134
910Percentage of total population with access to mains water supply 0.56 0.23 531 0.57 0.22 237 0.58 0.22 129
911Population 20666.34 30856.32 531 21223.95 28814.73 245 22455.95 32247.85 133
912Percentage of women as proportion of total population 0.49 0.01 530 0.49 0.01 239 0.49 0.01 128
913Gender ratio of hours worked – Formal Workers 1.05 0.80 555 1.09 0.82 246 1.08 0.83 134
914Gender ratio of age – Formal Workers 1.00 0.09 555 1.00 0.11 246 0.99 0.05 134
915Percentage of municipal population less than 1 year old 1.95 0.46 555 1.95 0.47 246 1.96 0.46 134
916Percentage of municipal population 1 to 4 years old 8.12 1.67 555 8.16 1.72 246 8.12 1.77 134
917Percentage of municipal population 49 to 59 years old 7.58 1.55 555 7.49 1.47 246 7.53 1.51 134
918Percentage of municipal population 60 to 69 years old 5.32 1.33 555 5.31 1.25 246 5.26 1.25 134
919Percentage of municipal population 70 to 79 years old 3.04 0.98 555 3.02 0.95 246 3.00 0.95 134
920Dependent Variables
921Gender ratio - Average income in terms of minimum salaries 0.83 0.15 555 0.81 0.14 246 0.82 0.13 134
922Gender ratio - Number of workers earning up to 0.5 minimum salaries 2.22 2.86 408 2.27 3.30 182 2.30 2.67 103
923Gender ratio - Number of workers earning between 0.51 and 1 minimum salary 1.57 1.21 553 1.60 1.27 245 1.64 1.24 133
924Gender ratio - Number of workers earning between 1.01 and 1.5 minimum salaries 1.45 0.98 554 1.58 1.30 246 1.48 0.76 134
925Gender ratio - Number of workers earning between 10.01 and 15 minimum salaries 0.61 0.93 470 0.59 0.91 210 0.54 0.76 115
926Gender ratio - Number of workers earning between 15.01 and 20 minimum salaries 0.47 0.64 398 0.46 0.63 175 0.49 0.61 98
927Gender ratio - Number of workers earning over 20 minimum salaries 0.36 0.63 414 0.29 0.48 184 0.25 0.31 99
928Table 2B
929Descriptive statistics and t-tests by elected.
930Variable Margin of victory Between -5% and 5%
931Woman Man Diff.
932Average Std. Dev. Obs. Average Std. Dev. Obs.
933Covariates
934Mayors who completed primary education but not high school 0.13 0.34 202 0.31 0.46 205 0.17***
935Mayors who completed high school but not higher education 0.39 0.49 202 0.33 0.47 205 -0.06
936Mayors who completed higher education 0.46 0.50 202 0.34 0.47 205 −0.11**
937Worker´s Party (PT) 0.05 0.23 202 0.04 0.20 205 −0.01
938Brazilian Social Democratic Party (PSDB) 0.16 0.37 202 0.13 0.34 205 −0.03
939Front Liberal Party (PFL) 0.20 0.40 202 0.21 0.41 205 0.01
940Percentage of total population with access to main water supply 0.57 0.21 192 0.56 0.22 195 −0.01
941Population 18083.46 23603.24 201 22228.95 33441.27 204 4145.5
942Percentage of women in total population 0.49 0.01 193 0.49 .01 196 0.00
943Gender ratio of hours worked – Formal Workers 1.05 0.76 202 1.06 .77 205 0.01
944Gender ratio of age – Formal Workers 1.01 0.07 202 1.01 .08 205 0.00
945Percentage of municipal population less than 1 year old 1.97 0.46 145 1.98 .49 160 0.01
946Percentage of municipal population 1 to 4 years old 8.24 1.77 145 8.22 1.80 160 −0.02
947Percentage of municipal population 49 to 59 years old 7.57 1.51 145 7.48 1.51 160 −0.08
948Percentage of municipal population 60 to 69 years old 5.29 1.28 145 5.26 1.24 160 −0.03
949Percentage of municipal population 70 to 79 years old 3.06 0.98 145 3.00 0.92 160 −0.06
950Dependent Variables
951Gender ratio - Average income in terms of minimum salaries 0.84 0.18 202 0.83 0.16 205 −0.01
952Gender ratio - Number of workers earning up to 0.5 minimum salaries 2.36 2.51 148 2.07 2.37 159 −0.30
953Gender ratio - Number of workers earning between 0.51 and 1 minimum salary 1.50 1.07 202 1.59 1.06 204 0.08
954Gender ratio - Number of workers earning between 1.01 and 1.5 minimum salaries 1.30 0.59 201 1.40 0.75 205 0.10
955Gender ratio - Number of workers earning between 10.01 and 15 minimum salaries 0.55 0.65 167 0.55 0.74 163 0.01
956Gender ratio - Number of workers earning between 15.01 and 20 minimum salaries 0.48 0.62 155 0.39 0.64 141 −0.08
957Gender ratio - Number of workers earning over 20 minimum salaries 0.30 0.42 148 0.38 0.73 145 0.08
958Note: * p< 0.01, ** p< 0.05, * p< 0.1.
959P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
9608
961election than lose it. Potential manipulation of the selection mechanism
962would occur if there were a jump in frequency of the running variable
963(percentage margin of votes) close to the cut-off point. Based on the
964procedure proposed by McCrary (2008), we test the null hypothesis of
965continuity of the density of the running variable (i.e., margin of votes)
966against the hypothesis of an interruption at the cut-off point. Our results
967indicate that the null hypothesis cannot be rejected (t= −0.1556),
968indicating no discontinuity around the cut-off point. Therefore, there is
969no discontinuity around the cut-off point.
970We used a second procedure to investigate whether municipal
971characteristics affected the electoral result. We also use here the
972method proposed by Calonico et al. (2014). The main idea is that there
973is no “jump” in these variables in elections with a zero margin of votes
974(i.e., close to the cut-off point). If there is a jump, then municipal
975characteristics might explain the electoral result (the environment). For
976instance, a higher percentage of women as a proportion of the total
977population when a woman wins a close-run election could be the factor
978explaining the victory of a woman over a man. We need, therefore,
979results without a “jump” in these variables. In other words, we need
980balanced results for the covariates (see Table 3A).
981The municipal characteristics are well-balanced; that is, we observe
982no significant differences between the treatment and the control group on
983municipal covariates. The variables we used included: (a) percentage of
984the total population with access to the public drinking water supply
985system;, (b) the total population of the municipality; (c) women as a
986percentage of the total population; (d) the ratio of hours worked in formal
987organizations, by gender; (e) the ratio of ages in formal organizations, by
988gender; (f) the percentage of the population<1year old; (g) the per-
989centage of the municipal population 1 to 4years old (children can affect
990women's progress in the labor market); (h) the percentage of the muni-
991cipal population 49 to 59years old, the percentage of the municipal po-
992pulation 60 to 69years old; and (i) the percentage of the municipal po-
993pulation 70 to 79years old (the retirement age of women and men may
994differ, given that women live longer than men in Brazil, IBGE).
995We also examine the differences in the observable characteristics of
996candidates for mayor. This assessment provides a means for
997determining which characteristics differ between male and female
998candidates (see Table 3B).
999Between the examined characteristics of elected mayors, which in-
1000clude education level and party affiliation (i.e., Worker's Party-PT,
1001Brazilian Social Democratic Party-PSDB, and Liberal Front Party-PFL),
1002we find no significant differences. Parties may have either a favorable
1003strategy for electing women candidates or a bias against women can-
1004didates (Kittilson, 2006; Krook, 2006).
1005Imbens and Lemieux (2008) and Eggers et al. (2015) did not es-
1006tablish the importance of covariate balance in the individual char-
1007acteristics of same-gender candidates (i.e., women/winners compared
1008with women/losers and men/losers compared with men/winners) in
1009the sharp regression discontinuity design. In this study, we compare
1010same-gender characteristics. We adopt this procedure to check whether
1011the women/men elected differ from other candidates of the same sex.
1012The premise of this verification is to exclude the bias that may be at-
1013tributed to “super” women/men winning (a woman with characteristics
1014that are superior to those of other women in her group – between
1015winners and losers – which could explain the electoral result; the same
1016is valid for men) (see Table 3C).
1017The individual characteristics of same-gender candidates appear
1018well-balanced and the candidates appear to be very similar. We find a
1019small discrepancy between male candidates. Among male candidates,
1020two of the six characteristics are not balanced at the 10% level. We have
1021fewer winning men than losing men in the PT (Worker's Party) and
1022fewer winning men who completed higher education. We have two
1023other parties (PFL and PSDB) and two other education levels (i.e.,
1024completed primary education and completed high school) that are ba-
1025lanced between the candidates. There is no strong evidence of “super”
1026candidates, and this residual difference between male candidates does
1027not affect the internal validity of the quasi-experiment. 4
10280 10 20 30
1029y c n e u q e r f e t u l o s b A
1030-.6 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 .6
1031Percentage margin of votes
1032bin=2% bin=1%
1033bin=0.05%
1034Frequency
10350 1 2 3 4
1036Density Estimate
1037-1 -.5 0 .5 1
1038Percentage margin of votes
1039McCrary test
1040Fig. 1. Frequency and McCrary test.
10414 According to Guala (2012), internal validity is threatened in an experiment when
1042confounding variables compete with the “jump” in the zero margin of victory to explain
1043the dependent variables. The non-dominant difference (two out of seven) between male
1044candidates does not explain the victory of women over men.
1045P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
10469
1047The third procedure investigates the possibility that the observed
1048change could be driven by movements in the labor market and the
1049management of previous mayors (including gender differences which
1050will have an electoral impact) in a municipality, instead of local female
1051leadership. Therefore, we analyzed whether the observed change in
1052dependent variables occurred before the 2000 election, which can be
1053considered as a placebo test. In those cases in which significant change
1054is observed, we ignore these variables in the main result because the
1055result was identified previously: y t−1 →y t. Thus, it is neither x t →y t nor
1056x t →y t+τ. We also use the method proposed by Calonico et al. (2014) to
1057investigate this issue.
1058A. Direct effect of a female leader in both public and private organi-
1059zations
1060The best data strategy for identifying the institutional heterogeneity
1061of the public and private environments is to investigate the effect of a
1062female leader as mayor, vis-à-vis a male leader with an electoral victory
1063in another municipality, in two samples. The natural way to investigate
1064the division in organizations is to build a categorical variable, a dummy
1065with a value equal to 1 for the public sector and zero otherwise; we then
1066interact the dummy variable with the treatment dummy (D it ) and re-
1067estimate the initial estimate with the interaction. The interaction cap-
1068tures the difference in the treatment group in public organizations vis-à-
1069vis private organizations in the same group. The problem here is that
1070the regression discontinuity estimate presupposes that Margin it is equal
1071to zero (see Eq. (2)) and the covariates does not produce any effect on
1072the treatment dummy (see Lee & Lemieux, 2010), including the inter-
1073action effect.
1074Given the above result, we decided to show our empirical results in
1075two samples, separated in the sequence: one public and the other pri-
1076vate. Table 4 shows the lagged results (i.e., previous results on our
1077dependent variables) using the regression discontinuity strategy for two
1078distinct samples.
1079We find only one of the variables with significant results in the
1080private sample: the ratio of female to male workers earning up to 0.5
1081minimum salaries in the sample of private organizations (see Table 4).
1082We do not consider significant results in the main investigation for this
1083variable because a lagged variable result is not a result in a quasi-ex-
1084periment (internal validity).
1085Main results
1086We present two sets of results. The first set considers public orga-
1087nizations and the second set considers private organizations. In each set
1088we show the gender changes in municipal organizations considering:
1089(a) the term immediately after a woman is elected as the mayor,
1090compared with another municipality in which a man wins; (b) the
1091subsequent term (four years later); and (c) the same subsequent term
1092when the same woman as in the quasi-experiment is reelected. We
1093adopt this procedure because it is important to consider “the time ef-
1094fect”, given that a leader's influence may not be immediate, especially
1095for the type of dependent variable we are studying. More specifically, in
1096the case of our study, time is important because the leaders are chosen
1097from a “mimicked” randomized experiment for public organizations
1098and they need some time to deploy their preferences. They also have to
1099choose top managers and the latter have to choose middle managers.
1100Finally, middle managers have to choose lower positions in the orga-
1101nizations influenced by a female in the leader position (mayor); this
1102cascading influence requires time.
1103The interpretation of the pro- (or anti-) woman result is as follows:
1104An increase (or decrease) in the average number of minimum salaries
1105received by women as compared to men is favorable (or unfavorable) to
1106women. The interpretation of the gender ratio result is slightly more
1107complicated given that all earnings are fixed in terms of minimum
1108salaries: An increase (or decrease) in the ratio of female to male workers
1109Table 3A
1110Balance test for municipal characteristics.
1111Women vs. Men Elections
1112Regression discontinuity estimate
1113Percentage of total
1114population with access
1115to mains water supply
1116Population Percentage of women
1117as proportion of total
1118population
1119Gender ratio of
1120hours worked –
1121Formal Workers
1122Gender ratio of
1123age – Formal
1124Workers
1125Percentage of
1126municipal population
1127less than 1 year old
1128Percentage of
1129municipal
1130population 1 to 4
1131years old
1132Percentage of
1133municipal
1134population 49 to 59
1135years old
1136Percentage of
1137municipal
1138population 60 to 69
1139years old
1140Percentage of
1141municipal population
114270 to 79 years old
1143Running Variable:
1144Margin of Victory
1145−0.0750 −1,418 −0.000808 −0.0781 0.00213 −0.0385 0.00454 0.0754 −0.0121 0.0570
1146(0.0566) (5,620) (0.00326) (0.157) (0.0143) (0.0921) (0.348) (0.322) (0.252) (0.215)
1147Observations 531 551 530 555 555 555 555 555 555 555
1148Note: Standard errors in parentheses; *** p<0.01, ** p< 0.05, * p< 0.1; 1. Local-Polynomial Regression-Discontinuity (RD) point estimators with Robust Confidence Intervals proposed in Calonico et al. (2014); 2. CCT for bandwidth selector
1149proposed by Calonico et al. (2014). The last line of each variable is the number of observations.
1150P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
115110
1152earning low salaries is unfavorable (or favorable) to women (low po-
1153sitions). In contrast, an increase (or decrease) in the ratio of female to
1154male workers earning intermediate/high salaries is favorable (or un-
1155favorable) to women (middle and top managers).
1156First, we will describe the results with a non-robust change. For the
1157period immediately after a woman is elected (first for public firms), we
1158observed one significant difference in the sample of public organiza-
1159tions. There is a reduction in the ratio of female to male workers
1160earning salaries between 1.01 and 1.5 times the minimum salary, a
1161favorable result for women.
1162We observe three significant results in the second column of Table 5
1163(the next term of mayors for public organizations, regardless of whether
1164the mayor is standing for reelection, or not). All the results are con-
1165centrated in average salaries and upwards. There is a reduction in the
1166ratio of female to male workers in the following cases: earning between
116710.01 and 15 minimum salaries and between 15.01 and 20 minimum
1168salaries, an anti-woman results. However, there is an increase in the
1169ratio of female to male workers who receive earnings >20 minimum
1170salaries in public organizations, a pro-woman results. Given these
1171contradictory results, we infer there to be no clear result, either pro- or
1172anti-women.
1173In the second set of results (private organizations), in the fourth and
1174fifth columns of Table 5 we report the effect of local female leadership
1175in private organizations. There are two significant results. In the fourth
1176column, we observe an increase, on average, in minimum salaries fa-
1177vorable to women. In the fifth column, there is a reduction in the ratio
1178of female to male workers earning between 15.01 and 20 minimum
1179salaries.
1180Finally, our most impressive change occurs in public organizations
1181when the female leader is reelected.
1182Reelected woman
1183The third and sixth columns show results when a woman is re-
1184elected. We find no significant results in private organizations (sixth
1185column). The results in the third column indicate significant increases
1186in five of seven cases (what is observed, on average, is the result of the
1187composition effect – from different groups): (a) the ratio of female to
1188male workers earning salaries between 0.51 and 1 times the minimum
1189salary – low positions; (b) the ratio of female to male workers earning
1190salaries between 1.01 and 1.5 times the minimum salary – low posi-
1191tions; (c) the ratio of female to male workers earning salaries between
119210.1 and 15 times the minimum salary – middle managers; and (d) the
1193ratio of female to male workers earning salaries over 20 times the
1194minimum salary – top managers. However, we find a contradictory
1195result again; whereas conditions are non-favorable to women at the
1196bottom (low earnings), the results are very favorable at the top of the
1197organization (with both intermediate and high earnings).
1198To confirm these results, we need to observe the graphic visuali-
1199zation of the discontinuities, following the recommendations of Imbens
1200and Lemieux (2008) and Eggers et al. (2015). We do so for the reelected
1201mayoral result.
1202A. Graphic analysis of (low and high) earnings
1203Fig. 2 reports the impact when the earnings of workers are for low
1204positions (left side: the ratio of female to male workers earning between
12050.51 and 1 minimum salaries; and right side: the ratio of female to male
1206workers earning between 1.01 and 1.5 minimum salaries).
1207Fig. 3 shows this impact when the earnings of workers are for
1208middle managers (left side: the ratio of female to male workers earning
1209intermediate salaries of between 10.01 and 15 minimum salaries) and
1210Table 3B
1211Balance test for candidate characteristics.
1212Women vs. Men Elections
1213Regression discontinuity estimate
1214Worker´s Party
1215(PT)
1216Brazilian Social
1217Democratic Party (PSDB)
1218Front Liberal
1219Party (PFL)
1220Completed primary
1221education but not high
1222school
1223Completed high school but
1224not higher education
1225Completed higher
1226education
1227Running Variable:
1228Margin of Victory
1229−0.0126 0.151 −0.0968 −0.159 −0.0175 0.171
1230(0.0140) (0.0955) (0.0899) (0.102) (0.111) (0.108)
1231Observations 555 555 555 555 555 555
1232Note: Standard errors in parentheses; *** p< 0.01, ** p< 0.05, * p <0.1; 1. Local-Polynomial Regression-Discontinuity (RD) point estimators with Robust Confidence Intervals
1233proposed in Calonico et al. (2014); 2. CCT for bandwidth selector proposed by Calonico et al. (2014). The last line for each variable is the number of observations.
1234Table 3C
1235Balance test with same candidates of quasi-experiment, same gender.
1236Regression discontinuity estimate
1237Worker’s Party
1238(PT)
1239Front Liberal Party
1240(PFL)
1241Brazilian Social
1242Democratic Party (PSDB)
1243Completed primary
1244education but not high
1245school
1246Completed high school but
1247not higher education
1248Completed higher
1249education
1250Between Women (Female winner vs. Female loser)
1251Running Variable:
1252Margin of Victory
1253−0.00780 −0.000232 −0.0115 −0.0623 0.0293 −0.0246
1254(0.0522) (0.0564) (0.0660) (0.0595) (0.107) (0.0991)
1255Observations 668 668 668 668 668 668
1256Between Men (Male winner vs. Male loser)
1257Running Variable:
1258Margin of Victory
1259−0.0832* 0.00949 −0.00902 0.0843 −0.0181 −0.165*
1260(0.0449) (0.0714) (0.0483) (0.0607) (0.0904) (0.0961)
1261Observations 668 668 668 668 668 668
1262Note: Standard errors in parentheses; *** p< 0.01, ** p< 0.05, * p <0.1; 1. Local-Polynomial Regression-Discontinuity (RD) point estimators with Robust Confidence Intervals
1263proposed in Calonico et al. (2014); 2. CCT for bandwidth selector proposed by Calonico et al. (2014). The last line for each variable is the number of observations.
1264P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
126511
1266top managers (right side: the ratio of female to male workers earning
1267top salaries of over 20 minimum salaries).
1268Unlike studies using a regression discontinuity design that report
1269average points in different bandwidths, we opt to plot the result of each
1270municipality. Doing so allows us to see whether there is discontinuity
1271around the cut-off point or not, and to observe the inequality in earn-
1272ings in public and private organizations in detail. For instance, in Fig. 2
1273we can see that the number of women with low earnings is higher than
1274the number of low-earning men (above 1; we have a value equal to 1
1275when the number of female workers is equal to the number of male
1276workers in the observed earning's range). In Fig. 3, with both inter-
1277mediate and high earnings, we identify those municipalities in which
1278the number of women with both intermediate and high earnings (below
12791) is lower than the number of men in the same earnings bands. There is
1280also a large group of municipalities in which there are no women re-
1281ceiving intermediate/high earnings, regardless of women's victories.
1282The solid (control group) and dash (treatment group) lines plot the
1283predicted global values from a local polynomial regression (i.e., Kernel
1284Epanechnikov) with second degree separately on either side of the fe-
1285male win-lose cut-off point (an election with a small margin of victory
1286equals zero).
1287In Fig. 2, we observe no discontinuity for either measure. Therefore,
1288the observed anti-women situation in the regression discontinuity es-
1289timate for the low earnings groups – low positions – is not visually
1290confirmed. As we mentioned, it is important to confirm graphically the
1291result obtained statistically (Eggers et al., 2015; Imbens & Lemieux,
12922008).
1293In contrast, the observed difference in regression discontinuity es-
1294timates is visually confirmed for both groups: middle managers and top
1295managers around the cut-off point. When women are elected as the
1296mayor they make a difference to these groups when they remain in
1297office for more than one term; that is, when they are reelected. The
1298gender ratios for both measures in these municipalities are higher
1299around the cut-off point, indicating a jump. It is noteworthy that there
1300are more women earning more than men until, approximately, a 15%
1301margin of victory for women. After this point, the ratio remains stable
1302and falls less than when a woman loses because the “winning female”
1303event begins to be a “rare event”: Women are running against men for
1304mayor. The exception is an outlier with a margin of victory of 40% for
1305the intermediate group (see the figure on the right side). Indeed, when
1306the number of events is higher, that is, close to the cut-off point on the
1307right side, we observe a number of women with higher earnings.
1308As for the low positions' result not being confirmed, we observe only
1309favorable results for women who earn both middle and top manager
1310salaries. When women were elected by way of a quasi-experiment
1311(2000) and reelected for a second term (2005–2008) this result led to an
1312increase in the number of women in both top and middle management
1313positions in public organizations between 2005 and 2008. Following
1314our data strategy, this result appears restricted to public organizations
1315because we observe no significant effect in private organizations. If we
1316had observed any significant results in private organizations, we would
1317have had to perform the same procedure as previously (graphs, etc.).
1318Only then could we interpret the results for private organizations.
1319The results in the middle managers' group (between 10.01 and 15
1320minimum salaries) do not eliminate the possibility of a role model in-
1321fluence. If we had obtained results only for the top group, then the
1322isolated impact of command would be more plausible.
1323B. Additional Robustness Checks
1324We use two additional robustness checks to confirm the results for
1325middle/top managers: a new graph excluding municipal outliers, and
1326Table 4
1327Income situation of Female Workers in firms at the municipal level (lagged variables).
1328Outcome variables Public
1329Organizations
1330Private
1331Organizations
1332Regression discontinuity estimate
1333Lagged variables Lagged variables
1334Gender ratio Female Workers/Male Workers – average
1335income in terms of minimum salaries
1336−0.0756 −0.147
1337(0.08) (0.127)
1338281 392
1339Female Workers/ Workers earning up to 0.5
1340minimum salary
13411.14 2.917**
1342(1.058) (1.283)
1343136 221
1344Female Workers/Workers earning between
13450.51 and 1 minimum salary
13460.0435 −0.0391
1347(0.40) (0.787)
1348298 380
1349Female Workers/Male Workers earning
1350between 1.01 and 1.5 minimum salaries
1351−0.254 0.0563
1352(0.19) (0.589)
1353206 373
1354Female Workers/Male Workers earning
1355between 10.01 and 15 minimum salaries
13560.120 −0.134
1357(0.16) (0.418)
1358164 306
1359Female Workers/Male Workers earning
1360between 15.01 and 20 minimum salaries
13610.0526 −0.352
1362(0.237) (0.373)
1363178 273
1364Female Workers/Male Workers earning over
136520 minimum salaries
13660.084 2.284
1367(0.16) (3.033)
1368162 283
1369Note: Standard errors in parentheses; *** p <0.01, ** p< 0.05, * p <0.1; 1. Local-Polynomial Regression-Discontinuity (RD) point estimators with Robust Confidence Intervals
1370proposed in Calonico et al. (2014); 2. CCT for bandwidth selector proposed by Calonico et al. (2014). The last line for each variable is the number of observations considered.
1371P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
137212
13730 1 2 3 4
1374Female Workers / Male Workers earning between 0.5 and 1 minimum salary
1375-.4 -.2 0 .2 .4
1376Margin of Woman Victory
1377Municipal Defeated Woman Municipal Victorious Woman
1378Local Polynomial Regression Local Polynomial Regression
13790 1 2 3 4
1380Female Workers / Male Workers earning between 1.01 and 1.5 minimum salary
1381-.4 -.2 0 .2 .4
1382Margin of Woman Victory
1383Municipal Defeated Woman
1384Municipal Victorious Woman
1385Local Polynomial Regression Local Polynomial Regression
1386Fig. 2. Impact of elected women mayors on gender ratio in firms (Low Earning).
1387Mayor election (2000) – Municipal Gender ratio information of municipalities with reelected mayors (2005–2008).
1388Table 5
1389Income situation of Female Workers in firms at the municipal level.
1390Public Organizations Private Organizations
1391Regression discontinuity estimate Regression discontinuity estimate
1392Outcome Variables Term immediately after
1393election
1394Next, Term Reelected Term immediately after
1395election
1396Next, Term Reelected
1397Gender ratio Female Workers/Male Workers – average income in
1398terms of minimum salaries
13990.011 0.006 0.0523 0.145* −0.0529 −0.203
1400(0.0291) (0.0371) (0.0987) (0.0865) (0.0705) (0.277)
1401351 268 41 541 560 67
1402Female Workers/Male Workers earning up to 0.5
1403minimum salaries
1404−0.14 0.67 −0.573 0.584 −1.566 −2.084
1405(0.607) (0.65) (0.92) (1.391) (1.649) (2.293)
1406214 274 38 353 390 42
1407Female Workers/Male Workers earning between
14080.51 and 1 minimum salaries
14090.206 0.122 0.480* −0.101 0.756 0.454
1410(0.283) (0.230) (0.25) (0.464) (1.071) (0.431)
1411390 268 22 530 553 66
1412Female Workers/Male Workers earning between
14131.01 and 1.5 minimum salaries
1414−0.40** 0.0155 0.54** 0.132 0.217 1.300
1415(0.19) (0.138) (0.80) (0.222) (0.166) (0.804)
1416391 325 40 526 554 66
1417Female Workers/Male Workers earning between
141810.01 and 15 minimum salaries
1419−0.112 −0.338* 2.33** 0.278 −0.0171 5.137
1420(0.182) (0.195) (1.079) (0.631) (0.222) (4.651)
1421276 201 20 432 410 45
1422Female Workers/Male Workers earning between
142315.01 and 20 minimum salaries
14240.16 −0.383* 0.0764 0.0615 −0.637** 0.884
1425(0.16) (0.22) (0.40) (0.527) (0.300) (0.673)
1426256 204 33 374 362 37
1427Female Workers/Male Workers earning over 20
1428minimum salaries
1429−0.134 0.5** 1.11* −0.158 −0.133 −0.769
1430(0.112) (0.24) (0.69) (0.241) (0.367) (0.928)
1431201 214 28 379 351 37
1432Note: Standard errors in parentheses; *** p< 0.01, ** p< 0.05, * p <0.1; 1. Local-Polynomial Regression-Discontinuity (RD) point estimators with Robust Confidence Intervals
1433proposed in Calonico et al. (2014); 2. CCT for bandwidth selector proposed by Calonico et al. (2014). The last line for each variable is the number of observations considered.
1434P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
143513
1436the same regression discontinuity estimate with different bandwidths.
1437Because the discontinuity results could be attributed to municipal
1438outliers, that is, municipalities in which the number of female workers
1439compared with the number of male workers is extremely high for both
1440intermediate and top salaries, we re-create Fig. 3 (Fig. A.1 in the online
1441appendix) excluding those municipalities with gender ratios above 1.5.
1442Graphically, the results around the cut-off point are practically the same
1443when these outliers are excluded.
1444We re-perform the econometric exercise with different bandwidths
1445(Eggers et al., 2015Imbens & Lemieux, 2008). Calonico et al. (2014) use
1446an optimal bandwidth (estimate the best size of bandwidth regarding
1447the distribution of points in the sample): For the ratio of female to male
1448workers earning between 10.01 and 15 minimum salaries it is 0.142,
1449and for the ratio of female to male workers earning over 20 minimum
1450salaries it is 0.112. Using the same methodology (Calonico et al., 2014),
1451we fix different optimal bandwidths (larger and smaller around the
1452optimal bandwidth) and observe no difference in the main results
1453(Table A.1 in the online appendix). Therefore, our results are robust for
1454different bandwidths.
1455Discussion of the results
1456Stereotypically speaking, in many countries, including Brazil,
1457women are thought to have characteristics that do not fit the image of
1458an ideal leader. Top management positions require achievement-or-
1459iented and agent-like traits as well as emotional toughness; such de-
1460scriptions are usually thought to be the antithesis of what are com-
1461monly-perceived to be feminine qualities (Hoyt, 2005; Hoyt &
1462Blascovich, 2007). If leadership is achieved, women may react in one of
1463two ways: they may “undermine” the careers of other women (QBP), or
1464they may become “models”, with other female individuals tending to
1465shape their patterns of behavior on these women (RM).
1466Organizations' exposure to female leaders may block the rise of
1467other women; alternatively, it may improve women's aspirations and
1468encourage them to enter historically male-dominated environments.
1469Each alternative may lead to a different result in organizations: either
1470increase or reduce the gender difference in organizations (hypothesis
1471H1). Only a robust empirical exercise can help shed light on this issue.
1472The primary goal of this research was to examine whether female
1473leadership helps reduce gender difference in organizations. We pre-
1474dicted that women's exposure in public organizations when a leader has
1475“command” is effective for reducing gender difference. These predic-
1476tions were empirically tested, and when a woman was elected mayor in
1477a quasi-experiment, we observed an increase in women in the ratio of
1478top-managers because of the leader's choice. Middle managers depend
1479on the internal dynamic of organizations: they are chosen by top-
1480managers and top-managers are “influenced” by female leaders when
1481choosing these positions. Thus, our results suggest that the QBP does
1482not exist, or is less than the command and RM effect in public organi-
1483zations.
1484We observed the most robust effects favorable to women when the
1485same woman was reelected, that is, being in power for two consecutive
1486terms. Consequently, we emphasize that time is an important dimen-
1487sion in leadership studies (see Antonakis et al., 2012; Shamir, 2011).
1488Shamir (2011) draws attention to the neglect of time-related con-
1489siderations in leadership studies. According to Shamir, relationships
1490between followers and leaders occur over time and it takes time for the
1491majority of leaders' inputs to produce outcomes.
1492We do not obtain the same results in private organizations. These
1493results are somewhat similar to Bertrand's work. Bertrand et al. (2014)
1494investigate whether the Norwegian company reform legislation suc-
1495ceeded in its primary objective of reducing gender disparities in the
1496corporate sector. To address this disparity, in December 2003 Norway
1497passed a law requiring there to be a minimum 40% of each gender on
1498the boards of directors of public limited companies. According to
1499Bertrand et al. (2014), whereas the gender gap in wages in Norway
1500was< 14% on average among full time workers in 2000, only 5% of
1501board members were women, and their annual earnings were 20%
1502lower than those of male board members. Their results, however, show
1503that in the short run the reform has had very little discernible impact on
1504women in business, beyond its direct effect on newly-appointed female
1505board members. We do not claim that the result obtained is the result of
1506the QBP effect because we observe no changes in the position of top
1507managers in private organizations. Thus, there is no positive effect for
1508private firms, which makes sense, at least in the short to medium term,
1509given that majors do not have much command over private
15100 .5 1 1.5 2 2.5
1511Female Workers / Male Workers earning between 10.01 and 15 minimum salary
1512-.4 -.2 0 .2 .4
1513Margin of Woman Victory
1514Municipal Defeated Woman Municipal Victorious Woman
1515Local Polynomial Regression
1516Local Polynomial Regression
15170 .5 1 1.5 2 2.5
1518Female Workers / Male Workers earning over 20 minimum salary
1519-.4 -.2 0 .2 .4
1520Margin of Woman Victory
1521Municipal Defeated Woman
1522Municipal Victorious Woman
1523Local Polynomial Regression Local Polynomial Regression
1524Fig. 3. Impact of elected women mayors on gender ratio in firms (Intermediate and High Earning).
1525Mayor election (2000) – Municipal Gender ratio information of municipalities with reelected mayors (2005–2008).
1526P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
152714
1528organizations and RM over gender-related outcomes in private orga-
1529nizations.
1530By using the same method, we also examined the impact on public
1531policy (e.g., the Municipal Planning with Women policies, Municipal
1532Women Police Station, Municipal Center for Women Assistance, and
1533Municipal Public Daycare policies) to check for a possible influence by
1534municipal government on the main results. For example, public daycare
1535should help women in the labor market. We observe no difference in
1536public policies between female and male mayors. The results can be
1537requested from the authors.
1538Overall our results add to the research on female leadership by in-
1539corporating managerial discretion theory (Finkelstein & Hambrick,
15401990; Finkelstein & Peteraf, 2007; Hambrick & Finkelstein, 1987) and
1541the RM effect on people's aspirations and the self-perceptions of women
1542(Bandura, 1977).
1543Conclusions
1544The present research examined whether female leadership per se
1545helps reduce gender difference in organizations. Women belong to a
1546social group that has been historically subject to barriers when it comes
1547to occupying positions in society that have been traditionally male-
1548dominated. As women find it difficult to attain top leadership roles in
1549society, when they become leaders they prove to be often capable of
1550matching, or even outperforming men (Hoyt, 2005Hoyt & Blascovich,
15512007).
1552By focusing on reducing gender inequality, two possible situations
1553can occur when women reach the top that can be of benefit to other
1554women: (a) females who, like a queen bee, succeed in male-dominated
1555settings may also play a negative role in the advancement of their fe-
1556male subordinates (Derks et al., 2011; Derks et al., 2011; Faniko et al.,
15572016); (b) female leaders directly help by making pro-female choices
1558and work as a role model in some way to have a positive impact on
1559women's positions within organizations—doing so, in turn, may have an
1560impact on other women's aspirations or their self-perception, by way of
1561a process of social comparison (Bandura, 1977).
1562The environment also contributes to improving the status of women.
1563The power, discretion, and influence that is characteristic of a mayor
1564(with command over the public sector) provide a choice set to imple-
1565ment leader preferences. Women in other organizational settings,
1566therefore, need to have similar decision-making power and managerial
1567discretion in organizations in order to bring about favorable changes for
1568women themselves.
1569We tested the hypothesis that female leaders reduce gender differ-
1570ence in public organizations and our most robust result in the quasi-
1571experiment showed that when a woman was elected mayor, the number
1572of women top-managers increased compared to the number of men in
1573the same category. Because we also found a positive result at the in-
1574termediate level – middle managers - and this occurs in the internal
1575dynamic of organizations (influence), the results speak for the role
1576model explanation: the queen bee phenomenon observed in literature is
1577either non-existent or less impactful than the role model effect. The
1578results of changes in the number of women earning low salaries when
1579compared to men were not very strong (lower positions); that is, the
1580result was observed in the estimate, but was not confirmed in the
1581graphic analysis.
1582We also evaluated these results in private organizations. There were
1583no perceived improvements for women, suggesting that in that context,
1584the influence via role model effects did not occur. Of course, we did not
1585directly examine what happens when a female top-manager is exo-
1586genously appointed in a private firm, so we cannot speak for the dis-
1587cretion that top managers may have in private organizations and if they
1588would impose their pro-female preferences, as the female leaders in
1589public organizations did.
1590In terms of theoretical implications, this research contributes to our
1591understanding of the effect of leadership in organizations. More
1592specifically, these empirical results indicate a need for the female leader
1593to be in a favorable institutional environment having a lot of asym-
1594metrical power and discretion (i.e., she must have control over other
1595people, or valued recourses can impose their preferences), which is an
1596additional variable in understanding the impact of leaders on the out-
1597comes process (Finkelstein & Hambrick, 1990; Finkelstein & Peteraf,
15982007; Rucker et al., 2010; Sturm & Antonakis, 2015).
1599In empirical terms, this research overcame the omnipresent problem
1600of endogeneity, which is often ignored by management and applied
1601psychology researchers undertaking observation work (Antonakis et al.,
16022010; Fischer et al., 2017). For example, such works include Hoyt
1603(2005), Hoyt and Blascovich (2007), Derks et al. (2011), Derks et al.
1604(2011), Faniko et al. (2016) and Derks et al. (2015). We used compe-
1605titive gender elections with a small margin of victory between the first-
1606place (i.e., female) and second-place (i.e., male) candidates. Unlike the
1607majority of leadership experiments, which use laboratory evidence (see
1608Antonakis, Bastardoz, Jacquart, & Shamir, 2016; Eden, 2017), our in-
1609vestigation was carried out into public and private organizations in
1610different Brazilian municipalities, which produces a result that not only
1611is interlay valid, but one that has external validity.
1612The queen bee phenomenon might well exist in business, govern-
1613ment and politics as a result of gender inequality, but previous findings
1614cannot definitively make any claims that this phenomenon exists be-
1615cause it has generally not been properly causally identified in previous
1616research, or it has relied on idiosyncratic, selective samples, or un-
1617generalizable case studies. Thus, given the lack of rigor in previous
1618research, and on the basis of our findings, it appears that the queen bee
1619phenomenon may simple be a myth. Note that gender stereotypes are
1620becoming fuzzier with time and this across a wide array of settings
1621(Koenig, Eagly, Mitchell, & Ristikari, 2011). We are reasonably con-
1622fident, therefore, that our results are not unique to Brazil and will
1623generalize to other contexts given the ubiquity of male-female power
1624differentials in consequential settings and that the leader stereotype is
1625still mostly defined in male terms (Koenig et al., 2011). As long as
1626conditions are favorable for top-level female leaders, we are likely to
1627observe similar results in other settings. Future research should con-
1628sider ways of retesting for the existence of this phenomenon in private
1629firms using properly-specified causal designs.
1630As our results would suggest, the threat of a Queen Bee “sting” is
1631either non-existent or less than the effects of command using their
1632discretion to make pro-female choices or having a positive influence
1633from being a role model. Female leaders are not, as they are often
1634characterized in the popular press, “bitches” (Khazan, 2017) or “tyr-
1635annical” leaders (Drexler, 2013); such characterizations are very re-
1636grettable and contribute to the general derogation of women. Our re-
1637sults contradict these characterizations. In favorable institutional
1638environments that afford female leaders with the needed power and
1639managerial discretion, female leaders are “benevolent” and create op-
1640portunities and a pro-female condition for other women. For female
1641leaders in such environments our results speak for a top-level female
1642leader acting like a powerful, but stately and distinguished leader; more
1643like a “Regal Leader” than a Queen Bee.
1644Appendix A. Supplementary data
1645Supplementary data to this article can be found online at https://
1646doi.org/10.1016/j.leaqua.2018.03.002.
1647References
1648Antonakis, J., Bastardoz, N., Jacquart, P., & Shamir, B. (2016). Charisma: An ill-defined
1649and ill-measured gift. Annual Review of Organizational Psychology and Organizational
1650Behavior, 3, 293–319.
1651Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2010). On making causal claims: A
1652review and recommendations. The Leadership Quarterly, 21, 1086–1120.
1653Antonakis, J., Day, D. V., & Schyns, B. (2012). Leadership and individual differences: At
1654the cusp of a renaissance. The Leadership Quarterly, 23, 643–650.
1655P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
165615
1657Ashton, M. C. (2007). Individual differences and personality. Boston: Elsevier Academic
1658Press.
1659Atkeson, L. R. (2003). Not all cues are created equal: The conditional impact of female
1660candidates on political engagement. Journal of Politics, 65(4), 1040–1061.
1661Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change.
1662Psychological Review, 84(2), 191.
1663Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Inc:
1664Prentice-Hall.
1665Bandura, A. (1992). Social cognitive theory of social referencing. Social referencing and the
1666social construction of reality in infancy (pp. 175–208). Springer US.
1667Bandura, A. (1999). Social cognitive theory of personality. Handbook of personality: Theory
1668and research, 154–196.
1669Bandura, A. (2005). The evolution of social cognitive theory. Great Minds in Management,
16709–35.
1671Barbosa, K., & Ferreira, F. (2017). Democracy, government hiring, and quality of local public
1672goods. (INSPER Working Paper).
1673Beaman, L. A., Chattopadhyay, R., Duflo, E., Pande, R., & Topalova, P. (2008). Powerful
1674women: Does exposure reduce bias? (No. w14198). National Bureau of Economic:
1675Research.
1676Bertrand, M., Black, S. E., Jensen, S., & Lleras-Muney, A. (2014). Breaking the glass ceiling?
1677The effect of board quotas on female labor market outcomes in Norway (No. w20256).
1678National Bureau of Economic: Research.
1679Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014). Robust data-driven inference in the
1680regression-discontinuity design. Stata Journal, 14(4), 909–946.
1681Campbell, D. E., & Wolbrecht, C. (2006). See Jane run: Women politicians as role models
1682for adolescents. Journal of Politics, 68(2), 233–247.
1683Carroll, S. J. (1994). Women as candidates in American politics. Indiana University Press.
1684Census, I. B. G. E.. http://www.censo2010.ibge.gov.br Retrieved from.
1685Coppedge, M. (1997). A classification of Latin American political parties. Working paper
1686number 244. The Helen Kellogg Institute for International Studies.
1687Cox, G. W., & McCubbins, M. D. (1986). Electoral politics as a redistributive game. The
1688Journal of Politics, 48(2), 370–389.
1689Deloitte Touche Tohmatsu Limited (2017). Women in the boardroom: A global perspective
1690(5th edition). . Retrieved from www2.deloitte.com/.
1691Derks, B., Ellemers, N., Van Laar, C., & De Groot, K. (2011). Do sexist organizational
1692cultures create the queen bee? British Journal of Social Psychology, 50(3), 519–535.
1693Derks, B., Laar, C., Ellemers, N., & Raghoe, G. (2015). Extending the queen bee effect:
1694How Hindustani workers cope with disadvantage by distancing the self from the
1695group. Journal of Social Issues, 71(3), 476–496.
1696Derks, B., Van Laar, C., & Ellemers, N. (2016). The queen bee phenomenon: Why women
1697leaders distance themselves from junior women. The Leadership Quarterly, 27(3),
1698456–469.
1699Derks, B., Van Laar, C., Ellemers, N., & De Groot, K. (2011). Gender-bias primes elicit
1700queen-bee responses among senior policewomen. Psychological Science, 22(10),
17011243–1249.
1702Dixit, A., & Londregan, J. (1996). The determinants of success of special interests in re-
1703distributive politics. The Journal of Politics, 58(4), 1132–1155.
1704Drexler, P. (2013). The tyranny of the queen bee. The Wall Street Journal. Retrieved from
1705http://online.wsj.com/news/articles/
1706SB10001424127887323884304578328271526080496.
1707Eagly, A. H., & Karau, S. J. (2002). Role congruity theory of prejudice toward female
1708leaders. Psychological Review, 109(3), 573.
1709Eden, D. (2017). Field experiments in organizations. Annual Review of Organizational
1710Psychology and Organizational Behavior, 4, 91–122.
1711Eggers, A. C., Fowler, A., Hainmueller, J., Hall, A. B., & Snyder, J. M. (2015). On the
1712validity of the regression discontinuity design for estimating electoral effects: New
1713evidence from over 40,000 close races. American Journal of Political Science, 59(1),
1714259–274.
1715Ellemers, N., Heuvel, H., Gilder, D., Maass, A., & Bonvini, A. (2004). The under-
1716representation of women in science: Differential commitment or the queen bee syn-
1717drome? British Journal of Social Psychology, 43(3), 315–338.
1718Faniko, K., Ellemers, N., & Derks, B. (2016). Queen bees and alpha males: Are successful
1719women more competitive than successful men? European Journal of Social Psychology,
172046(7), 903–913.
1721Finkelstein, S., & Hambrick, D. C. (1990). Top-management-team tenure and organiza-
1722tional outcomes: The moderating role of managerial discretion. Administrative Science
1723Quarterly, 484–503.
1724Finkelstein, S., & Peteraf, M. A. (2007). Managerial activities: A missing link in man-
1725agerial discretion theory. Strategic Organization, 5(3), 237–248.
1726Fischer, T., Dietz, J., & Antonakis, J. (2017). Leadership process models: A review and
1727synthesis. Journal of Management, 43(6), 1726–1753.
1728Fleeson, W. (2001). Toward a structure- and process-integrated view of personality: Traits
1729as density distributions of states. Journal of Personality and Social Psychology, 80(6),
17301011–1027.
1731Fleeson, W., & Gallagher, P. (2009). The implications of Big Five standing for the dis-
1732tribution of trait manifestation in behavior: Fifteen experience-sampling studies and a
1733meta-analysis. Journal of Personality and Social Psychology, 97(6), 1097–1114.
1734Fox, R. L. (1997). Gender dynamics in congressional elections. Vol. 2. Sage.
1735Fujiwara, T. (2011). A regression discontinuity test of strategic voting and Duverger's law.
1736Quarterly Journal of Political Science, Vol. 6(3–4), 197–233.
1737Geddes, B. (1994). Politician's dilemma: Building state capacity in Latin America. University
1738of California Press.
1739Gordin, J. P. (2002). The political and partisan determinants of patronage in Latin
1740America 1960–1994: A comparative perspective. European Journal of Political
1741Research, 41(4), 513–549.
1742Graziano, L. (1976). A conceptual framework for the study of clientelistic behavior.
1743European Journal of Political Research, 4(2), 149–174.
1744Guala, F. (2012). Reciprocity: Weak or strong? What punishment experiments do (and do
1745not) demonstrate. Behavioral and Brain Sciences, 35(01), 1–15.
1746Hambrick, D. C., & Finkelstein, S. (1987). Managerial discretion: A bridge between polar
1747views of organizational outcomes. Research in Organizational Behavior.
1748Hansen, S. B. (1997). Talking about politics: Gender and contextual effects on political
1749proselytizing. The Journal of Politics, 59(01), 73–103.
1750Hoyt, C. L. (2005). The role of leadership efficacy and stereotype activation in women's
1751identification with leadership. Journal of Leadership & Organizational Studies, 11(4),
17522–14.
1753Hoyt, C. L., & Blascovich, J. (2007). Leadership efficacy and women leaders' responses to
1754stereotype activation. Group Processes & Intergroup Relations, 10(4), 595–616.
1755Hoyt, C. L., & Simon, S. (2011). Female leaders injurious or inspiring role models for
1756women? Psychology of Women Quarterly, 35(1), 143–157.
1757Imbens, G., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice.
1758Journal of Econometrics, 142(2), 615–635.
1759Kanter, R. M. (1977). Men and women of the corporation. Publishers of New York245–266.
1760Kanter, R. M. (1987). Men and women of the corporation revisited. Management Review,
176176(3), 14.
1762Khazan, O. (2017). Why do women bully each other at work? The atlantic magazine.
1763Retrieved from http://www.theatlantic.com/magazine/.
1764Kittilson, M. C. (2006). Challenging parties, changing parliaments: Women and elected office
1765in contemporary Western Europe. Columbus, OH: Ohio State University.
1766Koch, J. (1997). Candidate gender and women's psychological engagement in politics.
1767American Politics Research, 25(1), 118–133.
1768Koenig, A. M., Eagly, A. H., Mitchell, A. A., & Ristikari, T. (2011). Are leader stereotypes
1769masculine? A meta-analysis of three research paradigms. Psychological Bulletin,
1770137(4), 616.
1771Krook, M. L. (2006). Reforming representation: The diffusion of candidate gender quotas
1772worldwide. Politics and Gender, 2(3), 303–327.
1773Lee, D. S. (2001). The electoral advantage to incumbency and voters' valuation of politicians'
1774experience: A regression discontinuity analysis of elections to the US. (No. w8441).
1775National Bureau of Economic Research.
1776Lee, D. S., & Card, D. (2008). Regression discontinuity inference with specification error.
1777Journal of Econometrics, 142(2), 655–674.
1778Lee, D. S., & Lemieux, T. (2010). Regression discontinuity design in economics. Journal of
1779Economic Literature, 48, 281–355.
1780Lee, D. S., Moretti, E., & Butler, M. J. (2004). Do voters affect or elect policies? Evidence
1781from the US house. The Quarterly Journal of Economics, 807–859.
1782Lindbeck, A., & Weibull, J. W. (1993). A model of political equilibrium in a representative
1783democracy. Journal of Public Economics, 51(2), 195–209.
1784Mavin, S. (2006). Venus envy: Problematizing solidarity behaviour and queen bees.
1785Women in Management Review, 21(4), 264–276.
1786Mavin, S. (2008). Queen bees, wannabees and afraid to bees: No more ‘best enemies’ for
1787women in management? British Journal of Management, 19(s1).
1788McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity
1789design: A density test. Journal of Econometrics, 142(2), 698–714.
1790Miguel, L. F. (2003). Capital Político e Carreira Eleitoral: Algumas Variáveis na eleição.
1791Revista de Sociologia e Política, 20(20), 115–134.
1792Miller, A. R. (2017). Women and Leadership. In S. L. Averett, L. M. Argys, & S. D. Hoffman
1793(Eds.). Oxford handbook on women and the economyNew York: Oxford University
1794Press. Forthcoming. Available at SSRN https://ssrn.com/abstract=2961252.
1795Moreton, C. (2015). Margaret Thatcher: A new book explores the Iron Lady's religious
1796faith, and reveals how she modelled herself on Joan Crawford. The independent.
1797Retrieved from www.independent.co.uk/.
1798Nieva, V. F., & Gutek, B. A. (1981). Women and work: A psychological perspective. Praeger
1799Publishers.
1800O'Leary, V., & Ryan, M. M. (1994). Women bosses: Counting the changes or changes that
1801count. Women in management: A developing presence (pp. 63–78). London: Routledge.
1802Parker, S. K., Williams, H. M., & Turner, N. (2006). Modeling the antecedents of proactive
1803behavior at work. Journal of Applied Psychology, 91(3), 636–652.
1804Reality Check team (2018). Queen bees: Do women hinder the progress of other women? BBC
1805News. Retrieved from www.bbc.com/.
1806Rucker, D. D., Dubois, D., & Galinsky, A. D. (2010). Generous paupers and stingy princes:
1807Power drives consumer spending on self versus others. Journal of Consumer Research,
180837(6), 1015–1029.
1809Sapiro, V., & Johnston Conover, P. (1997). The variable gender basis of electoral politics:
1810Gender and context in the 1992 US election. British Journal of Political Science, 27(04),
1811497–523.
1812Schein, V. E., Mueller, R., Lituchy, T., & Liu, J. (1996). Think manager–think male: A
1813global phenomenon? Journal of Organizational Behavior, 33–41.
1814Schlozman, K. L., Burns, N., & Verba, S. (1994). Gender and the pathways to participa-
1815tion: The role of resources. The Journal of Politics, 56(04), 963–990.
1816Schneider, B. R. (1991). Politics within the state: Elite bureaucrats and industrial policy in
1817authoritarian Brazil. University of Pittsburgh Press.
1818Shamir, B. (2011). Leadership takes time: Some implications of (not) taking time seriously
1819in leadership research. The Leadership Quarterly, 22(2), 307–315.
1820Staines, G., Tavris, C., & Jayaratne, T. E. (1974). The queen bee syndrome. Psychology
1821Today, 7(8), 55–60.
1822Sturm, R. E., & Antonakis, J. (2015). Interpersonal power: A review, critique, and re-
1823search agenda. Journal of Management, 41(1), 136–163.
1824Wolbrecht, C., & Campbell, D. E. (2007). Leading by example: Female members of par-
1825liament as political role models. American Journal of Political Science, 51(4), 921–939.
1826P.R. Arvate et al. The Leadership Quarterly xxx (xxxx) xxx–xxx
182716