DDo competent women receive unfavorable treatment?
Yuki Takahashi ∗ Latest VersionDecember 9, 2020
Abstract
Do competent women receive unfavorable treatment than equally competent men? I study thisquestion in a laboratory experiment where unfavorable treatment has material consequences.I find that neither men nor women treat competent women less favorably; if anything, bothmen and women treat competent women slightly more favorably than equally competentmen. The findings provide a piece of evidence that competent women may not necessarilyreceive unfavorable treatment in settings with material consequences, which may shed newlight on hiring and promotion practices in labor markets.
JEL Classification:
C91, D91, J16, M51
Keywords: competence, gender bias, labor markets, laboratory experiment ∗ Department of Economics, University of Bologna. Email: [email protected] . I am grateful to MariaBigoni, Natalia Montinari, and Siri Isaksson whose feedback was essential for this project. I am also gratefulto participants of the experiment for their participation and cooperation. Ingvild Almås, Laura Anderlucci,Tiziano Arduini, Francesca Barigozzi, Teodora Boneva, Enrico Cantoni, Giovanna d’Adda, Chiara Natalie Focacci,Margherita Fort, Catalina Franco, Astrid Kunze, Fabio Landini, Pascal Langenbach, Annalisa Loviglio, ValeriaMaggian, Joshua Miller, Paola Profeta, Eugenio Proto, Tommaso Sonno, Sigrid Suetens, Alessandro Tavoni,Bertil Tungodden, ESA Experimental Methods Discussion group, and the University of Bologna’s PhD studentsall provided many helpful comments. This paper also benefited from participants’ comments at the AppliedYoung Economist Webinar, the BEEN Meeting, seminars at Ca’ Foscari University, the NHH, and the Universityof Bologna. Veronica Rattini and oTree help & discussion group kindly answered my questions about oTreeprogramming. Lorenzo Golinelli provided excellent technical and administrative assistance. The pre-analysis planis available at the OSF registry: https://osf.io/ypsmx. The experimental instructions are available in the onlineappendix: https://yukitakahashi1.github.io/files/CareerProgressionApp.pdf. a r X i v : . [ ec on . GN ] D ec Introduction
A literature argues that people consider competent women as less likable than equally competentmen (Heilman 2001; Rudman and Phelan 2008). This is also a view shared by several top femalecorporate executives. However, it is unclear whether being less likable has practical implications;that is, whether competent women receive unfavorable treatment in decisions such as hiringand promotion. Indeed, this question has been explored mostly by means of questionnaires andhypothetical decisions (Heilman et al. 2004; Phelan, Moss-Racusin, and Rudman 2008; Rudmanand Fairchild 2004; Rudman 1998; Rudman et al. 2012).Evidence from decisions with material consequences mainly comes from audit studies andis mixed: while Quadlin (2018) finds unfavorable treatment, Ceci and Williams (2015) andWilliams and Ceci (2015) do not. One possible reason for this mixed evidence is employers’wrong prior belief about competent women’s personality which tends to be negative as evidencedby the literature: because employers have to work with their employees for a long period of time,they want to hire people whom they are comfortable to work with. However, their prior must beupdated once the employers see the actual job applicants in the interview. Also, in promotiondecisions, employers or managers know a potential candidate very well and their prior beliefmust be irrelevant.In this paper, I tackle this question by means of a controlled laboratory experiment. Iuse dictator game allocation as a measure of favorable and unfavorable treatment with clearmaterial consequences and exogenously vary the recipient’s gender and competence. I measurecompetence by an IQ test, an attribute people care most about (Eil and Rao 2011; Zimmermann2020). In the experiment, participants first work on an incentivized IQ test. After the test,participants are randomly assigned to a group of six and receive a ranking of their IQ withintheir group. Once they answer the comprehension questions about their IQ rank, three of thesix members are randomly chosen to be dictators and play three rounds of dictator game withthe other three members chosen to be recipients, observing the recipients’ facial photos and firstnames – both of which convey information about gender – and the IQ ranks. Using dictator IQ fixed effects and exploiting random grouping of participants to addressthe endogeneity of participants’ IQ and recipients’ gender, I do not find a significant differencebetween dictators’ allocation to competent women and to competent men; if anything, dictatorsallocate slightly more to competent women. The point estimate of the difference is positiveand statistically indistinguishable from 0. The lower bound of the difference is -3.7% of thedictator endowment, which is quantitatively much smaller (2.4-3.1 times smaller) in absolute
1. In her book
Lean In: Women, Work, and the Will to Lead , the Facebook’s Chief Operating Officer SherylSandberg expresses her view as follows: “If a woman is competent, she does not seem nice enough. If a womanseems really nice, she is considered more nice than competent. Since people want to hire and promote those whoare both competent and nice, this crates a huge stumbling block for women” (Sandberg 2013).2. The experimental design, the hypotheses, and the empirical strategy are pre-registered at the OSF registry:https://osf.io/ypsmx. However, there are a number of changes to the pre-analysis plan discussed in appendix A.3. The use of photos follows recent literature and allows the dictators to infer the gender of the recipients in anatural way as they would do in their daily life (Babcock et al. 2017; Coffman 2014; Isaksson 2018), but I addressthe possibility that recipients’ gender-specific characteristics (e.g. women may smile more often in a photo) affectdictators’ allocation. Several alternative explanations are inconsistent with the results; most importantly, theresults are not due to experimental manipulation failure, ex-post randomization failure, wrongidentification assumptions, or lack of statistical power. These findings suggest that competentwomen do not receive unfavorable treatment in decisions involving material consequences suchas hiring and promotion.This paper mainly relates to two strands of literature. The first focuses on the tradeoffwomen face between being competent and being likable. The literature finds that people perceivefemale leaders (Heilman, Block, and Martell 1995; Heilman and Okimoto 2007; Rudman andKilianski 2000) and competent women (Heilman et al. 2004; Rudman 1998) negatively. Italso finds that people evaluate competent women negatively, but these results are obtained inset-ups without real consequences (Phelan, Moss-Racusin, and Rudman 2008; Rudman andFairchild 2004; Rudman et al. 2012). However, the studies about evaluations towards competentwomen with real consequences find mixed evidence: while Quadlin (2018) finds top-performingfemale college students less favorable treatment in hiring than equally qualified male students,Ceci and Williams (2015) and Williams and Ceci (2015) find qualified female applicants forassistant professor positions receive equal or more favorable treatment than equally qualifiedmale applicants. My results suggest that the employers’ prior belief about competent womenmay be driving these mixed findings.When the consequence of their evaluation is not immediately clear, people seem to evaluatewomen in traditionally male occupations more critically: Boring (2017) and Mengel, Sauermann,and Zölitz (2019) find that female university instructors receive lower student evaluation. Thereis also evidence that female economists’ work are undervalued (Koffi 2019; Sarsons et al. 2020)and female university faculty are less likely to get promotion (De Paola, Ponzo, and Scoppa 2018).However, these critical evaluations may simply reflect the lack of women in these occupationsand thus people do not have enough prior information about women’s competence, ratherthan taste-based discrimination. Sarsons (2019) finds that female surgeons receive a morenegative evaluation for their failure and Ditonto (2017) finds that voters care more about femalepoliticians’ competence than male politicians’ competence. Also, Bohren, Imas, and Rosenberg(2019) find that while women initially receive lower credits than men in their contributions to anonline mathematics discussion forum, they receive higher credits than men after they accumulateenough positive evaluations. My findings are compatible with the explanation that people donot have enough prior information about women’s competence, and they give fair evaluations towomen once they show they are competent.
4. While dictators only see the recipients’ IQ relative to theirs and thus the competence measure is relative totheirs, dictators do not see their IQ at the time they play dictator games. Indeed, in the real world, we do nothave an absolute measure of other people’s competence but evaluate relative to some benchmark. Nevertheless, Iprovide evidence that relative and absolute competence distinction does not matter for my results. Experiment
The experiment consists of two parts as shown in figure 1; instructions for each part are onlydelivered at the end of the previous part. Participants earn a participation fee of 2.5€ for theirparticipation. Experimental instructions are available in the online appendix.
Figure 1: Overview of the experiment
Notes:
This figure shows an overview of the experiment discussed in detail in section 2.1.
Pre-experiment: Random desk assignment & photo taking
After registration at the laboratory entrance, participants are randomly assigned to a desk.Before the start of part 1, participants take their facial photos at a photo booth and enter theirfirst name on their computer. After that, we experimenters go to each participant’s desk to checkthat their photo and first name match them to ensure all participants that other participants’photos and first names are real, following Isaksson (2018).
Part 1: IQ test
In part 1, participants work on an incentivized 9 IQ test questions for 9 minutes. I use Bilkeret al. (2012)’s form A 9-item Raven test which predicts one’s IQ measured with the full-lengthRaven test with more than 90% accuracy. Participants receive 0.5€ for each correct answer.They receive information about how many IQ test questions they have solved correctly only atthe end of the experiment. I use IQ as the measure of competence because previous studies findit is an attribute people care most about (Eil and Rao 2011; Zimmermann 2020).After the IQ test, participants make an incentivized guess on the number of IQ test questionsthey have solved correctly: they receive 0.5€ if their guess is correct. The answer to this questionmeasures their over-confidence level. They receive feedback on this guess only at the end of theexperiment.Following Eil and Rao (2011), six participants are randomly grouped, and they are informedof the ranking of their IQ relative to other group members. Ties are broken randomly. Theythen have to answer a set of comprehension questions as shown in figure 2 in order to proceedto the next part.
Part 2: Dictator game
In part 2, three participants in each group are randomly chosen to become dictators and the otherthree participants become recipients. Dictators are paired with the three recipients in their groupone by one in a random order, receive an endowment, and play a dictator game. When they playthe dictator game, dictators observe the recipients’ facial photo and first name and IQ rank. The3 igure 2: IQ rank assignment and the comprehension questions
Feedback
Among your 6 group members including you, you received
Rank 4 .Among your 6 group members, how many people performed better than you in the IQ test?Among your 6 group members, how many people performed worse than you in the IQ test?Next
General instructions
Please turn off your mobile phone.Please do not communicate with other participants.Please only use paper and pencil.Once you understand the instructions or enter your decisions, please click “Next” to proceed unless instructedotherwise.If you have any questions, please raise your hand at any time.
Debug info
Basic info
ID in group Group Round number Participant P6 Participant labelSession code vx84ysv2
Notes:
This figure shows an example of the IQ rank assignment and the comprehension questions. In thisexample, the participant was ranked 4th from the top within a group of 6 participants. Thus, the answer tothe first question is 3 (three participants performed better in the IQ test) and the second question is 2 (twoparticipants performed worse in the IQ test). use of photo allows me to convey information about gender of other participants in a naturalway as in the recent literature (Babcock et al. 2017; Coffman 2014; Isaksson 2018). Dictatorsare also told that their allocation decisions are anonymous except for the experimenters: theyare told that their allocation is paid to the recipients as a “top-up” to their earnings. Dictatorsdecide allocation by moving a cursor on a slider where the cursor is initially hidden to preventanchoring, as shown in figure 3. I use a cursor to make it more cognitively demanding to figureout fair allocation, which is shown to increase more self-interested decisions (Exley and Kessler2019). I also vary the endowment across rounds to make each dictator game less repetitive:7€ for 1st and 3rd rounds, 5€ for 2nd round. At the end of the experiment, one out of threeallocations is randomly chosen for each participant as earnings for this part. I also collect an indirect measure of dictators’ beliefs on how many IQ test questions thepaired recipients have solved correctly. To prevent the belief elicitation to affect or be affectedby the dictator game, I exploit the random assignment of participants to dictators and recipients(derived from the random desk assignment) and use recipients’ beliefs as a proxy for dictators’beliefs. Specifically, while dictators are playing the dictator game, recipients are paired withthe other two recipients in the same group one by one in random order and make incentivizedguesses on how many IQ test questions they have solved correctly, observing the other tworecipients’ facial photo, first name, and IQ rank. Each correct guess gives them 0.5€.
Post-experiment: Questionnaire
After the dictator game and guessing are over, participants are told their earnings from the IQtest, dictator game, and the guesses. Before receiving their earnings, participants answer a short
5. To address the non-anonymity of showing facial photo and first name, I ask participants how well they knowthe paired participants on a scale of 4 (did not know at all, saw before, knew but not very well, knew very well). Iask this question twice to make sure they do not answer randomly: right after the three dictator games or twoguesses and in the post-experimental questionnaire.6. For each dictator for each round, one of the three recipients in the same group is randomly chosen withreplacement and the dictator allocates the endowment between themselves and the recipient. Thus, it is possiblethat two dictators play dictator game with the same recipient in the same round. At the end of the dictatorgames, each participant has three allocations, and one of which is randomly chosen for payment. igure 3: Dictator’s allocation screen Round 1 of 3
Neve
Rank 5
You have received for this round.You have been paired with Neve .Please allocate the endowment between yourself and Neve. When you click the line below, a cursor appears. You can movethe cursor by dragging it. Please move the cursor to your preferred position to determine the allocation.
You Neve
Next
General instructions
Please turn off your mobile phone.Please do not communicate with other participants.Please only use paper and pencil.Once you understand the instructions or enter your decisions, please click “Next” to proceed unless instructedotherwise.If you have any questions, please raise your hand at any time.
Debug info
Basic info
ID in group Group Round number Participant P1 Participant labelSession code edtlog7n
Notes:
This figure shows an example of a dictator’s allocation screen. In this example, the dictator is playingthe first round and paired with a recipient whose first name is Neve with IQ rank 5. questionnaire about their demographics that are used for balance tests and robustness checks.
The experiment was computerized and programmed with oTree (Chen, Schonger, and Wickens2016), and conducted in English during November-December 2019 at the Bologna Laboratoryfor Experiments in Social Science (BLESS). I recruited 390 students of the University of Bolognavia ORSEE (Greiner 2015) who (i) were born in Italy, (ii) available to participate in Englishexperiments, and (iii) had not participated in gender-related experiments in the past (as far as Icould check). The number of participants was based on the power simulation in the pre-analysisplan to achieve 80% power. The average length of a session was 70 minutes including registration and payment. Theaverage payment per participant was about 10€ including the participation fee and 1.5€ ofgratuity for photo use in another experiment (which I asked for recipients only). I ran 24 sessionsin total and the number of participants in each session varied from 12 to 30 and was a multipleof 6.I limit participants to Italy-born students so that their first name and photo do not signal
7. I exclude the 1st session data because of the problem discussed in appendix A. Nevertheless, the resultsincluding the 1st session data give me the same conclusions and are available upon request. and whom the dictator declared they knew them“very well” at least once.These data screenings leave me 390 participants, 195 dictators, and 558 observations (withdictators’ allocation as the unit of observations). I estimate the following equation with OLS:
Allocate ij = β + β IQHigher ij + β F emale j + β IQHigher ij ∗ F emale j + IQF E i + X ij γ + (cid:15) ij (1)where each variable is defined as follows:• Allocate ij ∈ [0 , i ’s allocation to recipient j as a fraction of endowment.• IQHigher ij ∈ { , } : an indicator variable equals 1 if recipient j ’s IQ is higher thandictator i .• F emale j ∈ { , } : an indicator variable equals 1 if recipient j is female.• IQF E i := P l =2 θ l e li : fixed effects for the dictators’ IQ (number of IQ test questions theyhave solved correctly), where e li ∈ { , } is an indicator variable equals 1 if dictator i ’s IQis l=1,...,9, 0 otherwise.• X ij : a set of additional covariates to increase statistical power and to address potentialimbalance. • (cid:15) ij : omitted factors that are correlated with dictator i ’s allocation to recipient j conditionalon covariates.Dictator’s IQ fixed effects is included following Zimmermann (2020) so that the coefficients inequation 1 capture allocation differences due to the recipients’ IQ, not the dictators’. I clusterstandard error at dictator level (Liang and Zeger 1986) and apply Pustejovsky and Tipton(2018)’s small cluster bias adjustment to address potential inflation of the type I error rate dueto moderate cluster size.Table 1 shows what the coefficients in equation 1 identify. β identifies the allocation differenceto male recipients with higher and lower IQ which captures dictators’ distributional preference,among other effects. β identifies the allocation difference to female and male recipients withlower IQ, namely every difference due to the recipients being female (e.g. women smile moreand dictators like to give more to smiling people due to closer social distance). β identifies theinteraction of these two effects. Therefore, the allocation difference between female and male
8. Although it is easy to distinguish Italian and non-Italian sounding names, to make sure not to misclassify, Iasked the laboratory manager who was native Italian to check participants’ first names after each session.9. The covariates include dictator characteristics (age, gender dummy, region of origin dummy, social sciencemajor dummy, STEM major dummy, post-bachelor dummy, over-confidence level), recipient characteristics (age,region of origin dummy), round fixed effects, and fixed effects for proximity between the dictator and the recipient.The full description of the covariates is in appendix B.10. This is because people with different IQ (cognitive ability) may have a different distributional preference.For example, Almås et al. (2017) find that people from a low socio-economics family – which can be correlatedwith their cognitive ability – hold stronger egalitarian views than people from a middle or a high socio-economicfamily. Fisman et al. (2015) find that students in a top US law school – who presumably are smarter than averageUS citizens – are more meritocratic and more efficiency-oriented than average US citizens. β + β , while the main effect of interest, the sameallocation difference after controlling for the recipients’ any gender-specific effects, is identifiedby β . Note that only the relative IQ matters because dictators only observe the recipients’ IQrelative to themselves (and I control for dictators’ IQ). Later, I will elaborate on this point more. Table 1: Dictator’s allocation identified by equation 1
Recipient’s gender
Female Male
Recipient’sIQ
Higher β + β + β + β β + β Lower β + β β Notes:
This table shows what the coefficients in equation 1 identify. Each cell represents dictator’s allocationto recipients with higher (first row) or lower (second row) IQ and whose gender is female (first column) ormale (second column).
Summary statistics
Table 2 summarizes the data after excluding participants and observa-tions discussed in subsection 2.2. Looking at panels A and B, participants’ average IQ level(number of IQ test questions solved correctly) is about 7 (with a maximum 9) and gender isroughly balanced. Also, dictators took nearly 2 minutes to solve the feedback questions on theirIQ rank. Looking at panel C, most dictators did not know the paired recipients (after excludingpairs in which dictator knew the recipient “very well”). Looking at panel D, an average dictatorallocated to paired recipients 40% of their endowment and variation in allocation within each IQis as large as overall variation in allocation. The latter indicates that there is enough variationin allocation I can exploit in my empirical specification (which uses dictator’s IQ fixed effects).Figure C1 shows empirical density (panel A) and empirical distribution (panel B) of dictators’allocation to further elaborate panel D of table 2. First, panel A shows that nearly 45% ofdictators have chosen equal allocation. Second, the empirical distribution of giving in panelB resembles the empirical distribution of allocation in Bohnet and Frey (1999)’s one-wayidentification treatment which also shows recipients’ face to the dictators.
Balance tests
For coefficients in equation 1 to have causal interpretation, the dictator’s IQrank must be exogenous conditional on the dictator’s IQ fixed effects. Table C1 presents evidencefor this claim. Also, I have to make sure that randomization was successful ex-post so thatdictators face recipients of different gender and IQ in a balanced way conditional on the dictator’sIQ fixed effects. Tables C2 and C3 present evidence supporting this claim.
11. Demeaned standard deviation is sample standard deviation of (cid:94)
Allocate ik = Allocate ik − Allocate k , where Allocate ik is allocation by dictator i whose IQ is k and Allocate k = P i ∈ k Allocate ik is average allocation bydictators with IQ k . able 2: Summary statistics: Dictator data Mean SD
Panel A: Dictators
IQ level 6.69 1.23IQ rank 3.58 1.67Age 23.47 2.72Female 0.53 0.50From Emilia-Romagna 0.18 0.39Humanities 0.46 0.50Social sciences 0.19 0.40STEM 0.35 0.48Post bachelor 0.46 0.50Overconfidence 0.43 0.76Time on feedback (sec.) 107.60 95.60Observations 195
Panel B: Paired recipients
IQ level 6.84 1.16IQ rank 3.42 1.74IQ higher 0.53 0.50Age 23.35 2.77Female 0.47 0.50From Emilia-Romagna 0.20 0.40Observations 558
Panel C: Proximity
Did not know at all 0.96 0.19Knew but not well 0.03 0.17Saw before 0.01 0.09Observations 558
Panel D: Dictator’s allocation (fraction of endowment)
Allocation 0.40 0.24Allocation (demeaned) 0.24Observations 558
Notes:
This table shows summary statistics for the full sample: the dictators’ and the paired recipients’characteristics, how well dictators knew the paired recipients, and dictators’ allocation. Recipients whosename is non-Italian sounding and whom the dictator declared they knew them “very well” at least oneare not included. Standard deviation of demeaned allocation is calculated as sample standard deviationof (cid:94)
Allocate ik = Allocate ik − Allocate k , where Allocate ik is allocation by dictator i whose IQ is k and Allocate k = P i ∈ k Allocate ik is average allocation of dictators with IQ k . Manipulation check
Figure 4 provides evidence that dictators respond differently to therecipients’ gender and IQ information: it shows dictators’ average allocation for each category ofrecipients – female recipients with higher IQ, female recipients with lower IQ, male recipientswith higher IQ, and male recipients with lower IQ – along with their 95% confidence intervals.Looking at panel A, we see that dictators allocate most to female recipients with higher IQ,more to male recipients with higher IQ, and slightly more to female recipients with lower IQ– compared to male recipients with lower IQ. In addition, the allocations to female recipientswith higher IQ and male recipients with lower IQ are statistically different at 5% level and the8llocations to female recipients with higher and lower IQ are marginally statistically different at10%.
Figure 4: Dictators’ allocation by the recipients’ category * ** F r a c t i on o f endo w m en t Panel A: All dictators (N=558) ** *** F e m a l e − I Q h i ghe r F e m a l e − I Q l o w e r M a l e − I Q h i ghe r M a l e − I Q l o w e r F r a c t i on o f endo w m en t Panel B: Male dictators (N=260) F e m a l e − I Q h i ghe r F e m a l e − I Q l o w e r M a l e − I Q h i ghe r M a l e − I Q l o w e r F r a c t i on o f endo w m en t Panel C: Female dictators (N=298)
Notes:
This figure shows dictators’ allocation as a fraction of endowment by recipients’ category along withtheir 95% confidence intervals for all dictators (panel A), male dictators (panel B), and female dictators(panel C). Confidence intervals are calculated with the standard errors clustered at the dictator level withPustejovsky and Tipton (2018)’s small cluster bias adjustment. Horizontal lines over categories indicatestatistically significant differences. Unit of observation: dictator’s allocation. Significance levels: * 10%, **5%, and *** 1%.
Panel B, which shows male dictators’ average allocation for each category of recipients,presents the same pattern as panel A but the differences are larger. In addition, some differencesare more statistically significant despite the smaller sample size: the allocations to femalerecipients with higher IQ and male recipients with lower IQ are statistically different at 5% sodo the allocations to female recipients with higher and lower IQ. Also, the allocations to malerecipients with higher and lower IQ are marginally statistically different at 10%.On the other hand, female dictators’ average allocation for each category of recipientspresented in panel C shows a rather stark difference between female and male dictators. Whilemale dictators discriminate more based on ability and gender, female dictators do not. Indeed, all9he allocation differences are statistically insignificant even at 10%. This observation is consistentwith the existing literature that women are more inequality averse (Croson and Gneezy 2009)but inconsistent with Cappelen, Falch, and Tungodden (2019) who find that women dislikemale losers more than men. In addition, female dictators’ allocation is overall higher than maledictators, consistent with existing dictator game experiments (Engel 2011). The differences inobservable characteristics between female and male dictators reported in panel A of table C4are also consistent with the existing literature. Table 3 presents the results with all dictators. Column 1 presents estimate without controllingfor dictator’s IQ and shows the direction of the bias without including dictator IQ fixed effects:although statistically insignificant, dictators with lower IQ allocate more to recipients withhigher IQ regardless of the recipients’ gender as shown by the coefficient estimate on
IQHigher ij ,biasing the estimate upwards. From columns 2 to 5, I gradually increase the number of covariatesto check the robustness of my main specification in column 5. They show that the coefficientestimates are stable across 4 columns.Looking at column 5, the coefficient estimate on IQHigher ij ∗ F emale j is positive andstatistically insignificant. To give a statistical claim about the insignificance, I use dualitybetween hypothesis testing and confidence interval (Casella and Berger 2001) and examinewhat effect size we can reject and whether it is quantitatively important as typically done inepistemology (e.g. Chaisemartin and Chaisemartin 2020). Thus turning to the 95% confidenceinterval reported below the standard error estimate, the negative end is about -0.037, suggestingthat we can reject the effect size lower than -3.7% of the dictator endowment at 5% significancelevel. This value is very small, about 2.4-3.1 times smaller than the effect size of typicaldictator game experiments that examine the role of social distance with university students (e.g.,Brañas-Garza et al. 2010; Charness and Gneezy 2008; Leider et al. 2010). While OLS only picks up the average effect, these observations hold also in distribution.Panel A of figure 5 presents empirical CDFs of dictators’ allocation for each recipient category,demeaned by the dictator’s IQ fixed effects to give a causal interpretation. The figure showsthat the CDF of dictators’ allocation to female recipients with higher IQ (solid blue line) almost
12. Table C4 presents the same summary statistics as table 2 but separately for female and male dictators andtheir differences. It shows that female dictators are more likely to major in humanities, less likely to major insocial sciences and STEM, less overconfident, and tend to allocate more to recipients – characteristics consistentwith the literature on gender differences.13. Charness and Gneezy (2008) examine how informing the recipient’s family name increases the dictators’giving using a university student sample, and find an 8.9% increase in giving as a fraction of endowment. Leideret al. (2010) find using a university student sample that dictators increase giving by 11.42% as a fraction ofendowment for their friends relative to someone living in the same student dormitory. Brañas-Garza et al. (2010)also find using a university student sample that dictators give about 10% more of their endowment to friendsrelative to other students in the same class.14. Dictators’ allocation is demeaned for dictators’ IQ level so that the CDFs correspond to the regressionresults: (cid:94)
Allocate ik = Allocate ik − Allocate k + Allocate , where
Allocate ik is allocation by dictator i whose IQis k , Allocate k = P i ∈ k Allocate ik is average allocation of dictators with IQ k , and Allocate = P i Allocate ik isaverage allocation by all dictators. Allocate is added to re-center the allocation. Although this re-centering leavesa few observations outside the 0-1 range, they do not alter the results and thus are trimmed for ease of visualinspection. able 3: The role of the recipients’ gender and IQ in dictators’ allocation: Alldictators Outcome: Dictator’s allocation (fraction of endowment)(1) (2) (3) (4) (5)IQHigher 0.031 0.011 0.013 0.005 0.006(0.031) (0.033) (0.033) (0.033) (0.034)[-0.030, 0.093] [-0.054, 0.075] [-0.053, 0.078] [-0.059, 0.070] [-0.061, 0.072]Female 0.018 0.014 0.014 0.007 0.006(0.027) (0.027) (0.027) (0.026) (0.026)[-0.037, 0.072] [-0.040, 0.067] [-0.040, 0.068] [-0.044, 0.058] [-0.045, 0.057]IQHigherxFemale 0.024 0.027 0.026 0.034 0.035(0.037) (0.037) (0.037) (0.036) (0.037)[-0.048, 0.097] [-0.045, 0.100] [-0.048, 0.099] [-0.037, 0.105] [-0.037, 0.107]Dictator IQ FE - (cid:51) (cid:51) (cid:51) (cid:51)
Round FE - - (cid:51) (cid:51) (cid:51)
Proximity FE - - (cid:51) (cid:51) (cid:51)
Dictator controls - - - (cid:51) (cid:51)
Recipient controls - - - - (cid:51)
Female+IQHigherxFemale 0.042 0.041 0.04 0.041 0.041(0.026) (0.026) (0.026) (0.026) (0.026)[-0.009, 0.093] [-0.01, 0.092] [-0.012, 0.091] [-0.01, 0.092] [-0.011, 0.093]Outcome Mean 0.403 0.403 0.403 0.403 0.403Outcome SD 0.239 0.239 0.239 0.239 0.239R-squared 0.011 0.025 0.028 0.079 0.086Observations 558 558 558 558 558Clusters 195 195 195 195 195
Notes:
This table shows OLS estimates of the role of the recipients’ gender and IQ in dictators’ allocation.The outcome variable is dictators’ allocation as a fraction of endowment. The main specification is column5 which includes all covariates (see the main text for detail). Columns 2-4 provide robustness of the mainspecification by excluding some covariates and column 1 shows bias of not including dictator IQ fixed effects.Joint statistical significance of coefficient estimate on Female+IQHigherxFemale is calculated using t-test.The standard error (in parenthesis) and the 95% confidence interval (in bracket) are reported below eachcoefficient estimate. The standard errors are clustered at the dictator level with Pustejovsky and Tipton(2018)’s small cluster bias adjustment. R-squared is net of the dictator IQ fixed effects. Unit of observation:dictator’s allocation. Significance levels: * 10%, ** 5%, and *** 1%. always lies to the right of the other CDFs (although all CDFs are statistically indistinguishablefrom each other at 5% significance level), suggesting people do not treat competent womenunfavorably than competent men.The results also hold separately for male and female dictators. Column 1 of table 4 presentsresults with male dictators only and column 2 results with female dictators only, both will fullcontrol. First, the coefficient estimate on
IQHigher ij ∗ F emale j is positive and statisticallyinsignificant both for male and female dictators. Second, while the 95% confidence interval iswider due to the reduction of sample size by about half, we can still reject at 5% significance levelthe effect size lower than -9.0% for male dictators and -3.5% for female dictators. -9.0% is stillthe magnitude of the effect size of typical dictator game experiments. As with the full sampleestimate, these observations also hold in distribution as reported in panel B (male dictators) andin panel C (female dictators) of figure 5. For both male dictators and female dictators, the CDFof dictators’ allocation to female recipients with higher IQ (solid blue line) almost always lieson the right of the other CDFs (although all CDFs are statistically indistinguishable from eachother at 5% significance level), suggesting that neither men nor women do not treat competent11 igure 5: CDFs of dictators’ allocation by the recipients’ category (demeaned) C u m u l a t i v e p r obab ili t y Panel A: All dictators (N=558, Kruskal−Wallis simulated p−value=0.372) C u m u l a t i v e p r obab ili t y Panel B: Male dictators (N=260, Kruskal−Wallis simulated p−value=0.120) C u m u l a t i v e p r obab ili t y Panel C: Female dictators (N=298, Kruskal−Wallis simulated p−value=0.221)
Recipient Female−IQ higher Female−IQ lower Male−IQ higher Male−IQ lower
Notes:
These figures show the empirical distribution of demeaned dictators’ allocation by recipients’ categoryfor all dictators (panel A), male dictators (panel B), and female dictators (panel C). Demeaning was donewith respect to the dictators’ IQ so that the CDFs have causal interpretation: (cid:94)
Allocate ik = Allocate ik − Allocate k + Allocate , where
Allocate ik is allocation by dictator i whose IQ is k , Allocate k = P i ∈ k Allocate ik is average allocation of dictators with IQ k , and Allocate = P i Allocate ik is average allocation by all dictators. Allocate is added to re-center the allocation. Values below 0 and above 1 are trimmed for ease of visualinspection, but including those observations does not alter my results (there are only a few observationsoutside 0-1 range). Kruskal-Wallis simulated p-values are calculated using randomization inference (Young2019) to address arbitrary dependency among observations with 2000 draws under the null hypothesis of nolocation difference (i.e. all CDFs coincide). Unit of observation: dictator’s allocation.
Table 4: The role of the recipients’ gender and IQ in dictators’ allocation:Robustness checks
Outcome: Dictator’s allocation (fraction of endowment) Belief on IQSample: Male Female Over-confident Non-over-confident Evaluator(1) (2) (3) (4) (5)IQHigher 0.048 -0.049 0.032 -0.032 0.232(0.055) (0.042) (0.048) (0.049) (0.303)[-0.062, 0.158] [-0.132, 0.034] [-0.065, 0.128] [-0.130, 0.067] [-0.371, 0.834]Female 0.014 -0.014 -0.007 0.013 -0.352(0.034) (0.037) (0.033) (0.042) (0.292)[-0.054, 0.082] [-0.089, 0.061] [-0.073, 0.060] [-0.072, 0.098] [-0.931, 0.226]IQHigherxFemale 0.031 0.057 0.038 0.046 0.512(0.061) (0.046) (0.050) (0.054) (0.392)[-0.090, 0.152] [-0.035, 0.148] [-0.060, 0.136] [-0.063, 0.154] [-0.261, 1.286]Female+IQHigherxFemale 0.045 0.042 0.031 0.059 0.16(0.047) (0.029) (0.035) (0.038) (0.257)[-0.048, 0.138] [-0.015, 0.1] [-0.037, 0.1] [-0.016, 0.133] [-0.346, 0.666]Outcome Mean 0.369 0.432 0.385 0.427 6.342Outcome SD 0.253 0.223 0.241 0.235 1.89R-squared 0.151 0.084 0.112 0.151 0.097Observations 260 298 325 233 368Clusters 91 104 115 80 193
Notes:
This table shows OLS estimates of the role of the recipients’ gender and IQ in dictators’ allocation formale and female dictators (columns 1-2), overconfident and non-overconfident dictators (columns 3-4), andevaluators’ belief on the recipients’ IQ (column 5). The outcome variable is dictators’ allocation as a fractionof endowment in columns 1-2 and evaluators’ belief on the recipients’ IQ level in column 5. All specificationsinclude dictator IQ fixed effects, round fixed effects, proximity fixed effects, dictator (or evaluator) controls,and recipient controls, except columns 1 and 2 where dictator’s gender dummy is excluded and columns3-4 where dictator’s overconfidence measure is excluded. Joint statistical significance of coefficient estimateon Female+IQHigherxFemale is calculated using t-test. The standard error (in parenthesis) and the 95%confidence interval (in bracket) are reported below each coefficient estimate. The standard errors are clusteredat the dictator or the evaluator level with Pustejovsky and Tipton (2018)’s small cluster bias adjustment.R-squared is net of the dictator IQ fixed effects. Unit of observation: dictator’s allocation (columns 1-2) andevaluator’s belief (column 3). Significance levels: * 10%, ** 5%, and *** 1%.
While dictators only see the recipients’ IQ relative to theirs and thus the competencemeasure is relative to theirs, dictators do not see their IQ at the time they play dictator games.Indeed, in the real world, we do not have an absolute measure of other people’s competence butevaluate relative to some benchmark. Yet, if anything, overconfident people is likely to considerpeople whose competence is higher than themselves as more competent in absolute terms thannon-overconfident people do, after controlling for their actual competence.In columns 3-4 of table 4, I present the results separately for overconfident dictators (dictatorswho guess their IQ higher than their actual IQ, column 3) and non-overconfident dictators(dictators who guess their IQ equal to or lower than their actual IQ, column 4). For bothtypes of dictators, the coefficient estimate on
IQHigher ij ∗ F emale j is positive and statisticallyinsignificant and the lower bound of the estimate at 5% significance level is almost identical(-6.0% for over-confident dictators and -6.3% for non-overconfident dictators). Thus, relative orabsolute does not matter for my main results. 13 .3 Alternative explanations Female dictators’ in-group preference
One competing explanation is female dictators’favoritism towards people who belong to the same social group, or in-group preference (Tajfel andTurner 1979), which biases my β estimates upward. However, this explanation is inconsistentwith the data. First, I use the difference in allocation between lower IQ female and male recipientsas a control group, which eliminates the recipients’ gender-specific allocation preference foranalysis with female dictators. Second, the results with male dictators who do not have anin-group preference towards female recipients still reject the effect size lower than that of atypical dictator game experiment studying the effect of social distance using a university studentsample. Distaste against lower IQ male recipients
Although I use the difference in allocationto lower IQ female and male recipients to control for any recipient gender-specific allocationpreference, this may not be a clean control because people may have a negative bias againstunder-performing men (Cappelen, Falch, and Tungodden 2019; Moss-Racusin, Phelan, andRudman 2010). This explanation is also inconsistent with the data. First, the allocationdifference between higher IQ female and male recipients without using lower IQ female-maleallocation differences, estimate of β + β , still suggests the same conclusion: we can reject atthe 5% significance level the effect size lower than -1.1% of dictator endowment for all dictators(table 3, column 5), lower than -4.8% for male dictators (table 4, column 1), lower than -1.5%for female dictators (table 4). Second, while these single-difference estimates do not control forthe recipients’ gender-specific allocation preference, Cappelen, Falch, and Tungodden (2019) findthat the distaste mostly comes from women and the results with male dictators only should notbe affected by this distaste. A wrong belief that female recipients are less competent
My empirical specificationcompares female and male recipients with higher IQ. The identification fails if dictators considerfemale recipients as less competent than male recipients even if they have a higher IQ thandictators. Although this is unlikely, Fiske et al. (2002) find that people consider women as lesscompetent than men where the competence measures include intelligence.This explanation indeed does not apply to my sample. Column 5 of table 4 presents resultsfrom a regression where I replace dictators’ allocation with recipients’ belief (whom I callevaluator) about the other recipients’ IQ level which proxies dictators’ belief. Recipients’ beliefis a valid proxy for dictators’ belief by the random assignment of participants to dictators andrecipients and that both dictators and recipients face the same environment until the start ofthe dictator game. The estimate of β is positive albeit statistically insignificant, suggestingdictators do not believe that higher IQ female recipients are less competent than higher IQ malerecipients.This belief analysis, however, points to a potentially interesting difference in people’s beliefupdating process about women’s and men’s competence: people may update women’s competence
15. Table C5 presents evidence that recipients and dictators do not differ in their observable characteristics andcharacteristics of paired recipients. β . Experimental manipulation failure
The effect size becomes null if dictators do not respondto the recipients’ gender and IQ information. However, dictators in my sample do respond tothe recipients’ gender and IQ information in statistically significant ways as we already see infigure 4. Ex-post randomization failure
My empirical specification cannot detect causal effects ifeither (i) dictators’ IQ rank is endogenous even conditional on the dictators’ IQ fixed effects or(ii) dictators of specific characteristics face recipients with a specific gender or/and with loweror higher IQ. However, both concerns are addressed by a random desk assignment. Also, wesee that even ex-post, the random assignment was successful in tables C1, C2, and C3. Whilethe recipients’ region of origin is unbalanced (table C3, column 10) – which can happen by thedefinition of type I error – I include recipients’ region of origin dummy in my main specificationwhich controls the imbalance nonparametrically. Table C6 presents results for various subsamplesand we can still reject the effect size lower than -4.3% to -8.7% at 5% significance level, whichfurther addresses concerns for ex-post imbalance. Last, while I pool all the higher and lowerIQ recipients despite that dictators can also see the IQ rank differences, figure C2 shows thattaking into account the IQ rank differences does not alter the results. Wrong identification assumptions
Any causal inference relies on several assumptions, sofailure to reject the null hypothesis of no effect can be no effect, but can also be that someidentification assumptions are wrong. However, aside from those discussed thus far, I do notmake any significant assumptions because my empirical specification is a simple double difference-in-means. I also apply Pustejovsky and Tipton (2018)’s small cluster bias adjustment to addressthe finite-sample bias of the standard error. Thus, it is unlikely that the failure to reject thenull can be attributed to some implausible identification assumptions. Note that I also show Ican reject a very small effect size using confidence intervals.
Lack of statistical power
When the power is low (type II error rate is high), the confidenceinterval becomes wider. However, my confidence interval can reject a very small effect size at a5% significance level. Also, while there is an ex-post minimum detectable effect estimate, it issimply 2.8 times the standard error and mostly useful for cross-study comparison (McKenzie andOzier 2019); the information used in the confidence interval is strictly larger than the informationused in the ex-post minimum detectable effect.
16. In table C6, column 1 excludes dictators with IQ rank 1 and 6 who never face recipients with lower / higherIQ. Column 2 excludes dictator-recipient pairs in which the dictator knows the recipients even a little and column3 pairs in which the dictator saw the recipients before.17. Figure C2 shows OLS estimates of equation 1 but splitting
IQHigher ij into 6 separate dummies indicatingthe recipients’ IQ rank differences relative to the dictators’. The lower/higher the recipient’s IQ, the morenegative/positive their IQ rank difference. For brevity, the figure only plots the coefficient estimates on theinteraction terms between the 6 separate IQHigher ij and F emale j , ˆ β along with their 95% confidence intervals. Conclusion
This paper examines whether competent women receive unfavorable treatment compared tocompetent men. Using dictator game giving as a measure of favorable and unfavorable treatmentand exogenously varying gender and competence measured by an IQ test, I show that people treatcompetent women no less favorably than competent men; if anything, people treat competentwomen slightly more favorably. The lower bound of my estimate is -3.7% of dictator endowment,which is much smaller than the effect size of dictator game experiments studying the role ofsocial distance. I also show that experimental manipulation is successful, randomization wassuccessful even ex-post, identification assumptions are plausible, and the experiment has sufficientstatistical power.This paper contributes to the literature in two ways. First, I provide evidence that, in thestylized environment where unfavorable treatment has material consequences, the argument thatwomen face a tradeoff between being competent and being likable does not hold. This suggeststhat competent women may receive fair treatment in hiring and promotion if the results areexternally valid. Second, while several studies show that women are more critically evaluated intraditionally male occupations, my results indicate that a plausible explanation for this evidenceis people’s lack of enough prior about women’s competence in these occupations rather thantaste-based gender discrimination.Indeed, there is ample evidence that female leaders (Chakraborty and Serra 2019; Håkansson2020) and competitors (Datta Gupta, Poulsen, and Villeval 2013) receive more aggressivetreatments and receive less support by men (Born, Ranehill, and Sandberg 2020). My studyis silent to gender discrimination where there are intense interactions and competition amongwomen and men; there is evidence that men hold motivated gender bias (Sinclair and Kunda2000) and it is a topic of future research. Still, my results apply to vertical relationships such asworkers vs. managers and employees vs. employers and provide a piece of evidence in consideringhiring and promotion practices in labor markets.16 eferences
Almås, Ingvild, Alexander W Cappelen, Kjell G Salvanes, Erik Ø Sørensen, and Bertil Tungodden.2017. “Fairness and family background.”
Politics, Philosophy & Economics
16, no. 2 (1,2017): 117–131.Babcock, Linda, María P. Recalde, Lise Vesterlund, and Laurie Weingart. 2017. “GenderDifferences in Accepting and Receiving Requests for Tasks with Low Promotability.”
American Economic Review
107 (3): 714–747.Bell, Robert M., and Daniel F. McCaffrey. 2002. “Bias Reduction in Standard Errors for LinearRegression with Multi-Stage Samples.”
Survey Methodology
28 (2): 169–181.Bilker, Warren B., John A. Hansen, Colleen M. Brensinger, Jan Richard, Raquel E. Gur, andRuben C. Gur. 2012. “Development of Abbreviated Nine-Item Forms of the Raven’s StandardProgressive Matrices Test.”
Assessment
19, no. 3 (1, 2012): 354–369.Bohnet, Iris, and Bruno S. Frey. 1999. “Social Distance and Other-Regarding Behavior inDictator Games: Comment.”
American Economic Review
89 (1): 335–339.Bohren, J. Aislinn, Alex Imas, and Michael Rosenberg. 2019. “The Dynamics of Discrimination:Theory and Evidence.”
American Economic Review
109 (10): 3395–3436.Boring, Anne. 2017. “Gender biases in student evaluations of teaching.”
Journal of PublicEconomics
145 (1, 2017): 27–41.Born, Andreas, Eva Ranehill, and Anna Sandberg. 2020. “Gender and Willingness to Lead: Doesthe Gender Composition of Teams Matter?”
The Review of Economics and Statistics (10,2020).Brañas-Garza, Pablo, Ramón Cobo-Reyes, María Paz Espinosa, Natalia Jiménez, JaromírKovářík, and Giovanni Ponti. 2010. “Altruism and social integration.”
Games and EconomicBehavior
69, no. 2 (1, 2010): 249–257.Cappelen, Alexander, Ranveig Falch, and Bertil Tungodden. 2019.
The Boy Crisis: ExperimentalEvidence on the Acceptance of Males Falling Behind.
HCEO Working Paper 2019-014.Casella, George, and Roger L. Berger. 2001.
Statistical Inference.
Frontiersin Psychology
ClinicalInfectious Diseases.
Chakraborty, Priyanka, and Danila Serra. 2019.
Gender differences in top leadership roles: Doesworker backlash matter?
Working Paper.Charness, Gary, and Uri Gneezy. 2008. “What’s in a name? Anonymity and social distance indictator and ultimatum games.”
Journal of Economic Behavior & Organization
68, no. 1 (1,2008): 29–35. 17hen, Daniel L., Martin Schonger, and Chris Wickens. 2016. “oTree—An open-source platformfor laboratory, online, and field experiments.”
Journal of Behavioral and ExperimentalFinance
The Quarterly Journal of Economics
Journal of EconomicLiterature
47 (2): 448–474.Datta Gupta, Nabanita, Anders Poulsen, and Marie Claire Villeval. 2013. “Gender Matchingand Competitiveness: Experimental Evidence.”
Economic Inquiry
51 (1): 816–835.De Paola, Maria, Michela Ponzo, and Vincenzo Scoppa. 2018. “Are Men Given Priority for TopJobs? Investigating the Glass Ceiling in Italian Academia.”
Journal of Human Capital
Political Behavior
39, no. 2 (1, 2017): 301–325.Eil, David, and Justin M. Rao. 2011. “The Good News-Bad News Effect: Asymmetric Processingof Objective Information about Yourself.”
American Economic Journal: Microeconomics
Experimental Economics
14, no. 4 (1,2011): 583–610.Exley, Christine L., and Judd B. Kessler. 2019.
Motivated Errors.
Working Paper.Fiske, Susan T., Amy J. C. Cuddy, Peter Glick, and Jun Xu. 2002. “A model of (often mixed)stereotype content: Competence and warmth respectively follow from perceived status andcompetition.”
Journal of personality and social psychology
82 (6): 878–902.Fisman, Raymond, Pamela Jakiela, Shachar Kariv, and Daniel Markovits. 2015. “The distribu-tional preferences of an elite.”
Science
Journal of the Economic Science Association
1, no. 1 (1, 2015): 114–125.Håkansson, Sandra. 2020. “Do women pay a higher price for power? Gender bias in politicalviolence in Sweden.”
Journal of Politics.
Heilman, Madeline E. 2001. “Description and Prescription: How Gender Stereotypes PreventWomen’s Ascent Up the Organizational Ladder.”
Journal of Social Issues
57, no. 4 (1, 2001):657–674.Heilman, Madeline E., Caryn J. Block, and Richard F. Martell. 1995. “Sex stereotypes: Do theyinfluence perceptions of managers?”
Journal of Social Behavior and Personality
10 (6):237–252.Heilman, Madeline E., and Tyler G. Okimoto. 2007. “Why are women penalized for success atmale tasks?: the implied communality deficit.”
The Journal of Applied Psychology
92 (1):81–92.Heilman, Madeline E., Aaron S. Wallen, Daniella Fuchs, and Melinda M. Tamkins. 2004.“Penalties for success: reactions to women who succeed at male gender-typed tasks.”
TheJournal of Applied Psychology
89 (3): 416–427.18saksson, Siri. 2018.
It Takes Two: Gender Differences in Group Work.
Working Paper. 14, 2018.Koffi, Marlène. 2019.
Innovative Ideas and Gender Inequality.
Working Paper.Leider, Stephen, Tanya Rosenblat, Markus M. Möbius, and Quoc-Anh Do. 2010. “What do weExpect from our Friends?”
Journal of the European Economic Association
Biometrika
73, no. 1 (1, 1986): 13–22.MacKinnon, James G, and Halbert White. 1985. “Some heteroskedasticity-consistent covariancematrix estimators with improved finite sample properties.”
Journal of Econometrics
29, no.3 (1, 1985): 305–325.McKenzie, David, and Owen Ozier. 2019.
Why ex-post power using estimated effect sizes is bad,but an ex-post MDE is not.
Development Impact. The World Bank, 16, 2019.Mengel, Friederike, Jan Sauermann, and Ulf Zölitz. 2019. “Gender Bias in Teaching Evaluations.”
Journal of the European Economic Association.
Moss-Racusin, Corinne A., Julie E. Phelan, and Laurie A. Rudman. 2010. “When men breakthe gender rules: Status incongruity and backlash against modest men.”
Psychology of Men& Masculinity
11 (2): 140–151.Phelan, Julie E., Corinne A. Moss-Racusin, and Laurie A. Rudman. 2008. “Competent yet Out inthe Cold: Shifting Criteria for Hiring Reflect Backlash Toward Agentic Women.”
Psychologyof Women Quarterly
32 (4): 406–413.Pustejovsky, James E., and Elizabeth Tipton. 2018. “Small-Sample Methods for Cluster-RobustVariance Estimation and Hypothesis Testing in Fixed Effects Models.”
Journal of Business& Economic Statistics
36, no. 4 (2, 2018): 672–683.Quadlin, Natasha. 2018. “The Mark of a Woman’s Record: Gender and Academic Performancein Hiring.”
American Sociological Review
83, no. 2 (1, 2018): 331–360.Rudman, Laurie A. 1998. “Self-promotion as a risk factor for women: The costs and benefits ofcounterstereotypical impression management.”
Journal of Personality and Social Psychology
74 (3): 629–645.Rudman, Laurie A., and Kimberly Fairchild. 2004. “Reactions to counterstereotypic behavior:the role of backlash in cultural stereotype maintenance.”
Journal of Personality and SocialPsychology
87 (2): 157–176.Rudman, Laurie A., and Stephen E. Kilianski. 2000. “Implicit and Explicit Attitudes TowardFemale Authority.”
Personality and Social Psychology Bulletin
26, no. 11 (1, 2000): 1315–1328.Rudman, Laurie A., Corinne A. Moss-Racusin, Julie E. Phelan, and Sanne Nauts. 2012. “Statusincongruity and backlash effects: Defending the gender hierarchy motivates prejudice againstfemale leaders.”
Journal of Experimental Social Psychology
48, no. 1 (1, 2012): 165–179.Rudman, Laurie A., and Julie E. Phelan. 2008. “Backlash effects for disconfirming genderstereotypes in organizations.”
Research in Organizational Behavior
28 (1, 2008): 61–79.Sandberg, Sheryl. 2013.
Lean In: Women, Work, and the Will to Lead.
New York, NY: Knopf,12, 2013. 19arsons, Heather. 2019.
Interpreting Signals in the Labor Market: Evidence from Medical Referrals.
Working Paper. 16, 2019.Sarsons, Heather, Klarita Gërxhani, Ernesto Reuben, and Arthur Schram. 2020. “GenderDifferences in Recognition for Group Work.”
Journal of Political Economy.
Sinclair, Lisa, and Ziva Kunda. 2000. “Motivated Stereotyping of Women: She’s Fine if ShePraised Me but Incompetent if She Criticized Me.”
Personality and Social PsychologyBulletin
26, no. 11 (1, 2000): 1329–1342.Tajfel, Henry, and John Turner. 1979. “An integrative theory of intergroup conflict.” In
Thesocial psychology of intergroup relations, edited by William G. Austin and Stephen Worchel,33–47. Monterey, CA: Brooks Cole Publishing.Williams, Wendy M., and Stephen J. Ceci. 2015. “National hiring experiments reveal 2:1 facultypreference for women on STEM tenure track.”
Proceedings of the National Academy ofSciences
The Quarterly Journal of Economics
American Economic Review
110 (2): 337–361. 20 ppendix A Changes to the pre-analysis plan
In the initial design, recipients finished all the tasks except the post-questionnaire and left thelaboratory before dictators receive their IQ rank, so that dictators could play dictator gamewithout recipients in the same room. The allocation to the recipients was paid electronically asa “participation fee” for the online post-questionnaire which was sent to recipients via emailafter the session was over. However, as I ran the 1st session with this initial design with 24participants, dictators had to wait idly for about 20-30 minutes until recipients left the laboratoryand dictators seemed to have lost concentration during this period: about half of the dictatorscould not answer the comprehension questions about their IQ rank. Thus, I changed the designand let recipients stay in the laboratory while dictators played the dictator game. I looked at the1st session data before making this change. I exclude the 1st session data in the analysis, butresults including the 1st session data delivers the same conclusion and available upon request.Also, the oTree code and instructions used for the 1st session are available upon request.I also made the following minor changes after the 1st session:1. I reduced participation fee from 3€ to 2.5€ because participants earned more than Iexpected in the IQ test.2. I added more explanation to the instructions on how the IQ rank was assigned and how toallocate endowment in the dictator game.3. I asked participants’ major by simply choosing among the choices from humanities, socialsciences, natural sciences/mathematics, medicine, and engineering and letting them typein their degree program name for a check, instead of letting them access to the Universityof Bologna’s degree program website. This is because the computers in the laboratorysometimes did not accept iframe or prevented a pop-up to another website due to thesecurity setting.Other changes are the following:
Interpretation and focus :1. I rephrased smartness as competence to better place my results in the literature.2. I mainly discussed results for question 3.
Analysis :3. I corrected the definition of
Lower ij . Consequently, I renamed it as IQHigher ij to makethe meaning clearer.4. I added distributional analysis (in figure 5) to also examine whether the results hold alsoin distribution.5. I used lm_robust instead of vcovCR to apply Pustejovsky and Tipton (2018)’s small clusterbias adjustment because vcovCR did not make degrees of freedom adjustment.6. I included in female and male dictator regressions STEM major dummy and Emilia-Romagna dummy because excluding them in regressions where the sample is conditionedby gender made little sense. The results are invariant to the exclusion of these covariates.7. I divided dictators’ allocation by dictator endowment to facilitate the interpretation of theregression results (this does not affect my results because of the round fixed effects).21 ppendix B Description of covariates X ij in the main specification (equation 1) includes the following variables:Dictator characteristics• Age i ∈ N : dictator i ’s age.• F emale i ∈ { , } : an indicator variable equals 1 if dictator i is female, 0 otherwise.• F romEmiliaRomagna i ∈ { , } : an indicator variable equals 1 if dictator i is fromEmilia-Romagna region (where the University of Bologna is located), 0 otherwise.• SocialSciences i ∈ { , } : an indicator variable equals 1 if dictator i ’s major is socialsciences, 0 otherwise.• ST EM i ∈ { , } : an indicator variable equals 1 if dictator i ’s major is natural sci-ences/mathematics, engineering, or medicine; 0 otherwise.• P ostBachelor i ∈ { , } : an indicator variable equals 1 if dictator i ’s degree programis either master/post-bachelor, in the 4th year or beyond of bachelor-master combinedprogram, or PhD, 0 otherwise. • OverConf idence i ∈ {− , , } : degree of dictator i ’s overconfidence. It is equal to − i ’s guess about the number of IQ test questions they correctly solved is lowerthan the actual number, 0 if equal to the actual number, and 1 if higher than the actualnumber.Recipient characteristics• Age j ∈ N : recipient j ’s age.• F romEmiliaRomagna j ∈ { , } : an indicator variable equals 1 if recipient j is fromEmilia-Romagna region, 0 otherwise.Fixed effects• P l =2 r l : round fixed effects where r l ∈ { , } is an indicator variable equals 1 if the roundis equal to l=1,2,3, 0 otherwise.• P l =2 q lij : proximity fixed effects where q lij ∈ { , } is an indicator variable showing theproximity between dictator i and recipient j , and equals 1 if dictator i does not knowrecipient j at all (l=1), has seen before (l=2), knows but not very well (l=3).
18. In Italy, bachelor is a 3 year program. ppendix C Additional figures and tables Figure C1: Density and distribution of the dictators’ allocation
Giving in the dictator game (fraction of endowment) D en s i t y Panel A: Density (N=558)
Giving in the dictator game (fraction of endowment) C u m u l a t i v e p r obab ili t y Panel B: Distribution (N=558)
Notes:
These figures show the empirical density (panel A) and the empirical distribution (panel B) of thedictators’ allocation as a fraction of endowment. Recipients whose name is non-Italian sounding and whomthe dictator declared they knew them “very well” at least once are excluded. Unit of observation: dictator’sallocation. igure C2: The role of the recipients’ IQ and gender in dictators’ allocation:Taking into account for IQ rank differences −0.10.00.10.2 +3 +2 +1 −1 −2 −3 Recipient's relative IQ rank b ^ Notes:
This figure shows OLS estimates of the role of recipient’s gender and IQ in dictators’ allocation thattakes into account for the IQ rank differences dictators observe by splitting
IQHigher ij into 6 separatedummies indicating the recipients’ IQ rank differences relative to the dictators’. The lower/higher therecipient’s IQ, the more negative/positive their IQ rank difference. The specification includes dictator IQfixed effects, round fixed effects, proximity fixed effects, dictator controls, and recipient controls. The outcomevariable is dictators’ allocation as a fraction of endowment. For brevity, the figure only plots the coefficientestimates on the interaction terms between the 6 separate IQHigher ij and F emale j , ˆ β , along with their95% confidence intervals, which is calculated with standard errors clustered at dictator level with Pustejovskyand Tipton (2018)’s small cluster bias adjustment. Unit of observation: dictator’s allocation. Table C1: Balance test: IQ rank
Outcome: Age Female From Emilia-Romagna Human-ities Socialsciences STEM Postbachelor Over-confidence(1) (2) (3) (4) (5) (6) (7) (8)IQ rank = 2 0.010 0.221* 0.074 -0.095 0.034 0.061 0.151 0.146(0.796) (0.128) (0.104) (0.130) (0.088) (0.130) (0.127) (0.200)IQ rank = 3 -0.300 0.139 -0.007 -0.101 0.183 -0.081 0.183 0.160(0.776) (0.143) (0.103) (0.142) (0.120) (0.137) (0.137) (0.241)IQ rank = 4 -0.536 0.094 0.138 -0.146 0.101 0.045 0.187 0.430*(0.894) (0.148) (0.116) (0.148) (0.123) (0.148) (0.145) (0.258)IQ rank = 5 0.534 0.092 0.062 -0.220 0.166 0.054 0.061 0.158(0.959) (0.165) (0.128) (0.175) (0.128) (0.165) (0.156) (0.271)IQ rank = 6 -0.040 0.070 0.021 -0.368* 0.442*** -0.074 0.013 0.346(1.093) (0.191) (0.147) (0.201) (0.162) (0.173) (0.191) (0.306)Dictator IQ FE (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51)
F statistic 0.571 0.634 0.704 0.697 1.91* 0.626 0.739 0.83R-squared 0.040 0.067 0.040 0.042 0.074 0.062 0.027 0.032Observations 195 195 195 195 195 195 195 195
Notes:
This table shows balance across dictators with different IQ ranks. The estimates are obtained byrunning OLS regression of various dictator characteristics on IQ rank dummies with dictator IQ fixed effects.The F statistic shows the joint significance of IQ rank = 2 to IQ rank = 6 dummies. HC2 heteroskedasticity-robust standard errors (MacKinnon and White 1985) with Bell and McCaffrey (2002)’s small sample biasadjustment are reported below each coefficient estimate. R-squared is net of dictator IQ fixed effects. Unit ofobservation: dictator. Significance levels: * 10%, ** 5%, and *** 1%. able C2: Balance test: Recipient’s category Outcome: Age Female From Emilia-Romagna Human-ities Socialsciences STEM Postbachelor Over-confidence(1) (2) (3) (4) (5) (6) (7) (8)IQHigher -0.429 0.001 0.105** -0.065 0.106** -0.041 -0.071 0.063(0.350) (0.064) (0.048) (0.065) (0.051) (0.060) (0.063) (0.107)Female -0.228 0.060 0.080* -0.026 0.015 0.011 -0.043 0.040(0.336) (0.059) (0.048) (0.057) (0.046) (0.057) (0.060) (0.090)IQHigherxFemale 0.431 0.010 -0.148** 0.014 -0.063 0.049 0.069 -0.051(0.458) (0.082) (0.064) (0.081) (0.062) (0.079) (0.084) (0.129)Dictator IQ FE (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51)
F statistic 0.522 1.078 2.074 0.505 1.731 0.661 0.417 0.119R-squared 0.029 0.052 0.034 0.025 0.028 0.050 0.014 0.007Observations 558 558 558 558 558 558 558 558Clusters 195 195 195 195 195 195 195 195
Notes:
This table shows that dictators were matched recipients of different gender and IQ in a balanced wayeven ex-post. The estimates are obtained by running OLS regression of various dictator characteristics oncovariates of interest with dictator IQ fixed effects. The F statistic shows the joint significance of all covariates.The standard errors are clustered at the dictator level with Pustejovsky and Tipton (2018)’s small clusterbias adjustment are reported below each coefficient estimate. R-squared is net of dictator IQ fixed effects.Unit of observation: dictator-recipient pair. Significance levels: * 10%, ** 5%, and *** 1%.
Table C3: Balance test: Recipient’s category (cont.)
Outcome: Age(recipient) From Emilia-Romagna(recipient) Dictatorgameround 1 Dictatorgameround 2 Dictatorgameround 3 Did notknowat all Sawbefore Knewbut notvery well(9) (10) (11) (12) (13) (14) (15) (16)IQHigher -0.792** 0.188*** -0.084 -0.026 0.110* -0.002 0.008 -0.006(0.374) (0.050) (0.065) (0.064) (0.061) (0.026) (0.022) (0.018)Female -0.284 0.025 -0.084 0.037 0.047 0.020 -0.011 -0.009(0.344) (0.038) (0.062) (0.058) (0.059) (0.020) (0.017) (0.010)IQHigherxFemale 0.626 -0.100 0.137 -0.084 -0.053 -0.020 0.005 0.014(0.462) (0.062) (0.084) (0.079) (0.084) (0.026) (0.025) (0.020)Dictator IQ FE (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51) (cid:51)
F statistic 1.537 5.51*** 0.941 0.89 1.207 0.666 0.415 1.071R-squared 0.013 0.041 0.006 0.006 0.007 0.047 0.014 0.074Observations 558 558 558 558 558 558 558 558Clusters 195 195 195 195 195 195 195 195
Notes:
This table shows that dictators were matched recipients of different gender and IQ in a balanced wayeven ex-post. The estimates are obtained by running OLS regression of various recipient characteristics andround and proximity dummies on covariates of interest with dictator IQ fixed effects. The F statistic showsthe joint significance of all covariates. The standard errors are clustered at the dictator level with Pustejovskyand Tipton (2018)’s small cluster bias adjustment are reported below each coefficient estimate. R-squared isnet of dictator IQ fixed effects. Unit of observation: dictator-recipient pair. Significance levels: * 10%, ** 5%,and *** 1%. able C4: Summary statistics: Dictator data by gender Female Male DifferenceMean SD Mean SD p-value
Panel A: Dictators
IQ level 6.52 1.20 6.89 1.24 0.04IQ rank 3.83 1.59 3.31 1.73 0.03Age 23.68 2.62 23.23 2.81 0.25From Emilia-Romagna 0.18 0.39 0.19 0.39 0.94Humanities 0.58 0.50 0.32 0.47 0.00Social sciences 0.15 0.36 0.24 0.43 0.13STEM 0.27 0.45 0.44 0.50 0.01Post bachelor 0.53 0.50 0.37 0.49 0.03Overconfidence 0.31 0.78 0.56 0.72 0.02Time on feedback (sec.) 107.67 89.88 107.52 102.26 0.99Observations 104 91
Panel B: Paired recipients
IQ level 6.77 1.19 6.91 1.12 0.15IQ rank 3.39 1.75 3.45 1.74 0.72IQ higher 0.57 0.50 0.48 0.50 0.03Age 23.17 2.57 23.55 2.98 0.12Female 0.50 0.50 0.43 0.50 0.10From Emilia-Romagna 0.15 0.36 0.25 0.43 0.01Observations 298 260
Panel C: Proximity
Did not know at all 0.98 0.15 0.95 0.23 0.07Knew but not well 0.02 0.15 0.03 0.18 0.44Saw before 0.00 0.00 0.02 0.14 0.02Observations 298 260
Panel D: Dictator’s allocation (fraction of endowment)
Allocation 0.43 0.22 0.37 0.25 0.00Allocation (demeaned) 0.22 0.25Observations 298 260
Notes:
This table shows summary statistics separately for female and male dictators: the dictators’ and thepaired recipients’ characteristics, how well dictators knew the paired recipients, and dictators’ allocation.Recipients whose name is non-Italian sounding and whom the dictator declared they knew them “very well”at least one are not included. Standard deviation of demeaned allocation is calculated as sample standarddeviation of (cid:94)
Allocate ik = Allocate ik − Allocate k , where Allocate ik is allocation by dictator i whose IQ is k and Allocate k = P i ∈ k Allocate ik is average allocation of dictators with IQ k . P-values for difference in meansare calculated with the two-sample t-test with HC2 heteroskedasticity-robust standard errors (MacKinnonand White 1985) with Bell and McCaffrey (2002)’s small sample bias adjustment. able C5: Summary statistics: Evaluator data vs. dictator data Evaluator Dictator DifferenceMean SD Mean SD p-value
Panel A: Evaluator / Dictator
IQ level 6.84 1.14 6.69 1.23 0.21IQ rank 3.40 1.74 3.58 1.67 0.30Age 23.34 2.78 23.47 2.72 0.63From Emilia-Romagna 0.20 0.40 0.18 0.39 0.76Humanities 0.34 0.48 0.46 0.50 0.02Social sciences 0.27 0.44 0.19 0.40 0.08STEM 0.39 0.49 0.35 0.48 0.42Post bachelor 0.49 0.50 0.46 0.50 0.48Overconfidence 0.49 0.75 0.43 0.76 0.42Time on feedback (sec.) 93.26 83.96 107.60 95.60 0.12Observations 193 195
Panel B: Paired recipients
IQ level 6.84 1.16 6.84 1.16 1.00IQ rank 3.42 1.74 3.42 1.74 0.98IQ higher 0.50 0.50 0.53 0.50 0.46Age 23.35 2.80 23.35 2.77 0.99Female 0.47 0.50 0.47 0.50 0.99From Emilia-Romagna 0.19 0.40 0.20 0.40 0.87Observations 368 558
Panel C: Proximity
Did not know at all 0.98 0.14 0.96 0.19 0.08Knew but not well 0.02 0.14 0.03 0.17 0.34Saw before 0.00 0.00 0.01 0.09 0.03Observations 368 558
Panel D: Belief on the recipient’s IQ
Belief on IQ level 6.34 1.89Belief on IQ level (demeaned) 1.87Observations 368
Notes:
This table shows summary statistics for the evaluators and dictators: the evaluators’/dictators’ andthe paired recipients’ characteristics, how well evaluators/dictators knew the paired recipients, and evaluators’belief. Recipients whose name is non-Italian sounding and whom the dictator declared they knew them “verywell” at least one are not included. P-values for difference in means are calculated with the two-sample t-testwith HC2 heteroskedasticity-robust standard errors (MacKinnon and White 1985) with Bell and McCaffrey(2002)’s small sample bias adjustment. able C6: The role of the recipients’ gender and IQ in the dictators’ allocation:Further robustness checks Outcome: Dictator’s allocation (fraction of endowment)Sample: ExcludingIQ rank 1 and 6 Excludingproximity 3 Excludingproximity 2 and 3(1) (2) (3)IQHigher 0.006 0.011 0.005(0.036) (0.033) (0.034)[-0.065, 0.077] [-0.056, 0.077] [-0.062, 0.073]Female 0.019 0.006 0.008(0.029) (0.026) (0.027)[-0.040, 0.077] [-0.045, 0.058] [-0.046, 0.062]IQHigherxFemale 0.001 0.029 0.029(0.044) (0.037) (0.038)[-0.087, 0.088] [-0.043, 0.102] [-0.047, 0.104]Female+IQHigherxFemale 0.019 0.036 0.037(0.034) (0.026) (0.026)[-0.047, 0.086] [-0.016, 0.087] [-0.015, 0.088]Outcome Mean 0.412 0.402 0.404Outcome SD 0.234 0.239 0.239R-squared 0.102 0.083 0.079Observations 386 553 537Clusters 135 195 194
Notes: