Tuesday, March 6, 2018

Can Female Role Models Reduce the Gender Gap in Science? Evidence from Classroom Interventions in French High Schools with Clémentine Van Effenterre

Women are underrepresented among technical degree holders and STEM workers, and the gender segregation in these occupations has remained fairly consistent over time. How can we design interventions that might help fix the gender gap in science? WAPPP Postdoctoral Fellow Clémantine Van Effenterre presented the results of her study on the effects of STEM role models in the classroom.


This study focused on French high school students in year 10 and year 12. At the end of year 10, students choose to specialize in a science, social science, or humanities track, conditional on performance in these areas. At the end of year 12, students sit for their graduation exam, the Baccalauréat, which is the gateway to pursuing higher education. In order to gain entry to certain higher education tracks, year 12 students can opt to pursue highly selective prep classes (classes préparatoires aux grandes écoles), including in science.

Women are slightly overrepresented in year 10 science courses, but are underrepresented in science in year 11 and year 12, as well as in the selective science classes préparatoires.


This study was a large-scale randomized experiment, with control groups and treatment groups of year 10 and year 12 students.  Students in treatment groups had a one-hour session with a female STEM role model, either a “professional” (private employees in a STEM field) or a “researcher” (PhD candidates or post-doctoral students in the sciences). The role models presented on career prospects and wage rates in the sciences versus the humanities, the diversity of jobs in the sciences, and the underrepresentation of women in STEM jobs.

In addition to sharing their own experiences, the role models showed two videos. The first emphasized that men and women have the same “science potential” and that there are no biologically-determined sex differences with respect to science aptitude. The second video focused on how much workers in STEM occupations enjoy their workplace culture and work/life balance, as well as the shorter-than-expected duration of many science degree programs.

Dr. Van Effenterre and her colleagues surveyed students on the intensity of their stereotypical views of STEM jobs and women in science and their choice of school tracks, and collected data on their grades and higher education admission outcomes.



Students in the treatment group, both male and female, changed their beliefs regarding careers in science and women in science. Female year 10 students were less likely to think that studying for a science degree takes a long time, that science jobs are dreary or solitary, or that it is hard to achieve work-life balance in STEM. With respect to women and science, the intervention emphasized underrepresentation and debunking biological essentialism. While students reduced their stereotypical beliefs in this area, they were more likely to think that “women like science less than men” and that “progress for women in science is slow.”

This is an interesting result: after the intervention, students are aware that men are disproportionately represented in science, but that the reason is not biologically determined. To make sense of this state of affairs, Dr. Van Effenterre and her colleagues hypothesize, students think that women must either not like science and not pursue it, or go into STEM fields and face discrimination or other obstacles to success in the field.

The effect sizes were larger for female students, but male students changed their beliefs too; the intervention may have been more salient for female students, but had an effect for all treated students.

School Tracks

This study revealed no change in track choices for year 10 students, but year 12 students in the treatment group significantly increased their likelihood of pursuing selective preparatory STEM programs: male students’ participation increased by 38% and female students’ by 20%, which are large increases over the low baseline for these selective programs.

Dr. Van Effenterre and her colleagues were initially puzzled by the lack of movement for year 10 students, but realized that student’s track choices are in some ways predetermined by past academic performance and other social factors, and are likely very hard to change with a small intervention. 


The role models may have impacted students’ beliefs and choices in a variety of ways: students may have been inspired to increase their effort to gain access to science jobs or may have experienced a boost in self-confidence or sense of fit in STEM roles after interacting with role models.

Dr. Van Effenterre and her colleagues examined the impact of the treatment on the grades the students obtained on the Baccalauréat, compared to students’ grades on other standardized exams. On average, there was no significant response to the treatment in terms of grades on the Baccalauréat for male or female students, and so the “increased effort” mechanism does not necessarily explain the observed effect.

Role models in the “professionals” category had a greater impact on both the decision to choose a science track and admittance to it compared to “researchers” for both male and female students. One limitation of this study was that the role models were only quasi-randomized, but it appears that the “professionals” may have been more relevant and compelling to students than the “researchers,” which presents additional questions about how to tailor the best type of role model for students to achieve the intended effect.


Experiences with female STEM role models can be an effective policy tool to change beliefs and outcomes for male and female students, even in a limited, one-hour intervention. Dr. Van Effenterre emphasized that closing gender gaps today impacts the next generation: if they can see it, they can be it.

Monday, February 12, 2018

The Influence of Sexual Orientation and Race on Gender Prescriptive Stereotypes with Sa-kiera Hudson

Gender stereotypes are powerful forces that shape collective expectations and are socialized from an early age; these stereotypes are not just descriptive, but also prescriptive, stating what sorts of things men and women should and should not do based on their gender.

Rule congruity theory helps to explain why these prescriptive stereotypes are often a source of backlash: men and women are evaluated negatively when the role that they occupy is incongruent with the stereotypes of their gender. Negative evaluations can come from both descriptive and prescriptive stereotypes – women can spark descriptive backlash in leadership roles if there are few other women in leadership, and prescriptive backlash if the traits she displays violates one’s notions of how women should behave.

Rule congruity theory has sparked a great deal of research, but researchers generally haven’t interrogated who or what we mean when we say “men” or “women.” Research on role congruity theory and gender generally presumes straight, white, middle-class individuals, which gives reason to believe that prescriptive stereotypes of women don’t fully encompass lesbian women or women of color.

Taking an intersectional perspective may reveal that societal expectations are unique based on other identities. Importantly, less than 20% of studies on race include women, and less than 10% of studies on gender include race: there is an enormous amount of work to be done in intersectional psychology.

The first WAPPP Seminar of the semester featured Sa-kiera Hudson, PhD Candidate in Psychology at Harvard University and WAPPP Fellow, presenting research on intersectional identities and gender stereotypes. In particular, Hudson questions whether rule congruity theory is adequate to explain intersectional identities and, if not, how the theory should be modified.

Study 1

Hudson’s first study incorporated sexual orientation as an additional identity. Existing literature suggests that in the realm of stereotypes, gay men and straight women are presumed to be similar, as are lesbian women and straight men, because of their similar sexual preferences.

In this study, individuals were given a list of 70 traits, some of which were more stereotypically masculine, feminine, or neutral (e.g. “assertive,” “warm and kind,” “honest”), and asked to rate whether that trait was desirable for straight and gay men and women, for men and women without a sexual orientation identifier, and for straight or gay people, without specifying gender.

Hudson hypothesized three expected patterns: stability in gender stereotypes, evidence of heterocentrism, and evidence of gender inversion.

1. Stability in gender stereotypes: Hudson’s findings were substantially the same as a similar study from 2002, which indicates that prescriptive gender stereotypes are relatively stable and haven’t changed much in the last 15 years.

2. Evidence of heterocentrism: There was no trait for which respondents’ rankings of desirable traits for a “man” deviated from those for a “straight man,” nor “woman” from “straight woman” or “person” from “straight person.” This finding suggests that when individuals are rating a person without a sexual orientation identifier, they presume straightness. While this isn’t surprising, it is useful: very few studies directly demonstrate heteronormativity in judgments like these.

3. Gender inversion: Hudson expected to see evidence of gender inversion, similarities between gay men and straight women and between lesbian women and straight men, in addition to differences between gay and straight men that mirror the differences between straight men and straight women. Of the 70 traits, there was evidence of gender inversion for 9: for example, it was less desirable for straight women to be assertive compared to straight men, but more desirable for lesbian women to be assertive compared to gay men.

An additional 12 traits showed evidence of gender asymmetry, where desirability ratings for men or women were the same regardless of sexual orientation, but diverged for the opposite gender. For example, cleanliness was equally desirable for gay and straight men, but for women, there was more pressure for straight women to be clean than for lesbian women. Thirty-four traits displayed evidence of sexual orientation asymmetry – there was gender differentiation for straight and gay men and women, but no difference between gay men and lesbian women. This finding suggests that gender might not matter as much as sexual orientation in terms of how we perceive gay men and lesbian women.

In addition, Hudson presented three other interesting wrinkles from this study:

  • It is more desirable for lesbians to be assertive, competitive, and forceful compared to straight women, but there was no difference in trait desirability for business sense, career-oriented, or leadership ability, which suggests that it is acceptable for lesbian women to display dominant personality traits, but not in ways that might disturb the status quo of gender hierarchy.
  • The differences between straight and gay men are much larger than the differences between straight and lesbian women, which suggests that gay men may lose some of the status that comes with being male by virtue of their sexual orientation, while lesbian women don’t gain much status above that accorded to straight women.
  • Finally, Hudson’s study revealed some evidence that homosexuality is associated with low status – study respondents found it less desirable for gay men and lesbian women to be loyal, dependable, honest, or enthusiastic, traits that are typically considered neutral and positive. 
Study 2

Hudson’s second study replicated the first study’s methods using race rather than sexual orientation. There are different stereotypes of men and women of different races; for example, white women are stereotyped as submissive and feminine, black women as confident and aggressive. What studies do exist demonstrate that black and white women are treated differently with respect to normative traits – black women are not sanctioned as heavily for “aggressive leadership behaviors” as black men and white women. Something about the interaction of race and gender stereotypes specific to black women may mediate backlash.

Just as in the first study, Hudson found stability of gender stereotypes over time and evidence of eurocentrism: when respondents rated traits for a person without a race marker, their answers most closely matched those given for white individuals, suggesting that they viewed white as the default.

Hudson initially hypothesized that this study would yield evidence of gendered race theory, which posits that black individuals are masculinized. However, the results produced limited evidence for gendered race theory! For black men, it was less desirable to be masculine and more desirable to be feminine compared to the control male target. For black women, it was less desirable to be feminine and more desirable to be masculine compared to the control female target, suggesting that both black men and women are distanced from their gender stereotypes. For 38 traits, there was little gender differentiation among black targets but a significant amount for white targets, again suggesting that gender is not as important as race for stereotypes regarding men and women of color.

Both of these studies revealed that the prescriptive norms for gays and lesbians and for black men and women are mostly feminized norms, which suggests that status might play a role in how gender prescriptive stereotypes are generated for intersectional targets – these stereotypes are less about gender per se and more about relative status. The results of these studies underscore just how much intersectionality matters for contemporary research and the importance of updating existing theories accordingly.

Saturday, December 30, 2017

Equal but Inequitable: Who Benefits from Gender-Neutral Tenure Clock Stopping Policies with Kelly Bedard

In many highly educated professions, we see a pattern in which the number of female workers drops off over time, as does the share of women in the top earners cohort. Many of these professional tracks have an “up-or-out” quality, like making partner at a law firm or earning tenure at a university. This phenomenon is consistent with women entering lower-income fields or leaving them altogether: what mechanism might explain this effect, and what other implications might this pattern have? The final WAPPP seminar of the semester featured Kelly Bedard, Department Chair and Professor of Economics at the University of California, Santa Barbara, describing the results of a study on tenure clock-stopping policies in major university economics departments.

Over the last few decades, many universities have instituted “clock-stopping policies” for tenure. If a young academic has a child before receiving tenure, they may have an extra year before their tenure evaluation – rather than the usual six years to tenure, a new parent could have seven years of work evaluated, with their tenure clock stopped for one year.  Having children is a productivity shock that, though these policies, is taken into account in gauging lifetime productivity.

These clock-stopping policies come in two-flavors: in early iterations, only women were eligible for clock-stopping; more recent policies are gender neutral, such that anyone who has a child can stop their clock. Some universities started with a female-only policy and then switched to gender-neutral. Professor Bedard’s research asks: what are the distributional consequences of these policies? How might these policies change the way people think about publishing and fertility strategies?

The simple story would be to consider two individuals, Person A and Person B, going up for tenure. Person A, a woman, has a child, stops her clock, and does no research for a year. Person B, a man, has a child, stops his clock, but does not change anything about his productivity during that year. Under a gender-neutral clock-stopping policy, Person A was not hurt in her tenure evaluation—her productivity was the same in the other years, even with a year of no research. This is precisely what the policy was designed to do, to ensure that faculty who take parental leave are not harmed in their tenure evaluations by their decision to do so. Person B, however, benefited from the policy, with an additional year to write papers that might factor into his tenure decision. Other faculty who did not have children and did not stop their clocks would be just as unaffected as Person A. This benefit to Person B, while the policy is net neutral for everyone else, is something of an odd result – this is not what the policy was designed to do.

If tenure decisions are imperfect – that is, not made strictly on the basis of the six-year timeline, excluding any time in which the clock was stopped – Person B is now pushing up the bar for tenure with his additional year of productivity. Person B will still be able to achieve tenure, but Person A and anyone who has not had a child will have a more difficult time getting over that threshold. This gender-neutral clock-stopping policy therefore has the possibility for some odd distributional consequences, which may be compounded by the additional productivity costs of being pregnant or departmental variation in teaching relief or other parental leave policies.

In addition to tenure decisions, these policies may also affect publishing decisions. Economists tend to focus on the top five journals, which each have a long lag in time to publication. These journals have a low likelihood of acceptance, but a high return if one’s article is published. Lower-ranked journals have a higher likelihood, but a lower return. Clock stopping adds an additional time period to this publishing game – with only one time period, a researcher might decide to send a manuscript to a lower-ranked journal for the added likelihood of publication. Without a top-five journal article, they may not receive tenure at their current institution, but likely will somewhere else. With two time periods, faculty are incentivized to send manuscripts first to the top-ranked journals; if the manuscript is rejected, it could always be sent to a lower-ranked journal in the second time period. If this strategy works, there will be obvious gains to those who stop their clocks versus those who don’t.

There are multiple avenues through which this effect could be occurring: clock-stopping could change an individual’s number of papers and where they are published and could affect tenure decisions and ultimately fertility decisions, which themselves may have publishing or productivity effects.

To test these possibilities, Professor Bedard assembled two unique datasets, one of the clock-stopping policies for each of the top 50 economics departments, as well as an exhaustive list of all assistant professors from 1980-2010. Using course catalogs, online CVs, and LinkedIn, she assembled a dataset of 1600 assistant professors who started their careers at a top 50 economics department who published at least two papers in eight years, along with their complete job histories and publication records.

The first considered metric was probability that an individual would get tenure at the school at which they started their career. In general, tenuring rates are fairly flat for both men and women, though the average rates are higher for men. What happens with the introduction of clock-stopping policies? Accounting for a series of controls, including flexible gender-specific time dummies and gender-specific university fixed effects, female-only clock-stopping policies have almost no effect on tenuring rates. However, gender-neutral clock-stopping policies have an enormous effect: with a gender-neutral clock-stopping policy, male tenure rates rise by 18 percentage points, and female tenure rates fall by 19 percentage points.

What is the mechanism for this effect? Fortunately, these clock-stopping policies do not impact the likelihood of getting tenure somewhere; instead, women seem to be moving and getting tenure at schools other than where they start. Using the complete job histories, Professor Bedard suggests that, conditional on getting tenure somewhere, once clock-stopping policies enter into place, fewer female faculty move to lower-ranked departments. While there is very little upward movement, this means that female faculty are more likely to move laterally, receiving tenure at a school ranked similarly to the one at which they started.

What are the effects of these policies on publishing? Over the first seven years of their careers, 40-50% of female faculty are not publishing in the top five journals, and those that do publish on average one article. Male faculty are somewhat more likely to publish in the top five journals, but still not to a significant degree. Comparing schools with clock-stopping policies (of either flavor) versus those without, over time female faculty at schools with clock-stopping experience, on average, a drop in publishing. By contrast, men experience a consistent increase in publishing success, such that men have, on average, half a publication more in a top 5 journal. Given that most people only publish one article in a top five journal, clock-stopping has a big effect for men with major impacts for publishing success.

The first version of Professor Bedard’s paper did not contain any analysis on fertility, but so many people wanted to know that she added an additional section. One major difficulty of studying effects on fertility is acquiring the data: most of the collected data is administrative, and few faculty report whether they have children on their CVs. Instead, Professor Bedard sent emails to each of the faculty members for which she had emails – around 85-90% of the dataset – asking whether the individual had children and, if so, in what year the children were born. The response rate (around 65%) was surprisingly high, but of course was entirely non-random; those who had left the profession were unlikely to respond, while those at higher-ranked schools were far more likely to respond. At schools with a gender-neutral clock-stopping policy, about 56% of pre-tenure faculty have a child, compared to 46% of those without a gender-neutral clock-stopping policy. Based on these data, it appears that clock-stopping does not change total fertility, but may change the timing of fertility.

These results for tenure, publishing, and fertility illuminate the side effects of clock-stopping policies and how they might eliminate burdens for women but confer additional benefits on men.

Sunday, November 26, 2017

Betting the House: How Assets Influence Marriage Selection, Marriage Stability, and Child Investments with Corinne Low

How do we understand the marriage contract from an economic standpoint? Just as value is created, divided, and exchanged in the labor market, so it is on the marriage market. Last week’s WAPPP seminar featured Corinne Low, Assistant Professor of Business Economics and Public Policy at the Wharton School at the University of Pennsylvania, as she described the results of her study on marriage and homeownership as a lens into understanding what is valuable about the marriage contract, who reaps value from the marriage contract, and as policies change, how that value is changing over time.

While marriage was previously the universally adopted or aspired to family arrangement, this is no longer the case. There has been a recent increase in cohabitation and nonmarital fertility. However, this change does not necessarily mean that tastes or social values have changed; rather, Professor Low wonders, what in the marriage equation has changed to make people optimize their preferences differently? How has this changed been applied throughout society, where we see a tremendous amount of stratification in rates of marriage based on income, level of education, and race?

The key question in this line of inquiry is what made marriage valuable in the first place, and what made it retain value for some populations and lose value for others?

The marriage contract and nonmarital cohabitation have converged in recent years: the security of marriage has decreased with the advent of unilateral divorce, whereas nonmarital relationships have become more secure with the enforcement of paternity rights and responsibilities outside of marriage. However, there remains one key difference: only in marriage are assets divided up upon divorce. If one person in a cohabitating couple is making all of the mortgage payments, if that relationship breaks up, the other gets nothing. However, in a marriage contract that asset is divided or, very often, assigned to the mother (throughout the presentation, Professor Low referred to this effect in terms of husbands and wives; while this is not universally true, it is disproportionately the case and was used illustratively).

There is, therefore, an important distinction between income, which will be divided in both marital and nonmarital relationships in the form of child support, and asset-sharing, which is unique to the marriage contract.

Home-purchasing, according to Professor Low, may be an important way for husbands to commit to their wives. Because she will be guaranteed at least a share of that valuable asset if the relationship goes south, in her economic calculus it makes sense to invest in the human capital of the children, even if it reduces her own income. Similarly, because the house is at risk, it disincentivizes the husband from unilaterally seeking a divorce. “Betting the House” may be a way of securing the marriage contract.

Investing in children has a cost – if a parent spends time investing in children, they have less time to invest in their own human capital. Generally mothers are the ones making these large investments – women tend to experience a decline in wages after having children, and economically speaking it makes sense for the lower-income partner to invest in children. While marriage used to provide some security for these investments, the marriage contract got more precarious as divorce became easier and more common. However, high-asset individuals can still offer some of this insurance, because those assets will be divided and can provide security for child human capital investments. As such, marriage is really only useful for people who have assets: the decision has less to do with a desire to invest in child human capital than with an ability to secure that investment with assets that can back the marriage contract.

Now that paternity rights and responsibilities are enforced outside of marriage, nonmarital relationships are just as good as marital relationships for individuals without the assets to buy a house – only those with assets get the extra security for their marriage contract. As such, people without assets will flee from marriage, and those with assets will continue to get married. Professor Low and her colleagues set up a model in which nonmarital couples, upon splitting up, would either receive no asset transfer, 1/3 of what the asset transfer in a marital relationship would be, or 2/3 of the marital relationship asset transfer. Under each of these conditions, how do individuals’ preferences change with respect to marital and nonmarital relationships? With no nonmarital transfer, people without assets are divided between marital and nonmarital relationships. At the 1/3 transfer level, people without assets significantly increase their nonmarital fertility. At the 2/3 level, still less than what a spouse would get if a marriage split up, people without assets entirely abandon marriage. At each stage, people with assets choose marriage.

This hypothetical bears out in the data: looking at a longitudinal data set of marriages and average home prices in the year of marriage, those who got married in an era of high housing prices were less able to purchase a home when they got married, and therefore they were less able to secure their child human capital investment. While there are certainly other explanations available, couples facing high housing prices at the time of their marriage tend to have fewer children, and their children are more likely to be held back in school. Assets provide commitment value that allows for greater specialization in the home and more optimal investments in children, which makes the marriage contract more valuable.

This data demonstrates that legal changes (unilateral divorce and paternity acknowledgement) and concordant economic change can be responsible for stratification in marital decisions, rather than “tastes” or preferences. This finding has significant implications for policymakers, as the racial gap in assets and homeownership is much larger than the racial gap in income, in part because of redlining and mortgage discrimination practices. The gap in marriage among racial groups that has been attributed to tastes or values can actually be attributed to homeownership or the ability to get a mortgage.  If we take marriage seriously as an economic contract, it is evident that policies that change the contracting environment have deep implications. This model provides a contracting explanation for why marriage might lead to greater child investment, without relying on arguments about preferences.

Wednesday, November 15, 2017

Gender Differences in Self and Other-Competition with Johanna Mollerstrom

There is a well-established literature demonstrating that women are less likely to choose to enter competitive settings, even when they are just as able as their would-be competitors. While most of these experiments have been conducted in a laboratory setting, different views toward competition have obvious implications for life outside the lab, including education and career trajectories. However, there has not been a similar investigation about women’s attitudes toward self-competition – are women more or less willing to compete against their own previous scores over time? Johanna Mollerstrom, Professor of Economics at Humboldt University, presented the results of three recent experiments on self- and other-competition.

In existing studies, there are generally two mechanisms underlying women’s differential interest in competition: women may be more risk-averse, or they may be less (over-?) confident than men. With self-competition, risk aversion and confidence take on slightly different valences: it’s still risky to compete against yourself, but risk based on the uncertainty of who you’ll be competing against is alleviated. Similarly, confidence may shift when individuals only have to assess their own capacity to improve on previous rounds. There are reasons to believe that self-competition may be different.


The first study was a simple lab experiment designed to test whether there is a gender difference in willingness to self-compete. Participants were asked to add up as many sets of five two-digit numbers as possible in a five-minute period. In the first round, participants were paid a piece rate of $1 for every correct problem. In the second round, participants were either randomly matched with another participant for a competition round or competed against their own first-round score. The winner of the round (either defeating their opponent or their first-round score) got $2 per correct problem. In the third round, participants got a choice between the piece rate and competition, and then were asked a few questions about their risk aversion and confidence. Half of the participants in round three could choose to opt into another other-competition round, while the other half competed against their own scores in round two.

In the other-competition condition, women were less willing to compete than equally-able men, but the difference disappears when controlling for risk aversion and confidence. This finding replicates many past experiments. In the self-competition condition, we see much the same result; risk aversion, in particular, seems to moderate any observed gender difference in willingness to compete. However, this particular experiment had a smaller sample size (N = 200), and a larger sample may indicate that there is no gender difference in willingness to self-compete at all.


The second study recruited a larger sample on MTurk and replicated the first study with a few modifications. The task was changed to a captcha-style exercise rather than addition, and the rounds were shortened to 90 seconds to conform to other online experiments. In addition, the researchers added two additional treatments, manipulating whether other-competitors were matched with someone of the same or opposite sex, while holding constant that they were matched with someone who answered the same number of questions correctly in the first round.

The results of this experiment demonstrated that, again, women were less willing to compete against others, but that the gender difference was moderated by differences in confidence and risk aversion. This experiment revealed a much more precise result for self-competition: there is no significant gender difference in willing to self-compete, though confidence and risk aversion still impact this result. It seems that confidence may play a larger role in decision to compete against others than onself.

Interestingly, when matched with a competitor with the same gender and ability level, women are still less willing to compete than men. In other studies, researchers have found that female-only groups are more willing to compete, but this was not the case in this experiment.

Further, when matched with a competitor of the opposite gender and same ability level, the researchers did not observe a significant gender difference in willingness to compete. It may be that this condition provided better information – competitors did not have to think about how good their competition was, which may have provided a confidence boost (or alleviated some of the risk of competition) and eliminated the gender difference. In both the self- and other-competition conditions, men were less risk-averse than women. However, men were more confident in the other-competition condition; in the self-competition condition, there was no gender difference in confidence.


The first two studies permitted participants to choose whether they wanted to compete; in the third study, participants were forced to compete, but were permitted to choose between self- and other-competition.

In treatment one, participants got the piece rate, then either self-competition or other-competition in randomized order. Finally, participants were able to choose between each of the three options. In treatment two, the fourth round was set up so that participants had to compete, but could chose self- or other-competition. In treatment three, participants didn’t experience self-competition at all: they did two piece rate rounds, and then could choose piece rate or other-competition. In treatment four, participants complete two piece rate rounds, then self-competition, and then in the final round could opt to compete against others.

In treatment one, a large proportion of participants (of both genders!) chose the piece rate. The two competition types seem to be roughly equally popular. In treatment two, when forced to compete, self-competition was much more popular than other-competition. This finding supports the hypothesis that individuals are better able to assess their own capabilities and performance, and may be able to intuit that they can improve a great deal in another round of self-competition.

Having experienced self-competition does not change women’s propensity to choose other-competition. Interestingly, having experienced self-competition makes men less likely to choose other-competition. While this finding may have something to do with how men think about confidence, the researchers did not have an ex ante hypothesis that would explain this result, which is an interesting avenue for future research.

The researchers also sought to decompose confidence by giving each participant a ratio of “how many tasks do you think you completed correctly” over “how many tasks do you think others completed correctly.” In other-competition, men were significantly overconfident and women were significantly underconfident. However, in self-competition, there was no significant gender difference in confidence. Men are more sure that they will beat another person that they are sure that they will beat themselves, while women are more confident in their ability to improve over time rather than beat another person. Men and women are about equally good at assessing their own performance, but women tend to overestimate (and men to underestimate) how well others do. This finding may have relevance for giving feedback: to encourage women, it may be better to tell them “others are not as good as you think” than “you’re better than you think.”

The researchers also examined causal attributions of performance, asking participants on a 1-10 scale whether their performance was “only due to factors I could control” to “only things I could not control.” In self-competition, there was no gender difference in causal attribution – unsurprisingly, all participants attributed self-competition more to factors they could control. In the other-competition condition, women were slightly more likely to think that the result was due to uncontrollable factors. This effect was especially pronounced for participants who believed that they’d won the round: women who believe that they’d won thought that their victory was based on factors that they could not control, such as luck in who they were matched to compete against.

In terms of future research directions, Professor Mollerstrom and her colleagues are interested in whether, if participants know they will be competing against themselves, they undercompete in round one. If so, this finding may limit the contexts in which self-competition would be appropriate. Alternatively, properly calibrated pre-set goals may be an effective middle ground between self- and other-competition. There is a great deal of room to think about how to use competitive institutions – not just other-competition—in more gender-neutral ways that nevertheless preserve the performance-boosting properties of competition.

New Ways of Thinking About Gender and Leadership Effectiveness with Aparna Joshi

The challenges that women face in entering and performing effectively in leadership roles have been widely documented: recent research indicates that the gender gap in pay and promotion is fourteen times greater than gaps in performance. The disconnect between rewards and performance is particularly large in high-prestige occupations. Aparna Joshi, Arnold Family Professor of Management at Penn State University, presented two studies on gender and leadership effectiveness, examining the issue from the perspective of both men and women in leadership roles.

Professor Joshi’s research focuses on gender as a social construction, rather than biological sex differences.  Much of the work that has been done about sex effects in senior management roles focuses on women, but men have gender too! Focusing on sex also may obscure variability among women and men. Through this lens, Professor Joshi investigates whether women are rewarded when they are able to navigate through masculinized environments and whether men are punished when they cannot.

Gender and Leadership in the U.S. Congress

Politics in the United States is a highly masculine context; while women have been making gains in representation, women legislators make up approximately 20% of the U.S. Congress. Professor Joshi’s first study focused on when female politicians have been successful in bringing about change: to what extent are women legislators able to get bills passed? What types of bills are they more or less successful at getting passed? Are women legislators more successful when they highlight their distinctiveness, or when they try to assimilate with the dominant group?

To test these questions, Professor Joshi utilized a sixteen-year longitudinal data set of all Congressional bills introduced from 1993 to 2008, controlling for the president in office and the majority party in Congress. In general, men were more effective than women in getting bills passed (850 to 133) and introduced a larger number of bills overall (approximately 12,000 of 70,000 were introduced by women). Professor Joshi and her colleagues examined the content of these bills and coded them on a range from more female-identified (education, healthcare) to more male-identified (defense, fiscal policy).

The results of this analysis reveal that women legislators are as good as men at getting neutral (neither male- nor female-identified) bills passed.  Men legislators were also about as effective as women at getting female-identified bills; though many men don’t think that they have the credibility to speak about “women’s issues,” they nevertheless demonstrated legislative success. The most significant finding from this sample was that women legislators were particularly successful in getting male-identified bills passed.  This finding indicates that for women in male-dominated environments, distancing themselves from their distinctive identity group may correlate with greater success.

There was also an interesting effect over the tenure of female legislators that seems to indicate both a novelty premium for new legislators and a credibility premium for more senior legislators in passing female-identified bills. This finding presents an exciting avenue for future research!

Gender and Leadership among Fortune 500 CEOs

Professor Joshi also presented a study to problematize the “think manager think male” paradigm that tends to treat men as a monolithic bloc rather than examining variability among men. This study focused on masculinity as a source of variability for men: do different levels of masculinity and femininity predict differential performance or pay among men?

To test this question, Professor Joshi and her colleagues coded interview videos of male Fortune 500 CEOs for masculine and feminine traits and compared these findings to their pay in the first year of being CEO.  Even within this highly select sample of male CEOs, there was a range of femininity and masculinity that further reinforces the idea that men are not all the same.

Among this sample, femininity did not have significant effects on pay or performance.  Controlling for the firm’s prior performance, CEOs that were androgynous were the highest performing in the sample. However, CEOs that were masculine or androgynous were the highest paid in the sample. These findings indicate that while greater masculinity doesn’t correlate with higher performance, it does relate to higher pay.

This study demonstrates, first, that there is gender variability among men, even at the CEO level. Secondly, variations in masculinity can have costs for men. Rather than affecting only women, the “think manager/think male” paradigm also has implications for men navigating masculinized working environments. Further research on gender and leadership effectiveness should be sure to take into account gendered effects on both men and women. 

Tuesday, November 7, 2017

Intimate Violence and Sexual Violence in Chinese Societies with Jia Xue

WAPPP Fellow Jia Xue, PhD Candidate in Social Welfare at the University of Pennsylvania, presented her work on domestic violence, social media, and mental health in the Chinese context.  

To begin, she described two cases of domestic violence in China: in the first, a Chinese woman, married at 26, was severely beaten by her husband. She called the police at least eight times, but they refused to intervene on the grounds that domestic violence was a “family matter.” The woman ultimately died in 2008 after ten months of marriage because of the abuse. In the other, an American woman married a Chinese man who was a famous English teacher. When her husband began abusing her, she also reported to the police who said that they could not assist her. She decided to post pictures of her injuries on Weibo, often described as China’s Twitter. Within an hour, her post had been retweeted 10,000 times and had received 4,000 comments. With the help of activists on social media, she launched a landmark case against her husband and received compensation for the abuse.

While these two cases differ on many grounds, the use of social media to speak out against abuse caught Jia’s eye. Weibo was established in 2010 and grew quickly, so that by the time the American woman chose to speak out in 2011, there was an established social media community to bring public scrutiny to her case. Social media, according to Jia, has changed the public from consumers of information to producers of information, telling the government and the media what they think should be on the policy agenda.

The growth of social media in China has been astronomical. More than half of all internet users are Chinese, and almost 90% of the Chinese population owns a mobile phone. Weibo is the leading Chinese social network and boasts more than 600 million users.

While many terms are used to describe domestic violence in the United States – including intimate partner violence, violence against women, and gender-based violence – domestic violence is the sole term used in China, and refers specifically to physical harm perpetrated by current family members. Domestic violence in the Chinese context does not include violence by former partners or among same-sex couples. While there is no national representative survey of domestic violence in China, the All-China Women’s Federation estimates that one-third of married women have experienced domestic violence in their lifetime.

Jia pointed out a twenty-year lag between the U.S. and China in terms of domestic violence legislation. The U.S. first passed the Violence Against Women Act in 1994. The Fourth World Conference on Women in Beijing in 1995 raised awareness about the issue of domestic violence, but the first provincial regulation against it was not released until 2000, in Hunan. In 2001, domestic violence was mentioned for the first time in a national-level law. Finally, after many years of advocacy, a national domestic violence law was passed in 2015, to be implemented in March 2016.

Jia’s research is at the intersection of violence and social media, using computational, big-data approaches to examine domestic violence. One of her current projects is to explore (re)tweeting behavior of domestic violence topics on Twitter. Using topic modeling to summarize and categorize tweets, it became clear that organizational Twitter accounts discuss domestic violence in the context of advocacy, while individual Twitter accounts are more likely to discuss high-profile individual cases, like those about sports or movie stars. 

In a separate study, she examined the correlation between use of traditional or contemporary media and prevalence rates of domestic violence. Survey respondents who were more likely to read books experienced fewer incidents of violence, while those who had higher levels of internet use reported more domestic violence. Jia hypothesized that more time spent online reduces communication between couples, which leads to marital tension, though the reverse could be true: in relationships with a higher incidence of violence, individuals may spend more time on the internet to be away from their spouse. The seminar audience also pointed out that there could be a generational effect: older survey respondents may be less likely to use the internet and less likely to categorize certain behaviors as domestic violence. Alternatively, there could be salience bias in this sample: survey respondents who spend more time online and consume more international news may have learned more about the issue of domestic violence and have a different view of the issue than those who consume more traditional media sources.

Jia described one final study on using social media to explore the impact of domestic violence on mental health in China. One limitation of existing studies is that domestic violence is unpredictable; it is difficult to measure mental health before and after an incident of domestic violence. With Weibo, researchers have a corpus of posts from individuals over a long period of time and can isolate messages before and after a given incident to examine the impacts of domestic violence on depression, suicidal ideation, and life satisfaction.  Jia and her colleagues amassed data from 1.6 million users, a total of 5 billion posts. The researchers selected posts based on nine types of domestic violence keywords and then manually screened posts to ensure that the poster was describing a personal incident. Ultimately, 644 posts were identified as real domestic violence cases. The next step was for the researchers to identify whether the post described the poster’s first experience of domestic violence, including if the post clearly stated that this was the first experience (approximately 70% of posts) or that the poster had never experienced domestic violence before (15%). Ultimately, the researchers ended up with 232 unique Weibo posts. Finally, the researchers selected a group of 232 Weibo users with no domestic violence experience and matched them to the existing pool based on gender and location.

For the four weeks prior to and after the first-time incident of domestic violence, the researchers used natural language processing and behavioral features to create a prediction model of users’ levels of depression, suicidal ideation, and life satisfaction.  The two matched groups did not show any significant differences in mental health before the domestic violence incidents, and the non-victim group did not show any significant changes in mental health over the eight-week period.  For Weibo users who experience intimate partner violence, 40% were more depressed after the incident, but there were no significant impacts on suicidal ideation or life satisfaction. Posters who experienced abuse from their parents demonstrated lower life satisfaction after the first-time event. Those who were exposed to domestic violence but were not themselves the victim had a higher rate of suicidal ideation and lower life satisfaction.

There are certain limitations of this study: researchers have to rely on profile information, and there may be a selection bias for posters who are willing to describe their experience on social media. However, this study was able to show short-term mental health impacts after incidents of domestic violence using innovative methods. These research findings may spur additional research or policy interventions for prevention.