CHAPTER 6

What Makes Women the “Silent Sex” When Their Status Is Low?

Early in my career, I went to numerous meetings where I was the only woman present. I would want to contribute to the conversation but would think, if I say that, everybody will think it’s really stupid.

—MADELEINE ALBRIGHT¹

MADELEINE ALBRIGHT IS one of the most accomplished women in recent US history. She holds a doctorate from Columbia University, speaks multiple languages fluently, served as the US ambassador to the United Nations, and was the first female secretary of state. Recently, she was awarded the Presidential Medal of Freedom, the highest civilian honor in the country, granted to individuals who have made “an especially meritorious contribution to the security or national interests of the United States.” Clearly, she is not defined by some want of insights or knowledge or some inability to express herself. And yet, as the epigraph indicates, even the most ambitious and accomplished women can feel less confident and speak up less often when women are scarce and the group follows a masculine norm of discussion. Describing her experiences in these early-career meetings, Dr. Albright goes on to say, “And then a man would say exactly what I had in mind and the other participants would find it brilliant.” In other words, what women have to contribute but choose not to articulate may be of great value to the group, and there is no good reason for women to hold back.

Yet in the previous chapter, we showed that women do in fact speak less than men, sometimes dramatically less. However, the size of the gender gap depends upon the nature of the group—specifically, its gender composition and its decision rule. We also showed that inequalities in speaking behavior are closely related to inequalities in authority within the group. Speech participation affects how others in the group assess their fellow group members. These key findings highlight a major theme of our book: group-level factors can profoundly alter the way men and women participate in deliberating groups and the influence they amass from their participation.

However, we want to know not only how much women and men are speaking, but also which women and men talk and which do not, and whether the answers to those questions change as the group-level context changes. We are especially interested in whether we can account for some of the mechanisms through which this inequality of speaking behavior emerges in the conditions that are especially damaging to women. To that end we turn now to exploring how the characteristics, beliefs, and preferences of men and women differ and how those characteristics, beliefs, and preferences are associated with their speaking behavior across the different conditions.

THE TRINITY OF GENDERED ATTITUDES

In chapter 3, we asked what might hold women back from exercising their voice in political discourse. There we noted that existing literature raises three possible individual-level explanations for women’s lower levels of participation: they may have less confidence than men; they may dislike conflict and competition; and they may be more empathetic and sensitive to social bonds, thus frustrated by situations where such bonds are weak or absent.

The most likely suspect is a constellation of beliefs, attitudes, and scripts related to confidence. To briefly recap our extensive discussion from chapter 3: society creates expectations based on gender, teaching its denizens that women are naturally suited to feminine roles—namely, physical attractiveness and nurturance; that women thus have less competence in areas not clearly marked as their domain; and that women who overstep these bounds deserve condemnation or at least dislike when they take overt leadership actions, assert themselves, or otherwise violate the dictates of nurturance. As a consequence, women tend to have less confidence than men, to assert themselves to a lesser, and less overt, degree, and to engage in leadership actions gingerly and with compensating behaviors that signal nurturance and connection. These are not necessarily conscious beliefs, but they tend to be well internalized and pervasive.

For example, even very accomplished women tend to have lower self-perceptions about their abilities than men, rating themselves lower on measures of competence than comparable groups of men (Beyer and Bowden 1997). In a meta-analysis of a large number of studies, Kling and colleagues (1999) find that men tend to adopt a more “self-congratulatory” view of themselves and have persistently greater levels of self-esteem than women. Women are less likely than men to view themselves as capable and efficacious in the political arena. In Mansbridge’s interviews in Selby, Vermont, women were much less likely than men to believe that “the town pays any attention to what people like you think” or to believe that they have “any say about what the town does” (1983, 311). As Fox and Lawless (2011) summarize, this tendency for women to undervalue their skills and abilities can persist even among women who have already achieved high levels of success.

Not only are women more likely to rate themselves lower on measures of confidence, self-esteem, or self-efficacy, their confidence in group settings also tends to be more sensitive to social cues, including group context. As Linda Babcock and Sara Laschever put it, “women’s feelings of self-confidence fluctuate more than men’s in response to the specifics of a situation” (2003, 142). Susan Hansen (1997) argues that settings in which politics is discussed can be especially alienating contexts for women, with the “content and style of political discourse” being off-putting. When describing deliberative settings more generally, Hibbing and Theiss-Morse (2002) assert, “The chorus in the interest-group pluralism heaven may sing with a decidedly upper-class accent, but in direct deliberation heaven it sings with a decidedly white, male, educated, confident, blowhard accent” (203). Therefore, we expect that women will not only tend to express lower levels of confidence and self-efficacy prior to deliberation, but they will also be especially sensitive to group-level cues—how does the group discussion proceed, what are the norms of interaction, how do others in the group respond to women’s initial comments? Are the speakers mostly confident “blowhards” or not? Such group-level features may have a profound effect on women, especially those who come into the conversation harboring worries about their ability to participate successfully. As the female doctor we quoted in chapter 3 described her experiences on charity committees, “You get your cues right away. I will make comments about things, but it seems that no one hears me or no one agrees with me. And then I clam up.”

Related to both confidence and increased sensitivity to the social context is women’s tendency to have greater difficulty than men rebounding from negative feedback. As we summarized in chapter 3, previous work finds that men are more likely than women to disregard negative cues, while women are strongly influenced by them (see Roberts 1991). In our earlier review of the literature, for example, we mentioned the study by Horvath, Beaudin, and Wright (1992), which found that men were much more likely to persist as economics majors when they received grades lower than B in an introductory economics course. And fMRI studies show that women are less likely than men to show resilience to negative feedback about their performance (Montague et al. 2012).

Jane Mansbridge’s interviews with residents of Selby, Vermont, show that even the prospect of negative feedback can cause women to withdraw from participation. For example, Edith Hurley describes her concerns that others will judge her negatively at the town meeting: “I don’t like to get up in town meeting and say, well, this and that … well, everybody’s looking, or doing something, and they’ll say [whisper], ‘She’s a fool!’” Florence Johnson, who has never attended a town meeting, cites similar worries when explaining why she and others don’t attend: “If you go there, and you speak up, they make fun of you for speaking up and so on, and I guess people just don’t want to go and be made fun of” (Mansbridge 1983, 61).

In chapter 3 we also raised a second possibility: perhaps women’s relative silence has less to do with confidence and resilience and more to do with the level of conflict in the group. In a study of deliberative groups discussing health care reform, Esterling, Fung, and Lee (2010) find that moderate levels of disagreement can be productive for participants and result in a rich, healthy exchange of views. High levels of disagreement, however, result in participants viewing the group as dysfunctional. The research we summarized in chapter 3 suggests that men and women may have differing preferences for conflict and that women prefer more consensual, less competitive forms of group interaction. Jane Mansbridge’s in-depth interviews in Selby, Vermont, again illustrate the point. Edith Hurley stays away from the Selby town meeting not only because she is worried about criticism, but also in part because of the disagreements she expects to face. “You get in a lot of hubbub … people get quarreling,” Hurley confides, and an older woman concurs: “I just don’t like disagreeable situations” (Mansbridge 1983, 65). Altogether, Mansbridge reports, nearly one-quarter of the residents she interviewed expressed some concern about the conflictual nature of face-to-face interactions in town hall meetings (65).² Perhaps, then, it is the conflict-averse women who withdraw from the conversation when they confront more competitive, conflict-heavy norms of discussion and interaction, the sorts of norms the gender literature tells us are especially likely to be found in groups where men outnumber women.

Our third pillar of the trinity of gendered attitudes involves the social bonds among group members. As we discussed in chapter 3, Susan Beck’s (2001) study of gender and municipal councils concludes that women are especially attuned to the level of collegiality and fellow-feeling among council members, and Lyn Kathlene’s (2001) study of Colorado state legislators finds that women were much more likely to use words that emphasized social-connectedness instead of autonomy. Similarly, Karpowitz and Frost (2007) present evidence that women are more likely than men to make communitarian arguments when giving testimony before the town council. Once again, Mansbridge’s Selby interviews illustrate the point well. One of the larger concerns she identifies is how friendships and other such close personal connections will be affected by the arguments and debates found at the town council meeting. As Phyllis Gunn laments, “They get so darned personal at town meeting!” (Mansbridge 1983, 63). Put differently, Phyllis Gunn is concerned about how the sometimes-conflictual norms of interaction at the town meeting will intrude upon the social bonds among the townspeople. Such a reaction could be the result of conflict aversion alone, but it could also be tied to a desire for a more empathetic form of communication in which emotional connections are more carefully cultivated. Thus this line of research suggests that empathy may be a key explanatory variable. Women who are oriented toward greater prosociality will withdraw from conversations, this approach predicts, when the norms of interaction neglect social bonds and the connections among deliberators.

Our review of previous work thus highlights three distinct mechanisms for women’s verbal participation: confidence and resilience to negative feedback, aversion to conflict, and feelings of empathy or prosociality. Each of these is tied to the cues women receive through group norms. Does the group welcome all views, or is it a competitive environment where only the most confident may want to venture into the fray? How much conflict is present in the group? How sensitive is the group to social bonds, and do the group members send cues that they care about their personal connections with each other?

The existing literature leads us to expect that these mechanisms will make a difference even when controlling for other factors that are commonly thought to be associated with increased participation, such as income, age, or education. Women’s lower level of confidence and increased sensitivity to social signals within the group is not merely a function of education or income. Even well-compensated, well-educated women can experience these effects. In other words, we have reason to expect that factors like confidence, comfort with disagreement, or empathy can affect the level of women’s participation in discussion even when we have controlled for women’s socioeconomic status. In this chapter, our aim is to understand the extent to which confidence, conflict aversion, and empathy help to explain gender disparities across different conditions. That is, we want to know not only what mechanisms help to explain the participation of men and women but also how those mechanisms are affected by the experimental conditions.

Prior to deliberation, we asked each participant a number of questions about their beliefs and preferences. As we saw in the previous chapter, this allows us to explore how individuals with differing predeliberation attitudes behave during group deliberation. Participants privately indicated their level of agreement or disagreement with several different statements on a 5-point Likert scale ranging from strongly disagree to strongly agree. We presented participants with statements designed to tap precisely the mechanisms we discussed in the earlier chapters. For example, our measure of confidence includes two traditional measures of internal political efficacy—“I feel that I have a pretty good understanding of the important political issues facing us today” and “Sometimes politics and the government seem so complicated that a person like me can’t really understand what is going on” (reverse coded). In addition, we asked participants to respond to statements designed to measure their sense of confidence in articulating themselves to others—“I am frequently frustrated by my inability to express my opinions to others” (reverse coded)—and their ability to effectively take part in group deliberations—“I am capable of participating effectively in group discussions about important political issues.” To these we also add two statements tapping a general sense of self-assurance—“In general, I do better on most things than most people” and “I am confident in my abilities, even when confronting tasks I haven’t done before.” These questions scale together well (α = 0.71), and factor analysis shows that they load highly on a single factor. Together, they comprise a scale designed to measure predeliberation confidence, especially confidence in one’s ability to talk with others about politics.³ For ease of analysis, we recode the scale to run from 0 to 1, with high scores indicating greater confidence.

We measure comfort with disagreement using an item inspired by Goldstein’s (1999) conflict communication scale (see also Mutz and Reeves 2005). Participants indicated their level of agreement or disagreement with the statement, “I feel uneasy and uncomfortable when people argue about politics.” Similarly, our measure of empathy is an item drawn from Caprara and colleagues’ (2005) “prosocialness” scale. This item—“I easily put myself in the shoes of those who are in discomfort”—captures the concern for the feelings and thoughts of others in the group, and the tendency to tune in, and accommodate, to their needs. Caprara and colleagues (2005) show that this item tends to produce significant differences in empathy between men and women, thus making it especially helpful for our purposes.⁴ Our empathy measure also captures well the general differences in prosocial behavior between men and women identified by other researchers (Eagly and Crowley 1986).

Participants also spent part of the time prior to deliberation reviewing information about various principles of redistribution provided to them in the participant handbook. We wanted to be sure that all members of the group shared at least a minimal understanding of the purposes and values behind different approaches to redistribution and how those different approaches might work in the context of the experiment. To test their understanding, we administered a brief, eleven-item quiz designed to review the four main principles of redistribution that were discussed in the participant handbook. As respondents completed the quiz, we gave them feedback about whether each answer was correct or not. For those who answered incorrectly, we offered an additional opportunity to respond, and then we provided the correct answer, along with an explanation of why it was correct (and why other options were wrong). Thus the quiz gave each respondent clear cues about their understanding of the principles to be discussed by the group. By the end of the quiz, all respondents should have achieved roughly equal knowledge of the definitions and purposes of the principles as well as how each principle would work if the group were to choose it during deliberation. In this sense the quiz score is not a pure measure of knowledge held during discussion, as all participants either answered correctly or were provided the correct answers.

The quiz does, however, give us a measure of how often respondents received negative cues about their understanding of the topic to be discussed by the group. We constructed a “negative quiz feedback score” variable that shows the proportion of questions for which participants received negative feedback. The fact that we provided feedback along the way means that those who answered incorrectly on the first try received negative signals about their performance. Higher negative quiz feedback scores thus indicate not only lower initial levels of performance (a fact we hoped to remedy by providing the correct answer and explanations) but also—and more importantly for our purposes—more negative cues about their understanding of the topic of group discussion. Those who answered most quiz questions correctly on the first try received very little negative feedback, and those who made more mistakes during the quiz received considerably more negative feedback. Our review of the gender literature tells us that women might be especially sensitive to these sorts of negative cues about their performance. The variable is scaled from 0 to 1, with the value representing the proportion of quiz questions for which the participant received negative cues. The lowest score thus means the participant received no negative cues, and higher scores indicate higher doses of negative cues about competence.

So far we have described the factors that we treat as the core aspects of gender. In addition, however, we want to control on related factors that are less centrally part and parcel of gender. Included among these is the now-familiar variable of egalitarianism (see the chapter 4 appendix for question wording and the online appendix for additional details). In addition, prior to debriefing, respondents indicated their age, income, education level, partisanship, and liberal-conservative ideology. The partisanship and ideology questions are standard questions asked in many political surveys, and full descriptions are available in the online appendix.⁵

Our analysis shows some differences between men and women with respect to many of the attitudes and characteristics we asked our participants to report. In table 6.1, each of the variables of interest other than age has been recoded from 0 to 1 for ease of comparison and interpretation.⁶ As can be seen in the table, the mean differences are statistically significant but tend to be small to moderate in size, ranging from a difference of 3 percentage points to as many as 11—results that are fully consistent with larger studies of individual-level gender differences (Sapiro 2003). For example, the female advantage in education is small (and does not quite achieve traditional levels of statistical significance), with both the average man and average woman in the sample having experienced some college, though having not yet completed their degrees. Similarly, the male advantage in quiz scores amounts to little more than three-fourths of a question on an eleven-point quiz, and the vast majority of both men and women score at least nine out of eleven. In general, we find the expected gender differences in basic demographic characteristics (income, education, age), social-psychological characteristics (confidence in discussion contexts, comfort with disagreement, and empathy), and political attitudes (egalitarianism and liberalism). While small to moderate in size, all of these differences run in the directions previous work on gender differences would lead us to anticipate. In this sense, our sample appears to reflect well the larger tendencies of men and women in the population.

Nonetheless, sometimes even moderate average differences can mark substantial gender disparities in the way the variables are distributed. Most important for our purposes, only 32% of women score above the sample median on our confidence measure. If we use the overall sample’s median score as a divider between high and low levels of confidence, more than 60% of men qualify as “highly confident.” That is, men are twice as likely as women to score above the median of confidence. These differences are also found at the extremes of the distribution. More than 16% of men in our sample score in the 90^th percentile or higher on our confidence measure, while less than 4% of women reach the top decile: a fourfold difference. Conversely, nearly 16% of women score in the bottom decile on the confidence measure, while less than 5% of men qualify as having such very low levels of confidence. All of these results converge on a clear conclusion: male participants in our experiment are much more likely to be found at the high end of the confidence measure, while women are found disproportionately at the low end of the scale.

We found meaningful differences between men and women in the three key factors we expect may be implicated in women’s lower participation levels. In addition, although we found small differences in negative feedback received (the quiz score), we expect that men and women will react differently to the presence of such negative responses. The key question, moreover, is whether the conditions of deliberation activate these gendered differences or suppress them.

A NOTE ABOUT METHOD

Given this set of differences between men and women, our question is how these attributes are related to women’s speaking behavior and, more importantly, how those relationships change across the experimental conditions. At the outset, we want to make clear that our analysis here is different from the experimental approach we have adopted to this point (and to which we will return in subsequent chapters). The attributes and mechanisms we will discuss in this chapter have not been experimentally manipulated. However, random assignment means they should be roughly equally distributed across the conditions (a fact that chapter appendix table A6.1 confirms). Thus our experimental research design allows us to investigate how the relationship between an attribute of the person and their talk time changes across the experimental conditions. As we explore how the connections between individual characteristics and talk time vary across different types of groups, we can gain insights into the mechanisms through which the gender dynamics we discovered in the previous chapter are produced. For example, if the relationship between confidence and talk time is the same across the experimental conditions, then confidence cannot be a key mechanism through which the differences in speaking behavior are produced. If, however, confidence becomes a more important predictor of talk in the group-level conditions where women talk least, then we have evidence that the group-level factors are affecting talk time in part by moderating the relationship between confidence and speaking behavior.⁷

The set of mechanisms we explore in this chapter also allows us to tease out differences between those factors that are, we argue, related to gender but distinct, such as demographic characteristics like income, education, or age, and characteristics that get at deeper differences between men and women. Although gender differences exist with respect to the demographic characteristics (women have lower incomes, and in our sample they are slightly older and better educated), we do not assume that such differences capture core distinctions between the genders, and we want to be sure that they are not spuriously driving the gender differences we have presented to this point. In the previous chapter, we showed that gender gaps in authority and speaking behavior were not spuriously related to preferences for principles of redistribution, agreement with the group’s predeliberation preferences, or egalitarianism. In the present analysis, we will continue to control for these differences in political preferences, but we will train our attention more directly on attributes like confidence, comfort with disagreement, and empathy. These variables capture a constellation of concepts that go beyond socioeconomic status or issue attitudes. Thus our primary focus will be on attributes that are, as we have laid out in the opening chapters, more centrally associated with gender identities.

Our estimation strategy is to regress Proportion Talk, the main variable from the previous chapter, on these attributes, while also controlling for demographic characteristics and relevant attitudes, particularly liberalism and egalitarianism. We then compare the coefficients across the experimental conditions. Because we have a limited number of men and women in the conditions with gender tokens, we pool together the conditions in which women are a minority (one-woman and two-women conditions) as well as those conditions where women are a majority (three-women and four-women conditions). While we expect the most meaningful differences among mixed-gender groups to be between the one-woman and four-women conditions, our sample size is too limited to run regressions with a full set of controls on gender tokens only. Pooling the conditions yields a sufficiently large sample size to run the models with a full set of controls. This approach allows us to test the conditions where women are more disadvantaged, relative to men—majority rule with one to two women and unanimous rule in groups where women predominate—against the other groups. Because we do not expect differences across decision rules among gender homogenous groups, and again to maximize our statistical power, we pool majority and unanimous conditions for groups composed exclusively of men or women. Thus our analysis includes five separate regressions for each gender—majority rule with a majority of women, majority rule with a minority of women, unanimous rule with a majority of women, unanimous rule with a minority of women, and gender homogeneous groups (male or female enclaves). As in the previous chapter, we also control for experimental location, given that participants were randomly assigned within, but not across, locations.

Tables 6.2 and 6.3 present the results of these OLS regressions and show how each of our key variables of interest—confidence, comfort with disagreement, empathy, and negative quiz feedback—is related to Proportion Talk in each condition. Table 6.2 includes results for women, and table 6.3 shows results for men. Controls for demographic characteristics and for liberalism are included. We employ liberalism instead of egalitarianism in these models because our measure of empathy is significantly correlated with egalitarianism (p = 0.02), but not with liberalism (p = 0.50).⁸ For the sake of parsimony in presentation, these regressions include all of our main variables of interest together, but the results are very similar if we analyze confidence, quiz score, comfort with disagreement, or empathy separately, adding only controls for the demographic characteristics and liberalism to the models.

THE SHIFTING EFFECTS OF WOMEN’S CONFIDENCE

We begin with the effects of confidence, displayed in the first row of table 6.2. Confirming our expectation, table 6.2 shows that women’s confidence in their ability to articulate opinions and participate in group discussion is strongly related to women’s proportion of the group’s talk time. But as expected, the table shows that the magnitude of the confidence effect varies across the conditions. For example, for women, the magnitude of the confidence coefficient is more than 3.5 times larger in majority-rule groups with few women than in majority-rule groups with many women. And the coefficient in the majority-rule groups with few women is over 13 times larger than in the unanimous-rule groups with few women. When the circumstances of discussion shortchange women’s authority, and women are severely disempowered in terms of talk time, confidence matters much more.⁹

Figure 6.1 illustrates our findings by showing the confidence coefficient point estimates for women and men across the mixed-gender experimental conditions. For women, we see a pattern of strong interaction between gender composition and decision rule: the effect of confidence is large and statistically significant (p < 0.03) in majority-rule groups with few women, but much smaller and not close to statistical significance in unanimous groups with few women. In groups with many women, the pattern is reversed: confidence matters more in groups with many women, with a statistically significant coefficient that is nearly twice the size of the coefficients for majority-rule groups. For women, confidence matters most in the conditions where women’s status is lowest, and the effect of confidence is greatest in those groups where women tend to face the greatest inequalities in speaking behavior—majority rule with few women and unanimous rule with many women.

As the figure shows, confidence is important for men too, but the pattern highlights large differences across decision rules and no prominent interaction between decision rule and gender composition. Although high-confidence men speak more than low-confidence men under majority rule, figure 6.1 makes clear that gender composition makes no difference to this effect. Confidence differentiates among men under majority rule, regardless of their proportion. And by the same token, it has no effect under unanimous rule, whether women are the majority or minority. For example, confidence coefficients for men in the unanimous groups are extremely small and never statistically different from 0. While confidence affects women strongly where their status is lowest, and not at all where their status is highest, it does nothing like this for men.¹⁰ Confidence does not affect men’s participation more where men have low status (or where they have high status, for that matter). That is, for women (but not for men) the effect of confidence is highly sensitive to the interaction of decision rule and gender composition.

The formal test of significant difference between the coefficients often achieves or comes close to statistical significance, despite the fact that the sample size in some of our regressions is limited (see online appendix, table C6.1). Take, for example, the difference between the role of confidence when minority women operate under majority rule rather than unanimous rule. The coefficients differ by well over 30 percentage points, and the difference is significant at the 90% confidence level (p < 0.07, two-tailed test).¹¹ The difference in the effect of confidence between minority and majority women under majority rule is smaller and not significant, but is in the expected direction. And in a simplified model in which we compare only the one-woman condition with the four-women conditions under majority rule, the difference in the effect of confidence increases by 43% from the coefficient that compares minority and majority women. These results are evidence that confidence matters more for women when women’s status is low and their Proportion Talk lags behind that of men. When the group context leaves women at a disadvantage, those who score lower on the confidence scale tend to talk far less than those who feel more self-assured about their ability to participate in the group.

These differences across conditions can be seen even more clearly in figure 6.2, which shows the predicted Proportion Talk among extremely low-confidence women, defined as those who score in the bottom decile in confidence, in each of our experimental conditions. The figure shows that in the most disempowering condition for women (majority rule with very few women in the group), the least confident women are predicted to take up only 8% of the conversation—less than half of the 20% standard of equality—and much less than low-confidence women in all the other conditions. The differences between the predicted values for the lowest-confidence women in the one-woman to two-women/majority-rule condition and the estimates for all the other conditions are significant at p < 0.05 (two-tailed). Low-confidence women rarely reach equality under any condition, but they are especially unlikely to do so when they are outnumbered by men in groups deciding by majority rule. When women’s status is lower, low-confidence women tend to drop out of the discussion. They do not leave the room, but they account for less than half of their “fair share” of the conversation.

To sharpen the point, we compare low-confidence women with high-confidence women and with low-confidence men, in the setting where women’s authority is lowest: majority-rule, minority-female groups. Figure 6.3 shows the raw percentage of men or women reaching the 20% standard of equality of talk. Where men have high status—under majority rule and high numbers—even low-confidence men reach equality of talk about half the time, which is about the same percentage as high-confidence men. Thus, although high-confidence men speak more than low-confidence men in this condition (as figure 6.1 showed), even low-confidence men regularly achieve equality of talk in the group. For women, the pattern is very different. Only 15% of low-confidence women reach equality, compared to 40% of their more confident counterparts.¹² In other words, confident women are dramatically more likely to reach the standard of equal participation we have been using. Even these more confident women reach equality less than half of the time, however. Figure 6.3 shows that high-confidence women reach equality less often than low-confidence men. When gender is a marker of status, the power of gender is strong enough that it overcomes one of our model’s strongest predictors of individuals’ speech. Where women’s status is low, confidence matters not at all to men’s ability to achieve equality in the group but makes a large difference to women’s ability.

To this point we have shown that confidence matters for women’s participation, but the size of the effect varies substantially across the different conditions. The level of authority, instantiated by the interaction of decision rule and gender composition, affects the speaking behavior of women in part by magnifying or muting the effect of confidence. The same is not true for men.

Importantly, these results are not a function of differences across the conditions in the distribution of confidence among women. Random assignment assures that confidence is roughly equally distributed across the conditions, and chapter appendix table A6.1 provides evidence that confidence and other characteristics are roughly equally distributed within the experiment. Thus our experimental research design reassures us in our explanation for why the behavior of women changes across the conditions; the effects of confidence also vary with the treatments.

CONFIDENCE AND RESILIENCE

Related to confidence is the effect of negative feedback about the predeliberation quiz. Recall that prior to deliberation, all participants took a quiz designed to test their understanding of the basic principles of redistribution that the group would discuss. If participants missed a quiz question, they were told that their answer was incorrect (and were ultimately given the correct answer). Figure 6.4 shows the effects of this negative feedback to the respondents about their answers on the quiz, as drawn from table 6.2 and table 6.3.

The figure makes clear that as predicted, women’s patterns of resilience to negative feedback differ across the experimental conditions. Unanimous rule protects women regardless of their number, while under majority rule, women require large numbers to avoid quiescence from negative feedback. Specifically, women are not affected by learning that they gave incorrect answers under unanimous rule, regardless of gender composition. Put differently, unanimous rule mutes the effects of negative feedback on women, so that the effect of the feedback does not achieve statistical significance whether few or many women are in the group. Women benefit from unanimous rule in part because it makes them more resilient to negative inferences about their competence.

Under majority rule, however, women’s numbers do matter. Negative feedback depresses the average woman’s talk substantially (p < 0.04) when women are the gender minority. In fact, that is by far the worst condition for women. However, the effect of negative feedback is much smaller and no longer statistically significant when women comprise the majority.¹³ Although not shown in the figure, the effect of negative feedback is also small and not significant in female enclaves (table 6.2). In sum, the interaction hypothesis receives partial support—being many empowers women, but unanimous rule helps them even when they are few.

Unlike the expectations of gender role theory, men also experience adverse effects from negative feedback about their performance. Under majority rule, the effects on a man of learning that he made mistakes is the mirror image of the effect on the average woman. Negative feedback has a large and strongly significant effect on men in groups with many women.¹⁴ Negative feedback also affects men under unanimous rule, regardless of the gender composition of the group. Unanimous rule does not protect men from the adverse effects of negative performance as it does women. In sum, men are most sensitive to negative implications about their competence when women are most empowered and not sensitive when women are least empowered. They react in between these extremes under unanimous rule, regardless of numbers.

The key finding about resilience in the face of negative feedback about one’s performance is that receiving negative messages about competence can reduce the verbal participation of either men or women in groups where they are the gender minority under majority rule. Under that decision rule, being a member of the lower-status gender can powerfully enhance the negative effect of bad performance on speech. Unanimous rule helps women but not men, though men are better off there than as the gender minority under majority rule.

Having shown that negative feedback and confidence are especially important for women in majority-rule groups with few women, we now ask how the two mechanisms work together. By virtue of random assignment, the experimental conditions include women at both high and low confidence levels who received high, average, or low levels of negative feedback on the quiz. Those with high levels of negative feedback are defined as those who scored in the lowest quartile on the quiz; they received negative feedback on five or more questions on the eleven-item quiz. Average levels of negative feedback include participants who scored in the middle quartiles on the quiz, receiving negative feedback between one and four times. Low levels of negative feedback are defined as receiving no negative feedback. These are participants who received perfect scores on the quiz.

Figure 6.5 presents raw average Proportion Talk for women in the condition where they are least empowered: majority-rule groups with one to two women. It shows that among low-confidence women, only those who score at the highest levels on the quiz (the top quartile of participants) exhibit a Proportion Talk even close to the 20% line of equality, and even among that group, average talk does not quite reach equality. Among the larger group of women who receive average or above average levels of negative feedback, talk time drops to half the line of equality or lower. For high-confidence women, negative feedback can also matter, but it takes a great deal of it—five or more incorrect questions on the quiz—before talk time drops well below the standard of equality. High-confidence women who score at average levels on the quiz do slightly less well than low-confidence women who received little negative feedback, and high-confidence women who scored very well on the quiz tend to participate well above the standard of equality.¹⁵ Thus high levels of negative feedback can depress the verbal participation of both high- and low-confidence women, and even low-confidence women who score moderately well on the quiz, receiving negative feedback between one and four times, tend to remain quiet during the group discussion. The key to increased participation for women in groups where they are most disadvantaged is a combination of confidence and the absence of negative signals about their competence. The most talkative women are highly confident and received positive reassurance during the quiz. And these women not only achieved equality of participation but also asserted themselves at levels well beyond the standard of equality.

HOW OTHER EFFECTS SHIFT ACROSS THE CONDITIONS OF DELIBERATION

Two other variables that we identified in chapter 3 as potential explanations for women’s reluctance to speak up—comfort with disagreement and empathy—have very little effect on women’s level of participation in the discussion. Neither variable is significant in any model of women’s Proportion Talk, nor are there substantial disparities in the effects of these variables across the experimental conditions.¹⁶ Women who are more comfortable with disagreement are as likely to speak as women who are less so. Similarly, women who are more empathetic are neither more nor less likely to speak, though the lack of effect may simply be the result of comparatively little variation among women on our empathy measure.¹⁷ We hesitate to draw definitive conclusions from these null results given that each concept is measured by only one item. In addition, our tests here only address the possibility that gender matters as an individual difference in levels or effects of the person’s stable level of empathy or of conflict aversion. We do not intend here to ask whether the group dynamic of empathy, or the level of conflict in the discussion, matter to gender inequality. We reserve those questions for later chapters. However, our results pinpoint the source of the difference between men and women in conditions that disadvantage women in women’s lower levels of confidence and resilience and not in their aversion to conflict or empathy.

The results so far have dealt with core gendered concepts, but it is worth pausing to see if the regressions in table 6.2 turned up any other interesting patterns. Because our models include a control for the participant’s liberalism, we can also explore how the relationship between issue attitudes and speaking behavior changes across conditions. We know from the last chapter that the participant’s commitment to egalitarianism does not make the effect of the conditions evaporate, but it may be that liberalism is more closely connected to speaking behavior in some conditions than in others. In chapter appendix figure A6.1, we see the familiar interaction between gender composition and decision rule for women’s (but not men’s) liberalism. When women’s status is low and their verbal participation lags behind that of men, more liberal women are the ones who are less likely to speak up.¹⁸ This effect is strongest in majority-rule groups with very few women. In the conditions where women’s status is greater, liberal women speak more than other women, though the coefficient is not statistically significant in unanimous groups with minority women or in gender homogeneous groups. For men, the effect of liberalism is typically very small and never statistically significant. To the extent that liberalism matters for men, it appears that liberal men also speak more in conditions where women’s status is greater, though this effect falls somewhat short of statistical significance.¹⁹

What can we glean from this? In our view, ideological self-placement, like the demographic characteristics we have included in our models, is distinct from attributes like confidence or empathy. Nonetheless, women do tend to be slightly more liberal than men, and in the conditions where women suffer the greatest inequalities of status and participation, women whose ideological commitments lean in this comparatively more “feminine” direction take up much less of the conversation. We will return to this issue in greater depth in the next chapter, when we explore the content of what women and men have to say, but for now, it does appear that groups dominated by men are less welcoming to at least some of women’s preferences.

Finally, the regressions in tables 6.2 and 6.3 also show that participation in the group discussion is not primarily driven by the socioeconomic variables. In the condition in which the gender disparity in talk time is the most profound—majority-rule groups with few women—women’s education is negatively related to voice. The best-educated women take up less of the group’s conversation than the least educated women. In all other conditions, the relationship is positive but small. In no conditions are income or age significantly related to talk time. Similarly, table 6.3 shows that in general, these demographic characteristics tend to have small effects that do not reach significance for male participation either.

In the political science literature on participation, socioeconomic factors have long been shown to be closely related to political participation, especially political acts that are more demanding than voting (Brady, Verba, and Schlozman 1995; Rosenstone and Hansen 1993; Verba and Nie 1972). No doubt such factors contribute to having the resources to attend meetings or other deliberative activities. But the choice to attend may be different from the patterns of talk once the meeting has begun, and within our deliberating groups, the participation of men and women is not primarily driven by socioeconomic status. Of course, our sample is quite well educated, relative to the population as a whole, so the full range of educational achievements found in nationally representative samples is not included. The vast majority of our respondents have at least some college education, and only one woman (and no men) had less than a high school education. In this sense, our results should not be seen as a refutation of the well-established socioeconomic status model. By the same token, however, our sample is more like the men and women who, previous research tells us, actually do show up at meetings or are more likely to take part in other political events—those with higher levels of education and civic skills. Our data show that within this particular deliberative setting, it is not the case that women are talking less in some conditions because they are less educated, have fewer economic resources, or are younger. In general, we find that these factors make very little difference, and when they do matter, such as in the conditions where women are most disadvantaged, the best-educated women withdraw from the conversation—just the opposite of what the socioeconomic status model would have predicted.

Our findings support the notion that norms of participation in majority-rule groups with few women are detrimental to women because women are more deterred by lack of confidence. Women who start out with lower levels of confidence and those who receive negative cues about their competence before deliberation are most adversely affected by circumstances in which women are minorities under majority rule. In addition, we also found that women whose ideologies lean in a more liberal direction speak less in these groups. While confidence also affects men, the pattern is quite different for them. Among men, confidence matters on the other side of the distribution—rather than depressing the participation of low-confidence men, predominantly male groups under majority rule elevate the participation of the more confident men. In addition, this condition prompts a masculine pattern in which less-empathetic men are especially likely to speak. Further evidence that the same circumstances prompt quite different patterns among men and women is the finding that negative feedback on the quiz has little effect on men’s speaking behavior.

These trends are not found in majority-rule groups with many women or in unanimous groups with few women. Male-dominated groups deciding by majority rule are groups where floor time goes to the confident, the less empathetic, the less liberal, and those who are best able to overcome negative feedback. All of those mechanisms favor men over women.

CONCLUSION

The last chapter showed that women are the “silent sex,” in a manner of speaking. In the settings that characterize most arenas of politics and public affairs, and in many other formal discussions that take place in civic organizations, work teams, and other common venues, women are not a majority, and the norm of interaction has masculine characteristics. Numbers and norms of interaction combine to deter women from fully expressing their thoughts. In this chapter, we took off from this point of departure, asking why women speak less than men in such settings. Just as importantly, we looked for clues to tell us why women overcome the difficulties when they are placed in other circumstances. Specifically, we asked what it is about being a woman that makes participation less likely.

The evidence we presented points toward several aspects of gender as culprits. Most importantly, confidence has much to do with women’s relative quiescence. Many more women fall into the low-confidence than the high-confidence category. Women are less likely than men to be confident in their abilities to participate effectively in group discussion, and when the group-level factors are stacked against them—such as when they are outnumbered under majority rule—such women reach equality only rarely. Women behave consistently with society’s expectation that it is not proper for women to take actions of overt agency and assertion, particularly vis-à-vis men. Women are assigned particular roles, and those roles carry less power and authority than the roles of men. Gender not only divides people into biological categories, it also ranks the social worth of these categories. As the sociologist Cecelia Ridgeway put it, “The signature of status beliefs … is that they continue to link the higher status group with greater overall competence and with whatever specific skills are most valued by the society at that time. The evaluative content of gender stereotypes has changed in recent years, with perceptions of women becoming more positive, but the essential hierarchical element has remained: Men are still evaluated more favorably in the socially important area of instrumental competence” (Ridgeway 2001, 639).

By implication, situations where women interact with men cue these expectations—unless the dynamic of interaction actively addresses women’s sense of inferiority. And indeed, we found that conditions that position women in a lower status by virtue of the combination of small numbers and a rule that disempowers those small numbers are especially bad for women with low confidence.

But this stable trait can also be affected by negative cues about women’s authority or competence. Even minor negative signals, such as one or two incorrect answers on an eleven-item quiz, can lower women’s average contribution from near equality to half that rate (as we showed in figure 6.5). However, these trait and signal effects only kick in where women’s status is at its ebb—under majority rule, when women are few in number. This apparent lack of resiliency, however, is not particular to women. Men experience the same, in a mirror image. Men are unaffected by negative feedback when women’s status is lowest, and they are most strongly affected in groups where women’s status is highest—majority rule with majority women. That is, for both genders the effect of negative quiz feedback in groups deciding by majority rule is greatest when participants are part of the gender minority under a decision rule that empowers the majority. Furthermore, unanimous rule helps both sexes, but women more. In the case of negative feedback about performance, women are no more disadvantaged than men. By implication, it is the stable trait of low confidence, more than the effects of mild levels of negative performance feedback, that accounts for women’s lower speech participation under conditions of low status.

Confidence and resilience are not the whole story, however. Even in conditions of high status, most confident women do not reach equal participation. Also important is the ideology of the person. In the conditions where women’s status is lowest, women with more liberal ideologies seem more hesitant to speak up. Given that women’s issue preferences tend to be comparatively more liberal than men’s, this finding suggests that some types of groups may be less hospitable to women’s views. To explore that possibility more directly, we must analyze what men and women actually say during the group discussion. It is to the issue content of speech that we now turn.

 

---

¹ http://www.realsimple.com/work-life/life-strategies/inspiration-motivation/how-to-tactfully-speak-your-mind-00100000081879/index.html.

² Mansbridge does not report whether there are gender differences in her sample with respect to conflict avoidance.

³ Full question wording and descriptive statistics are available in the online appendix.

⁴ Ideally, we would have included the full conflict avoidance and empathy scales, but because of concerns about completion time for the full experiment, we only had space for a limited number of items.

⁵ We placed the party and ideology variables at the end of the session in an effort to avoid priming partisan identities prior to deliberation.

⁶ The table shows predicted values from an OLS model that includes gender and controls for experimental location.

⁷ Of course, the causal story would be even stronger with an experimental manipulation of confidence, something that could be undertaken in future work.

⁸ Findings are very similar if we control for egalitarianism instead, but reducing the correlation allows for a cleaner test of the role of empathy.

⁹ Importantly, the confidence results replicate when the individual items in the confidence scale are substituted for the full scale, and the sizes of the confidence coefficients are always larger than the items tapping empathy and comfort with disagreement. The relationships between confidence and talk time are strongest for the following two items: “I am frequently frustrated by my ability to express my opinions to others” (reverse coded, p < 0.01) and “I am capable of participating effectively in group discussions about important political issues” (p < 0.01). So it is not the case that the findings are driven by the fact that confidence is measured with a scale and the others with a single item. In addition, of the two items with the strongest connection to talk time, one has a direct reference to politics, while the other does not.

¹⁰ One could argue that men’s status is “lowest” in the one condition where women achieve equality—majority rule with many women, particularly in groups with four women. For men, the largest confidence coefficient is in majority-rule groups with a majority of women, but this coefficient is not significantly larger than the effect under majority-rule groups with few women. For men, the more profound pattern is that confidence matters for talk time under majority rule, but not under unanimity.

¹¹ See online appendix table C6.1 for the regressions from which these formal tests were drawn.

¹² As a further point of comparison, low-confidence women in groups with few women reach equality close to 30% of the time when the group decides by unanimity—still a relatively low number, but nearly double the percentage of low-confidence women who reach equality under majority rule.

¹³ The difference in the coefficients does not reach statistical significance, however. Formal tests of the differences in the coefficients are shown in the online appendix table C6.2.

¹⁴ We can compare more directly the relative size of the effects on women and on men in a formal test of the differences in the coefficients (see online appendix table C6.2). In groups where women’s status is lowest, women, but not men, are adversely affected by negative feedback. As table C6.2 shows, the gender difference in the size of the negative feedback coefficients is marginally significant, but only using a more generous one-tailed test (p < 0.10). When women’s status is highest, under majority rule and a preponderance of women, negative feedback depresses men’s participation to a much greater extent than women (p = 0.03, two-tailed).

¹⁵ This high average is important but only suggestive because only two participants were in this category—one is from our outlier group, and she takes up 43% of the conversation. The other woman in this category takes up 21% of the group talk time.

¹⁶ Among men, we find in the two conditions where men significantly outparticipate women—majority rule with few women or unanimous rule with many women—more empathetic men speak less. The more talkative men are those who are less likely to express prosocial attitudes.

¹⁷ More than 75% of women agree or strongly agree that they “easily put themselves in the shoes of others.”

¹⁸ The trends are similar, though somewhat smaller in magnitude, in models that control for egalitarianism.

¹⁹ In majority rule groups with many women, the coefficient for liberal men is positive (0.26) but does not reach significance (p < 0.14, two-tailed test). Because the number of men in this condition is small, the large coefficient is still informative. See table 6.2 for details.