Why Women’s Sidelining from Debates Dealing with “Masculine” Topics Is Detrimental to Representative Democracy

For women, holding more seats in parliament and being actively involved in legislative decision-making is important for symbolic, substantial, and strategic reasons. It is an opportunity to counteract preexisting social and systemic barriers by signaling their skills and expertise (Fernandes, Debus, and Bäck Reference Fernandes, Debus and Bäck2021) and acting as role models (Catalano Reference Catalano2009; Wolbrecht and Campbell Reference Wolbrecht and Campbell2007). More substantially, it is expected to empower female MPs, enabling them to “make a difference” for women as a group. Depending on the explanandum of choice, “to make a difference” may refer to a range of desirable outcomes, including policy decisions (e.g., Atchison and Down Reference Atchison and Down2009; Atkinson Reference Atkinson2020), offices (Krook and O’Brien Reference Krook and O’Brien2012), and “politics as a workplace perspective” more generally (Dahlerup Reference Dahlerup1988, Reference Dahlerup2006). Lastly, parliamentary debates may be exploited by parties and MPs for strategic purposes (Slapin and Proksch Reference Slapin, Proksch, Martin, Saalfeld and Strøm2014). Taking the floor increases MPs’ chances to express their positions for inter- and intraparty purposes and to gain visibility in the eyes of the media and voters (Maltzman and Sigelman Reference Maltzman and Sigelman1996). Considering this, it is therefore important to assess gendered speech patterns, given that who has access to the floor may ultimately affect the legitimacy of the democratic system (Bäck and Debus Reference Bäck and Debus2019).

Scholars committed to assessing the presence of a gender gap in legislative debates have focused mainly on speechmaking. Previous research on developed countries suggests that female MPs tend to speak less, though imbalances decrease or vanish when focusing on issues that are canonically associated with women (Bäck, Debus, and Müller Reference Bäck, Debus and Müller2014). Moreover, even when no differences in speechmaking are detected, female MPs tend to deliver shorter speeches (Bäck, Debus, and Fernandes Reference Bäck, Debus and Fernandes2021). In this study, we focus on the gender gap in issue attention ( H₁ ) and language usage ( H₂ and H₃ ), as we think these two dimensions may shed light on gendered speech patterns in legislative settings, fruitfully complementing studies on speechmaking. In detail, we aim to detect the existence—if any—of a gender gap in issue attention and language usage among female and male MPs and to investigate how things change as female MPs grow from a few tokens into a considerable minority. To pursue these goals, we take as our point of departure the social role theory of sex differences and similarities (Eagly Reference Eagly1987; Eagly and Wood Reference Eagly and Wood1999, Reference Eagly, Wood, Van Lange, Kruglanski and Higgins2011), according to which men and women behave differently in many circumstances because of societal gender stereotypes. Gender stereotypes, in turn, form as people observe women and men enacting specific social roles in the contingent society in which they live and infer that the sexes possess the corresponding dispositions. These dispositions are (wrongly) thought to be stable and natural, rather than contingent and cultural. For example, in postindustrial societies, men are more likely than women to be employed in fatigue-based roles, while women are more likely than men to fill caretaking roles at home and at work. This division of labor is the outcome of the interaction between physical sex differences (men’s greater strength and women’s reproductive abilities) and the context-dependent and over-time changing demands of the social and economic environment. Gender stereotypes arise when people observe these contingent social roles and assume that they reflect men’s and women’s intrinsic traits (a process known as “correspondence inference”; see Gilbert and Malone Reference Gilbert and Malone1995). Eagly and Wood (Reference Eagly, Wood, Van Lange, Kruglanski and Higgins2011) maintain that gender stereotypes influence behavior in role-appropriate directions through gender identity and others’ stereotypical expectations. The authors explain that gender stereotypes may influence people’s self-concepts by becoming gender identities that activate self-regulatory processes. Moreover, gender stereotypes influence behavior by shaping others’ expectations. In this sense, they act as self-fulfilling prophecies: as people tend to sanction women and men for deviating from gender stereotypes, they tend to perpetuate those stereotypes by exacting costs from men and women who deviate from the norm. This expectation is supported by recent research showing that candidates’ conformity to gender stereotypes plays a role in voter turnout and election results (Anzia and Bernhard Reference Anzia and Bernhard2022).

In this respect, the social role theory typically contrasts “agentic” and “communal” attributes: men, more than women, are thought to be agentic—namely, masterful, assertive, competitive, and dominant—and they are rewarded for behaving in accordance with these perceived characteristics. Women, more than men, are thought to be communal—namely, friendly, unselfish, concerned with others, and emotionally expressive and, again, they are rewarded if they behave in accordance with these stereotypical traits. Crucially for our study, in the legislative setting, this logic should induce female MPs to specialize in the “communal issues” that are perceived to be linked to care activities and to the private sphere of the home and family, such as welfare, health care, and education. Male MPs, instead, are induced to specialize in the “agentic issues” that are perceived to be tied to the public sphere, such as economic planning, national security, and technological issues. Alternative labels used by the literature in the field are “low-profile,” “soft,” or “feminine” issues versus “high-profile,” “hard,” or “masculine” issues (e.g., Fernandes, Debus, and Bäck Reference Fernandes, Debus and Bäck2021). In what follows, we keep the communal versus agentic opposition.

We start from the assumption that women’s sidelining from debates dealing with agentic issues is detrimental to the functioning of representative democracy, as women deserve equal opportunities to have their say in all policy domains (Fernandes, Debus, and Bäck Reference Fernandes, Debus and Bäck2021; Greene and O’Brien Reference Greene and O’Brien2016). Thus, in the following sections, we discuss whether and how contextual changes occurring in parliament may overcome systematic differences in issue attention among MPs, and how female MPs may behave—that is, how they may speak—when they enter a previously male-owned issue.

The Naive Adaptation of the “Critical Mass” Concept from Physics to Politics

In this regard, interesting insights come from the use of the “critical mass” concept in gender and politics research, which can be traced back to the seminal contributions of Kanter (Reference Kanter1977a, Reference Kanter1977b) and Dahlerup (Reference Dahlerup1988). As Dahlerup explains (Reference Dahlerup2006, 512), “Originally, the term critical mass was borrowed from nuclear physics, where it refers to the quantity needed to start a chain reaction, an irreversible turning point, a take-off into a new situation or process.” By analogy, the critical mass concept has been metaphorically applied to the political realm to refer to the proportion of women needed to trigger an irreversible qualitative shift in a legislative assembly.

The baseline prediction derived from this naive adaptation of the critical mass concept from physics to politics reads roughly as follows: women in parliament are unlikely to have an impact on political and policy outcomes until they overcome a certain threshold, whose value is around 30%, after which they are able and willing to reveal their sincere preferences and coalesce. Since the 1980s, this version of the original contributions by Kanter and Dahlerup has been reframed in more deterministic and normative terms by feminist movements, activists, politicians, the media, and international organizations to convey the message not only that “numbers, or rather percentages, count” (Dahlerup Reference Dahlerup1988, 275–76), but also that numbers alone may be enough to empower women in legislative assemblies. Accordingly, this condensed version of the so-called critical mass theory (CMT) has been widely and successfully used to theoretically justify and practically support the adoption of measures to bring more women into political office.

Although acknowledging the positive effect that this rendering of the CMT has achieved around the world, many scholars have criticized it on various grounds, increasingly questioning its relevance for studying women’s political representation (e.g., Beckwith Reference Beckwith2007). The first discarded element is the idea that a single percentage—this esoteric-flavored 30%—holds the key to all women’s representation needs. Indeed, already in the late 1970s, empirical investigations highlighted that “thresholds are situation-specific” (Granovetter Reference Granovetter1978) as they result from the configuration of costs and benefits associated by female MPs with different actions in specific situations. More recently, machine-learning-based research has identified critical mass intervals for different policy areas, rather than a single critical mass point (Funk, Paul, and Philips Reference Funk, Paul and Philips2022). Other discarded elements are the ideas of irreversibility and unceasing progress, which were challenged and falsified by the backlash against women’s representation in Eastern and Central European countries. Indeed, rising numbers of women in powerful positions may generate hostile reactions among male legislators (e.g., Hawkesworth Reference Hawkesworth2003).

Still, these critiques do not imply that the (relative) number of women in legislative assemblies is totally meaningless. They ask, instead, for a closer and more exhaustive reading of the two founding authors—Kanter and Dahlerup—to uncover ambiguities and shortcomings in their original formulations without neglecting their theoretical richness. Moreover, they suggest a more guarded approach, one that allows for multiple possibilities in the relationship between women’s (relative) numerical presence in legislative assemblies and women-friendly outcomes (Childs and Krook Reference Childs and Krook2006, Reference Childs and Krook2008, Reference Childs and Krook2009).

Why We Apply the Critical Mass Argument to Parliamentary Groups Rather than to the Plenary Assembly

Besides the magic 30% tipping point and the idea of irreversibility, key aspects of this rendering of the CMT that have aroused criticism are its simplified account of legislatures inner workings and its underestimation of the ideological differences that exist among female MPs. Building on these critiques, we investigate the relationship between women’s descriptive representation and the gender gap in issue attention (i.e., gendered specialization of MPs across issues according to gender stereotypes) by looking at the parliamentary group level rather than at the plenary assembly level, thus differentiating our approach from that traditionally employed in the CMT literature. In detail, our first hypothesis reads as follows:

We justify the choice to test H₁ at the parliamentary group level on both theoretical and empirical grounds. As for the theory, it allows us to better account for the constraining and enabling characteristics of legislative contexts and their inner workings (Childs and Krook Reference Childs and Krook2006). Legislative studies have explained how newcomers in a legislative assembly are pushed to learn and conform to relevant formal and informal norms of conduct through a process of socialization with the veteran members of their parliamentary group (Asher Reference Asher1973). While this holds for every new MP, gender and politics scholars added that these norms tend to reflect a bias toward men’s experiences and authority (Hawkesworth Reference Hawkesworth2003). Accordingly, this process of socialization is likely to compel female newcomers to conform to existing legislative practices in ways that undermine their ability to work for women-friendly outcomes.

Moreover, parliamentary groups differ in their party culture and inner organization, and these differences may affect women members’ legislative behavior. Different mechanisms of candidate selection and enforcement of party discipline across parliamentary groups may affect the kinds of women who are elected, their behavior in the legislative process, and the policy positions they advocate (Cowley and Childs Reference Cowley and Childs2003). It is well known that MPs achieve the ability to challenge the party line—if they want to—the more they can count on strong local ties and individual support bases, and the more they gain seniority and material and immaterial resources (Tavits Reference Tavits2009). Usually being newcomers, and having fewer resources on average, distancing themselves from the party line is likely to be a harder choice for female MPs, and even harder if they are affiliated with parties in which candidate selection is strongly centralized and there is strong internal discipline. Empirical evidence sustaining this line of thought has been uncovered in different European countries, from the United Kingdom to Italy (Cowley and Childs Reference Cowley and Childs2003; Papavero and Zucchini Reference Papavero and Zucchini2018). Furthermore, in multiparty systems like the Italian one, cross-party cooperation is uncommon and usually considered suspicious. Accordingly, women MPs are more likely to affect decision-making processes first by trying to build consensus among their male colleagues in the parliamentary group (Dahlerup Reference Dahlerup2006).

Lastly, the focus on parliamentary groups helps us avoiding the risk of “essentializing women,” namely, treating them as a monolithic entity with a shared set of interests and beliefs because of their biological nature and downplaying their ideological differences in the name of their shared experience of marginalization (Catalano Reference Catalano2009). Indeed, women display crosscutting identities (e.g., ethnicity, income, party affiliation, feminist identity) that, in turn, affect their worldviews and policy preferences. Among these identities, party affiliation and ideology are crucial: for example, being affiliated with a party that has a more (less) progressive ideology is likely to aid (hinder) the substantive representation of women by providing female MPs with a more (less) supportive environment (Grey Reference Grey2006).

As for the empirics, the choice to test H₁ at the parliamentary group level allows us to better address the main inferential difficulty that any analysis of the effect of women presence in legislative bodies on political and policy outcomes faces: the fact that women’s institutional presence is increasing almost monotonically over time, which makes it hard to disentangle its effect from that of the social climate and value system, which are also changing over time. This empirical problem, however, is less salient when working at the parliamentary group level because there is generally a large variability in the share of women among parliamentary groups. This means that we can still compare the behavior of MPs who belong to parliamentary groups having higher or lower shares of women within the same legislature and time period (the correlation between the standard deviation of the average women’s share in parliamentary groups by legislature and year is indeed a reassuring 0.35). We will return to this point later.

Women’s Language Usage as Proxy for Their Capability to Put Forward a Different Agenda

Note, however, that finding empirical support for H₁ is not sufficient to say that as women become more numerous, they will form strategic coalitions with one another and promote legislation in line with women’s concerns. Indeed, observing a decreasing gender gap in issue attention when women increase their institutional presence would simply tell us that female MPs have been able to (partially) overcome the segregation of topics by gender, accessing previously male-owned agentic issues. Notwithstanding the desirability of such an outcome, one could still reconnect it to a mere mechanical consequence of women increased institutional presence: given that plenary time is finite and parliamentary activities need to cover a wide range of issues, for example, smaller parliamentary groups with a large share of women could be “forced” to leave them the floor on agentic issues as well. In short, observing a decreasing gender gap in issue attention would tell us nothing about how female MPs behave and what they substantially stand for when they take the floor in previously male-owned issues.

To address this point, we engage in a close reading of the original contributions by Kanter (Reference Kanter1977a, Reference Kanter1977b) and Dahlerup (Reference Dahlerup1988) and integrate their theorizations with two other prolific research streams in political science. Following Childs and Krook’s (Reference Childs and Krook2009) suggestion, we formulate H₂ and H₃ by bridging Kanter’s theory of tokenism, as enriched and extended by Dahlerup to the study of women in politics, with the literature on (1) the process of socialization to shared norms in legislative assemblies (Asher Reference Asher1973; Bowler, Farrell, and Katz Reference Bowler, Farrell and Katz1999), and (2) collective action and “activation thresholds” (to reclaim the jargon employed in Granovetter Reference Granovetter1978 and Kuran Reference Kuran1991). H₂ regards how women are expected to speak when they are in minority situations and pushed to conform to gender stereotypes; H₃ regards how this may change as they grow more numerous.

Let us start from the conjectures made by Kanter when describing what happens to women who occupy the position of tokens within a “skewed group” with a large preponderance of men. Willingly or not, the proportional rarity of tokens makes them likely to be treated as symbols rather than individuals. Tokenism generates the perceptual phenomenon of higher visibility within the group,Footnote ¹ which, in turn, generates pressure on tokens’ performance. Kanter explains that tokens may react to performance pressure in two ways.

First and most commonly, tokens seek to reduce their exposure by conforming. In this respect, we maintain that tokens act exactly like the newcomers entering a legislative body mentioned in the previous subsection, who are pushed to conform to relevant norms through a process of socialization with veteran members (the first stream of literature). As we discussed, as these norms are gendered to women’s detriment, this socialization process is likely to undermine female MPs’ involvement in legislative decision-making processes. Notably for this study, it has been shown that newcomers also conform in their speech style (Decadri and Boussalis Reference Decadri and Boussalis2020). When the presence of female MPs in parliamentary groups is scarce, the standard and shared language will be that of men, who are the “dominants.” Following this logic, when a female MP enters a parliamentary group and takes the floor on a previously male-owned agentic issue, she might be induced to adapt her language to that of her male colleagues, to show that she is conforming.

However, Kanter speculates about a second, less common, reaction to heightened visibility: overachievement. Aware of the performance pressures, some tokens are likely to stand out vividly, “to promote themselves and their work at every opportunity and let those around them know how well they were doing” (Kanter Reference Kanter1977a, 974). We maintain such a different reaction resonates with the second stream of literature. Indeed, female MPs can be characterized by different “activation thresholds,” where a threshold is simply “that point where the perceived benefits to an individual of doing the thing in question exceed the perceived costs” (Granovetter Reference Granovetter1978). The few females within a parliamentary group mostly dominated by males are likely to present activation thresholds much lower than the average. After all, they had to break the glass ceiling to get elected in a very unfavorable environment—thus, it seems safe to assume they have an extraordinary passion for politics and very strong commitments and policy preferences. These female newcomers are expected to be strong enough not to silence their policy preferences, and thus not to adapt their language to that of their male colleagues.

The idea that a few highly motivated individuals may be sufficient to achieve women-friendly outcomes—that “numbers may matter less than the presence of ‘women-identified-women”’ (Childs and Krook Reference Childs and Krook2009, 137)—is not new in gender politics research. Notably, Kanter (Reference Kanter1977b, 987) specifies that the problems of tokenism might be overcome also by tokens who are highly identified with their own social group. Moreover, Dahlerup draws attention to “critical acts” that can empower minority members (e.g., the recruitment of other women, the introduction of quotas). The adoption of these critical acts relies, again, on the existence of minority members willing to mobilize the resources of the organization or institution to improve the situation for themselves and the whole minority group (Dahlerup Reference Dahlerup1988, 296). Childs and Krook (Reference Childs and Krook2006) label these highly motivated minority members “critical actors”: namely, legislators (not necessarily women) who initiate policy proposals on their own and/or play a central role in mobilizing others for women-friendly policy change, regardless of the number of female representatives (Childs and Withey Reference Childs and Withey2006). Moreover, the literature on collective action adds that in smaller groups, coordination among individuals is easier, while free-riding is harder (Olson Reference Olson1971). This explains why smaller groups of women may be more effective than larger ones in achieving common goals. Indeed, these small, cohesive, and well-coordinated groups of female MPs may specialize in women’s concerns without appearing to undermine male domination (Crowley Reference Crowley2004).

Considering this discussion, we formulate two alternative hypotheses regarding women’s language usage, when they occupy the position of tokens in parliamentary groups with a large preponderance of men:

But what can we expect when the share of women increases—that is, when we move from “skewed” to “tilted” groups? Here as well, we can depict two alternative (but not necessarily mutually exclusive) hypotheses. On one hand, we would expect to observe an increasing convergence between female and male MPs’ policy priorities, and thus in the lexical choices they make to communicate them. Indeed, when the percentage of female MPs increases, we expect a “regression to the mean effect” in terms of activation thresholds, as it is unrealistic to think that large groups consist of a majority of highly motivated critical actors. Consequently, female MPs conform to masculine practices and adapt their vocabulary to that of male colleagues, undermining their ability to integrate their own stances into policy making.

This process of convergence between female and male language usage may happen for several reasons. Starting from “pessimistic” explanations, the process of learning and commitment to shared norms may activate itself, thus pushing women to conform to the positions taken by men on several issues. Worse, backlash effects may manifest: male MPs may enact several tactics to silence women and blocking their opportunities to freely articulate their own views (Hawkesworth Reference Hawkesworth2003; Heath, Schwindt-Bayer, and Taylor-Robinson Reference Heath, Roseanna and Taylor‐Robinson2005). For example, in the case of New Zealand, Grey (Reference Grey2006) found that male politicians reacted to the growing presence of female MPs by increasing their statements against feminist ideals. Such a backlash seems to have made female MPs more hesitant to advocate for women as a group in parliamentary debates. Kathlene (Reference Kathlene, Duerst-Lahti and Kelly1995) found similar results in the case of United States.

Moreover, as theorized by Kanter (Reference Kanter1977b, 966), with an increase in their relative numbers, minority members start to become individuals differentiated from each other. This may result in the election of a more internally heterogeneous group of female MPs, which may include women wanting to consolidate traditional gender roles (Tremblay and Pelletier Reference Tremblay and Pelletier2001) or willing to represent constituents other than women (Grey Reference Grey2006). Indeed, in larger groups coordination among individuals becomes harder, while free-riding becomes easier: in our case, thinking that enough female colleagues will continue to lobby on behalf of women as a group, some female MP owning less pronounced women-friendly preferences may decide to free-ride to foster her own priorities (Childs and Krook Reference Childs and Krook2009; Schwindt‐Bayer Reference Schwindt‐Bayer2006).

Lastly, there is also an “optimistic” explanation behind the convergence between female and male language usage. When female MPs increase their relative numbers, they may influence the behavior of their male colleagues, modifying the overall political culture in a more women-friendly direction (Dahlerup Reference Dahlerup2006) and thus leading male MPs to show more interest in women’s concerns (Bratton Reference Bratton2005). However, it is worth noting that male MPs may also decide to embed women’s concerns in their agenda for less noble strategic reasons: to occupying previously female-owned issues or restricting the margin of maneuver of female colleagues (Reingold Reference Reingold, Wolbrecht, Beckwith and Baldez2008). Considering this discussion, no matter which of these mechanisms are in place, they all lead to the same expectation regarding women’s language usage as their share in parliamentary groups increases:

On the other hand, Kanter’s most cited conjecture about how tokens’ behavior may change when their distribution becomes less extreme (i.e., in “tilted” groups)—namely, that “with an increase in relative numbers, minority members are potentially allies, can form coalitions, and can affect the culture of the group” (Kanter Reference Kanter1977b, 966)—points to an alternative to the scenario depicted in H_3a . It suggests that the convergence trend ( H_3a ), if present, ceases when female MPs become a considerable minority. At this point, they are in fact capable of forming strategic coalitions with other women to stand out, introducing their stances and constructing new standards of communication that they can now share with a more conspicuous group of peers. Moreover, larger numbers may enhance critical actors’ opportunities for critical acts as formerly passive legislators may positively respond to their calls to action, lending important momentum to women-friendly outcomes (Childs and Krook Reference Childs and Krook2009). Accordingly, we can also expect the vocabulary used by female and male MPs when addressing the same agentic issue to become more and more dissimilar as women’s numbers in parliamentary groups increase:

Note that, contrary to H₁ , which suggests a straightforward linear relationship connecting women share in parliamentary groups with the outcome (i.e., issue attention), when we consider how female MPs speak about a given agentic issue, a more complex (i.e., nonlinear) pattern may arise from the hypotheses just discussed. Indeed, by combining H_2b , with H_3a and H_3b , a U-shaped pattern may manifest, with women using a different language from men when talking about an agentic issue within a parliamentary group at the two possible “extremes”: when they are either few or relatively a lot.

The Italian Case

We test our hypotheses by running a longitudinal case study on Italy (1948–2020). Both theoretical and empirical reasons justify our choice. Italy evolved from being one of the countries in Europe with the lowest female presence in parliament to overcoming, over the past 15 years, the 30% threshold for female representation in parliaments set by international standards. Moreover, in this long time frame, the Italian political scenario witnessed dramatic changes in the parties represented in parliament and in their ideological positions, in the electoral law (with the introduction of legislative quotas in 2017), and in the formal and informal rules concerning the way political institutions operate (Russo and Verzichelli Reference Russo and Verzichelli2016).

Lastly, Italy constitutes an interesting case to test our expectations as, despite legislative quotas and the equality promotion rhetoric, the obstacles that women must face to access the political arena still outnumber the opportunities (Guadagnini Reference Guadagnini and Lovenduski2005). Regulations for balancing women’s representation partially failed because of their inaccurate design, strategic choices operated by party leaders (Sampugnaro and Montemagno Reference Sampugnaro and Montemagno2020), and voters’ electoral behavior. Female candidates must still fight in very competitive elections to win seats and discount a notorious “gender penalty” (Pansaldi and Pinto Reference Pansaldi and Pinto2020) which seems reinforced by media still promoting a collective imaginary of women as sexual objects or as mothers and housewives (Belluati, Piccio, and Sampugnaro Reference Belluati, Piccio and Sampugnaro2020).

A Novel Data Set on the Italian Chamber of Deputies

We collect all parliamentary speeches in the Italian Chamber of Deputies from the first legislature of the Italian Republic after the Constituent Assembly (1948) until February of the last republican legislature (2020). This provides information on more than 7,000 MPs over a period of 72 years. Such a long time frame allows us to evaluate the static snapshot suggested by the social role theory in its dynamic evolution, assessing whether women’s presence in parliamentary groups moderates gender gaps in issue attention ( H₁ ) and increases female MPs’ chances of drawing attention to specific stances ( H₂ and H₃ ). Indeed, if the average percentage of women within parliamentary groups has increased markedly over time, its variance across parliamentary groups has increased as well (see Figure 1). This represents an ideal scenario for testing our hypotheses.

Figure 1. Average percentage and standard deviation of women in parliamentary groups over time.

Our data set is the result of a process that involved data acquisition, data cleaning, and speech association with MPs. We collected plenary sessions transcripts from the Italian Parliament website, which are only available as PDF files in the considered time frame.Footnote ² We performed a data-cleaning step to recognize and isolate each MP’s speech throughout the transcripts and to drop speeches deemed not relevant because of their purely procedural content. Finally, given that session transcripts are not published with a systematic association between MPs and their speeches, we applied a data-matching algorithm to appropriately recognize each deputy’s name in the transcripts and associate it to their metadata. The data set contains almost 1.3 million speeches associated with Italian deputies.

Measuring Issue Attention

Our dependent variable is MPs’ attention across issues when they speak in plenary sessions, which we operationalize using a newly developed keyword-assisted topic model, keyATM (Eshima, Imai, and Sasaki Reference Eshima, Imai and Sasaki2023). One of the main advantages of using keyword-assisted classification is that researchers can label keyword topics before fitting the model. As Eshima, Imai, and Sasaki (Reference Eshima, Imai and Sasaki2023) explain, researchers must rely on their substantive knowledge of the context under scrutiny to exogenously define the topics and the corresponding keywords. When choosing each topic and topic’s keywords, is thus essential to select a reliable source.

In our case, the identification and labeling of our topics are guided by the list of standing committees of the Italian Chamber of Deputies as defined by the Rules of Procedure of the Chamber of Deputies (Rule 22).Footnote ³ The procedure generates a concise but nonetheless comprehensive list of 11 policy issues typically discussed in democratic national assemblies.Footnote ⁴ Then, to define the relevant set of keywords for each topic, we refer to the official description of each standing committee’s scope of competence, as specified by the circular issued by the president of the Chamber of Deputies on October 16, 1996, and select the most salient words.Footnote ⁵ This documentation provides an official exogenous source, which remains unaffected by the authors’ post hoc interpretation. Given the nature of the documentation used to define topics labels and keywords, topic keywords remain fixed. Still, to guarantee that our topic model can efficiently embed over-time changes in words meaning and usage, we estimate a dynamic version of keyATM in which the association between words and topic can change over time. The complete list of keywords, as well as the official committee descriptions, are available in Appendix A and Appendix B, respectively.

Table 1 presents the 11 labeled topics together with the base form of the respective sets of keywords and their classification as communal versus agentic issues according to the literature in the field (see Bäck and Debus Reference Bäck and Debus2019; Chaqués-Bonafont and Cristancho Reference Chaqués-Bonafont and Cristancho2022 for a similar classification). Among the issues considered, Productive Activities stands out as typical of the Italian definition of parliamentary committees, which separates it from the Labor and Social Policy Committee. The area of competence of the Productive Activities Committee includes industrial policy, trade, consumer protection, and business development cooperation. The Labor and Social Policy Committee has competence over social security, income support, and vocational training policies, as well as health care and protection of family, children, and the elderly.

Table 1. Labeled topics and keywords

Note: A complete list of keywords is reported in Table 2A of Appendix B.

Before estimating our semisupervised model, we apply standard preprocessing techniques like text tokenization and removal of punctuation, symbols, and stop words. We then collapse every utterance by MP and by year. This results in 33,500 observations, each representing the totality of an MP’s oral communication over the course of a year. We train the keyATM model on the yearly MPs’ speeches and obtain our main quantity of interests: per document topic proportions across all the 11 labeled topics, which reflect the main themes covered in each document.

Following best practices in automated content analysis, we propose a few validation exercises of our model to gauge its internal, external, and face validity (Grimmer and Stewart Reference Grimmer and Stewart2013). As for internal validity, we scrutinize the 20 terms with the highest membership weighting for each keyword topic and assess their consistency with their respective topic label. Our review concludes that the top 20 terms are all consistent with their respective topic labels except for the topic Transport, whose top words include many procedural terms. As for external validity, we estimate the correlation of our topics with those produced by the Italian Legislative Speech Dataset, a validated expert-based manual content analysis of Italian Parliament investiture speeches (Ceron, Curini, and Negri Reference Ceron, Curini and Negri2019; Curini Reference Curini2011). In this case as well, we obtain good levels of correlation except for the topic Transport. Finally, we gauge the compatibility of the over-time evolution of our topics’ saliency with the historical evolution of these same topics in the public debate. All our topics, except for Transport, exhibit a dynamic that is coherent with historical evolution. Based on these three validation exercises, we opt to gauge valid all our topics except for Transport, which we remove from our analysis (see Appendix D).Footnote ⁶

Variables and Model Specification

The 10 topics’ proportions constitute our dependent variables, each measuring an MP’s issue attention to different policy areas on a yearly basis. This allows us to directly test H₁ . Moving to explanatory variables, we include a dummy variable called Female that equals 1 for females MPs and 0 for males,Footnote ⁷ and a variable called Women Group that measures the share of women in each parliamentary group. As far as control variables are concerned, we consider time shocks that could alter MPs’ tendency to talk about a certain topic during a specific historical moment, each MP’s committee assignment, seniority, and, finally, her party’s left-right ideology. We capture time shocks using year fixed effects; MPs’ committee assignment using a dummy variable equal to 1 if the MP belongs to the committee in charge of the topic under scrutiny and 0 otherwise (e.g., when Y is Defense, this variable is equal to 1 for MPs in the Defense Committee, 0 for the others); seniority using the total number of terms served by an MP; and the MP’s party’s left-right ideology using a dummy variable that distinguishes between left- and right-wing parliamentary groups (see Curini and Pinto Reference Curini and Pinto2017 for a similar operationalization).

Given the nature of our 10 dependent variables, we estimate a series of Zellner’s seemingly unrelated regressions with robust standard errors clustered at the MP level,Footnote ⁸ which allow us to jointly estimate a set of regressions that exhibit a contemporaneous correlation between error terms and dependent variables (Cameron and Trivedi Reference Cameron and Trivedi2010). The first model specification is the most parsimonious and simply checks whether the Italian case conforms to the gendered specialization of MPs across issues suggested by the social role theory through the variable Female (Model 1). The second model tests H₁ by interacting Female with Women Group (Model 2). Control variables are embedded in every model specification. Notice also that several alternative model specifications with additional control variables have been run to assess the robustness of our findings (see the Robustness Checks section and Appendix G).

Robustness Checks

As anticipated in the theoretical section, the fact that women’s institutional presence is increasing almost monotonically over time makes it hard to disentangle its effect from that of the social climate and value system, which are also changing over time. This is less relevant when focusing, as we do, on MPs’ behavior at the parliamentary group—rather than at the plenary—level. Still, in this latter case, it cannot be lightly dismissed, especially once the observational nature of our data is acknowledged.

We support our main argument ( H₁ ) in three distinct ways. First, we propose a first-difference version of Model 2 focused on female parliamentarians only, whereby we collapse our observations by parliamentary group and legislature. In this model, the dependent variable is the (by-legislature) first-differenced average saliency of each agentic topic in the speeches of all women belonging to a given parliamentary group. Accordingly, our main explanatory variable is now the (by-legislature) first-differenced share of women in a given parliamentary group. We control for each topic saliency’s lagged value and for group size, and we add dummies indicating the passage of the divorce and referendum laws. Assuming that value systems and culture are relatively stable from on legislature to the next, this approach allows us to keep temporal dynamics under control, although at the expense of granularity. Results are in line with those of Model 2 (see Figure 4 and Table 8A in Appendix F).

Figure 4. First-difference version of Model 2: Women only (Model 4).

Second, we estimate Model 2 on a subsample of our original data set spanning 1975 to 2020, to test H₁ on a period that is culturally more homogeneous with respect to gender roles, stereotypes, and social norms. Indeed, since the late 1960s, Italy has experienced a gradual but constant change in the customs of its society and family organization, which used to place women in a subordinate position. Such a change achieved a pivotal point with the approval of the divorce law in 1974, and it has been described as “the anthropological change of the Italian middle classes” (Pasolini Reference Pasolini1974, 35). As shown in Figure 5, the results that we obtain are again in line with those we get when we focus on the whole sample (1948–2020) (see Table 9A in Appendix F).

Figure 5. Marginal effect of Female on topic saliency, conditional on Women Group (kernel distribution in blue): Speeches from 1975 to 2020 (Model 3). Note: Marginsplots representing the effect of Female on topics’ saliency, conditional on Women Group. Marginal effect of Female on the left-side y-axes, kernel distribution of Women Group on right-side y-axes, Women Group average values on x-axes. Marginal effects with 95% confidence interval in light blue, kernel distribution in gray.

Third, we run our analysis controlling for a linear time trend, which should plausibly capture over-time changes in all other features—for example, society’s values and cultural norms—that are likely to covary with the increase in women’s representation but are not captured by our controls. Figure 6 and Table 10A in Appendix F show that our results hold. While not completely solving the issues related to an over-time evolving value system, we believe the empirical evidence provided by our robustness checks increases our confidence in H₁ ’s plausibility.

Figure 6. Marginal effect of Female on topic saliency, conditional on Women Group (kernel distribution in blue): Controlling for a linear time trend. Note: Marginsplots representing the effect of Female on topics’ saliency, conditional on Women Group. Marginal effect of Female on the left-side y-axes, kernel distribution of Women Group on right-side y-axes, Women Group average values on x-axes. Marginal effects with 95% confidence interval in light blue, Kernel distribution in gray.

The Gender Specificity Index

Our gender specificity index is built as follows. First, for each topic, we computed a vocabulary that includes all the terms that are most frequently used when talking about it, starting from our initial set of topic keywords, augmented by other words whose scope of meaning can be considered semantically close to the topic keywords.Footnote ¹¹ For each topic, we selected the top 500 of these “closest-in-meaning words.” Second, we ranked vocabularies’ keywords, for each topic, according to their usage by male and female MPs. Based on these rankings, we then defined the set of words that is gender-specific for each topic. More in details, given a topic, a gender-specific word is a topic keyword whose position in female and male rankings differs more than a fixed threshold (i.e., 100 positions). For example, “patriarchal” is a female-specific word since it is ranked 248th in the female-topic ranking and 425th in the male-topic ranking, while “mechanical-engineering” is a male-specific word since it is ranked 65th in the male-topic ranking and 560th in the female-topic ranking. However, “modernize” is not gender specific since it is ranked 311st in the male-topic ranking and 339th in the female-topic ranking.

Finally, for each topic, we calculated the gender specificity index as the ratio between the number of gender-specific words and the size of the topic vocabulary: for instance, a value of 0.2 for the topic Productive Activities would imply that 20% of the vocabulary of that topic is gender specific. As a result, any increase in the gender specificity index corresponds to an increase in the gender-specific language used when discussing that topic. Given our theoretical interest, we selected those speeches whose speaker belongs to a parliamentary group with a percentage of women ranging between 5% and 35%. Indeed, for the Italian case, groups with these percentages of women represent the whole spectrum of theoretically relevant women-to-men proportions.

How Language Changes as Women’s Presence Increases: The U-Shaped Pattern

Figure 7 reports the levels of gender-specific language among parliamentary groups characterized by a different share of women. What we show in figure is the average value of the index for all parliamentary groups with a certain percentage of women: for example, for the 5% case, we calculate the index value for all parliamentary groups with up to 5% women, and then we take the average. For the period 1948–2020, the graphs show a curvilinear relationship between the increasing proportion of women in a parliamentary group and the level of language specificity exhibited in said group.Footnote ¹² Replicating our analysis for the period following 1975 does not change, once again, any of our conclusions (see Figure 9A in Appendix I).Footnote ¹³

Figure 7. Gender language specificity within topics: Speeches from 1975 to 2020.

Starting from the “skewed groups,” we notice that MPs’ language exhibits a high degree of specificity. This dynamic is in line with H_2b —thus, in contrast with H_2a —and suggests women accessing political groups with a strong preponderance of men tend to navigate role expectations based on gender stereotypes by overachieving. As the shares of women start increasing, though, the degree of language specificity decreases, too. This dynamic is in line with H_3a : as we move from “skewed” to “tilted” groups, we observe a decrease in language specificity. Whether for more pessimistic reasons—like a smaller proportion of women-identified-as-women—or more optimistic ones—like men changing their attitudes to express more women-friendly behavior—what we observe is in line with the idea of women and men converging in their attitudes and behavior. This dynamic only seems to be interrupted when the percentage of women is finally high enough to empower and enable female MPs to coalesce. In line with H_3b , we find empirical evidence supporting the idea that once women constitute a consistent minority, they stop acting as tokens and they stand out—diverge—to construct new standards of communication.

To give a more concrete idea of what variation in language polarization means in terms of words’ usage, we provide some examples on how the vocabulary employed by male and female MPs to talk about Productive Activities changes as the percentage of women in parliamentary group varies. In the examples, we specify proportions for transparency and clarity of presentation. Still, we point out that these percentages are not to be considered generalizable tipping points, but rather context-dependent thresholds (Funk, Paul, and Philips Reference Funk, Paul and Philips2022; Grey Reference Grey2006).

In parliamentary groups with a female presence equal or lower than 5%—skewed groups—there is a strong polarization of language: overall, words with a strong gender connotation are over 50%. When we look at these gender-specific words, women’s propensity to push forward specific subtopics in the more general Productive Activities domain emerges clearly. Indeed, women more often than men use words referring to occupational sectors characterized by a strong female presence (e.g., “paddy field,” “textile sector”), social welfare and assistance (e.g., “aging”), or the third sector more broadly. Moreover, they point to labor market contracts largely used for women (e.g., “seasonal workers”) or to negative social and labor market conditions (e.g., “illiteracy,” “pauperization,” “underemployment”), often using feminine nouns and adjectives (e.g., “fired women,” “unemployed women”). Lastly, they refer more often to productive activities linked to entertainment (e.g., “cinema,” “editorial,” “theatre”). Men, on the other hand, seem more interested in the metallurgical and raw materials sectors (e.g., “mechanical-engineering,” “Fincantieri,”Footnote ¹⁴, “oil”) and to the agriculture and livestock sectors (e.g., “granary,” “dairy industry”).

When we switch to groups with a female presence of around 20%, language polarization decreases remarkably: we reach the lowest point in the language polarization curve, with the percentage of gender-specific words decreasing to about 20% of the total. In this case, we observe how words, or synonyms of words, that were male-specific in groups with 5% of women, like “Fincantieri,” “female-farmer,” and “breeding farms,” are now equally employed by men and women.

Finally, when we switch to groups with a share of women between 30% and 35%—tilted groups—we observe a sharp increase in language polarization, with a percentage of words with a gender connotation that is higher than 60%. Men and women belonging to these groups talk similarly to those in groups with 5% women about their favorite subissues. Men more often use words related to the primary and secondary sectors (e.g., “horticulture,” “dairy,” “fishing”); women more frequently employ words related to the third sector and to welfare (e.g., “artistic,” “pauperization,” “schooling,” “marginalization”), with the notable inclusion of terms referring to women’s issues and health care (e.g., “gynecology”), maternity (e.g., “infants”), and equal opportunities (e.g., “patriarchal,” “female-citizen”).