Backlash and Moral Panic

One of the soft repression tactics that elites utilize when faced with a “crisis of hegemony” that challenges the legitimacy of the dominant ideology in society is moral panic (Hall et al. Reference Hall, Critcher, Jefferson, Clarke and Roberts2013). Moral panic happens when a social condition or a group of people is portrayed as a threat to societal values and interests by the political discourse and the media, while “moral barricades are manned by editors, bishops, politicians and other right-thinking people” (Cohen Reference Cohen2011, 1). In other words, elites respond to perceived threats against their power and hegemony by creating a moral panic against the challenging groups using their definitional power and influence over media that allows them to control the debate around controversial issues (Hall et al. Reference Hall, Critcher, Jefferson, Clarke and Roberts2013).

Using moral rhetoric, the elites portray the challenging groups as deviants who have violated the official morality of the society and are therefore detrimental to conservative values, norms, and social order (Becker Reference Becker1967; Cohen Reference Cohen1991). Moral panic is not necessarily a genuine panic, and one of its features is the small or negligible likelihood of social harm compared to the response of the dominant groups. The moral panic serves to orchestrate public opinion by creating negative emotions of fear and anger, and even inciting violence. It therefore justifies the disproportionate reaction of elites to a challenge that has risen due to serious structural problems that resolving them would undermine the dominant hegemony (Cohen Reference Cohen1991; Goode and Ben-Yehuda Reference Goode and Ben-Yehuda1994; Hall et al. Reference Hall, Critcher, Jefferson, Clarke and Roberts2013; McRobbie and Thornton Reference McRobbie and Thornton1995).

During the “mugging crisis” in the early 1970s in the US, a moral panic was created by the dominant groups including the police and government that portrayed the black community as the cause of increasing social problems. This moral panic allowed the elites to suppress a crisis of legitimacy through adopting increasingly harsh and authoritarian measures (Hall et al. Reference Hall, Critcher, Jefferson, Clarke and Roberts2013). Similarly, in Europe, a moral panic surrounding Middle Eastern immigrants led to the normalization of racism and provided justification for implementing restrictive immigration policies (Krzyżanowski Reference Krzyżanowski2020).

In the context of gender quotas and women’s presence in politics, dominant actors employ moral panic to depict women politicians and progressive policies as violating the official morality of society, traditional values, and institutions. These actors consist of elites such as male legislators as well as conservative civil society groups like religious fundamentalist societies that have been on the rise over the past few decades and have influence over media (Biroli and Caminotti Reference Biroli and Caminotti2020; Fowler et al. Reference Fowler, Youngs, Larok, Marczewski, Vijayan, Nodia, Shapovalova, Sombatpoonsiri, von Bülow and Zihnioğlu2018; Panchaud et al. Reference Panchaud, Keogh, Stillman, Awusabo-Asare, Motta, Sidze and Silvia Monzón2019; Pousadela and Perera Reference Pousadela and Perera2021). Furthermore, in African countries, the media landscape is largely controlled by the state or the political elites and serves dominant interests and ideology. In the case of Zimbabwe, for instance, the ruling party ZANU-PF controls Zimbabwe Broadcasting Corporation and did not issue new broadcasting licenses in the country (Mano Reference Mano, Flew, Iosifidis and Steemers2016; Sparks Reference Sparks2009).

The example of Uganda’s fiercely contested Marriage and Divorce Bill — aimed at regulating marriage, separation, and divorce — illustrates the strategic use of moral panic. In the highly publicized 2013 debate over the bill, an alliance comprising religious institutions, traditionalists, and proponents of the patriarchy, led by male legislators, successfully defeated the bill. These groups, such as the Uganda Joint Christian Council and the Inter-Religious Council of Uganda, argued that it would disrupt traditional gender roles and family values, promote sexual promiscuity, and tarnish the sanctity of marriage.Footnote ¹ The women politicians sponsoring the bill were branded prostitutes seeking to “spoil the institution of marriage so that everyone would become a prostitute” (Larok Reference Larok2018, 160). This rhetoric incited nation-wide opposition and protests against the bill, resulting in its defeat.

Women politicians themselves are often accused by their male rivals and in the news reports of violating gendered norms and values such as modesty and sexual propriety, disrespecting their husbands, and disregarding their families as well as their duties as wives and mothers once they enter politics. For instance, when a well-known Kenyan politician Gladys Boss Shollei got a divorce, the reports, apart from covering all the gory details of the scandal with sensationalized titles such as “politics and marriage make a highly volatile mix for women,” blamed her political career as the reason for the breakdown of her marriage and questioned Shollei’s leadership capability as she had failed to “nurture a family.”Footnote ² In another article that was widely circulated online, the husband of a woman politician lamented allowing his wife to go into politics as she had become a prostitute and had illicit relationships with her male colleagues.Footnote ³

As people seldom have direct contact with politicians, the public’s perceptions about them are predominantly shaped through mediated exposure (Bligh et al. Reference Bligh, Schlehofer, Casad and Gaffney2012; Chong and Druckman Reference Chong and Druckman2007; Reese, Gandy, and Grant Reference Reese, Gandy and Grant2001). The nature of this mediation, as documented by a robust scholarship, is gendered and exposes existing power dynamics in society (Aaldering and Van Der Pas Reference Aaldering and Van Der Pas2018; Courtney et al. Reference Courtney, Breen, McGing, McMenamin, O’Malley and Rafter2020; Gidengil and Everitt Reference Gidengil and Everitt1999, Reference Gidengil and Everitt2003; Goodyear-Grant Reference Goodyear-Grant2009). A comprehensive meta-analysis of 90 studies on gendered political news coverage confirms the presence of gender bias. Women politicians often face heightened scrutiny of their appearance and personal lives, receive more negative assessments of their viability, and encounter a greater emphasis on gender stereotypes in relation to their traits and policy issues (Van der Pas and Aaldering Reference Van der Pas and Aaldering2020).

This mediated exposure, influenced by dominant actors with definitional power, plays a crucial role in the construction of symbolic representation. The effect of symbolic representation resides in its capacity to evoke “emotional, affective, irrational psychological response” (Pitkin Reference Pitkin1967, 100). A symbol is politically constructed and associated with certain representative meanings by shaping social habits and practices through manipulating the emotions and attitudes of the public (Lombardo and Meier Reference Lombardo and Meier2014; Pitkin Reference Pitkin1967). Consequently, this process perpetuates patterns of domination and inequality, reinforcing privileged and marginalized positions within society (Connell Reference Connell2002; Dovi Reference Dovi2007).

From this perspective, the symbolic effect of women’s representation and the social meaning that their political presence generates largely depends on the affective responses that people have been trained to experience upon exposure to women politicians who “stand for” women citizens in general (Lombardo and Meier Reference Lombardo and Meier2014; Meier Reference Meier, Franceschet, Krook and Piscopo2012; Pitkin Reference Pitkin1967). Research indicates that the moral rhetoric of elites can strongly influence individuals’ opinions and attitudes by eliciting powerful emotions (Clifford et al. Reference Clifford, Jerit, Rainey and Motyl2015; Clifford and Jerit Reference Clifford and Jerit2013; Graham and Haidt Reference Graham, Haidt, Mikulincer and Shaver2012; Tamborini et al. Reference Tamborini, Prabhu, Lewis, Grizzard and Eden2018).

Through the use of moral rhetoric, therefore, elites can construct or reinforce the connection between a symbol and its meaning. Consequently, by generating a moral panic against women politicians and moralizing women’s presence in politics, elites establish a link between women’s political representation and either the violation or the necessity of upholding conservative values. The moralization of an issue often bears negative consequences as it has the potential to elicit strong negative emotions, provoke hostile attitudes, and lead to punitive actions (Kraaijeveld and Jamrozik Reference Kraaijeveld and Jamrozik2022; Ryan Reference Ryan2014). In this study, I employ moral foundations theory to measure the presence of conservative moral rhetoric in news coverage of women politicians.

Backlash and Type of Quota

There are different types of quota policies — including reserved-seat, legislated candidate, and party quota models — that provide elites with varying levels of control over implementing quotas provisions (Darhour and Dahlerup Reference Darhour and Dahlerup2013; Krook Reference Krook2007). Reserved-seat quotas allocate a specific number of legislative seats to women by law and ensure the electoral outcome (Dahlerup Reference Dahlerup2005). Legislated candidate and party quotas target the gender composition of candidate lists, with the difference that the former is binding for all parties, whereas the latter is voluntarily adopted at the parties’ discretion.

Reserved-seat models are typically adopted in a top-down manner and imposed on elites when a strong executive seeks to maintain their strategic interests and dominance, or in response to pressure from international actors rather than as a measure to address the under-representation of women (Chowdhury Reference Chowdhury2002; Darhour Reference Darhour, Darhour and Dahlerup2020; Krook Reference Krook2006; Muriaas and Wang Reference Muriaas and Wang2012). Party quotas, in contrast, are often presented as political parties’ commitment to promoting gender equality. However, scholars argue that the decision to adopt party quotas is driven by various factors, including pressure from the electorate, a desire to improve the party’s electability by attracting support from women voters (Tripp and Kang Reference Tripp and Kang2008), and an attempt to consolidate their power in relation to other candidates while maintaining control over the nomination process (Baldez Reference Baldez2004). Furthermore, evidence indicates that in single-party dominant countries such as South Africa and Namibia, the implementation of party quotas can lead to a significant increase in women’s presence in the legislature similar to the reserved seat model.

Elites in countries with party quotas have the ability to utilize informal tactics, such as allocating women to unwinnable positions, manipulating election outcomes, and maintaining existing power dynamics, even with the presence of quotas (Verge and Troupel Reference Verge and Troupel2011). In contrast, elites in countries with reserved-seat models do not have access to such strategies. As a result, I argue that elites in reserved-seat model countries are more likely to perceive quota reforms as a threat to their privileges. To counter this perceived threat, they may employ soft repression tactics, such as generating moral panic, to mobilize the public against women’s political participation. Based on this analysis, I propose the second hypothesis:

Sample Selection

In this study, I focus on 10 selected countries in sub-Saharan Africa, including Nigeria, Namibia, Zimbabwe, Zambia, Uganda, Ghana, Rwanda, South Africa, Tanzania, and Kenya. These countries were selected based on several key considerations, including the variation in the percentage of women’s representation, the diversity in the adoption and types of quota policies, and the availability of data. Some of the countries in the sample have witnessed an increase in women’s descriptive representation attributed primarily to the implementation of gender quotas (Barnes and Burchard Reference Barnes and Burchard2013; Bush and Zetterberg Reference Bush and Zetterberg2020; Edgell Reference Edgell2017; Hughes and Tripp Reference Hughes and Tripp2015).

The selected sample offers a wide range of variation in terms of women’s descriptive representation. For example, at the beginning of the study, women occupied approximately 4% or less of the seats in both Nigeria and Kenya. While the percentage has remained relatively unchanged in Nigeria, it significantly increased to more than 21% in Kenya after the implementation of the reserved-seat quota in 2013. These countries also exhibit diversity in the adoption and implementation of quota systems. Ghana, Nigeria, and Zambia have not adopted quotas, whereas Kenya, Zimbabwe, Rwanda, and Namibia implemented the policy during the study period (Clayton Reference Clayton2014). South Africa, Tanzania, and Uganda had quotas in place before 2000. Kenya, Zimbabwe, Rwanda, Tanzania, and Uganda have implemented the reserved-seats model, whereas South Africa and Namibia adopted voluntary party quotas. This variation provides an ideal environment for investigating the potential backlash effect against women politicians following the implementation of gender quotas.

To gather newspaper articles for content analysis, I utilized the Nexis Data Lab service, a cloud-based platform that offers access to news articles from around the world. However, it is important to note that the availability of articles for non-English-speaking countries in sub-Saharan Africa was limited within the archive, and the coverage in these counties spanned a short period of time. Given this consideration, a careful assessment of data availability led to the selection of 10 countries in the region to ensure variation in terms of the percentage of women’s representation and the presence/type of quota policies.

Although the selection of these 10 countries provides valuable insights, it is important to acknowledge certain limitations in case selection. The lack of data for non-English-speaking countries in sub-Saharan Africa may limit the generalizability of findings. Nevertheless, the selected countries offer a meaningful context for investigating the potential backlash effect against women politicians following the implementation of gender quotas. Future research can expand the case selection to include a broader range of countries and incorporate additional sources of data to further enhance our understanding of this phenomenon.

News Collection Process

To obtain news articles about politicians in sample countries, I first collected the names of more than 20,000 aspiring former and current members of parliaments in these countries through official outlets including parliament websites and election results published by governments. Then I identified their gender by consulting various sources, including governmental gazettes, news articles, and social media platforms. This dataset was then used to collect around 500,000 articles published between 2000 and 2020 in the sample countries from Nexis Data Lab service, a recently launched cloud-based platform that provides access to news around the globe. The selection criterion is that politicians should be discussed in the headline or the lead, which ensures they are the main focus of the article (O’Keeffe and McCarthy Reference O’Keeffe and McCarthy2010). Using the data on the gender of politicians, I dropped articles where names from both genders were included in the lead.

As the number of articles for men far exceeds those on women, I randomly selected an equal number of stories about men to match the size of the sample for women to create a balanced corpus and ensure that results are not driven by sample size (Hopp et al. Reference Hopp, Fisher, Cornell, Huskey and Weber2020). After removing duplicate titles and articles published by non-domestic news outlets, the final corpus included more than 150,000 articles in which the sample size for each gender is roughly the same across the corpus (a total of approximately 75,000 for men and women each) within each country. Supplementary Figure 1 in SI showcases the distribution of articles in sample countries.

Measuring Moral Rhetoric in Text

For measuring conservative moral language in news articles, I utilize the Extended Moral Foundation Dictionary (eMFD) developed by Hopp et al. (Reference Hopp, Fisher, Cornell, Huskey and Weber2020), which builds on the original and relatively small Moral Foundations Dictionary (MFD) (Graham, Haidt, and Nosek Reference Graham, Haidt and Nosek2009). The eMFD offers an open-source Python package (eMFDscore) for analyzing text.Footnote ⁶ The developers of eMFD constructed their dictionary by leveraging an extensively validated, crowd-sourced annotation procedure. Moreover, the eMFDscore is trained on a corpus exclusively build from newspaper articles, and it has been utilized in several recent empirical analyses (Hopp et al. Reference Hopp, Fisher, Cornell, Huskey and Weber2020; Malik et al. Reference Malik, Hopp, Chen and Weber2021). These advantages are important for the task of extracting moral rhetoric from news and make eMFD a well-suited method for this study.

The eMFD takes into consideration the context in which the word appears and detects latent, morally relevant information in the text, whereby words are assigned a probability of belonging to a foundation. For example, the word kill might belong to the foundation of care/harm with a probability of 80% and the foundation of loyalty with a probability of 30%. The open-source text analysis tool developed for scoring the documents — eMFDscore — allows the researcher to score the document using only the foundation with the highest probability (single vice-virtue method), therefore avoiding correlation among various foundations. Here, I use “bow emfd single vice-virtue” to extract moral language from the text corpus. Supplementary Table 1 in SI provides an example of the package’s output.

Dependent and Independent Variables

Dependent variable — The dependent variable is a [0, 1]-bounded conservative moral rhetoric score that measures the salience of binding moral foundations invoked in articles. Higher values of the dependent variable indicate higher salience of conservative moral rhetoric. Previous studies have measured moral rhetoric in terms of separate foundations (e.g., purity/degradation, care/harm). Recent research, however, suggests foundations cluster together, and invoking one of them triggers other intuitions as well. These studies, therefore, use aggregate moral foundation scores (Malik et al. Reference Malik, Hopp, Chen and Weber2021). Following this approach, after obtaining the scores for each binding foundation separately, I aggregate them to create the conservative moral rhetoric score to better capture the overall proportion of moral rhetoric in text. Figure 1 shows the annual trend of the conservative moral rhetoric score in news coverage of women politicians for each country in the sample over the duration of the study.

Figure 1. Trends in the salience of conservative moral rhetoric in news coverage of politicians, 2001–2021. The dashed lines indicate year of quota adoption. As can be seen in countries with reserved-seat quotas — Kenya or Uganda for instance — there is a surge in the salience of conservative moral rhetoric compared to men and relative to the pre-quota period. It is noteworthy that in the case of South Africa, although quotas were in place prior to 2000, the country’s ruling party — the African National Congress — adopted a 50% gender quota in national elections 2009 when we observe an upward trend in the salience of conservative moral rhetoric in the news of women politicians. In countries without quota, however, the salience slightly varies or decreases for women across the years. Note: The annual values of the conservative moral rhetoric score are created by summing up and normalizing the scores of individual binding foundations for all news articles published by domestic news outlets in one year in each country.

Independent variables — Main independent variables include the percentage of women’s descriptive representation, quota adoption, type of quota system, and the gender of the politician discussed in the article.

Data on the proportion of seats held by women in single or lower chambers of national parliaments is obtained through the Inter-Parliamentary Union archives.Footnote ⁷ Figure B in SI demonstrates the changes in the proportion of women in parliaments across countries in the sample. The gender of the politician is a binary variable that indicates whether the lawmaker discussed in the article is male or female.

The binary variable of quota adoption indicates whether and when a country has adopted the policy, the data on which is collected from the International IDEA.Footnote ⁸ The type of quota is a categorical variable with three levels of no quota, reserved seat, and party quota, which is also created using data from the International IDEA. As the other main independent variable is the percentage of women’s representation in single or lower chambers of parliament, quota type is determined based on the affirmative action implemented in the lower chamber in case of bicameral countries.

Empirical Strategy

In this study, I analyze cross-national, longitudinal, and hierarchical data from domestic newspapers in the sample countries using a random-effects model. I also estimate fixed-effects models and present the results — that are consistent with random-effects models — in the SI. Clustering of observations in longitudinal data introduces two types of variances for variables: within- and between-country that correspond to two different relationships in response to a variable (i.e., percentage of women in single or lower chambers of parliaments). Random-effects estimation combines both types of variation for estimating the effect of the explanatory variable on the outcome. It accounts for the dependence of observations in the same group — articles published within the same country — and therefore leads to unbiased estimates of standard errors.

Scholars have recently begun to use a hybrid method that disentangles within- and between-unit variation errors (Barr et al. Reference Barr, Levy, Scheepers and Tily2013; Bell and Jones Reference Bell and Jones2015; Certo, Withers, and Semadeni Reference Certo, Withers and Semadeni2017; Malik et al. Reference Malik, Hopp, Chen and Weber2021). In this method, the between variable is calculated by obtaining the cluster mean of the explanatory variable $ \overline{x_i} $ and within variable by obtaining the group-centered variation $ \left({x}_{i\mathrm{j}}-\overline{x_i}\right) $ as shown in Equation (1):

(1) $$ {y}_{ij}={\beta}_1\overline{x_i}+{\beta}_2\left({x}_{ij}-\overline{x_i}\right)+{u}_i+{e}_{ij} $$

where within-country variation — $ {\beta}_2 $ — cannot be correlated with country-level effects $ {u}_i $ , as $ {u}_i $ does not vary within countries. By disentangling the variation in the data, the coefficients capture the within-country effect, specifically how changes in women’s descriptive representation within a country correlate with the salience of conservative moral language in interaction with quota adoption or the type of quota system.

I run two multilevel models for testing my proposed hypotheses, as shown in Equations (2) and (3). In Equation 2, the dependent variable is conservative moral rhetoric index that measures the salience of moral language in observation (article) $ i $ within the publication $ j $ in country $ k $ at occasion (month) $ t $ . $ Woman. Representation{(demean)}_{kt} $ indicates the within-country variation in the percentage of women’s parliamentary presence in country $ k $ at occasion $ t $ , whereas $ Quota.{Adoption}_{kt} $ indicates whether country $ k $ has implemented quota policy at occasion $ t $ , and $ {Gender}_i $ refers to the gender of politicians discussed in the article. $ Woman. Representation{(mean)}_k $ measures the time-invariant mean of women’s parliamentary presence in country. In Equation 3, $ Quota.{Type}_{kt} $ is the type of quota implemented in country $ k $ at occasion $ t $ . $ {u}_k $ , $ {u}_j $ , and $ {u}_t $ are random effects for country, publication, and occasion (month), respectively.

(2) $$ {\displaystyle \begin{array}{c}\hskip-30.12em Conservative. Rhetoric. Scor{e}_{\left\{ ijkt\right\}}\\ {}\hskip10.12em ={\beta}_1 Woman. Representation{(demeaned)}_{\left\{ tk\right\}}+{\beta}_2 Quota. Adoptio{n}_{\left\{ tk\right\}}\\ {}\hskip-1.12em +{\beta}_3 Gende{r}_i+{\beta}_4 Woman. Representation{(mean)}_k\\ {}\hskip10.12em +{\beta}_5 Woman. Representation(demeaned) \times Quota. Adoptio n \times Gender\\ {}\hskip-18.0em +{u}_k+{u}_j+{u}_t+{e}_{\left\{ ijkt\right\}}\end{array}} $$

(3) $$ {\displaystyle \begin{array}{c}\hskip-30em Conservative. Rhetoric. Scor{e}_{\left\{ ijkt\right\}}\\ {}\hskip7em ={\beta}_1 Woman. Representation{(demeaned)}_{\left\{ tk\right\}}+{\beta}_2 Quota. Typ{e}_{\left\{ tk\right\}}\\ {}\hskip1em +\beta \_3\; Gender\_i+\beta \_4\; Woman. Representation{(mean)}_k\\ {}\hskip8em +\beta \_5\; Woman. Representation(demeaned) \times Quota. Typ e \times Gender\\ {}\hskip-18em +{u}_k+{u}_j+{u}_t+{e}_{\left\{ ijkt\right\}}\end{array}} $$

The main term of interest in both models is the three-way interaction, which indicates the effect of the increase in the percentage of women in the legislature on the salience of conservative moral rhetoric in news coverage of women compared to men, conditional on the adoption or the type of gender quotas. Analyzing articles for both men and women serves to capture the overall salience of conservative moral rhetoric in the news coverage of politicians in the country. There is no reason to assume that the proportion of moral rhetoric in the news about men is meaningfully associated with the percentage of women’s parliamentary presence. Therefore, the result of interaction would reveal whether changes in the salience of conservative moral rhetoric for women are the result of overall trends in the endorsement of moral foundations in the countries under study or associated with women’s presence in politics.