WHERE ARE ALL THE (FEMALE) ACTIVISTS?

To build our argument about cross-national levels of participation in #MeToo, we draw on three related literatures: (1) general mass mobilization, (2) digital activism, and (3) movement diffusion. First, what leads to higher levels of mass mobilization within a country? Cross-national work on the determinants of involvement in the women's movement or in women's offline protest is still relatively rare (Fallon and Rademacher Reference Fallon, Rademacher, Alexander, Bolzendahl and Jalalzai2018; Murdie and Peksen Reference Murdie and Peksen2015). Most studies focus on women's participation in North America and Western Europe only (Costain Reference Costain1992; Soule et al. Reference Soule, McAdam, McCarthy and Su1999). The studies that do exist, however, often find very similar patterns of women's participation in protest to that of protest that does not center on women's issues, especially when compared with other historically disadvantaged groups (Murdie and Peksen Reference Murdie and Peksen2015; Bell et al. Reference Bell, Murdie and Peksen2019).

The larger literature on protest offers many important arguments on the drivers of mobilization. First, early scholarship highlighted the role of grievances and feelings of relative deprivation, often connecting these ideas to the overall use of violence by the government (Chenoweth and Ulfelder Reference Chenoweth and Ulfelder2017; Gurr Reference Gurr1968). Repression and injustice can mobilize individuals to take to the streets. Research on women's protest has similarly found that a lack of respect for women's rights often accompanies mobilization (Bell, Murdie, and Peksen Reference Bell, Murdie and Peksen2019; Costain Reference Costain1992; Murdie and Peksen Reference Murdie and Peksen2015).

Second, staging or joining a protest is costly, requiring both human capital and material resources (McCarthy and Zald Reference McCarthy and Zald1977). Much research has found that the presence of established civil society groups, greater individual wealth, and an educated population make mobilization easier by reducing individual barriers to participation and increasing the resources through which groups can mobilize for protest events (Chenoweth and Lewis Reference Chenoweth and Lewis2013; Dalton, Van Sickle, and Weldon Reference Dalton, Van Sickle and Weldon2010; Murdie and Peksen Reference Murdie and Peksen2015). This is especially true for nonviolent protest mobilization.

Many studies on the role of globalization in protest focus on how the informational resources provided by globalization may aid in mobilization. For example, the spread of information and interpersonal connections can help with coordination and reduce duplication of effort, allowing country-level movements to learn from each other and share resources (Bell, Murdie, and Peksen Reference Bell, Murdie and Peksen2019). Globalization can also make it easier for transnational nongovernmental organizations to be involved within a country, increasing the ability of local actors to coordinate and plan protests (Keck and Sikkink Reference Keck and Sikkink1998; Murdie and Peksen Reference Murdie and Peksen2015; Olzak and Tsutsui Reference Olzak and Tsutsui1998).

Much research also stresses the importance of political opportunity structures. Protest is more likely when individuals believe the political structure will allow mobilization and that mobilization has some chance of being successful (Costain Reference Costain1992; Lichbach Reference Lichbach1998; Murdie and Peksen Reference Murdie and Peksen2015). The opportunity structure also colors the tactics used; more violent tactics are more likely in closed political systems (Meyer Reference Meyer2004; Schock Reference Schock2004). Some researchers have suggested a curvilinear relationship between protest mobilization and political opportunity structures; after a certain level of openness, citizens may move to more insider political tactics, such as directly engaging with legislators, as opposed to taking to the streets to demand political change (Dalton, Van Sickle and Weldon Reference Dalton, Van Sickle and Weldon2010; Meyer Reference Meyer2004). Nonetheless, protest is less likely in closed political systems.

DIGITAL ACTIVISM

The literature discussed here is useful in understanding the cross-national factors that make protest more likely in some countries. However, much of this literature predates the internet and is centered on both nonviolent and violent forms of street protest. Are the drivers of digital activism similar? To start, many scholars stress that digital activism is somewhat less costly than traditional mobilization. For example, Earl and Kimport (Reference Earl and Kimport2011, 90) contend that “when participation can occur online, participation will generally be less costly and thus should spur flash-style mobilizations in which large numbers of people participate.” Compared with organizing or participating in a march, retweeting a hashtag or adding your name to an online petition takes less time and energy. Some digital activism is anonymous, limiting the potential costs of reprisal. In line with these arguments, digital activism could revolutionize contentious politics (Hands Reference Hands2011; Van Laer and Van Aelst Reference Van Laer and Aelst2010). Internet access becomes a key driver of revolution.

When it comes to #MeToo activism, however, and other digital feminist activism, there are some heightened costs of participation. Mendes, Ringrose, and Keller's (Reference Mendes, Ringrose and Keller2018, 242) online survey of digital feminist activists found that 72% had faced hostile trolling on Twitter after posting feminist views; this trolling ranged from taunts or slurs to “violent and graphic rape and death threats.” The human rights organization Amnesty International (2018) released a report on “Toxic Twitter,” noting that violence toward and abuse of women online may include the sharing of private information such as pictures and home addresses. These threats may be more than what one would receive for attending a feminist rally, and, even though the threats happen online, the fear of physical violence is often justified.

In addition, participation in digital feminist activism is likely influenced by political opportunity structures: if individuals are going to be involved in a movement, there has to be some general belief that the movement will be successful and that the government will not repress their mobilization. In addition, online resources are often coupled with offline mobilization. In this regard, digital activism is synergistic with traditional protest (Della Porta and Mosca Reference Della Porta and Mosca2005). This is in line with Hill's (Reference Hill2013, 6) discussion of the role of domestic political structures in the digital age:

Given the widespread prevalence of sexual harassment and assault around the world, it is likely that digital feminist activism has motivations similar to the relative deprivation or grievance arguments from traditional scholarship. However, with digital feminist activism especially, the political opportunity structure may make it more or less likely for individuals to speak out about their grievances. In many regards, this is similar to the reporting of sexual assaults to police forces. Countries with supportive political and legal systems often have a higher level of reported rapes, not because of a higher overall level of offenses but because of the “improved legal and social position of women” (Van Hofer Reference Von Hofer2000, 85). We expect this logic to inform involvement with the #MeToo movement as well.

DIFFUSION OF MOVEMENTS

When focusing on #MeToo, we are interested not only in what drives activism or protests but also in what drives involvement in an international movement, specifically a movement originating in the United States. How does a movement spread across borders? Studies on the diffusion of movements suggest several diffusion mechanisms (Strang and Soule Reference Strang and Soule1998). From contagion to social learning processes, these mechanisms stress international influences and how domestic actors are affected by foreign influences. The notion of contagion effects views the diffusion process as unpredictable and spontaneous. Protestors in a country observe protests in other countries and mimic the collective action without careful planning.

Social learning provides a more nuanced explanation of diffusion processes. Rather than simply copying other countries’ protests, collective actors vigilantly learn strategies and tactics from other countries. Either way, protests in other countries create demonstration effects. Potential protestors and activists will observe a successful movement in a country and try to organize the movement in their country (Kuran Reference Kuran1991; Tarrow Reference Tarrow1994; Weyland Reference Weyland2012). Weyland (Reference Weyland2012) argues that since discontented people face tremendous uncertainty, they rely on cognitive heuristics before organizing protests, and other countries’ protests provide them with these heuristics. Because of these heuristics, potential collective actors are more observant of striking, vivid protest events in other countries, and they often overestimate their chances of winning in contentious movements and underestimate important differences in domestic conditions between them and other countries. Weyland (Reference Weyland2009) explains the anti-regime movement in Europe in 1848 as driven by demonstration effects. Many scholars use demonstration effects to explain the series of protests in Eastern Europe in the early 2000s (Beissinger Reference Beissinger2007; Gunitsky Reference Gunitsky2018) and the Arab Spring in the early 2010s (Bamert, Gilardi, and Wasserfallen Reference Bamert, Gilardi and Wasserfallen2015; Gunitsky Reference Gunitsky2018; Weyland Reference Weyland2012).

Studies of protest diffusion, accordingly, pay attention to the routes that transmit protest information to other countries. Transnational activists and international nongovernmental organizations (INGOs), economic and social ties, and the internet are discussed in the literature (Brancati and Lucardi Reference Brancati and Lucardi2019; Lotan et al. Reference Lotan, Graeff, Ananny, Gaffney, Pearce and Boyd2011; Strang and Soule Reference Strang and Soule1998; Weyland Reference Weyland2019). Transnational activists and INGOs not only spread news but also hand down new methods and tactics to potential protestors (Beissinger Reference Beissinger2007; Bunce and Wolchik Reference Bunce and Wolchik2006; Tarrow Reference Tarrow2005).

Economic relations between countries also play a significant role in disseminating information since close economic relations will increase the communications and interactions of economic actors between countries. For the same reason, social ties are considered important in spreading information. Because close countries are more likely to be connected by economic and social ties, the diffusion literature often considers geographic contiguity an important factor that makes diffusion more likely (Gunitsky Reference Gunitsky2014; Houle, Kayser, and Xiang Reference Houle, Kayser and Xiang2016; Teorell Reference Teorell2010).

Finally, the diffusion literature stresses the role of the internet in diffusion processes (Bamert, Gilardi, and Wasserfallen Reference Bamert, Gilardi and Wasserfallen2015; Lotan et al. Reference Lotan, Graeff, Ananny, Gaffney, Pearce and Boyd2011; Weyland Reference Weyland2019). During the Arab Spring, social media outlets such as Facebook and Twitter were frequently used to diffuse protest news and mobilize protestors (Bamert, Gilardi, and Wasserfallen Reference Bamert, Gilardi and Wasserfallen2015; Larson et al. Reference Larson, Nagler, Ronen and Tucker2019; Weyland Reference Weyland2019). Social media disseminates news very quickly—sometimes almost in real time. Weyland (Reference Weyland2019, 4) notes that after the Tunisian dictator had been removed by street protests, “within five minutes,” Egyptian activists called for action by Egyptian people on Twitter.

Demonstration effects, however, often overestimate the importance of international influence in the diffusion process. Potential protestors also take domestic political conditions into account (Brancati and Lucardi Reference Brancati and Lucardi2019; Hale Reference Hale2019; Kim Reference Kim2019; Skorge Reference Skorge2018). International influences will prompt people to take action only when these people have both the willingness and opportunity for mobilization. Countries differ in their political opportunity structures, and these differences will lead to different mobilization environments, under which people have different levels of willingness and opportunities for mobilization. In a more open society, potential collective actors will be endowed with more political and social resources. This means that mobilization is easier for them compared with actors in a less open society. However, discontented people in an open society will be endowed with a wider set of options to express their discontent, meaning they may not take the same action that people in the other country took. Therefore, without taking these domestic influences into account, diffusion studies cannot adequately explain why a movement diffuses into some countries but not in others.

Moreover, in terms of women's political engagement, domestic political conditions play significant roles (Kim Reference Kim2019; Skorge Reference Skorge2018). Kim (Reference Kim2019, 594), for example, finds that women under the presence of direct democracy more actively engage in politics since direct democracy “signals the system's openness to women's voice” and thus encourages women to participate in political activities. Participation in #MeToo is costly for women since it involves a certain level of sharing of their personal narratives and stories. In a more open political system, women will feel encouraged to share their stories and thus are more likely to engage in #MeToo. On the contrary, in a closed political system, even though women receive international signals to participate in #MeToo, they are less encouraged by their political system to openly speak out against authority figures, and thus less likely to participate in the movement. Similarly, in a closed political system, those outside the winning coalition may not see their voice as being effective in garnering political concessions. Therefore, speaking out against those with historically more power may be viewed as ineffective. Further, a closed political system is less likely to garner offline mobilization, limiting any synergistic multiplicative effect that the digital activism could bring (Hill Reference Hill2013). Given that all #MeToo participants could face hostile trolling, threats, and invasions of privacy as a result of their activism, the costs may be seen as greater than the expected benefits of activism in a closed system.

Taking these arguments together, we contend that domestic political opportunity is an important driver of #MeToo mobilization; countries with more open political structures should have more #MeToo involvement within their borders than countries with closed political structures. Our understanding of traditional protest has long supported the role of domestic political opportunity structures in increasing individual willingness to join a movement (Kim Reference Kim2019; Skorge Reference Skorge2018). Further, for digital activism, political opportunity structures will color whether individuals think that the movement will be successful and that their concerns will be taken seriously. Finally, although #MeToo is an international movement, domestic political opportunity still matters for protest diffusion. We contend that the importance of domestic political opportunity structures should hold, even accounting for the host of international factors that make digital activism more likely.

HYPOTHESES AND CONCEPTUALIZATION OF OPEN POLITICAL SYSTEMS

As discussed earlier, we contend that political opportunity structures drive #MeToo mobilization. Drawing on this discussion, we use five Varieties of Democracy (V-Dem) high-level democracy indices to capture slightly different conceptualizations of the level of openness in a political system (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019). Each of these conceptualizations becomes one of our testable hypotheses. First, the V-Dem electoral democracy index not only captures free and fair processes of elections but also includes freedom of expression and the existence of independent media outlets that allow citizens to get access to alternative information. We expect that as a country protects citizens’ political rights with free and fair elections and higher degrees of freedom of expression, individuals in the country will be more likely to engage in the #MeToo movement (H₁). This is because this system will reassure activists that their voices will be heard through independent media outlets and will be taken seriously by their political system. If individuals do not think their tweets will reverberate through the local political system, they will be less likely to participate.

Second, V-Dem's liberal democracy index captures whether a country protects minority rights and prevents the tyranny of the government and the majority (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019). We expect that as a country protects minority rights better and has a strong rule of law and independent judiciary, citizens will be more likely to participate in #MeToo (H₂). That is, the system will signal to individuals that their rights will be protected as they reveal their grievances in public. Further, a liberal democracy will provide the judicial mechanisms that could make #MeToo participation lead to local change.

Third, the participatory democracy index entails citizens’ active participation in political processes (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019). It not only emphasizes electoral participation but also stresses nonelectoral participation, such as participation in civil society organizations, direct democracy, and subnational political bodies (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019, 40). We expect that as a political system encourages citizens to be more active in political processes, women in the system will be more likely to engage in #MeToo (H₃). Civil society groups and past experiences with direct democracy may make individuals both more likely to know about the movement and more likely to participate in advocacy through digital media.

Fourth, V-Dem's deliberative democracy index captures whether there are meaningful dialogues before making decisions in a polity (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019, 41). This democracy index emphasizes the processes in which preferences are formed and final decisions are made. These processes should rely on public reasoning for the common good and should not rely on “emotional appeals, solidary attachments, parochial interests, or coercion” (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019, 41). We expect that as a country performs better in a deliberative process, people in the country will be more likely to engage in #MeToo tweets (H₄). A strong culture of deliberative democracy is likely to be associated with both in-person and online political discourse. Therefore, #MeToo tweets may be part of a regularized political dialogue.

Finally, we take the egalitarian principle of democracy into account. V-Dem's egalitarian democracy index captures whether social groups have equal rights and equal access to power and whether resources are distributed so that “no individual is so impoverished as to preclude their participation” (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Bernhard2019, 41; Sigman and Lindberg Reference Sigman and Lindberg2019, 600). We expect that if a country is more egalitarian, citizens in the country will be more likely to engage in #MeToo tweets (H₅). When power is shared across groups, individuals will be more likely to see #MeToo advocacy as a worthwhile way to express their rights, irrespective of any historical inequality to their social group.

DEPENDENT VARIABLE

We test our hypotheses using data collected from Twitter. The dependent variable is the number of #MeToo tweets in a country.Footnote ⁵ We extracted global tweets for two time periods. These time periods were chosen arbitrarily, in no way related to the strength of the activism or political events at the time. In the first round, we collected 17,289 global tweets from July 20 to July 25, 2019. In the second round, we collected another 17,922 global tweets from September 15 to September 21, 2019.Footnote ⁶ These tweets contain locations that the account owners specified. With these locations, we coded the countries where the account owners were located.Footnote ⁷ Then, we counted the tweets by country for each time period. We took the square root of our dependent variable because of overdispersion. Although our data only allow us to examine digital feminist activism for a short time, the time scope still enables us to explore why the movement has diffused into some countries but not in others.

The period examined is less than two years after the movement started in the United States in October 2017. During those years, the movement did not fade away online. In fact, a simple search on Twitter finds many recent #MeToo tweets. On the New York Times's website, the search term “#MeToo” found six news articles published in August 2019. Moreover, the intervening period, almost two years from the original movement in July 2017, provided enough time for people around the world to get the information on the movement and process the information. In the United States, where the movement started, the online movement has triggered much offline feminist actions. Offline accusations, lawsuits, and street marches have occurred. Although it is premature to assess whether the online activism and subsequent offline actions have been successful, one can still argue that these actions are making positive changes in the United States. As people around the world were able to observe the process and the outcome of those activities for two years, they had time to process the information and calculate their expected utility of engaging in the #MeToo movement. Therefore, the intervening time window is not long enough for the movement to fade away, but it is still long enough to spread information outside the United States. A similar argument can be made for studies that investigate women's rights street protest in the post–Cold War period (Hughes, Krook, and Paxton Reference Hughes, Krook and Paxton2015; Murdie and Peksen Reference Murdie and Peksen2015; Stienstra Reference Stienstra2016). Some time has passed since the early waves of feminism and the women's rights movement, allowing us to examine participation in an existing movement.

It is worth noting that we only examine tweets that include the English version of #MeToo. Many of the tweets we examine were in other languages but included the English hashtag. We hope future work will examine variations of #MeToo advocacy in other languages. Our key results hold with the addition of a control for whether a country has English as an official language.Footnote ⁸

ALTERNATIVE EXPLANATIONS

We also test alternative explanations and examine whether they account for the diffusion of digital feminist activism. All models include variables that account for the most common explanations of #MeToo activism. First, because the #MeToo movement is an internet-based movement, we control for the country's internet penetration rates, using 2018 data from the World Bank; this was the most recent year for which data existed when we ran our analyses Additionally, we include a dichotomous indicator for the United States because the movement started there. Naturally, the number of #MeToo tweets is the largest for the United States. We also include dichotomous indicators for three countries that are known to block Twitter: China, Iran, and North Korea.

We also examine aspects of globalization and look at how globalization correlates with #MeToo activism. First, political globalization can make diffusion more likely. INGOs and transnational activists are regarded as important actors who can diffuse movements into foreign countries. These actors can not only engage in the spread of information but also actively train domestic activists and mobilize people. Also, even though their role may not be as direct as that of INGOs and transnational activists, intergovernmental organizations (IGOs) can play a role in the diffusion process. As a country engages in international political cooperation through IGOs, that country is more likely to be open to foreign impact. We examine the role of INGOs, transnational activists, and IGOs and whether their existence or influence in a country increases the likelihood of the diffusion of #MeToo.Footnote ⁹ We use the KOF political globalization index data from 2016, the most recent available year, to do so (Gygli et al. Reference Gygli, Haelg, Potrafke and Sturm2019). Since the numbers of IGO memberships, INGOs, and foreign embassies are considered to calculate the levels of political globalization, this index enables us to test the effect of these organizations on the diffusion of #MeToo.

We also examine whether economic globalization can lead to the transnational diffusion of a movement. That is, as economic actors interact with each other more frequently, countries are more vulnerable to foreign influences, making digital feminist activism more likely. We use the KOF economic globalization index from 2016 to test whether economic globalization increases the probability of #MeToo diffusion. The index measures the level of economic globalization with trade and capital flows (Gygli et al. Reference Gygli, Haelg, Potrafke and Sturm2019).

Finally, social globalization can be claimed to increase the likelihood of #MeToo diffusion. The KOF social globalization index defines social globalization as “direct interactions among citizens” in different countries; “the actual inflow of ideas, knowledge, and images”; and the extent to which U.S. cultural products dominate (Gygli et al. Reference Gygli, Haelg, Potrafke and Sturm2019). These interactions and inflows of foreign ideas and U.S. culture can lead citizens in a country to be more vulnerable to the impact of #MeToo movement that was started in the United States.

We include other control variables as well. We add the natural log of population from the World Bank's 2018 data. The literature finds that a larger population provides a greater opportunity for collective action (Inclán Reference Inclán2008). We control for the proportion of seats held by women in national parliaments from the World Bank's 2018 data. We control for this because the proportion of women in formal institutions can affect the women's movement. It could be argued that as more women hold offices in formal institutions, women's rights are more likely to be protected through those formal institutions. Therefore, women are less likely to engage in digital activism and more likely to find formal routes to accuse their sexual harassers.

MODEL SPECIFICATION

We estimate Bayesian linear models. We argue that the Bayesian approach is advantageous for two reasons. First, we have two tweet samples collected in different time periods: one in July 2019 and the other in September 2019. A Bayesian approach enables us to be flexible in dealing with these two samples. One simple approach is to aggregate the two samples and estimate one set of models. However, the limitation of this approach is that simply aggregating the two samples fails to take the different time frames into account. We collected tweets with a two-month interval because we think that an idiosyncratic event in a country may cause a sudden increase of #MeToo tweets in the country. But by aggregating the two data sets, we cannot examine this time frame effect. Therefore, it is important to examine the two data sets separately, and the Bayesian approach helps us to do that in a more sophisticated way.

Intuitively, people update their prior beliefs after obtaining a new observation. From the data collected in July 2019, we learned information about the diffusion of #MeToo. Then, we obtained a new data set in September 2019. Therefore, we needed to update our knowledge on the diffusion of #MeToo based on the newly collected data. With the concept of Bayesian updating, we can analyze the September data and utilize our prior information obtained from the July data.

Theoretically, we are able to move from the prior probability to the posterior probability and update our prior beliefs with Bayes's theorem:

$$P( {\theta {\rm \vert }D} ) \propto {\rm \;}P( \theta ) P( {D{\rm \vert }\theta } )$$

where P(θ) is the prior probability mass function of θ before observing the data, P(θ|D) is the posterior probability mass function of θ given the data, and P(D|θ) is the likelihood function (Gill Reference Gill2014). Because Bayes's theorem enables us to incorporate prior information into our models, we are able to examine the two samples in the context of Bayesian updating.

Second, we only examine a one-year time frame (2019), and thus our sample contains 141 countries as observations. This is a relatively small sample in size. Bayesian models allow us to estimate parameters with our relatively small sample without biases. The Bayesian approach does not rely on asymptotic approximation. Rather, based on Bayes's theorem, we can make inferences that are conditional on the data we have. Therefore, small sample inferences are not different from larger sample inferences (Gelman and Hill Reference Gelman and Hill2007; Gill Reference Gill2014).

Taking advantage of the Bayesian approach, we estimate three sets of Bayesian linear models. First, we estimate Bayesian linear models with the July data using noninformative independent normal priors for the regression coefficients with a mean of zero and variance of 100. Then, we estimate another set of Bayesian linear models with the September data. In this estimation, we incorporate prior information learned from the July data. We use informative priors that are posterior distributions of coefficients in our July data models.Footnote ¹⁰ Finally, we aggregate the July and September samples and estimate Bayesian linear models using noninformative normal priors for the regression coefficients with a mean of zero and variance of 100 to compare our results from the nonaggregated samples with the aggregated samples. We use Markov chain Monte Carlo (MCMC) methods to calculate joint posterior distributions for model parameters. MCMC estimation is conducted using JAGS and the R package rjags (Plummer Reference Plummer2003, Reference Plummer2013). We conduct convergence diagnostics and do not find the absence of convergence.

We rescale our variables by standardizing them. We do this for a better visualization of posterior distributions in the figures.