In 2018, Amnesty International published a series of reports alleging that Twitter is toxic for female journalists, activists, and politicians. Amnesty’s first report was based on interviews and surveys with women about their experiences of abuse or harassment on social media. The second report, published in December, was based on crowdsourced data collection and machine learning. Amnesty concluded that, “the violence and abuse many women experience on Twitter has a detrimental effect on their right to express themselves equally, freely, and without fear.”
The result of Amnesty’s #ToxicTwitter campaign has been striking, with several leading news organizations positively covering Amnesty’s research. In March, Reuters reported that Twitter is “failing women,” and New York Magazine reported that “Twitter Violates Women’s Human Rights.” In December, Wired reported that “Twitter is Indeed Toxic for Women,” and the Financial Times reported that “Twitter shares tumble amid concern over ‘toxic’ content” after the company was labeled the “Harvey Weinstein of social media” by Citron Research following Amnesty’s report. Naturally, for many, the take-away is that “women have a dramatically different experience on Twitter than men,” a sentiment endorsed by Amnesty International.
There are at least two inconsistencies with the #ToxicTwitter reports. First, and most notably, missing from Amnesty’s analysis is a comparative benchmark, namely that of men’s experiences on Twitter. This is not an appeal to whataboutism but instead a logical point about the empirical claim that “Twitter has a detrimental effect on [women’s] right to express themselves equally.” (emphasis added)
In an NPR interview with the director of Amnesty Tech, NPR asked, “What about men? Did you compare to men?” Amnesty responded:
Yeah. We didn’t actually look at men in this study. And this is partly because this study is the third big piece of research that we’ve done on the phenomenon of violence against women online. And so it’s based off the fact that we already knew and we know that the way that women are targeted online is very different, and it’s very gendered. It’s stuff like doxing and hacking and violent rape threats. And so, in this case, we are very specifically interested in the experience of women because we know, just as offline, discrimination and violence against women is rife.
Amnesty assumed “that the way that women are targeted online is very different,” an assumption that was not challenged by the NPR interviewer and an assumption that is not supported by empirical evidence. Data suggest that men may experience toxicity online at similar or even higher rates than women. The Pew Research Center found that overall, more men experience online harassment than women (44 percent vs. 37 percent), and men are twice as likely to experience online harassment as a result of their political views (19 percent vs. 10 percent).
In addition, of the six categories of online harassment that Pew analyzed, men experience higher rates than women for two of the categories: they are more likely to be called offensive names (30 percent vs. 23 percent) and to receive physical threats online (12 percent vs. 8 percent). Sexual harassment is the only category where women experience higher rates than men (8 percent vs. 4 percent). Three categories of online harassment show no statistically significant differences by gender, including purposeful embarrassment, stalking, and sustained harassment.
We have good reason to believe that the Pew data are reliable given that, when the study populations intersect, the findings are consistent with Amnesty’s research. For example, according to Pew, “blacks who go online are especially likely to say that they have been called offensive names (38 percent compared with 28 percent of white internet users) or to say that someone has tried to purposefully embarrass them (34 percent vs. 23 percent).” These data are consistent with Amnesty’s findings, where nearly 9 percent of Twitter mentions toward black women were problematic, compared to only 5.5 percent toward white women.
Additionally, according to the Anti-Defamation League’s report on antisemitic targeting of journalists during the presidential election, 10 journalists received 83 percent of the antisemitic tweets analyzed. Of these 10 journalists, seven were men, including all of the top three. These studies provide prima facie evidence that the online experiences of women are not necessarily more toxic than those of men.
Secondly, Amnesty’s summary of its key findings is misleading. For example, Amnesty states that “women of colour (black, Asian, Latinx, and mixed-race women) were 34 percent more likely to be mentioned in abusive or problematic tweets than white women.” This statement is entirely driven by the difference in treatment between black and white women, not the treatment of Asian, Latinx, and mixed-race women.
While it’s true that black women receive more problematic and abusive tweets than white women, Latinx women receive fewer problematic and abusive tweets than white women (4 percent vs. 5.6 percent problematic; 0.8 percent vs. 1.2 percent abusive). The rates are equivalent between white, Asian, and mixed-race women (5.6 percent vs. 5.7 percent vs. 5.8 percent problematic; 1.2 percent vs. 1.2 percent vs. 1.3 percent abusive). By summarizing the findings as white women versus people of color, in the aggregate, the statistics will support a discrepancy, but this finding is driven purely by the fact that black women receive about 60 percent more problematic tweets and 84 percent more abusive tweets than all other women. A worthwhile headline could be that white, Asian, Latinx, and mixed-race women were all treated the same on Twitter. But this headline wouldn’t serve Amnesty’s mission.
Why did Amnesty omit men from their study? Why did their headline statistics misleadingly report the racial breakdown for problematic and abusive tweets? I believe social signaling provides an answer.
“While Philanthropy Needs Money to Survive, It Needs Status to Attract Money.”1
For decades, many scholars from different disciplinary backgrounds have argued in favor of a basic idea: that altruism is a prestige-seeking activity, rather than a truly “unselfish regard for or devotion to the welfare of others.” According to the signaling theory of altruism, individuals engage in altruistic behavior (such as donating or volunteering to charity) as costly signals—conscious or subconscious—that credibly develop their reputations as wealthy, kind, generous, compassionate, and empathic in order to gain respect, prestige, and social status.2
Nowhere is this point clearer than in elite competitive philanthropy. In Why the Wealthy Give, Francie Ostrower documents elite philanthropy in New York City, finding that charitable giving is an important indicator of class and social status. Elites mainly give to impress each other: “wealthy donors are generally more focused on their peers, rather than those outside their class, as the audience for their philanthropy.”
Our donations to charity are not necessarily intended to maximize the benefit to others or efficiently provide the most good. Research in the social sciences suggest that people often donate money to personally benefit. As economists have found, if people are primarily motivated by the output produced by the charity (such as reducing poverty, feeding the homeless, advancing human rights, and so on), then we should observe standard free-rider problems. However, it turns out that we do not, at least not as often as would be predicted:
Most empirical studies of survey or donation data find that on average the benefit appears to be private [personal] in nature. This suggests that the last dollar that we give to charity is not motivated by the nonprofit’s output. (emphasis added)
An important personal benefit that donors receive is a boost to their reputation, i.e. a status benefit. We heap moral praise on individuals who successfully signal their virtue, generosity, and compassion. And we do so publicly, through Nobel Peace Prizes, medals for bravery in wartime, name recognition on buildings and plaques, and more quotidian instances in our social lives. Psychologists and biologists have also found that we choose long-term partners based on their successful displays of kindness, generosity, empathy, and heroism. “We fall in love with…mental and moral traits.”
The disposition to participate in altruistic behavior has evolved to communicate an individual’s underlying socially attractive qualities, such as wealth and abundance of resources,3 compassion and empathy,4 prosociality5 or cooperative inclinations,6 and loyalty to group norms and values.7 Potential mates and allies socially reward individuals for their conspicuous altruism. Signalers donate money as a display of their social desirability, and the audience of the signaler evaluates the signal insofar as it communicates something useful about the signaler.
Charitable individuals receive status benefits, not for how much good their actions do, but for what the sacrificial action communicates about their character. This produces social incentives for individuals to enact their charity in a conspicuous manner rather than to quietly behave altruistically with the intention to be optimally beneficial. The signaler is more interested in being seen providing assistance and less interested in the effect of that assistance. “The helper benefits from the act of helping, and the benefits to others are incidental—a side effect.”8
Signalers can increase the cost of the donation by giving away more money, donating more time, or engaging in physically strenuous activity (e.g. charity runs, dumping ice-water on one’s head). The costlier the signal, the more morally virtuous the signaler is presumed to be by her audience, regardless of the effect of the costly behavior on the intended beneficiaries.9 When potential allies and mates observe signaling behavior, they are interested in gleaning information about what to expect in social interactions with the signaler. Information pertaining to the effectiveness of the signaler’s donation is not useful information for allies and mates; it typically has no bearing on their social interactions with signalers.
David Brooks, in his criticism of the “earning to give” strategy (in which somebody earns a high-paying salary in order to donate much of it to charity), unintentionally makes the case for the signaling theory of altruism. “When we evaluate our friends, we don’t just measure the consequences of their lives. We measure who they intrinsically are. We don’t merely want to know if they have done good. We want to know if they are good.”
More relevant information to the audience will pertain to how well the donation conforms with the values of the audience’s social community, the size of the donation, and the political message the donation sends. As Eric A. Posner states in Law and Social Norms, “If enough people care sufficiently about their reputations (for being generous or for being wealthy) …people’s charitable contributions will not be…sensitive to the well-being of the donee.”
A donation to the NRA Foundation, which operates as a tax-exempt civil rights charity, may signal loyalty, toughness, and virtue to a conservative audience in rural America, but such a donation would send the wrong signals to a liberal, metropolitan audience. Republican Senator Cindy Hyde-Smith told a conservative crowd in Mississippi during the 2018 Senate race: “The Second Amendment right, the right to bear arms, is one of my strongest beliefs. I have been an NRA member for many, many, many years. My daughter, on her second birthday, got a lifetime membership to the NRA. So anybody who has any question about that, put it to rest.”
In contrast, a donation to Planned Parenthood would send the wrong signal to conservatives. It is not surprising, then, that Republicans smear other Republican candidates by accusing them of supporting Planned Parenthood. Candidate Sen. Chris McDaniel accused Sen. Hyde-Smith of supporting Planned Parenthood during the 2018 Mississippi Senate race. Hyde-Smith was forced to clarify, in no uncertain terms, that “Planned Parenthood is one of the worst things that has ever happened to us,” despite there being only one Planned Parenthood clinic in the state of Mississippi, and it is a clinic that does not perform abortions. Virtually every politician uses these and other charities to signal something about themselves.
Signaling also explains the rarity of anonymous donations. In the 1990s, Glazer and Konrad calculated anonymous donations to non-profits on file at the Pittsburgh Business Library. They found that the highest anonymous donation rates were to the Pittsburgh Philharmonic at 1.29 percent, Carnegie Mellon University at 0.26 percent, and Yale Law School at 0.21 percent. I collected data on the International Rescue Committee’s donations in 2017, and less than 10 percent were anonymous. If people donate to charity, in large part to receive status benefits, it makes sense that few people donate anonymously.
Signaling explains the “watching eyes” effect. In experimental studies of donor behavior, researchers have consistently found that images of eyes nearby increase the probability of donating.10 This suggests “the existence of automatic cognitive mechanisms for detecting social gaze and regulating social behavior accordingly.”
Signaling can explain why so few donors research charities before contributing. As Kevin Simler and Robin Hanson describe in Elephant in the Brain, donors usually do not spend time researching which charity most effectively helps their cause of interest because doing so generates private information. The donor gets signaling credit when donating to a charity that is publicly known, such as large, wealthy charities with name recognition like Amnesty International, high-profile natural disasters, or local churches and schools well-known to an individual’s local community.
Even when donors research charities, they mainly do so to validate the donation they have already made. Only 6.5 percent of donors claim to do comparative research on how much charities are accomplishing before making a charitable contribution. Less than one percent of donors spend more than a day researching charities. In experimental settings, researchers have found that people often do not choose welfare-maximizing options, even when they are given information about effectiveness.
Donors Buy Virtue from Charities, and Charities Curate the Product
Human rights non-governmental organizations (NGOs)—like Amnesty International—provide a signaling service to their donors. Donors purchase this signaling service, paying for the ability to show the world they are prosocial, open, multi-cultural, compassionate, empathic, and politically liberal. The primary product on offer is a badge outwardly signaling that the wearer is a person who is associated with the broadly known values of the human rights NGO. For the donor, the benefit is prestige and status that comes with associating with the organization. The NGO, for doing its part, receives money, status, and authority.11
The NGO world is a crowded space. Donors have millions of charities from which to choose. An organization does not need to convince donors to change their minds to attract their donation. Instead, the NGO can convince donors that it represents their views and will provide assistance in signaling their commitment to these views and loyalty to their community.
Corporations have discovered the power of virtue-signaling. In a New York Times article, Paul Sullivan writes, “Firms learn that as they help charities, they also help their brands.” For example, Subaru chose “well-known, noncontroversial charities,” such as the American Society for the Prevention of Cruelty to Animals, the Make-A-Wish Foundation, and Meals on Wheels. On the other hand, Discovery Communications, which produces Shark Week, began a wild tiger conservation program.
Companies also signal their virtues in advertisements. Gillette’s viral commercial “We Believe: The Best Men Can Be” was a calculated virtue-branding effort destined to annoy some and attract others, a worthwhile trade for a declining brand. Amnesty International USA, in a rare endorsement of a corporation, tweeted in support of Gillette: “People are upset about the Gillette ad? Repeat after me: We want a world without #ToxicMasculinity.” The international Twitter account for Amnesty likewise supported the ad: “The [world] will be a better place without #ToxicMasculinity.” In a response to an NPR query, a Gillette spokesperson said, “No longer can companies ‘just advertis[e] product benefits.’ These days ‘brand-building’ also means taking a stand on important societal issues, controversial as they may be.”
A study on corporate social responsibility found that 87 percent of study respondents reported “they would buy a product because of something the company advocated.” However, if the company advocates (signals) the wrong cause, 76 percent said they would boycott the product. This consumer demand produces an incentive for companies to learn what their customers want to signal and enhance their brand through the power of signaling.12
Just as companies both sell products and brands, so too, do human rights NGOs. They produce ethical products (e.g. issue awareness campaigns) that function in part to enhance the donor’s reputation. Human rights NGOs compete in an overcrowded charity market by convincing donors they can offer the best signaling value. And donors pick and choose which charity they believe will best improve their self-presentation. For some donors, a well-known international human rights NGO like Amnesty International might signal the right traits and values, given their social community. For others, it might be their alma mater or their church.
When human rights NGOs are partly funded to sell virtue, we should expect inconsistencies in their application of universal human rights principles, as demonstrated by Amnesty’s #ToxicTwitter campaign. Not all human rights issues or victim profiles send the same signals, even to people who claim to be committed to the idea that they should. Intriguingly, altruism signals may be affected by the victim profile, or recipient type. Metzger and Günther demonstrate experimentally that demand among donors is highest for information about the recipient type and lowest for information about aid impact. In another experiment, Bachke, Alfnes, and Wik show that information on recipient type has the largest impact on donation behavior, with children receiving the most donations, followed by girls, women, boys, and lastly, men.
Causes pertaining to women’s rights constitute the most popular category of human rights philanthropy. According to the Advancing Human Rights Initiative, 28.4 percent of all human rights philanthropic grants went toward women and girls from 2011 to 2015 (beating all other categories, including children, refugees, LGBTQI, and indigenous peoples, among others). This backdrop offers an initial glance into the social environment in which human rights donors likely find themselves.
The social expectation in the human rights community is that women are oppressed, and men are the oppressors. Tania Reynolds and her colleagues demonstrate experimentally that subjects in the United States and South Korea tend to, by default, assume that women are sensitive and vulnerable victims while men are powerful wrongdoers. This widely shared baseline expectation produces social incentives for donors to support human rights NGOs that adhere to and confirm this public perception.
Charli Carpenter documents gender discrimination against men in the international human security sector. In Innocent Women and Children, Carpenter documents how international humanitarian organizations evacuated women and children from besieged Srebrenica, leaving behind adult civilian men, even though adult men were systematically targeted and slaughtered by the Bosnian Serb Army. In ‘Lost’ Causes by the same author, human rights NGOs refused to take up involuntary infant male circumcision as a human rights cause, despite its issue-complementarity with female genital mutilation. It was viewed as competing with and minimizing women’s problems.
The major distinction between human rights violations against women and human rights violations against men is not that the former is an objectively worthier cause than the latter but that the former is publicly understood in the cosmopolitan human rights community, and the latter is not. The signaling theory of altruism expects that public knowledge will be a more relevant factor in shaping an individual’s altruistic behavior than private knowledge.13
Amnesty’s recent signaling misfire upon its foray into the sex work issue helps to illustrate the point. In 2015, Amnesty announced a new policy in favor of fully decriminalizing sex work, arguing that no consenting adults involved in sex work should be criminalized—including the buyers (who are disproportionately men). This policy position angered many public feminists as well as many Amnesty donors, who argued instead for the decriminalization of the (female) sex worker but the continued criminalization of the (male) buyer. Many celebrities, writers, feminists, and NGOs publicly denounced Amnesty’s policy by arguing that prostitution is “sexist, racist, and classist,” and that Amnesty’s policy would “fuel rape culture.” As many as 500 members (donors) from Amnesty’s Sweden office left the organization after the policy roll-out.
Soon after the policy announcement, Amnesty essentially killed the campaign. They stopped issuing reports on sex work, and aside from a few meager attempts to respond to the backlash, they did not attempt to shift public opinion in favor of their policy. Amnesty’s new policy on decriminalizing sex work was a signaling failure. Regardless of the merits of Amnesty’s position, it signaled the opposite of what most human rights donors wish to signal. It jeopardized donors’ ability to signal their alliance with, and their compassion and empathy for women. The “recipient type” of Amnesty’s work was perceived by many to have shifted from women (the sex workers) to men (the buyers).
Signaling offers an elegant explanation for why we should expect to observe inconsistencies in the work of human rights NGOs. Although universality is a key property of human rights principles, we know from decades of work on evolutionary and moral psychology that individuals develop moral judgments based in large part on social expectations, and these social expectations do not always neatly correspond to an impartial treatment of one’s principles. To the extent that human rights NGOs are funded by selling reputation or virtue to donors, we should expect inconsistencies to be a regular feature in their work.
* * *
Jonathan Haidt offers useful advice in The Righteous Mind, which can be applied to the human rights NGO sector: “Moral reasoning is part of our lifelong struggle to win friends and influence people…You’ll misunderstand moral reasoning if you think about it as something people do by themselves in order to figure out the truth.” Most of the time, our underlying motives for behaving altruistically—to attract potential mates and allies—are hidden to us. We are not necessarily conscious of the signals we send and receive. Robert Trivers, author of Folly of Fools, refers to this as adaptive self-deception: “We deceive ourselves the better to deceive others.”
Amnesty focused its #ToxicTwitter campaign on women, rather than people, because of the signaling value donors could be expected to gain by associating with an organization that fights for women’s rights. To mention the (potentially worse) situation for men on Twitter would be to ship a product the donors didn’t ask for. By only studying women, Amnesty made headlines for multiple days in a row with entirely positive press, allowing their supporters to reap social credit for associating with Amnesty. By all accounts, their #ToxicTwitter campaign was a signaling hit. Based on the positive receptivity, we should expect more of the same.
1 Ostrower, Francie (1995) Why the Wealthy Give: The Culture of Elite Philanthropy.
2 Simler, Kevin and Robin Hanson (2018) Elephant in the Brain: Hidden Motives in Everyday Life.
3 Zahavi, Amotz and Avishag Zahavi (1997) The Handicap Principle: A Missing Piece of Darwin’s Puzzle.
4 Miller, Geoffrey (2000) The Mating Mind: How Sexual Choice Shaped the Evolution of Human Nature.
5 Tessman, Irwin (1995) “Human Altruism as a Courtship Display.”
6 Posner, Eric A. (2000) Law and Social Norms.
7 Levine, John M. and Richard L. Moreland (2002) “Group Reactions to Loyalty and Disloyalty.” In Advances in Group Processes, Vol. 19. Group Cohesion, Trust and Solidarity.
8 Zahavi, Amotz and Avishag Zahavi (1997) The Handicap Principle: A Missing Piece of Darwin’s Puzzle.
9 Olivola, Christopher Y (2011) “When Noble Means Hinder Noble Ends: The Benefits and Costs of a Preference for Martyrdom in Altruism.” The Science of Giving: Experimental Approaches to the Study of Charity. For example, see the discussion on the Doctors Without Borders experiment.
10 Kelsey, Caroline, Amrisha Vaish, and Tobias Grossmann (2018) “Eyes, More Than Other Facial Features, Enhance Real-World Donation Behavior.”
11 Stroup, Sarah and Wendy Wong (2017) The Authority Trap: Strategic Choices of International NGOs
12 Miller, Geoffrey (2010) Spent: Sex, Evolution, and Consumer Behavior.
13 Simler, Kevin and Robin Hanson (2018) Elephant in the Brain: Hidden Motives in Everyday Life.
Join the newsletter to receive the latest updates in your inbox.