The Case Against Social Media is Stronger Than You Think

A reply to Dan Williams

The Mob, 1935, by Carl Hoeckner

Note: this post makes the case against social media on political grounds. Those interested in the well-documented mental health harms of social media should read this post as well.

1. Introduction

The philosopher Dan Williams recently published two pieces on social media— “Scapegoating the Algorithm” at Asterisk Magazine, and “The Case Against Social Media is Weaker Than You Think” at his Substack. As their titles attest to, both argue that the case against social media, on epistemic and political grounds, has been considerably overstated.

I recently published a lengthy essay arguing the opposite: that the case against social media has, if anything, been understated. And so, especially given I’m new to Substack, Williams’s recent work gives me a welcome opportunity to pick a fight for engagement-farming purposes. Sadly for my subscriber count, I agree with Williams on quite a bit, and so this is going to be less combative polemicizing and more demonstrating that a range of serious worries about social media are capable of charting a course between the equivocal lines of evidence he underlines.

I am going to focus on the putative political impacts of social media—in particular its impact on political polarization—rather than the specifically epistemic ones. These are often conflated, but I think it is helpful to cleave them apart. It is possible to believe, as I in fact do, that social media has had a dangerous, incendiary effect on American politics, while also believing that related concerns about misinformation and conspiracy theorizing are somewhat (though not entirely) overblown. I may decide to write a short follow-up on the latter, but will put it to the side for now.

With respect to social media’s supposed contributions to political polarization, and in particular ‘affective polarization,’ Williams highlights four main lines of evidence that contradict the prevailing narrative. First, polarization has been rising for decades and began doing so well in advance of social media. Second, polarization has increased the most in recent years among those who use social media and the internet the least, those over 65. Third, trends in polarization have diverged in countries that all have widespread social media use. And fourth, several high-quality experimental studies have found a negligible effect of social media use on individuals’ levels of polarization.

If I can import a cliché from academic philosophy, it is sometimes said that all argumentative objections fall into one of two categories: “oh yeah?” or “so what?” I opt for a mix of both here. I begin with the “oh yeah?”: Williams’s evidence is much less convincing than it initially seems, and beyond ruling out an implausibly large spike in polarization, tells us little about how social media might be influencing its trajectory.

I then turn to my main focus, the “so what?”: even if Williams is right that social media has not significantly contributed to affective polarization in the U.S., this is consistent with it having plenty of other negative effects on American politics. To make this argument concrete, I highlight different lines of evidence to Williams in order to demonstrate first, that social media has indeed had a deeply concerning impact on our politics, and second, that one need not rely on the metric of affective polarization in particular to make this case. I refer to the alternative view I develop as an elite radicalization theory of online politics.

While I am an avid reader and admirer of Williams’s work, I think he is engaged in what has unfortunately become a fashionable form of academic contrarianism: arguing that despite what very much seem to be foundational, risk-laden changes in our social order, all we observe is in fact business-as-usual, and to the extent it seems otherwise, that is because of some mix of psychological bias and media overhype.

A similar contrarianism plagued early predictions of the threat posed by Donald Trump. Those critics have either relented or fallen quiet now. It is time for those who continue to minimize the downside risks of the digital media revolution to finally do the same.

2. Williams on social media and polarization

Let’s first run through Williams’s four main lines of evidence against social media-driven polarization in a bit more detail.

The first is that affective polarization has been rising since well before the advent social media. As Williams explains in “Scapegoating the Algorithm,” this trend begins in the late 20^th century and likely has its roots in a) the partisan realignment surrounding the Democratic party’s 60s-era embrace of racial integration, and the Republican party’s subsequent “Southern Strategy” to appeal to disaffected white opponents of integration; and b) the rise of a more combatively partisan media ecosystem in the aftermath of Reagan’s 1987 repeal of the “fairness doctrine,” which required broadcasters to present balanced perspectives on controversial public issues.

Source: Finkel et. al. (2020)

Williams’s second line of evidence is the study Boxell, Gentzkow, and Shapiro (2017), which found that, between 1996 and 2016, the demographic groups that used social media and the internet the least, those 65 and older, also saw the largest increase in polarization.

Williams’s third line of evidence is a later study by the same authors, Boxell, Gentzkow, and Shapiro (2024), which studied trends in affective polarization since the 1980s in 12 OECD countries and found that, while the U.S. exhibited the largest increase over this period, countries like Australia, Britain, Norway, Sweden, and Germany all saw polarization fall, including in the period after the initial spread of social media.

Williams’s fourth line of evidence is that several high-quality experimental studies have examined the relationship between social media and polarization at an individual level and found no evidence of a connection. Guess et. al. (2023a), Guess et. al. (2023b), Nyhan et. al. (2023), and Alcott et. al. (2024) all found that interventions meant to limit social media’s ostensible polarizing effects (such as replacing default with reverse-chronological feeds, hiding reshares, reducing exposure to content from like-minded sources, and deactivating social media entirely) had a negligible impact on users’ levels of polarization.

2.1. “Oh yeah?”

Taken together, this evidence indeed appears to contradict that social media has been a major contributor to American political polarization. But dig a bit deeper, and things get much messier and more uncertain.

The fact is that we have relatively little data about affective polarization since ~2010, when smartphone and social media use became widespread in the U.S. The vast majority of this data comes from the American National Election Studies (ANES) survey, whose flagship data set is only collected during presidential election years. As a result, we only have 4 data points since 2010, those for 2012, 2016, 2020, and 2024.

Source: ANES

It is difficult to draw strong conclusions from this data one way or the other. Besides the fact there isn’t much of it, rival-party feeling does appear to start falling faster around 2010 or so, which would be consistent with some social-media related effect. But I concede I’m just eyeballing here; more rigorous conclusions will have to await further research.

The data on international trends in affective polarization isn’t especially convincing either. The main study cited by Williams—Boxell, Gentzkow, and Shapiro (2024)—looks at affective polarization in 12 countries since the 80s, but for our purposes, it is only the post-2010-years that are relevant, since it is only after this point that social media would’ve had any notable effect. That in mind, the data here is also relatively unsatisfying:

Source: Boxell, Gentzkow, and Shapiro (2024)

With the exception of Germany, there are never more than 3 data points per country after 2010. When you consider on top of this that rates of social media penetration over time probably vary across these countries, it becomes clear that there is little we can learn through these comparisons without more context, or at least more time.

The bare fact of diverging trend lines also tells us little to begin with, since these trends might have been different absent the influence of social media. For instance, polarization might have risen more slowly in the U.S. and fallen more quickly in Germany. Sure, maybe this data does conflict with the most alarmist narrative about social media—that there was some massive spike in polarization after 2010—but that was never plausible to begin with. There is still more than enough room, even in light of this data, to be justifiably concerned about how social media is shaping American politics.

That in mind, let’s move on to Williams’s two other lines of evidence—that polarization has grown more in recent years among the elderly (those least likely to use social media), and that the experimental research on social media has not turned up any appreciable relationship with polarization.

The paper documenting higher polarization among the elderly—Boxell, Gentzkow, and Shapiro (2017)—makes a very important caveat that Williams does not mention. While the authors see themselves as ruling out the most straightforward version of the social media-driven polarization story, they acknowledge it is possible to construct an alternative account that they cannot rule out:

[Alternative accounts would need to address] the rapid increase in polarization among those with limited internet use and negligible use of social media. However, it is possible to construct such accounts. It may be that social media increases polarization among the young while some other factor increases it among the old. It may be that there are spillovers across demographic groups; young adults polarized through social media might in turn affect the views of older adults or might indirectly influence older adults through channels like the selection of politicians or the endogenous positioning of traditional media. (p. 3, bold my own)

While we should not fault Boxell, Gentzkow, and Shapiro for the scope of their specific project, it is obvious in my view that every one of these spillovers is in play. I suspect the most important is the possibility that social media is influencing the tone and coverage of traditional media, and in particular cable news. This should ring true to anyone familiar with right wing cable’s aggressive coverage of various social media-native controversies, like the recent American Apparel ad starring Sydney Sweeney, or the various sorority rush videos currently going viral on TikTok. If this kind of coverage increases polarization among the elderly, that would still in part be social media’s fault. In light of potential spillover effects, this evidence can at best rule out that social media has influenced all demographic groups equally, and just as immediately, rather than seeing its influence channeled from some groups to others.

But this kind of ‘channeling’ was always going to be part of the story. Social media’s intrinsically social nature guarantees it will have large spillover effects. Ignoring this is like assuming that, if you could only delete your social media accounts, that would insulate you from 100% of social media’s effects on your life. The obvious reason it won’t is because everyone around you will still be using social media, and their use will continue to affect you (e.g. by their telling you what’s happening online, by you feeling excluded from their group-chats, etc.). As long as social media represents the social world to us, and as long as we share our impressions of that social world with others offline, then its effects on individuals are going to ‘spill over’ to their peers.

It is therefore wrong to assume that, since old people don’t use social media (or use it much less), they are not being affected by it. Due to spillovers, those of them that watch cable news are arguably affected by it every day. Not only that, but their relative lack of experience with the modern information environment means their political attitudes are likely to be more sensitive to (even the indirect effects of) social media than those who use it more often.

As a general rule, very few strong conclusions about the aggregate impact of social media can be drawn from research that does not account for these kinds of spillovers.

Let’s now turn to Williams’s last line of evidence, the experimental research on social media and political attitudes. This is probably the strongest evidence we have against social media-driven polarization. And yet, think carefully about what it is really in a position to tell us, and the result is again underwhelming.

Of the experiments Williams cites, the longest lasted three months, and only one—Alcott et. al. (2024)—actually entailed the complete deactivation of social media, rather than just modifications to how and what content was displayed. Already, this should give us pause about interpreting this evidence to imply that social media’s supposed impacts on politics are overblown. On a more even-handed read, all the studies besides Alcott et. al. (2024) show is that feed design changes don’t impact users’ political attitudes over relatively short periods.

Even focusing on complete deactivation, though, spillover effects remain a serious problem. Like anyone else, social media users’ political attitudes originate from and are continually modified by a wide variety of sources: their families, their friends, prominent members of their community, other acquaintances, online news, print news, T.V. news, podcasts, talk shows, radio shows, and so on. If these other influences have themselves been influenced by social media, then deactivating participants’ Facebooks and Instagrams for six weeks—the intervention in Alcott et. al. (2024)—does little to isolate them from the effects of social media in general. They experience those effects anyway by continuing to communicate with other social media users.

This is an even greater concern than usual in the case of Alcott et. al. (2024) because their intervention took place in the six weeks before the 2020 election. Those who took part in the experiment would have been inundated by far more political communication than usual, communication which was almost certainly influenced in some form by social media.

The moral is again that spillover effects aren’t to be ignored. Given social media’s now-enormous role in shaping public discourse, interventions targeting personal use alone can at best isolate a small sliver of social media’s total impact. That means that even high-quality experiments on individuals tell us little about social media-driven polarization in aggregate.

3. “So what?”

Suppose all these objections are unimpeachably correct. I’m sure they aren’t, but let’s assume. That would still only amount to a case for uncertainty. It remains, then, to make the case that rather than only remaining cautious about social media’s effects on politics, we should be seriously concerned.

Williams is right that the evidence on social media and affective polarization will not get us there on its own. But affective polarization is hardly the only potential measure of social media’s harms. Consider the following claims Williams cites (in “The Case Against Social Media is Weaker than You Think”) as typical of the public concern over social media:

Alexandria Ocasio-Cortez describes Meta as “a cancer to democracy metastasizing into a global surveillance and propaganda machine for boosting authoritarian regimes and destroying civil society.”

Jonathan Haidt has argued that social media platforms “dissolved the mortar trust, belief in institutions, and shared stories that held a large and diverse secular democracy [in America] together.’’

Obama suggests that one of “the biggest reasons for democracies weakening is the profound change that’s taking place in how we communicate and consumer information.”

These are of course strong criticisms. But it is far from clear that they depend on social media being a major driver of affective polarization in particular. One of the issues with placing academic and non-academic discourses in conversation is that the terms of the former may not map cleanly onto the terms of the latter (and vice versa). In this case, there is no reason to assume claims like those above rest on the evidence in favor of social-media driven polarization, especially if there are other, more convincing lines of evidence available.

This is my primary complaint against Williams: there are other, more convincing lines of evidence available.

Even if social media is not increasing affective polarization, it is making our politics more angry, tribal, and violent—and to more than a sufficient degree to justify grave concerns like those of Ocasio-Cortez, Haidt, and Obama.

In what follows, I focus on two lines of evidence in particular: research showing that social media amplifies the reach of emotionally extreme content, and research showing that higher social media penetration in a given geographic area induces more extreme political behavior, such as protests and hate crimes. I synthesize both into an overarching view I refer to as an ‘elite radicalization‘ theory of online politics.

3.1. The elite radicalization theory

It is well-established in the research literature that our attention on social media gravitates toward content that is more emotionally extreme. Recent studies have found that posts that are sad, fearful, uncivil, morally and emotionally indignant, hostile towards out-groups, and negatively valenced in general spread much further than their more neutral counterparts.

Brady et. al. (2017), for instance, found that each additional word with “moral-emotional content” (e.g. “duty,” “fear,” “shame,” “war”) increased political Twitter posts’ diffusion factor by 20%. Brady et. al. (2019) later reproduced this finding using the tweets and retweets of over 500 presidential candidates and members of congress. Expressions of “moral anger and disgust,” the authors found, diffused particularly quickly.

That negative content spreads especially far online makes sense given some well-established properties of human psychology. Humans exhibit a broad-based psychological negativity bias as well a range of more specific attentional biases toward negative stimuli. This is because, like other organisms, we have evolved to be uniquely attuned to signals suggestive of danger. As a result, when given the opportunity to cycle through a variegated soup of thousands of digital signals daily, we tend to fixate on and amplify the most distressing ones.

This dynamic creates very strong incentives to prey on our negative emotions. Successful ‘attentional entrepreneurs’ online not only enjoy attention’s more immediate benefits like status or influence, but are often paid by platforms in proportion to the ad revenue they generate. In that case, if certain kinds of content reliably accrue the most attention, there are very strong incentives online to produce that content en masse.

That is exactly what has happened in recent years with sensationalized and excessively negative political content. The last decade and a half has seen the rise of a new class of political influencers who, empowered by social media’s unique incentive environment, have come to exert near-symphonic control over the fear, anger, and tribalism of large sectors of the American public. The phrase “political influencer” calls to mind names like Tucker Carlson and Candace Owens, but I mean it to refer to any content creator, pundit, journalist, or even politician with an active online presence oriented around the production of political content—so perhaps hundreds of thousands of users with followings of varying sizes. Critically, this group is not a random selection from social media’s overall user base, but skews wealthier and more educated, meaning its greater online influence is likely matched by greater offline influence as well.

Despite accounting for only a small slice of the online population, this new influencer class is coming to dominate the market for political communication. In the process, it is transforming America’s perception of itself, which, since America is a social entity constituted in part by its self-perceptions, just amounts to saying it is transforming America.

Robertson, del Rosario, and Van Bavel (2024) outline this group’s newfound influence in their recent paper “Inside the funhouse mirror factory: How social media distorts perceptions of norms.” It is worth quoting their analysis at length:

Online discussions are dominated by a surprisingly small, extremely vocal, and non-representative minority. Research on social media has found that, while only 3% of active accounts are toxic, they produce 33% of all content. Furthermore, 74% of all online conflicts are started in just 1% of communities, and 0.1% of users shared 80% of fake news. Not only does this extreme minority stir discontent, spread misinformation, and spark outrage online, they also bias the meta-perceptions of most users who passively “lurk” online. (p. 1, my italics)

[With respect to] political discussions [in particular], the people who post frequently on social media are often the most ideologically extreme. Indeed, 97% of political posts from Twitter/X come from just 10% of the most active users on social media, meaning that about 90% of the population’s political opinions are being represented by less than 3% of tweets online. This is a marked difference from offline polling data showing that most people are ideologically moderate, uninterested in politics, and avoid political discussions when they are able. (p. 2)

The result is that Americans are coming to see each other as much more politically extreme than they in fact are—or at least were. In the words of political scientist Henry Farrell, Americans have internalized a “malformed collective understanding,” in this case a vision of their collective identity that casts them as especially angry, pessimistic, and tribal.

The idea that this understanding is “malformed” is a tricky one, though. It invites the question: how malformed, and for how much longer? While in the past, survey evidence has indeed shown Americans to be moderate and relatively uninterested in politics, it is possible, and even probable, that social media is changing that. If in response to believing their peers are more politically extreme, people in turn become more extreme in reaction, then Farrell’s ‘malformed collective understanding’ may end up a self-fulfilling prophecy. Perhaps the more extreme content we ingest, the more likely it is we internalize this style of political communication and begin posting extreme content ourselves. There is already some limited evidence to support this possibility.

Kim et. al. (2021) found, for instance, that those exposed to toxic ‘featured’ comments on Facebook posts were more likely to post toxic comments later of their own volition. Brady et. al. (2021) found that Twitter users “conform their outrage expressions to the expressive norms of their social networks,” (p. 1) and become more likely to express outrage when their surrounding social network is itself more extreme (p. 8). And in an as-yet unpublished preprint, Brady et. al. (2025) found that in a simulated Twitter-like environment, “engagement-based algorithms” changed participants’ perception of descriptive and prescriptive norms surrounding “ingroup-aligned, moral and emotional political content,” and ultimately made them more intent on posting this content themselves.

These results seem consistent with widespread anecdotal evidence of once-moderate friends and family falling down political or conspiratorial ‘rabbit-holes,’ often via involvement with insular online communities (e.g. QANON). At a broader level, it is hard to deny that the expressive norms governing our political culture have changed (witness the much-discussed expansion of the Overton window), and that this has made extreme political speech (open Nazism, racism, misogyny, antisemitism and so on) more common.

If this is all downstream of a small and relatively well-off group of high frequency posters (some of our elected officials among them), that would suggest what we might call an ‘elite radicalization’ theory of online politics. The idea is that social media has empowered a (relatively) small group of political influencers who, in response to the perverse incentives created by attentional negativity bias, have disseminated ideas that make people more angry, fearful, and extremist.

Source: Me

As represented above, those ideas spread both via influencing other elites (think podcast and talk show hosts like Joe Rogan or Jon Stewart), as well as by influencing typical consumers of online political content directly. Regardless of the exact transmission pathway, social media’s vast webs of interaction ensure that whatever effects online political content has, its reach is sure to be very far.

Nor does this content just ‘bounce off’ political junkies already firm in their preexisting views. There is strong causal evidence that political influencers have the power not just to reflect, but to shape the attitudes of their followers. In a recent preprint, Rathje et. al. (2025) found in two different field experiments (n = 494, n = 1,133) that unfollowing “hyperpartisan social media influencers" on Twitter (they give the examples of Palmer Report on the left and Breitbart News on the right) “improved [participants’] recent feelings toward the out-party by 23.5% compared to the control group, with effects persisting for at least six months,” (p. 2). This should put to rest the idea that inflammatory political content is a purely ‘demand-side‘ phenomenon. Yes, this demand has always existed, but it is possible for the attitudinal effects of existing supply to increase it even further.

Perhaps this is not on its own cause for serious alarm, though. However unpleasant online political influencers’ ideas might be, the question is whether they transcend the digital realm to influence people’s real-world political behavior. If not, then maybe we can relax in the confidence that the apparent rise in extremist attitudes is confined to an annoying, but harmless class of digital ‘LARPers.’

With apologies to those in the market for this kind of apparently savvy, relieving narrative, there is a convincing base of evidence showing just the opposite. By amplifying extreme content online, social media does appear to catalyze more extreme political behavior offline as well. The following are quasi-experimental studies that take advantage of exogenous variation in how different social networks have spread in order to study their political impacts. Their findings are relatively univocal:

Bursztyn et. al. (2019) instrumented on variation in city-level penetration of VK (the dominant Russian social network) in Russia and found that greater social media use increased the prevalence of xenophobic attitudes, and lead to more hate crimes in cities with higher pre-existing levels of nationalism.

Sabatini and Sarracino (2019) instrumented on variation in broadband/high-speed internet coverage in Italy, and found that greater social media use significantly decreased measures of social, particularized, and institutional trust.

Enikolopov, Makarin, and Petrova (2020) also instrumented on variation in city-level penetration of VK in Russia and found that a 10% increase in penetration into a given city increased the probability of a protest there by 4.6% and the number of protesters by 19%.

Schaub and Morisi (2020) instrumented on variation in municipality-level broadband coverage in Germany and Italy and found that greater access to online communication platforms made Germans and Italians more likely to vote for the populist AfD and M5S parties, respectively.

Müller and Schwarz (2021) used a slightly different approach—instrumenting on variation in Facebook and internet outages in Germany, rather than access—and found that greater social media use predicted increased hate crimes against refugees.

Müller and Schwarz (2023) instrumented on variation in the number of Twitter users across counties induced by early adopters at the 2007 South by Southwest (SXSW) festival and found that higher Twitter use led to a sizable increase in anti-Muslim hate crimes during the 2016 Republican primary, likely driven by future president Donald Trump’s tweets. In particular, they found that “a one standard deviation-higher exposure to Twitter [was] associated with a 32 percent … increase in hate crimes.” (p. 272)

These studies’ geographic range underscores that their concerning effects are robust to meaningful changes in context. And since they aggregate data across whole countries and municipalities, they’re able to account for the kinds of spillover effects that individualized experiments ignore. At least within the Western world broadly construed, the spread of social media has consistently produced more extreme forms of political behavior, including xenophobic violence in the most severe cases, but also protests, shifting voting patterns, and presumably much more that scholars have yet to study in detail.

While the evidence on this front is more limited in the case of the United States, the graph below does not inspire much optimism—note the hinge point around the late aughts when social media (and smartphone) use first became pervasive.

Source: Journal of Democracy

Social media is not, then, just a vast machine for disseminating extreme ideas—the most successful of which seem to have a right-wing populist flavor—but also one for shaping practical politics, including in its highest-stakes moments. As the quasi-experimental research reinforces, it does so in large part by empowering elite political influencers who, far from being just ‘terminally online’ pests, have the power to genuinely refigure their home countries’ political terrain.

While much of the preexisting research on social media and politics is focused on the political right, it is worth flagging that the elite radicalization theory also helps explain the strand of more radical left-wing politics that has gained prominence over the last decade and a half. While, as a factual matter, left-wing political influencers are far less dangerous (see the preceding graph), it is no less true that they have successfully exploited attentional negativity bias for both personal and political gain, shaping our current political culture in the process. As it happens, I am more sympathetic to their views, and so think some of their impact has been positive. But that is my own personal view, not an implication of the theory.

3.2. Elite radicalization and two-party politics

One of the core advantages of the elite radicalization theory is that it can account for social media’s incendiary effects on politics without committing to any particular relationship between social media and political polarization. While I hope the theory will be independently interesting to readers, it is for this reason that it can minimally act as a “so what?” objection to Williams.

Note that other common accounts of social media’s relationship with politics build in some commitment to polarization from the outset. For instance, ‘echo-chamber’ or ‘filter-bubble’-based theories argue that social media sorts people along partisan lines, in which case it would make sense for those inside a given bubble to become more polarized against those outside.

But the elite radicalization theory makes no such commitment. Of the elite influencers empowered by social media, many have views orthogonal to the classical ideological divide between Republicans and Democrats. While the content they create is often angry, fear-mongering, and prejudicial, it is only occasionally aligned with one of the U.S.’s two major political parties. More often, it has a populist flavor that targets some nebulous group of elites spanning party lines. And much of it also straddles the divide between political and apolitical, weaving discussion of, say, the economy or the Epstein files into broader conversations about pop culture.

As a result, and contra much presumptive emphasis on polarization, online political content may decrease the intensity of Democrat or Republican party identification, not in spite of, but because it makes people more outraged about politics. That outrage, for instance, may make independents less likely to gravitate toward the Republican or Democratic party. It might also make those who are already party members dislike their party more, perhaps enough to leave it altogether. This would be consistent with the rising public distrust of both parties, the frequent infighting that now defines Democratic party politics, and the simmering tensions between the more nativist and more business and tech-oriented factions of the Republican party.

If social media does in fact decrease the incidence or intensity of major party identification overall, then affective polarization—since it is defined in terms of major party allegiances—would be the wrong metric to look at in order to understand social media’s effects on politics. While this is only speculation for now, the graph below showing a rising share of Americans identifying as independents is at the very least suggestive (again note the hinge point just before 2010).

Source: Gallup

None of this is to say that social media definitely doesn’t increase polarization. But at minimum, this is by no means necessary for the elite radicalization theory to succeed. With respect to ‘affective’ polarization in particular, the theory suggests that social media’s harms probably have more to do with the ‘affective’ part, than the ‘polarization’ part. In other words, while social media makes political discourse more affectively or emotionally intense, it may do so without making avowed Democrats or Republicans dislike each other more than they already did.

If the effects of social media go even further to the point of convincing a significant number of Americans to redefine as independents, that would not only make affective polarization less relevant, but might even slow its increase. This could happen if those convinced to leave their party exhibited higher-than-average-negativity towards the out-party when they were still members. That may seen counterintuitive, since presumably the Republicans (Democrats) who hate Democrats (Republicans) the most would be among those most loyal to the party, but it may be that high out-party hate is a proxy for greater disagreeableness generally, in which case those who are fed up enough with the two-party system to quit it may also be among those who once exhibited the most out-party hatred.

There is not yet sufficient evidence to confirm or deny any of these suggestions. But in general, academics and public commentators alike should be mindful that in addition to spreading political extremism, social media may also be shifting the underlying axis on which those extremes exist. This could have serious consequences for social-scientific measurements of political attitudes, not least that of affective polarization.

5. Conclusion

By incentivizing the creation of disproportionately negative and sensational content, and by in turn inducing more extreme, even violent political behavior, social media has almost certainly played a major role in the destabilizing political ructions of the last fifteen years—in particular in the U.S., but probably across Europe as well.

Elite political influencers, both those hailing originally from politics, television, and journalism, as well as those native to digital media platforms, have played a critical mediating role. It is they who generate much of the most provocative content, as well as who influence the outlooks and incentives of other elites, allowing their influence to spread well beyond the confines of the major social networking platforms.

While Williams is correct that we have not seen a large spike in affective polarization as a result, we have seen spikes in angry, fearful, and identitarian political speech. We have also seen large spikes in political violence. And though I will not pretend to have made the case for their direct causal connection to social media, we have also seen the rise of a more radical strain of progressive politics, as well as the once-fledgling, now all-powerful MAGA movement figureheaded by Donald Trump, both of whose successes seem hard to separate from the belligerently online politics of the 2010s. All of this has come to pass even as polarization has increased at an overall smooth rate.

As the digital media revolution continues to transform social life in the U.S. along almost every conceivable dimension, it is critical that we have a clear-eyed view of its effects. Williams is right to challenge certain aspects of the panic surrounding social media, such as the often confused and imprecise discourse surrounding “misinformation.” But when it comes to the technology’s effects on politics writ large, he has not taken a sufficiently holistic look at the evidence.