Americans have increasingly argued about when—and why—the U.S. has become sopolitically divided. Was polarization a slow burn stretching back decades, or a sudden rupture tied to more recent events?
A new large-scale empirical study suggests that almost all of the rise in U.S. polarization over political issues since the late 1980s occurred after 2008, driven by a sharp and sustained widening of ideological divides during the late 2000s and 2010s.
In the study, published in Royal Society Open Science, researchers introduce a machine-learning-based approach to measuring political issue polarization and apply it to decades of U.S. survey data, alongside comparative data from countries around the world.
Rather than relying on party labels or self-identified ideology, the study uses unsupervised learning techniques to detect how citizens’ policy views naturally cluster—and how far apart those clusters have drifted over time.
“Our study shows that 2008 was a major turning point for the divisions between left and right on many of the issues that define contemporary US politics,” co-author and professor at Cambridge University, Dr. Lee De-Wit, said in a press release. “Although sentiment varies by topic, the American public has moved left on many issues during the 21st century. This shift may surprise those familiar with the rightward turn of Republican leaders over the same period, as well as recent US headlines.”
In the study, researchers argue that polarization is not simply about people holding strong views, but about societies fracturing into distinct, internally coherent groups that increasingly clash across a wide range of issues simultaneously. By allowing machine learning algorithms to identify these groups directly from survey responses, researchers say polarization can be measured more objectively—and with far greater precision—than has previously been possible.
“There can be various reasons why voters feel they belong in a particular party, from media they consume to communities in which they live, even if many of their attitudes don’t fit,” co-author and professor of psychology at Cambridge, Dr David Young, explained “Clustering algorithms provide an objective, bottom‑up way to analyze shifts in political opinion beyond the ideologies of national parties and the self-identification of voters.”
A new way to measure polarization
Much of the debate over polarization has hinged on how it is measured. Traditional approaches often compare Democrats and Republicans, or liberals and conservatives, and track how far apart those groups are on specific issues.
However, those methods have drawbacks. Party coalitions shift over time, people “sort” themselves into parties that better match their views, and many citizens don’t identify strongly with any political label at all.
To address these issues, researchers used a machine learning algorithm called k-means clustering. Instead of asking which party someone belongs to, the algorithm looks at how individuals answer a wide range of policy questions and groups them into two clusters that are as internally similar—and as externally different—as possible.
In the U.S. analysis, the researchers applied this method to American National Election Studies data spanning from 1988 to 2024, covering issues ranging from abortion and traditional family values to government spending, economic inequality, and race-related policies.
The algorithm consistently identified two broad opinion clusters—roughly corresponding to left-leaning and right-leaning worldviews—but crucially, it did so without using party identification as an input.
Once those clusters were identified, the team measured polarization along three dimensions: how far apart the clusters were on average, how internally cohesive each cluster was, and how evenly sized the clusters were. Of these, one stood out.
The post-2008 polarization surge
Between 1988 and 2008, the distance between America’s two main opinion clusters barely budged. But after 2008, that changed dramatically. The separation between clusters rose sharply through 2020, before leveling off slightly in the most recent data.
“Separation increased between 1988 and 2024 by 0.14 scale points, from 0.22 to 0.36,” the authors report, noting that this represents a 64 percent increase overall—and that nearly all of it occurred after 2008.
Importantly, the other two dimensions of polarization remained largely stable. The clusters did not become more internally uniform, nor did one side come to numerically dominate the population. Instead, Americans increasingly sorted into two camps that were simply farther apart on a growing number of issues.
This pattern challenges narratives that frame polarization as a steady, decades-long climb. Instead, the data point to a prolonged plateau followed by a rapid escalation—a finding that aligns with broader social and technological shifts during the late 2000s, from the rise of social media to economic shocks and increasingly nationalized political conflict.
Not just parties, but packaged beliefs
One of the study’s more revealing insights is how polarization manifests across individual issues. Over time, the researchers found, the two clusters diverged on every issue they examined—not just a few high-profile flashpoints.
In some cases, both clusters moved in opposite directions, a classic picture of polarization. In others, both groups shifted in the same direction, but at different speeds, still increasing the distance between them.
The result is what researchers describe as “packaged” ideology: people on one side are now more likely to hold consistently aligned positions across social, economic, and cultural domains.
“What this effectively means is that whereas in 1988, people with generally right-leaning opinions did not necessarily have a more restrictive opinion on abortion than people with generally left-leaning opinions,” researchers explain. In contrast, “nowadays right-leaning views tend to be ‘packaged’ with restrictive views on abortion,” illustrating how polarization has increasingly bundled once-separate issues into coherent ideological camps.
That bundling effect helps explain why political disagreements today can feel so totalizing. It is not just that Americans disagree more, but that they disagree with the same people on nearly everything.
Machine learning meets political science
Beyond its conclusions about U.S. politics, the study is notable for drawing on machine learning and computational psychology to demonstrate how unsupervised algorithms can reveal latent structures in public opinion that traditional analyses may miss.
Clustering allows polarization to be measured without assuming in advance who the “sides” are—a critical advantage for cross-national comparisons.
In the paper’s second major analysis, the same approach is applied to survey data from more than 100 countries, revealing that cultural issues, rather than economic ones, tend to drive polarization globally, though in very different ways depending on a country’s level of development.
In the U.S. case, the technique also helps disentangle two often-confused phenomena: ideological sorting and genuine attitude divergence. Researchers found evidence that both have occurred. Americans have become better sorted into parties that reflect their views, but the views themselves have also moved further apart.
“In developing countries, we typically see a large culturally conservative majority and a culturally liberal minority. In highly developed countries, we tend to see socially liberal majorities, if only by a small margin,” Dr. Young said. “The United States is unusual in having a left and a right of roughly equal size. This has been the case for a long time, and it may help explain why polarisation in the US feels so intense.”
A turning point, or a pause?
One intriguing wrinkle in the data is what happens after 2020. The most recent election cycle shows a slight dip in measured separation, raising the possibility that the long surge in polarization may have slowed—or at least paused.
Researchers caution against concluding that Americans are becoming less politically divided. A single data point does not constitute a trend, and polarization levels remain far higher than they were prior to 2008. Even so, the pattern highlights that polarization is not an inevitable, ever-rising force, but one that ebbs and flows in response to political, social, and institutional pressures.
Ultimately, the study offers one of the clearest empirical demonstrations to date that U.S. political polarization did not rise gradually over the past four decades, but accelerated sharply after 2008.
What it does not yet explain is why that shift occurred—or which forces were most responsible for pushing Americans’ views further apart.
“In the past, someone with left-wing views on one issue might have held right-wing views on another. That’s rarer now,” said Dr. Young. He suggests that clearer ideological signaling from political leaders may play a role, adding, “I think this consolidation happens as people pick up clear cues from political elites about what goes together. It makes average positions more extreme, and widens gaps between US citizens.”
Nevertheless, researchers urge restraint in accepting that extreme polarization will be the new normal going forward. Rather, polarization has a history of inflection points, and with the right tools, it can be measured, understood, and perhaps—eventually—addressed.
“Political groups have moved further apart in recent years,” Dr. De-Wit added, “but these divisions are still fuzzy, and Americans are not split into entirely opposing tribes when it comes to the issues.”
Tim McMillan is a retired law enforcement executive, investigative reporter and co-founder of The Debrief. His writing typically focuses on defense, national security, the Intelligence Community and topics related to psychology. You can follow Tim on Twitter: @LtTimMcMillan. Tim can be reached by email: tim@thedebrief.org or through encrypted email: LtTimMcMillan@protonmail.com