Over the past several decades, Americans have become increasingly divided on political issues, but this trend does not appear to be occurring on a global scale. Recent papers published in Royal Society Open Science introduces a new way to measure these divisions using machine learning. The results reveal that polarization in the United States sharply spiked between 2008 and 2020, providing a new perspective on how political disagreements shape globally.
Political polarization generally refers to the division of society into two distinct and opposing groups. Researchers often try to measure this by looking at disagreements among people who support major political parties. Some scientists focus on emotional divisions, such as measuring how much opposing groups dislike each other.
Other researchers focus solely on policy disagreements, known as issue polarization. Traditional approaches to measuring issue polarization have some notable limitations. In the United States, for example, people may just be getting better at fitting their pre-existing beliefs into the correct political party.
This positioning process is called “sorting.” More sorting can make the country appear more divided, even if people’s actual opinions on the issues have not changed. Globally, it is difficult to measure polarization by comparing major political parties.
Many countries have two or more major political parties, while others have only one. To overcome these hurdles, researchers at the University of Cambridge have developed a new measurement tool. Researcher David Jack Young from the Department of Psychology led the study.
He was joined by Lee DeWitt, who heads the university’s political psychology lab. They wanted to track polarization based on people’s actual policy beliefs, rather than the political labels they chose for themselves. The research team turned to a machine learning technique called the K-means clustering algorithm.
This algorithm looks at large groups of people and automatically sorts them into clusters based purely on shared opinions. Imagine plotting everyone’s political opinions on a giant graph, and finding similar opinions close together and opposing opinions far apart. The algorithm finds the centers of these opinion clusters and assigns each person to the closest group.
It’s not about whether a person considers themselves liberal or conservative. After the algorithm forms the two groups, the researchers look at three different characteristics. The first is separation, which measures how far apart the average opinions of two groups are.
The second characteristic is dispersion. This examines how dispersed or cohesive the opinions are within each group. The third feature is size equality, which checks whether two groups contain similar numbers of people. Researchers argue that a society is highly polarized when it is divided into two groups that are widely separated but internally united and roughly equal in size.
In the first study, researchers applied the method to the United States. They analyzed more than 35,000 survey responses collected between 1988 and 2024. This data comes from the American National Election Study, a long-running public opinion project.
The survey asked Americans their opinions on 14 specific social and economic topics. These topics included the acceptability of abortion, the importance of traditional family ties, and whether the government should fund health insurance. Questions also addressed government efforts to address economic disparity and racial discrimination.
The researchers found that the gap between the two opinion groups widened by 64% during this period. In particular, the distance between clusters widened for each problem measured.
Certain social issues have shown particularly dramatic changes. In 1988, people with generally conservative views did not necessarily have restrictive views on abortion. By 2024, conservative opinion on a variety of topics had become firmly integrated with restrictive views on abortion and a strong emphasis on traditional family bonds.
On other topics, we found the two groups moving in diametrically opposite directions. Opinions about government-funded health insurance and the ongoing effects of racial inequality sharply divided the clusters. The groups have actively distanced themselves from each other for decades on these issues.
Most of this increase occurred during a specific period. Between 1988 and 2008, the distance between the two clusters remained relatively flat. But from 2008 to 2020, the two groups grew further apart.
“Our research shows that 2008 marked a major turning point for the left-right divide on many issues that define modern American politics,” DeWitt said. He noted that the population has generally moved to the left on many issues during this century. The data showed that left-wing groups have become substantially more progressive over the decades.
In contrast, the right-leaning cluster changed only slightly more conservatively. At the same time, the two groups remained internally cohesive and about the same size. The researchers also found that Americans are aligning their political labels more closely with their issue groups than ever before.
Young explained that ideological consistency is becoming more common. “It used to be that someone who had a left-wing view on one issue might have a right-wing view on another issue. That’s less the case now.”
This particular timeline calls into question the idea that polarization is just an inevitable feature of human psychology. If human nature were the only cause, divisions would probably increase at a constant rate. Rather, this rapid spike suggests that environmental factors, such as the financial crisis, changes in communication technology, and changes in political leadership, may be at play.
In the second study, the researchers expanded their scope to examine overall patterns. They analyzed survey responses from more than 173,000 people in 57 countries. This data comes from the World Values Survey and the European Values Survey, collected between 1999 and 2018.
This allowed the team to see if the trends observed in the United States were also occurring in other regions. They found no clear evidence that issue polarization is increasing globally. Intergroup separation has increased slightly globally, but cohesion within groups has also declined.
When we look at what is causing division globally, we find that cultural issues are the most powerful factor. Topics such as the acceptance of abortion, divorce, and homosexuality caused the greatest divisions between opinion groups. Differences over economic policy and democratic institutions did not divide the group as sharply.
The researchers also found that a country’s level of economic and social development changes how its population is divided. They measured this using the Human Development Index, a measure that takes into account life expectancy, education, and national income. In countries with low development scores, the algorithm typically found a culturally conservative majority and a culturally liberal minority.
In countries with higher development scores, the two groups tended to be more equal in size and more liberal overall. The reason the United States stands out globally is that the two opinion groups in the United States have long been roughly equal in size. Young noted that this even division may help explain why America’s political divisions feel so sharply divided.
The researchers also looked for social factors that predict how countries break up. They found that countries with higher ethnic diversity tended to have greater distance between groups on issues. This diversity was measured by the probability that two random citizens belong to different ethnic groups.
Furthermore, in countries with high wealth inequality, domestic opinion groups were less cohesive. This suggests that economic disparities can create more disagreements within political factions. Even people who agree on cultural issues may argue about economics if wealth is distributed unevenly.
Although the new measurement tool provides fresh insights, the researchers acknowledge that their study has some limitations. One limitation is that the clustering algorithm forces everyone into one of two groups. In reality, many people may not fit into either category.
Future research could consider alternative algorithms that allow individuals to remain unassigned if their views do not match the dominant cluster. Scientists can also test algorithms that classify populations into three or more groups. This has the potential to capture different power relations in countries with multiple political factions.
The study is also observational, meaning it can show a pattern but cannot prove its cause. It is impossible to say with certainty whether factors such as wealth inequality directly cause changes in public opinion. This association may be driven by unmeasured external factors.
Finally, the findings are limited to the specific research questions available in the dataset. A clearer picture of political polarization can be obtained by exploring how the nation is divided over a variety of contemporary issues.
The study, “A New Measure of Issue Polarization Using K-Means Clustering: US Trends and Predictors of Polarization Worldwide, 1988-2024,” was authored by David Jack Young, James Ackland, Andreas Kapounek, Jens Koed Madsen, Lara Jane Greening, and Lee de-Wit.
