New research finds a persistent and growing leftward tilt in the social sciences

A new study published in Theory and Society suggests that published research in the social sciences has leaned consistently to the political left for more than six decades. The findings indicate that this leftward tilt has grown stronger over time, particularly regarding social and cultural issues. This provides evidence that the academic publishing environment has grown increasingly uniform in its political orientation.

Past surveys have consistently shown that college and university faculty in the United States tend to identify with left-leaning political views. James Manzi, a researcher at the University of Oxford, wanted to know if this political preference actually appears in the published academic work itself. Prior to recent technological advances, analyzing the political content of hundreds of thousands of scientific texts was too expensive and time-consuming for human readers to accomplish.

Manzi decided to use artificial intelligence to read and code these massive amounts of text to see how academic outputs have shifted over a 65-year period. Looking at published output provides a stable way to see what disciplines consider important and how they frame problems. Instead of asking how individual scientists vote, this approach evaluates the actual ideas making their way into the scientific record.

The researcher collected exactly 599,194 English-language abstracts from social science articles published between 1960 and 2024. Abstracts are short summaries of research papers that appear at the beginning of an academic article. The text came from 367 academic journals representing eleven different disciplines, including economics, sociology, and political science.

To process this massive dataset, Manzi used a large language model, which is an artificial intelligence program that recognizes patterns in vast amounts of text, to act as a digital reader. To ensure the program evaluated every abstract consistently, the researcher provided a strict rubric based on the United States political spectrum as it existed in the year 2025. The program rated each abstract on a scale from zero to ten, where zero represented the far right, five was politically neutral, and ten represented the far left.

To help the program understand these ratings, the scientist gave it specific, modern political anchors. The instructions linked each number on the scale to the contemporary policy positions of recognizable political figures and think tanks. Using this rubric, the program answered specific questions about each abstract, including whether its claims were relevant to modern political debates and where its ideas would fall on today’s economic and cultural spectrums.

Out of the entire sample, the program identified 180,311 abstracts that directly discussed current political or social debates, and the scientist focused the main analysis on this subset.

Before running the main analysis, the researcher validated the artificial intelligence by having it score summaries of articles published by known political organizations. The computer program successfully placed these organizations on the correct side of the political spectrum. The researcher also tested the tool on theoretical mathematics papers, which correctly yielded no politically relevant results.

The main data showed that roughly 90 percent of the politically relevant social science articles leaned to the left. The average political stance of every single social science discipline remained left of center for every year during the entire measured period. The consistently left-leaning averages were driven largely by a near absence of right-leaning works in most disciplines.

Different academic fields did follow slightly different paths over time. Disciplines closely tied to public policy, such as economics and political science, showed a slight moderation toward the right during the 1970s and 1980s. After 1990, these policy-focused fields shifted back toward the left.

In contrast, disciplines focused on identity and culture, such as sociology and gender studies, showed a continuous leftward shift over the entire time frame. Around the year 2010, the leftward movement in several of these fields began to accelerate. The researcher noted that this acceleration looks like a continuation of a long-running trend rather than a sudden break from the past.

The scientist also observed that fields with stronger leftward orientations tended to have less political diversity among their published works. Over time, the social sciences became more ideologically uniform. As disciplines drifted further left, the variety of political viewpoints found in their abstracts shrank.

To understand the source of this ideological drift, the scientist looked at whether individuals were changing their minds or if the fields were simply bringing in different people. The data indicates that the leftward trajectory is primarily driven by new contributors entering the field with more progressive views. Older academics changing their views over the course of their careers played a much smaller role in the overall shift.

The analysis also split the ratings between economic issues and sociocultural issues. The artificial intelligence consistently rated abstracts as further left on social and cultural topics than on economic topics. This gap widened significantly over the decades.

This suggests that scientists grew increasingly progressive on cultural issues even when their economic views remained slightly closer to the center. It mirrors a broader societal trend where highly educated professionals often hold moderate economic views alongside progressive social ideals.

Readers should avoid misinterpreting these patterns as direct proof of intentional bias or the deliberate suppression of right-leaning views. The study measures the political tone of the final published work, but it does not explain why that tone exists. It is highly possible that scientists are increasingly studying topics like climate change or racial justice, which the artificial intelligence might naturally code as left-leaning.

The artificial intelligence also only looked at the short abstracts, not the full text of every scientific paper. Scientists often use abstracts as promotional summaries to catch the attention of journal editors. Researchers might rely on trendy political buzzwords to increase their chances of publication. This means the abstracts might sound more ideological than the actual data and methods found inside the full papers.

The study also applied a strict United States political framework to all English language abstracts. Because science is a global enterprise, a researcher in the United Kingdom or Australia might use concepts that read as progressive in America but are considered politically moderate in their home countries.

Finally, large language models carry their own biases based on the internet data used to train them. This internal programming might influence how the software defines political neutrality or interprets nuanced academic arguments.

The study, “The ideological orientation of academic social science research 1960–2024,” was authored by James Manzi.