Skip to main content
Official Logo of Columbia Business School
Academics
  • Visit Academics
  • Degree Programs
  • Admissions
  • Tuition & Financial Aid
  • Campus Life
  • Career Management
Faculty & Research
  • Visit Faculty & Research
  • Academic Divisions
  • Search the Directory
  • Research
  • Faculty Resources
  • Teaching Excellence
Executive Education
  • Visit Executive Education
  • For Organizations
  • For Individuals
  • Program Finder
  • Online Programs
  • Certificates
About Us
  • Visit About Us
  • CBS Directory
  • Events Calendar
  • Leadership
  • Our History
  • The CBS Experience
  • Newsroom
Alumni
  • Visit Alumni
  • Update Your Information
  • Lifetime Network
  • Alumni Benefits
  • Alumni Career Management
  • Women's Circle
  • Alumni Clubs
Insights
  • Visit Insights
  • Digital Future
  • Climate
  • Business & Society
  • Entrepreneurship
  • 21st Century Finance
  • Magazine
Business & Society, Economy & Policy

Can Machine Learning Detect Political Bias in Economics Papers?

Average Read Time:

New research out of Columbia Business School reveals that economists’ political leanings can be detected in their academic writing — and may even influence their empirical findings. 

Article Author(s)
  • Stephanie Walden
Published
May 30, 2025
Publication
Research In Brief
Jump to main content
Student in a library
Category
Thought Leadership
Topic(s)
Economics and Policy
Elections
Save Article

Download PDF

About the Researcher(s)

Bruce Kogut

Bruce Kogut

Sanford C. Bernstein & Co. Professor of Leadership and Ethics
Management Division
Academic Director of BAID
Hub Faculty

0%

Share
  • Share on Facebook
  • Share on Threads
  • Share on LinkedIn

When economists write academic papers about the effects of raising the minimum wage or changing tax policy, they’re supposed to let the data do the talking. But new research suggests that beneath the technical language and mathematical formulas, partisan ideologies may leave subtle but detectable fingerprints on such studies — and the policy recommendations they produce.

Bruce Kogut, the Sanford C. Bernstein & Co. Professor of Leadership and Ethics in the Management Division at Columbia Business School, has long been fascinated by how values and belief systems shape decision-making in academia and industry. This curiosity drove him to explore whether economists’ personal ideologies might also seep into their academic research. 

“In economics, researchers often find themselves in scenarios where one group thinks the answer is X and the other thinks it’s Y,” says Kogut. “We wanted to take a programmatic, research-driven approach to that discrepancy.”

While political bias in the media has been studied extensively, Kogut and co-authors Zubin Jelveh and Suresh Naidu wanted to know if they could detect partisan leanings in academic economics papers — works that are supposed to be rooted in scientific objectivity rather than political ideology. 

Their groundbreaking analysis revealed surprising answers. 

“There’s a correlation between political slant and research outcomes. Looking at the example of taxation, if you take the most left-wing versus the most right-wing researcher, given their slant, it will move the needle of the results by as much as 17%,” says Kogut. “That’s not minor — it’s a big number. But it is a less extreme difference than I think many would have forecasted.”

Key Takeaways:

  • Using machine learning analysis of academic papers, researchers found that economists’ political ideologies can be predicted from their writing style and word choices, even in technical academic work.
  • The study reveals inherent ideological sorting across economics specialties — with fields like labor economics trending liberal and finance or macroeconomics trending conservative.
  • Notably, the findings suggest that researchers’ political preferences may influence their research outcomes. As an example, when studying top-income tax rates, left-leaning economists’ methods led to optimal rates around 77%, while right-leaning economists’ approaches suggested rates closer to 60%.
  • The research highlights the need to acknowledge and account for potential ideological bias in economic research — especially in analyses that directly inform policy decisions.

Mining Academic Papers for Political Fingerprints

To investigate this topic, the researchers first identified published economists’ political orientations using public records of campaign contributions and signatures on partisan petitions to establish what researchers call “ground truth.”

The team then obtained the full text of more than 62,000 research articles published across 93 economics journals, as well as more than 17,000 working papers from the National Bureau of Economic Research. The papers spanned decades and included both published and pre-publication works.

To analyze word and phrase usage in the texts, the researchers applied natural language processing techniques like the random forests algorithm — an ensemble learning method that builds multiple decision trees and combines their outputs to improve predictive accuracy. The researchers then compared inferred political slants from the writing with the established ground truth. Throughout the endeavor, they also applied controls for accuracy, including adjustments for journal-specific language trends, topic-based biases, and methodological variations.

How ‘Left’ and ‘Right’ Language Appears in Academic Work

The results of this effort were striking. The researchers found they could accurately determine an economist’s political orientation based solely on their academic writing.

The research revealed telling patterns in word choice even within the same academic topics. When writing about education, for example, left-leaning economists more frequently used terms like Head Start (the federal program for children) and affirmative action, while right-leaning economists tended to use terms like human capital and cognitive skill — different language to describe aspects of the same field.

“There are really two different vocabularies for those on the left and those on the right,” says Kogut, adding that certain themes also correlated with ideological leanings. “Republicans care about the gold standard, for instance. They’re more focused on things like the ‘right to work,’ which in the United States means no unions. For Democrats, topics like gender and welfare show up more often.”

The study also found clear ideological sorting across different specialties within economics. “Some areas were more touched by slant. For example, labor economics is clearly more of a Democratic area,” he says. More conservative fields include finance and macroeconomics.

Notably, in fields that are policy-relevant, the research revealed that an economist’s predicted political ideology often correlates with their empirical findings. For instance, when studying the effects of minimum wage laws or unions, conservative-leaning economists were more likely to find negative economic impacts, while liberal-leaning economists found more positive or neutral effects.

The correlation was pronounced even when controlling for methodology, data sources, and other factors.

The Human Factor in Academic Objectivity

This research raises important questions about objectivity in economic research and its role in policymaking. While the research doesn’t claim that ideology directly causes specific research findings, it suggests that political preferences may influence how economists approach their work — from the methods they use to analyze data to the questions they choose to study in the first place.

At least in part, this can be chalked up to the very human tendency to gravitate toward research that aligns with one’s own values. “If you didn’t have those political beliefs, maybe you wouldn’t be drawn to the research — people tend to work on things they’re passionate about,” he says. “There’s a benefit to that, but there’s a danger that you’ll be biased in how you see things, too.”

Building on the insights from this project, Kogut is now turning his attention to another topical issue in academic integrity: p-hacking, or the manipulation of data analysis to achieve desired results. While economics has made strides in addressing p-hacking, there’s still work to be done in this and other arenas, such as medical clinical trials. With co-authors Jorge Guzman and Jelveh, Kogut has just completed new research on the effects of career pressures influencing the quality of analytical methods and statistical results in industry and academics.

There’s a simple principle behind both this recent research and the subsequent areas Kogut hopes to explore: Acknowledging the human factor is essential to improving scientific integrity. Rather than seeing the results of this study as undermining economics’ validity, Kogut believes it’s an opportunity to embrace greater transparency about these influences as part of the scientific process. 

“Science happens within a context of humans who are largely trying to do good science and also to succeed,” he says. “We should allocate even more effort into improving the quality of science and making it more transparent. Governance is advancing. Currently, economics is probably leading the efforts in social sciences to influence research quality through better practices and institutions. This remains probably an even greater challenge for other fields and disciplines. Doing this research keeps us well employed.”

 

Politics in research

 

Adapted from “Political Language in Economics” by Bruce Kogut and Suresh Naidu from Columbia University and Zubin Jelveh from the University of Maryland.

About the Researcher(s)

Bruce Kogut

Bruce Kogut

Sanford C. Bernstein & Co. Professor of Leadership and Ethics
Management Division
Academic Director of BAID
Hub Faculty

Related Articles

Insights
Management
Marketing
Marketplace
Research Findings
Technology
Date
July 15, 2025
A person smiling
Insights
Management
Marketing
Marketplace
Research Findings
Technology

How to Win More Bookings on Airbnb

A Columbia Business School study found that Airbnb hosts who smile in their profile photos see 3.5% more bookings. Smiling reduces customer uncertainty and increases trust, with implications for hospitality, e-commerce, and even online dating.
  • Read more about How to Win More Bookings on Airbnb about How to Win More Bookings on Airbnb
Business and Society
Future of Work
Leadership
The Workplace
Date
July 03, 2025
A team of workers
Business and Society
Future of Work
Leadership
The Workplace

How Team Diversity Prevents Collusion in the Workplace

Discover how team diversity and incentive design help prevent workplace collusion, based on new Columbia Business School research.
  • Read more about How Team Diversity Prevents Collusion in the Workplace about How Team Diversity Prevents Collusion in the Workplace
Date
June 17, 2025
Escalators in a building

How “Better World” Innovation Is Redefining Business Success

A new CBS-backed framework redefines innovation to prioritize sustainability, equity, and long-term value in today’s business landscape.
  • Read more about How “Better World” Innovation Is Redefining Business Success about How “Better World” Innovation Is Redefining Business Success
Algorithms
Artificial Intelligence
Business and Society
Digital Future
Entrepreneurship
Management
Date
May 05, 2025
Illustration of hands, resumes and laptop
Algorithms
Artificial Intelligence
Business and Society
Digital Future
Entrepreneurship
Management

Did AI Write That Pitch? The Impact of Generative AI on Hiring and Startup Evaluations

Research from Columbia Business School examines the challenges posed by generative AI in hiring and entrepreneurial pitching, offering insights into when AI helps — and when it hinders.

  • Read more about Did AI Write That Pitch? The Impact of Generative AI on Hiring and Startup Evaluations about Did AI Write That Pitch? The Impact of Generative AI on Hiring and Startup Evaluations

External CSS

Articles A11y button

Official Logo of Columbia Business School

Columbia University in the City of New York
665 West 130th Street, New York, NY 10027
Tel. 212-854-1100

Maps and Directions
    • Centers & Programs
    • Current Students
    • Corporate
    • Directory
    • Support Us
    • Recruiters & Partners
    • Faculty & Staff
    • Newsroom
    • Careers
    • Contact Us
    • Accessibility
    • Privacy & Policy Statements
Back to Top Upward arrow
TOP

© Columbia University

  • X
  • Instagram
  • Facebook
  • YouTube
  • LinkedIn