Visual bias in the news

11 Apr 2022

You may have heard that pictures are worth a thousand words, but did you imagine they influence how you read the news? Novel research by Giulia Caprini, to be presented at the Royal Economic Society's annual conference in April 2022, shows that the images employed to illustrate written news pieces can have a powerful influence, which media outlets may use to sway their audience.

Newsmakers, the study shows, choose as leading pictures images whose content reflects their political preferences. As a consequence, liberal and conservative newsmakers "speak" systematically different visual languages. This “visual bias” of news coverage, often less apparent than bias expressed in text, affects the way readers form opinions on issues in the news, and may reinforce the polarization of public opinion on political issues.

Focusing on the US media landscape, the study analyses 300,000 leading images from news published in 2020 by the main US news outlets. The researcher passed the pictures to computer vision algorithms to extract information on their content, such as the subjects and objects depicted, their pictorial attributes, or contextual aspects of each photo.

This information was then gathered to form a “visual vocabulary” and used to study the overall visual language of the different news sources. This analysis shows that liberal and conservative news outlets adopt very different visual jargon. Their visual language appeared politically divided, particularly in the coverage of politics, security, or health topics.

To investigate the effects of this partisan visual language on readers’ opinions, the researcher then set up an experiment. She recruited a nationally representative sample of 2,000 US residents to participate in an online survey.

Participants were shown news pieces on different topics, and asked to express their opinion on each topic, as in a regular opinion survey. All the written news contents (texts, headlines, bylines, sources, dates, etc.) were identical for every participant. Instead, the leading picture of each news piece varied randomly, among three alternatives: its visual language could resemble that of a Republican-leaning news source, a Democrat-leaning news source, or a politically neutral source.

The experimental results revealed that, by changing the pictures in the news, it is possible to influence respondents’ opinions on topics such as politics, the economy, security, or public health. In fact, readers whose news were illustrated by a “Republican-leaning” picture on average expressed a significantly more conservative opinion than readers whose news contained a “Democrat-leaning” one, despite reading the same written content.

For instance, relative to the Republican-leaning readers, the Democrat-leaning readers preferred a much lower budget for police forces (with a 4.5 billion $ difference), they were more dissatisfied with the pandemic management during Trump’s administration, they were more confident about lower future inflation, and they were more certain about Biden’s chances to revive the US-Iran nuclear deal.

Moreover, by considering participants' political leanings, the researcher found that readers of the news react more to pictures aligned with their political viewpoint. For instance, using as a reference point Republican affiliates who were exposed to neutral leading images, Republicans who were shown Republican-leaning images would be more distant (hence have a more extreme opinion) than Republicans exposed to Democrat-leaning pictures. The same dynamic applied also to the Democrat participants in the study.

These results indicate that news-leading pictures are more powerful in reinforcing prior beliefs than in promoting a different viewpoint. Given this, the presence of visual bias in news coverage causes public opinion to be more divided and polarized politically. This effect will be further amplified if people receive their news exclusively from like-minded sources, as often occurs when content is recommended by a social media algorithm.

Contact: Please contact RES to get in touch.

Notes to Editors:

The press release is highlighting research papers presented at the RES Annual Conference 2022 (#RES20220) for further information, please contact j.randalledwards@res.org.uk on 07970 201456 if you want the link to the full paper.

The Royal Economic Society’s popular 2022 virtual annual conference with in-person conference hubs, provides a forum for research, debate, and networking. The RES provides resources for economists and support for education and the training of students, teachers and researchers. 

The Royal Economic Society’s purpose is to promote the study of economic science. With over 6,500 members worldwide, RES is one of the oldest economic associations in the world. RES are a registered charity and membership is open to anyone who shares our aims and objectives.

The Society's two peer-reviewed journals - The Economic Journal and The Econometrics Journal – contain high quality papers from an international authorship. 

Please do tag RES on Twitter @RoyalEconSoc using #RES2022

Or further RES Social Media pages: Website  YouTube  LinkedIn