Working Paper: CEPR ID: DP16624
Authors: Elliott Ash; Ruben Durante; Mariia Grebenshchikova; Carlo Schwarz
Abstract: We propose and validate a new method to measure gender and ethnic stereotypes in news reports, using computer vision tools to assess the gender, race and ethnicity of individuals depicted in article images. Applying this approach to 700,000 web articles published in the New York Times and Fox News between 2000 and 2020, we find that males and whites are overrepresented relative to their population share, while women and Hispanics are underrepresented. Relating images to text, we find that news content perpetuates common stereotypes such as associating Blacks and Hispanics with low-skill jobs, crime, and poverty, and Asians with high-skill jobs and science. Analyzing news coverage of specific jobs, we show that racial stereotypes hold even after controlling for the actual share of a group in a given occupation. Finally, we document that group representation in the news is influenced by the gender and ethnic identity of authors and editors.
Keywords: stereotypes; gender; race; media; computer vision; text analysis
JEL Codes: L82; J15; J16; Z1; C45
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| males and whites are overrepresented in news images relative to their population share (J15) | potential bias in media representation (J15) |
| news content perpetuates stereotypes (Z13) | associating blacks and Hispanics with low-skill jobs, crime, and poverty (J79) |
| strong association between the topics of crime and poverty and images of blacks and Hispanics (K42) | stereotypes persist in media portrayal (J15) |
| gender and ethnic identity of authors and editors influence representation of these groups in news images (J15) | diversity in newsroom staff can impact media biases (J15) |