Media Slant is Contagious: Here's Why

Table of Links Abstract and 1 Introduction 2. Data 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3.2. Classifying transcripts by TV source 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 4. Econometric Framework 4.1. Instrumental variables specification 4.2. Instrument first stage and validity 5. Results 5.1. Main results 5.2. Robustness checks 6. Mechanisms and Heterogeneity 6.1. Local vs. national or international news content 6.2. Cable news media slant polarizes local newspapers 7. Conclusion and References Online Appendices A. Data Appendix A.1. Newspaper articles A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.4. Included prime-time TV shows and A.5. Summary statistics B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B.3. Human validation of NLP model B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.6. Topics from the newspaper-based LDA model C. Results Appendix C.1. First stage results and C.2. Instrument exogeneity C.3. Placebo: Content similarity in 1995/96 C.4. OLS results C.5. Reduced form results C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.12. Mechanisms: Language features and topics C.13. Mechanisms: Descriptive Evidence on Demand Side C.14. Mechanisms: Slant contagion and polarization Abstract This paper examines the diffusion of media slant, specifically how partisan content from national cable news affects local newspapers in the U.S., 2005-2008. We use a text-based measure of cable news slant trained on content from Fox News Channel (FNC), CNN, and MSNBC to analyze how local newspapers adopt FNC’s slant over CNN/MSNBC’s. Our findings show that local news becomes more similar to FNC content in response to an exogenous increase in local FNC viewership. This shift is not limited to borrowing from cable news, but rather, local newspapers’ own content changes. Further, cable TV slant polarizes local news content. 1. Introduction Scientific consensus and conventional wisdom agree: Media bias is widespread. Media outlets often feature content in a way that is favorable to a particular political party or ideological perspective (e.g. Groseclose and Milyo, 2005; Harmon and Muenchen, 2009; Gentzkow and Shapiro, 2010; Puglisi and Snyder, 2011). These biases in reporting can impact public opinion (e.g., Chiang and Knight, 2011; Djourelova, 2023), and in particular, there is extensive evidence that higher exposure to Fox News (FNC) boosts Republican vote shares and other conservative interests (DellaVigna and Kaplan, 2007; Martin and Yurukoglu, 2017; Ash et al., 2021; Ash and Galletta, 2023). One unanswered question motivated by these findings is whether biased news messaging can spread to and distort other news outlets’ content. This issue is policy-relevant since media regulation in democracies typically aims at providing consumers with a competitive news market, with a wide, unrestricted choice among independent news sources. After all, diverse news sources only translate into diverse reporting if they are independent in their news making. This work studies cross-media influence in the context of U.S. cable TV news channels – Fox News, CNN, and MSNBC – which have a well-documented partisan slant. When there is higher viewership of a cable news channel among a newspaper’s readers, is the associated partisan slant reflected in that newspaper’s content? Our word defines slant in relative terms – whether a piece of content resembles Fox News rather than CNN or MSNBC. The first step in answering this question is to measure the similarity between cable news and local newspaper content. For this purpose, we build a corpus of 24 million article snippets from 600+ U.S. local newspapers in the United States from 2005 through 2008. We combine these texts with transcripts of 40,000 episodes from Fox News, CNN, and MSNBC. We use this parallel corpus to construct a novel measure of cable news slant – that is, we train a machine-learning model to predict, for a given body of text, whether it resembles the language by the relatively conservative network (FNC), rather than language by the relatively liberal networks CNN or MSNBC. We validate the model and the associated predictions with human annotations. We apply our model to the local newspaper article corpus. For each article, we have a text-based metric reflecting the similarity to content from FNC shows relative to CNN and MSNBC shows. We aggregate the article similarities at the newspaper level. We add metadata on newspaper circulation, television channel positioning, ratings, and political and demographic covariates. Then, we investigate whether relative similarity to content in a cable news network increases in response to higher viewership in a newspaper’s market. Cross-sectional estimates of this relationship would likely be confounded, for example, by more ideologically conservative counties having both higher Fox News viewership and more conservative local reporting. For causal estimates, we exploit exogenous variation in cable news exposure across counties coming from variation in the relative channel numbering of the three cable networks (Martin and Yurukoglu, 2017). We provide a number of checks to validate the relevance of the first stage and the instrument exogeneity. In particular, the instrument is uncorrelated with other local characteristics predictive of viewership or of the relevant dimensions of local newspaper content. We find that media slant is contagious. Higher cable news network viewership increases the influence of a network’s content on local newspaper articles: a one-standarddeviation increase in a county’s FNC viewership (relative to averaged CNN and MSNBC viewership) would increase the similarity of the local newspaper’s content to FNC’s content by 0.31 standard deviations. Our estimated local average treatment effects survive various specification checks, including controls for local demographics, local cable television market characteristics, and text readability metrics (e.g., word length). The results are robust to alternative design choices in sampling, weighting, and instrument construction. Turning to the mechanisms, we ask whether local newspapers weave the FNC- or CNN/MSNBC-like slant into their original reporting (i.e., a shift in their own reporting) or whether the effects could be driven by copy-pasting. First, we devise a topic-based procedure to distinguish local news from non-local (that is, national or international) news. Analyzing local and non-local news separately, we find that cable news influences both kinds of content. Since cable TV shows cover national or international stories, the observed diffusion of media slant is not only direct borrowing of content from the cable TV channels. Instead, cable news exposure also shifts the original local reporting of the newspapers. Finally, we investigate whether cable news has polarized local news content. We split newspapers into three groups: those that have historically endorsed Democrats, those who have historically endorsed Republicans, and those without or with mixed endorsements. In response to exposure to FNC, historically Republican newspapers became more conservative (FNC-like), while historically Democrat newspapers became more liberal (CNN/MSNBC-like). Thus, media slant from cable news seems to encourage outlets to re-position themselves on the ideological spectrum following a more partisan consumer base. Cable news has remade news landscapes and increased political polarization in local news discourse. These findings add to the literature in political science and political economy on biased media (e.g., Ashworth and Shotts, 2010; Prat, 2018).[1] This literature provides good evidence that mass media shift election outcomes and readers’ policy preferences. First, Gentzkow et al. (2011) and Drago et al. (2014) report that the opening of local newspapers boosts voter turnout. Chiang and Knight (2011) show that a newspaper endorsement for a presidential candidate shifts voting intentions in favor of this candidate. Djourelova (2023) shows, for the case of immigration and border security, that the language used in newspapers can causally shift readers’ policy preferences. Beyond the United States, Enikolopov et al. (2011) find that Russian voters with access to an independent television station are more supportive of anti-Putin parties.[2] Regarding Fox News in particular, there is a body of research documenting its political and societal impact – beyond shifting votes (DellaVigna and Kaplan, 2007; Martin and Yurukoglu, 2017; Ash et al., 2021; Li and Martin, 2022). It has also been shown that cable news can affect voter knowledge (Hopkins and Ladd, 2014; Schroeder and Stone, 2015), fiscal policy decisions (Ash and Galletta, 2023), as well as behaviors during the COVID-19 pandemic (Bursztyn et al., 2021; Ash et al., 2020; Simonov et al., 2022). We add to this work by looking at the influence of partisan narratives using text analysis and looking at the spillover effects on other news outlets. Our main contribution to the debate on media bias lies in showing how news media outlets are interconnected. Recent contributions on cross-media influence document the influence of social media on traditional media (Cagé et al., 2020a; Hatte et al., 2021), and that news outlets copy-paste from each other extensively (Cagé et al., 2020b). We document explicitly how media bias from one media organization can causally spill over to other media organizations. Hence, our work identifies an additional potential channel through which partisan media affects political and social outcomes. The diffusion of national partisan priorities into local news is important because local newspapers are pivotal for citizen engagement and political accountability (e.g., Snyder and Strömberg, 2010). George and Waldfogel (2006) find that the market entry of a national media outlet (in their case, the New York Times) causes local outlets to focus more on local coverage. Martin and McCrain (2019) show that the acquisition of U.S. local TV stations by the national conglomerate Sinclair leads to an increased share of national as opposed to local content. Further, Mastrorocco and Ornaghi (2020) document that these acquisitions by Sinclair reduce coverage of local crime and subsequently lower crime clearance rates. We contribute to these debates by analyzing how higher exposure to slanted national cable news changes local content. Methodologically, our approach combines natural language processing (NLP), machine learning, and causal inference (see Gentzkow et al. 2019a and Ash and Hansen 2022 for overviews of text-data-based work in economics). Regarding text-as-data approaches to measuring partisanship, the most related work is Gentzkow and Shapiro (2010), showing a correlation in local news slant and local partisan preferences (see also Gentzkow et al., 2019b). Our innovation is in combining text-based slant measurement with a causal research design. More broadly, our work contributes to the long-lasting debate on the importance of (un)biased media in democratic politics – a topic that has become especially important in the current era of polarization in the U.S. and beyond. This paper is available on arxiv under CC 4.0 license. [1] For surveys on the empirical and theoretical literature, see Puglisi and Snyder (2015) and Gentzkow et al. (2015), respectively. [2] Prominent contributions on the mass media’s persuasive effects around the world include Adena et al. (2015), DellaVigna et al. (2014), or Yanagizawa-Drott (2014). Prat and Stroemberg (2013) and Stroemberg (2015) provide surveys on the mass media’ political effects. Authors: (1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch; (2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch; (3) Elliott Ash, ETH Zürich and ashe@ethz.ch. Table of Links Abstract and 1 Introduction 2. Data Abstract and 1 Introduction 2. Data 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3. Measuring Media Slant and 3.1. Text pre-processing and featurization 3.2. Classifying transcripts by TV source 3.2. Classifying transcripts by TV source 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 3.3. Text similarity between newspapers and TV stations and 3.4. Topic model 4. Econometric Framework 4.1. Instrumental variables specification 4.1. Instrumental variables specification 4.2. Instrument first stage and validity 4.2. Instrument first stage and validity 5. Results 5.1. Main results 5.1. Main results 5.2. Robustness checks 5.2. Robustness checks 6. Mechanisms and Heterogeneity 6.1. Local vs. national or international news content 6.1. Local vs. national or international news content 6.2. Cable news media slant polarizes local newspapers 6.2. Cable news media slant polarizes local newspapers 7. Conclusion and References 7. Conclusion and References Online Appendices Online Appendices A. Data Appendix A.1. Newspaper articles A.1. Newspaper articles A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.2. Alternative county matching of newspapers and A.3. Filtering of the article snippets A.4. Included prime-time TV shows and A.5. Summary statistics A.4. Included prime-time TV shows and A.5. Summary statistics B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B. Methods Appendix, B.1. Text pre-processing and B.2. Bigrams most predictive for FNC or CNN/MSNBC B.3. Human validation of NLP model B.3. Human validation of NLP model B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.4. Distribution of Fox News similarity in newspapers and B.5. Example articles by Fox News similarity B.6. Topics from the newspaper-based LDA model B.6. Topics from the newspaper-based LDA model C. Results Appendix C.1. First stage results and C.2. Instrument exogeneity C.1. First stage results and C.2. Instrument exogeneity C.3. Placebo: Content similarity in 1995/96 C.3. Placebo: Content similarity in 1995/96 C.4. OLS results C.4. OLS results C.5. Reduced form results C.5. Reduced form results C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.6. Sub-samples: Newspaper headquarters and other counties and C.7. Robustness: Alternative county matching C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.10. Robustness: Absolute and relative FNC viewership and C.11. Robustness: Dropping observations and clustering C.12. Mechanisms: Language features and topics C.12. Mechanisms: Language features and topics C.13. Mechanisms: Descriptive Evidence on Demand Side C.13. Mechanisms: Descriptive Evidence on Demand Side C.14. Mechanisms: Slant contagion and polarization C.14. Mechanisms: Slant contagion and polarization Abstract This paper examines the diffusion of media slant, specifically how partisan content from national cable news affects local newspapers in the U.S., 2005-2008. We use a text-based measure of cable news slant trained on content from Fox News Channel (FNC), CNN, and MSNBC to analyze how local newspapers adopt FNC’s slant over CNN/MSNBC’s. Our findings show that local news becomes more similar to FNC content in response to an exogenous increase in local FNC viewership. This shift is not limited to borrowing from cable news, but rather, local newspapers’ own content changes. Further, cable TV slant polarizes local news content. 1. Introduction Scientific consensus and conventional wisdom agree: Media bias is widespread. Media outlets often feature content in a way that is favorable to a particular political party or ideological perspective (e.g. Groseclose and Milyo, 2005; Harmon and Muenchen, 2009; Gentzkow and Shapiro, 2010; Puglisi and Snyder, 2011). These biases in reporting can impact public opinion (e.g., Chiang and Knight, 2011; Djourelova, 2023), and in particular, there is extensive evidence that higher exposure to Fox News (FNC) boosts Republican vote shares and other conservative interests (DellaVigna and Kaplan, 2007; Martin and Yurukoglu, 2017; Ash et al., 2021; Ash and Galletta, 2023). One unanswered question motivated by these findings is whether biased news messaging can spread to and distort other news outlets’ content. This issue is policy-relevant since media regulation in democracies typically aims at providing consumers with a competitive news market, with a wide, unrestricted choice among independent news sources. After all, diverse news sources only translate into diverse reporting if they are independent in their news making. This work studies cross-media influence in the context of U.S. cable TV news channels – Fox News, CNN, and MSNBC – which have a well-documented partisan slant. When there is higher viewership of a cable news channel among a newspaper’s readers, is the associated partisan slant reflected in that newspaper’s content? Our word defines slant in relative terms – whether a piece of content resembles Fox News rather than CNN or MSNBC. The first step in answering this question is to measure the similarity between cable news and local newspaper content. For this purpose, we build a corpus of 24 million article snippets from 600+ U.S. local newspapers in the United States from 2005 through 2008. We combine these texts with transcripts of 40,000 episodes from Fox News, CNN, and MSNBC. We use this parallel corpus to construct a novel measure of cable news slant – that is, we train a machine-learning model to predict, for a given body of text, whether it resembles the language by the relatively conservative network (FNC), rather than language by the relatively liberal networks CNN or MSNBC. We validate the model and the associated predictions with human annotations. We apply our model to the local newspaper article corpus. For each article, we have a text-based metric reflecting the similarity to content from FNC shows relative to CNN and MSNBC shows. We aggregate the article similarities at the newspaper level. We add metadata on newspaper circulation, television channel positioning, ratings, and political and demographic covariates. Then, we investigate whether relative similarity to content in a cable news network increases in response to higher viewership in a newspaper’s market. Cross-sectional estimates of this relationship would likely be confounded, for example, by more ideologically conservative counties having both higher Fox News viewership and more conservative local reporting. For causal estimates, we exploit exogenous variation in cable news exposure across counties coming from variation in the relative channel numbering of the three cable networks (Martin and Yurukoglu, 2017). We provide a number of checks to validate the relevance of the first stage and the instrument exogeneity. In particular, the instrument is uncorrelated with other local characteristics predictive of viewership or of the relevant dimensions of local newspaper content. We find that media slant is contagious. Higher cable news network viewership increases the influence of a network’s content on local newspaper articles: a one-standarddeviation increase in a county’s FNC viewership (relative to averaged CNN and MSNBC viewership) would increase the similarity of the local newspaper’s content to FNC’s content by 0.31 standard deviations. Our estimated local average treatment effects survive various specification checks, including controls for local demographics, local cable television market characteristics, and text readability metrics (e.g., word length). The results are robust to alternative design choices in sampling, weighting, and instrument construction. Turning to the mechanisms, we ask whether local newspapers weave the FNC- or CNN/MSNBC-like slant into their original reporting (i.e., a shift in their own reporting) or whether the effects could be driven by copy-pasting. First, we devise a topic-based procedure to distinguish local news from non-local (that is, national or international) news. Analyzing local and non-local news separately, we find that cable news influences both kinds of content. Since cable TV shows cover national or international stories, the observed diffusion of media slant is not only direct borrowing of content from the cable TV channels. Instead, cable news exposure also shifts the original local reporting of the newspapers. Finally, we investigate whether cable news has polarized local news content. We split newspapers into three groups: those that have historically endorsed Democrats, those who have historically endorsed Republicans, and those without or with mixed endorsements. In response to exposure to FNC, historically Republican newspapers became more conservative (FNC-like), while historically Democrat newspapers became more liberal (CNN/MSNBC-like). Thus, media slant from cable news seems to encourage outlets to re-position themselves on the ideological spectrum following a more partisan consumer base. Cable news has remade news landscapes and increased political polarization in local news discourse. These findings add to the literature in political science and political economy on biased media (e.g., Ashworth and Shotts, 2010; Prat, 2018).[1] This literature provides good evidence that mass media shift election outcomes and readers’ policy preferences. First, Gentzkow et al. (2011) and Drago et al. (2014) report that the opening of local newspapers boosts voter turnout. Chiang and Knight (2011) show that a newspaper endorsement for a presidential candidate shifts voting intentions in favor of this candidate. Djourelova (2023) shows, for the case of immigration and border security, that the language used in newspapers can causally shift readers’ policy preferences. Beyond the United States, Enikolopov et al. (2011) find that Russian voters with access to an independent television station are more supportive of anti-Putin parties.[2] Regarding Fox News in particular, there is a body of research documenting its political and societal impact – beyond shifting votes (DellaVigna and Kaplan, 2007; Martin and Yurukoglu, 2017; Ash et al., 2021; Li and Martin, 2022). It has also been shown that cable news can affect voter knowledge (Hopkins and Ladd, 2014; Schroeder and Stone, 2015), fiscal policy decisions (Ash and Galletta, 2023), as well as behaviors during the COVID-19 pandemic (Bursztyn et al., 2021; Ash et al., 2020; Simonov et al., 2022). We add to this work by looking at the influence of partisan narratives using text analysis and looking at the spillover effects on other news outlets. Our main contribution to the debate on media bias lies in showing how news media outlets are interconnected. Recent contributions on cross-media influence document the influence of social media on traditional media (Cagé et al., 2020a; Hatte et al., 2021), and that news outlets copy-paste from each other extensively (Cagé et al., 2020b). We document explicitly how media bias from one media organization can causally spill over to other media organizations. Hence, our work identifies an additional potential channel through which partisan media affects political and social outcomes. The diffusion of national partisan priorities into local news is important because local newspapers are pivotal for citizen engagement and political accountability (e.g., Snyder and Strömberg, 2010). George and Waldfogel (2006) find that the market entry of a national media outlet (in their case, the New York Times) causes local outlets to focus more on local coverage. Martin and McCrain (2019) show that the acquisition of U.S. local TV stations by the national conglomerate Sinclair leads to an increased share of national as opposed to local content. Further, Mastrorocco and Ornaghi (2020) document that these acquisitions by Sinclair reduce coverage of local crime and subsequently lower crime clearance rates. We contribute to these debates by analyzing how higher exposure to slanted national cable news changes local content. Methodologically, our approach combines natural language processing (NLP), machine learning, and causal inference (see Gentzkow et al. 2019a and Ash and Hansen 2022 for overviews of text-data-based work in economics). Regarding text-as-data approaches to measuring partisanship, the most related work is Gentzkow and Shapiro (2010), showing a correlation in local news slant and local partisan preferences (see also Gentzkow et al., 2019b). Our innovation is in combining text-based slant measurement with a causal research design. More broadly, our work contributes to the long-lasting debate on the importance of (un)biased media in democratic politics – a topic that has become especially important in the current era of polarization in the U.S. and beyond. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv [1] For surveys on the empirical and theoretical literature, see Puglisi and Snyder (2015) and Gentzkow et al. (2015), respectively. [2] Prominent contributions on the mass media’s persuasive effects around the world include Adena et al. (2015), DellaVigna et al. (2014), or Yanagizawa-Drott (2014). Prat and Stroemberg (2013) and Stroemberg (2015) provide surveys on the mass media’ political effects. Authors: (1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch; (2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch; (3) Elliott Ash, ETH Zürich and ashe@ethz.ch. Authors: Authors: (1) Philine Widmer, ETH Zürich and philine.widmer@gess.ethz.ch; (2) Sergio Galletta, ETH Zürich and sergio.galletta@gess.ethz.ch; (3) Elliott Ash, ETH Zürich and ashe@ethz.ch.

Media Slant is Contagious: Here's Why

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

A Comprehensive Overview of Image Generation Models: From GANs to Diffusion Techniques

How We Measured Cable News Slant: Researching Media Bias

Media Slant Research: Newspaper-County Observations

Multilingual Coarse Political Stance Classification of Media: Summary and Conclusions

Multilingual Coarse Political Stance Classification of Media: Distribution of Topics per Newspaper

A Comprehensive Overview of Image Generation Models: From GANs to Diffusion Techniques

How We Measured Cable News Slant: Researching Media Bias

Media Slant Research: Newspaper-County Observations

Multilingual Coarse Political Stance Classification of Media: Summary and Conclusions

Multilingual Coarse Political Stance Classification of Media: Distribution of Topics per Newspaper

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps