Media Slant is Contagious: The Main Results From Our Experiment

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

5. Results

5.1. Main results

Table 2 shows two-stage-least-squares estimates of the effect of higher FNC viewership on newspaper content similarity with FNC (the average probability that a snippet from a newspaper is predicted to be from FNC based on the bigrams it contains). [11] The right-hand side variable of interest is instrumented FNC viewership relative to averaged CNN and MSNBC viewership. All columns include state fixed effects and demographic controls. Column 2 also includes controls for the share of households with potential access to each of the three TV channels. Column 3 additionally controls for generic newspaper language features.

In all three columns, the estimated treatment effects are positive and statistically significant. The magnitudes across columns are highly similar, ranging from 0.31 in columns 1 and 2 to 0.32 in column 3. Thus, the channel and language controls do not change the estimates relative to the baseline in column 1 with only state fixed effects and demographic controls. All variables are standardized, so the interpretation is as follows: if Fox News viewership (relative to averaged CNN and MSNBC viewership) increases by one standard deviation in county j where newspaper i circulates, the similarity of i’s content with FNC increases by 0.31 standard deviations. [12] Alternatively, a onestandard-deviation decrease in the relative FNC channel position (11 relative positions) would increase slant by 0.03 standard deviations.

To interpret the magnitudes, note that the average slant difference between an FNC transcript snippet and a CNN/MSNBC transcript snippet in standardized units is 0.99 (the raw difference, in predicted probabilities, is 0.21). Given the estimated 2SLS coefficient (0.31), a one-standard-deviation decrease in the relative FNC channel position (11 positions) shifts slant towards FNC by about 3 percent of the difference between FNC and CNN/MSNBC.

[12] OLS estimates are reported in Appendix C.4.