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
6. Mechanisms and Heterogeneity
6.1. Local vs. national or international news content
6.2. Cable news media slant polarizes local newspapers
\ Online Appendices
A. Data Appendix
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.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.8. Robustness: Historical circulation weights and C.9. Robustness: Relative circulation weights
C.12. Mechanisms: Language features and topics
C.13. Mechanisms: Descriptive Evidence on Demand Side
C.14. Mechanisms: Slant contagion and polarization
5.2. Robustness checksOur first robustness checks rely on alternative samples. Table C.6 replicates the baseline estimates but only considers newspaper-county observations where the county coincides with where the immediate newspaper owner is based. The effect size doubles relative to the entire sample and is still significant despite a smaller sample. [13]
\ Second, we use alternative weighting schemes. The results are robust to weighting by circulation from the pre-FNC/MSNBC era, specifically 1995 (Table C.9). So, our main results are not driven by potential circulation changes due to cable news exposure. [14] Next, we weight observations by relative circulation shares by county, multiplied by the number of surveyed individuals for each county by Nielsen (Table C.10). In using the number of surveyed individuals, we follow Martin and Yurukoglu (2017). Similarly, we also weight observations by relative circulation shares by county, multiplied by county population (Table C.11). The results are again positive and statistically significant.
\ Third, we replicate the main results using different instrument specifications (Table C.12). Instead of FNC’s relative to CNN’s/MSNBC’s viewership combined, we take FNC viewership relative to CNN’s and MSNBC’s separately (columns 1/2 and columns 3/4, respectively), or just FNC viewership (columns 5/6). The estimates are positive and overall consistent with our main results, yet not significant in the FNC-vs-CNN specification.
\ Fourth, we rely on a different matching of newspapers to counties. We assign each newspaper to a main county based on its name and other metadata, producing a larger sample but with less detailed circulation data (see Section A.2 for details). The results are fully robust (Appendix C.7).
\ Finally, our results are still significant when clustering by state (Table C.13) and remain consistently significant when dropping newspapers individually (Figure C.5).
\
:::info This paper is available on arxiv under CC 4.0 license.
:::
[13] Meanwhile, the effect is smaller and insignificant when excluding headquarters counties (Table C.7).
\ [14] Our main analysis does not use the pre-FNC/MSNBC county-level circulation data because it is available for fewer outlets, resulting in half the sample size. Our baseline specification (contemporary circulation weights) is robust to using the subsample where 1995 circulation is available.
:::info Authors:
(1) Philine Widmer, ETH Zürich and [email protected];
(2) Sergio Galletta, ETH Zürich and [email protected];
(3) Elliott Ash, ETH Zürich and [email protected].
:::
\
All Rights Reserved. Copyright , Central Coast Communications, Inc.