Does Local News Stay Local?: Online Content Shifts in Sinclair-Acquired Stations

Technical Summary: Does Local News Stay Local? Online Content Shifts in Sinclair-Acquired Stations

Problem Statement
Local news outlets have historically served as reliable, non-partisan sources of community-focused information, with viewership positively correlated with cross-party voting and decreased political polarization. However, the Sinclair Broadcast Group, a major media conglomerate, has acquired numerous local stations over the last decade. Concerns exist that these acquisitions may alter local news coverage, potentially introducing national political biases and reducing community-specific reporting. While prior studies have examined broadcast transcripts or station websites, there is a lack of quantitative analysis focusing on digital content (specifically YouTube) and the linguistic shifts in politicized topics following such acquisitions. This paper investigates the effects of Sinclair acquisition on local news content, asking: (1) How does overall news coverage differ after purchase? and (2) How does coverage of politicized topics differ after purchase?

Methodology
The authors constructed a novel dataset comprising automated closed captions from YouTube channels of eight geographically diverse local news stations acquired by Sinclair, alongside two national outlets (CNN and Fox News) for comparison. The dataset spans sixteen years of video publishing, with purchase dates ranging from 2014 to 2017.

The analysis employs three primary computational linguistics methods:

1. Log-Odds Ratio with Dirichlet Prior ("Fightin' Words"): Used to identify words overrepresented in pre-purchase versus post-purchase transcripts. To mitigate temporal confounders, data was stratified by year (focusing on 2014–2016) to ensure paired comparisons.
2. Structured Topic Modeling (STM): A probabilistic model incorporating document metadata as covariates. Five models were trained: four to compare topic prevalence across time (pre/post) and political leaning (CNN/Fox), and a fifth to analyze how the "Sinclair vs. Non-Sinclair" status acts as a topical content covariate, allowing word distributions within topics to vary based on ownership.
3. Word Embedding Analysis: Separate Word2Vec models were trained on pre-purchase and post-purchase transcripts. The authors analyzed the nearest neighbors of seed words related to politicized issues (e.g., "racism," "abortion," "immigration") to detect shifts in semantic context. Additionally, embedding similarity was calculated between local news vectors and those from CNN and Fox to measure the degree of "nationalization" in language usage.

A controlled paired analysis was also conducted in the appendix, matching Sinclair-acquired stations with non-Sinclair stations in similar geographic regions to further isolate the effect of ownership from temporal trends.

Key Results

• Shift from Local to National Topics: Log-odds analysis reveals that post-acquisition stations increasingly use words associated with national politics (e.g., "president," "government," "federal") while decreasing usage of local-centric terms (e.g., "snow," "school," "downtown"). This shift persists even when controlling for election years.
• Topic Prevalence Changes: STM results indicate a clear divergence in topic coverage. Pre-purchase coverage is dominated by local concerns such as weather, local sports, schools, and community events. Post-purchase, there is a marked increase in topics related to national politics, including presidential candidates, the FBI, terrorism, and international conflicts (e.g., ISIS, Israel/Hamas). While local topics like police and crime remain prevalent, the framing often shifts toward national narratives.
• Semantic Polarization: Word embedding analysis shows that post-acquisition transcripts associate query words with more overtly political neighbors. For instance, "black" shifts from general color associations to terms like "africanamerican" and "movement," and "bias" shifts from culinary contexts to "implicit" and "prejudice."
• Convergence with National Outlets: Embedding similarity analysis demonstrates that after acquisition, local news stations use politicized keywords in contexts increasingly similar to both CNN and Fox News. This suggests a "nationalization" of local discourse, where local outlets adopt the semantic framing of polarized national media, regardless of specific right-wing or left-wing alignment in the embedding space.

Significance and Claims
The paper claims to provide compelling evidence that acquisition by the Sinclair Broadcast Group is associated with a shift in local news content from community-focused reporting to politicized national news. The authors argue that this trend contributes to the "nationalization" of local news, potentially increasing political polarization among local viewers by exposing them to content that mirrors national partisan divides.

The study highlights the utility of YouTube data as a novel source for analyzing content shifts in the digital news ecosystem. However, the authors remain modest regarding causal claims, acknowledging that while the timing of content shifts coincides with acquisitions, unobserved confounders (such as industry-wide trends in engagement or broader temporal shifts in news coverage) cannot be definitively ruled out. The work does not claim to prove that Sinclair causes viewer polarization directly but suggests that the observed changes in digital content provide a plausible mechanism for the viewer preference changes identified in prior literature.