Engagement vs. Commitment: The Economic Trade-Offs of Polarizing News Content

Shunyao Yan, Klaus M. Miller · 2026

Read Engagement vs. Commitment: The Economic Trade-Offs of Polarizing News Content by Shunyao Yan, Klaus M. Miller (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-18357.txt.

Introduction

Shunyao Yan and Klaus Miller ask in a 2026 study whether slanted news content that grabs focus also wins paying subscribers. Using newsroom supply shifts and election timing as levers, they find supply-driven polarization lifts time on site but not subscriptions. During high-salience election periods it cuts sign-ups and speeds churn. The work shows an asymmetric trade-off between engagement and commitment for digital news outlets.

Catalog record

CREATORShunyao Yan, Klaus M. Miller
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE106588
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.18357

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Engagement vs. Commitment: The Economic Trade-Offs of Polarizing News Content Shunyao Yan, Klaus M. Miller Abstract Content that drives engagement need not be the same content that drives willingness to pay. We study how polarizing content affects engagement (time on site) and commitment (subscriptions and retention) on a major news platform. We measure article-level polarization with deep-learning classifiers and large language models tailored to a multiparty system, and identify causal effects with two complementary instrumental variables: a Bartik instrument exploiting supply-side editorial variation, and an election instrument exploiting demand-side political salience. We find that supply-driven increases in polarizing content raise engagement but not subscriptions. During the high-salience election window, the same content reduces subscriptions and accelerates churn, with affective polarization driving the sharpest divergence. On the mechanism, we find evidence inconsistent with confirmation bias: three pre-determined ideology proxies do not moderate the engagement or subscription effects. By contrast, on ideological dimensions where the publisher covers both sides, exogenous shifts in the publisher’s supply of content opposite readers’ baseline ideology raise their consumption of that content, consistent with balanced consumption. These results document an asymmetric engagement-commitment trade-off for digital publishers: polarizing content reliably captures attention but does not convert to subscriptions, and actively damages commitment when political salience is elevated. Keywords: polarization, subscriptions, online media, news consumption, instrumental variables, natural language processing 1 Introduction News publishers earn revenue from two distinct sources, advertising on each page view and subscription on each retained reader ( 1 ; 6 ; 59 ; 19 ) . As digital advertising revenue has declined, subscription revenue has surpassed advertising at the median news organization ( 42 ) . Conversion from casual reader to paying subscriber is rare, and the lifetime value of a converted subscriber is large relative to the marginal advertising revenue from short-run engagement gains. Once acquired, news subscribers are unusually durable, with annual-plan retention at 71% for the median news brand and 83% at top performers ( 45 ) . Whether content that raises engagement also raises subscriptions and retention is therefore a first-order question for editorial strategy. Polarizing content is central to this tension. Broadly, we define polarizing content as content that conveys a stance corresponding to one side of a political spectrum or that employs emotionally charged us–versus–them language (we formalize this definition in Section 3). On the engagement side, the case is well established: polarizing language reliably attracts attention online ( 49 ; 14 ; 7 ; 10 ) , and engagement is closely linked to advertising revenue ( 30 ; 59 ) ; whether it also lifts subscriptions and retention, the margins on which publisher revenue now centrally depend, is an open question. If engagement and commitment diverge, the standard editorial heuristic to follow revealed engagement preferences is incomplete. Why might engagement and commitment diverge for polarizing news content? Prior literature suggests two distinct, and potentially competing, forces shaping consumers’ demand for news. One stream, rooted in confirmation bias theory, finds that consumers are drawn to “slanted” media outlets that align with and reinforce their existing political beliefs, suggesting that consumers respond positively to polarizing content, provided that it validates their worldview ( 28 ; 41 ) . Consumers’ attraction to such content has well-documented downstream effects on political, economic, and social behavior ( 25 ; 39 ; 54 ) . Yet a second stream complicates this picture by suggesting that consumers also have strong preferences for balanced and quality information. Empirical evidence shows that the average US consumer does not limit news consumption to a restricted set of ideologically similar outlets and, if anything, favors a balanced mix of sources ( 29 ; 33 ) . Even in contexts where ideological supply is heavily skewed, as in Russia, consumers’ revealed preferences appear driven more by non-ideological factors such as non-political content and third-party referrals than by ideological affinity ( 53 ) . These opposing tendencies, attraction to polarizing content alongside preferences for balance, motivate us to gauge commitment beyond engagement. To do so, we partner with a major European newspaper of record and track each reader simultaneously across engagement, subscription, and retention. Our panel covers every registered user over 40 consecutive weeks, including the focal country’s regularly scheduled federal election, and records each article a user reads together with their subscription and retention outcomes. Holding the publisher’s pricing and content delivery fixed throughout, exogenous variation in polarizing supply and demand identifies its causal effect on both engagement and commitment within the same user, isolating a within-publisher content margin that cross-platform and cross-outlet designs cannot reach. Our central finding is an asymmetric engagement-commitment trade-off: polarizing content lifts time on site but fails to convert to subscriptions, and actively damages commitment when political salience is elevated. In conducting our analysis, we overcome two main empirical challenges. The first lies in measuring polarization in news content. Existing text-based measures collapse polarization onto a single left–right label per article. Two features of our setting make that approach restrictive. First, polarization is multi-faceted. Affective polarization, the widening of in-group versus out-group affect, travels through emotionally charged language and has become a defining feature of contemporary polarization ( 35 ; 34 ; 13 ) , yet it is largely missed by lexical and stance metrics built around issue positions ( 28 ; 27 ) . Second, news articles can quote, paraphrase, or refute partisan language without endorsing it, while factual policy coverage can take a clear issue position without sounding like partisan discourse. A single article-level label therefore conflates two distinct signals: how closely the language aligns with party rhetoric, and what stance a reader would infer from the article as a whole. This distinction matters: engagement may respond to rhetorical cues and us–versus–them language, while subscription and retention may depend on whether readers perceive the article and publisher as balanced and credible. To address these measurement challenges, we produce two complementary article-level measures of polarization: a party-referenced measure of what is inherently polarizing in the language, based on a hierarchical BERT classifier (Bidirectional Encoder Representations from Transformers; a transformer-based deep-learning model) trained on parliamentary speech, and a reader-referenced measure of the polarization a reader would perceive, from a large language model (LLM) scoring each article along affective and three ideological dimensions. Section 3.3 details the framework. Pairing both reference systems on the same articles, which no prior measure does, lets us identify which dimension drives the engagement-commitment trade-off. The second challenge lies in identification: Isolating the causal effect of consuming polarizing content from selection bias is a well-known challenge in media effects research ( 24 ; 4 ) . Randomized experiments, while considered the gold standard for causal inference, face external validity concerns in this domain: for example, 16 show that participants tend to revert to their preferred media sources after an experiment, making it unclear whether experimental results generalize to real-world consumption. Our solution is a quasi-experimental instrumental-variable (IV) strategy that exploits plausibly exogenous variation from both the supply and the demand side. We construct two theoretically distinct instruments. First, a Bartik-style instrument (the Bartik instrument) interacts users’ stable topic preferences with weekly shifts in the publisher’s supply of polarizing content, driven by editorial planning and exogenous news events. Second, an election-driven

Read the complete text