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

This paper demonstrates that while polarizing news content reliably increases user engagement, it fails to drive subscriptions and can actively harm retention during high-salience political periods, revealing a critical economic trade-off for digital publishers where attention does not equate to commitment.

The Gist

Imagine a news website as a busy coffee shop. The owner makes money in two ways:

1. The "Window Shoppers" (Ads): People walking by glance at the shop window. The more people stop to look, the more the owner gets paid by advertisers.
2. The "Regulars" (Subscriptions): People who buy a monthly pass to sit inside, order coffee, and come back every day. This is the most valuable money the owner makes.

This paper asks a simple but tricky question: Does the kind of news that makes people stop and stare at the window (engagement) also make them want to buy a monthly pass (commitment)?

The answer, surprisingly, is no. In fact, it's often the opposite.

The "Rage Bait" Trap

The researchers studied a major European newspaper (similar to The New York Times or The Guardian) to see how "polarizing" news affects readers. Polarizing news is the kind that uses emotional, "us-versus-them" language—think of headlines that make you angry, excited, or feel like you're in a battle against the other side.

They found a "Polarization Trap":

• The Hook: When the newspaper published more of these emotional, divisive stories, people did stop and look. They spent more time on the site. This is great for the "window shoppers" (advertising revenue).
• The Breakup: However, this same content did not make people buy monthly passes. In fact, during big political events (like an election), reading these emotional stories actually made existing subscribers cancel their passes and leave.

The Analogy: Imagine a magician at a street fair. He does a flashy, loud trick that draws a huge crowd (engagement). Everyone stops to watch. But when he asks for donations to see the next show, the crowd walks away. Worse, if he keeps doing the same loud, aggressive tricks, the people who already bought tickets get annoyed and ask for a refund. The flashy trick captures attention, but it destroys the relationship needed to keep the customer.

How They Knew It Was True (The "Magic Trick" of Science)

Usually, it's hard to tell if polarizing news causes people to leave, or if people who are already angry just happen to click on polarizing news. To solve this, the researchers used two clever "natural experiments":

1. The Editor's Mood (Supply Side): They looked at times when editors, due to external news events, accidentally published more polarizing stories than usual. They saw that when this happened, people clicked more, but didn't subscribe.
2. The Election Shock (Demand Side): They compared readers in a country having a big election to readers in a neighboring country with the same language but no election. During the election week, the "home" readers were bombarded with political drama. The researchers found that this sudden dose of political polarization made those specific readers much more likely to cancel their subscriptions.

The "Confirmation Bias" Myth

A common belief is that people only want news that agrees with them (Confirmation Bias). The researchers tested this by checking if people who lean "Left" only clicked on "Left" polarizing news, and "Right" on "Right."

The Surprise: They didn't find this. Instead, they found that when the newspaper provided more of the opposite side's news, people actually read it. This suggests that most readers aren't just looking for an echo chamber; they are looking for a balanced diet of information. When the menu becomes too spicy or one-sided, they get sick of it and leave.

The Bottom Line for News Publishers

The paper concludes that for a news company trying to make money from subscriptions, chasing clicks with angry, divisive content is a bad business strategy.

• Short-term: It gets you more eyes on the page (good for ads).
• Long-term: It kills your subscription revenue. It turns "Regulars" into "Window Shoppers" and then into "Ex-Readers."

The authors calculate that the money lost from a single subscriber canceling their pass is 59 to 176 times greater than the tiny amount of ad money gained from the extra time that subscriber spent on the site.

In simple terms: If you want people to pay you to stay, don't try to make them angry just to get them to click. It's like trying to keep a friend by constantly starting arguments; they might listen for a minute, but they won't want to hang out with you forever.

Technical Summary

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

1. Problem Statement

Digital news publishers face a fundamental tension between two revenue streams: advertising, which relies on user engagement (time on site, page views), and subscriptions, which rely on user commitment (conversion and retention). While prior literature establishes that polarizing content reliably captures attention and drives engagement, it remains unclear whether such content translates into subscription revenue or retention. In fact, as advertising revenue declines and subscription models become dominant, the assumption that engagement proxies for publisher value is increasingly questionable.

The central research question is whether content that drives engagement also drives willingness to pay. Specifically, the authors investigate if polarizing content—defined as content conveying a political stance or employing emotionally charged "us-versus-them" language—creates an asymmetric trade-off where it boosts engagement but fails to convert to subscriptions, or even actively damages commitment.

2. Methodology

Empirical Setting and Data

The study utilizes a unique panel dataset from a major European "newspaper of record" (comparable in reputation to The New York Times or The Financial Times) covering 40 consecutive weeks, including a federal election in the focal country. The dataset includes:

• User-Level Data: 119,913 registered users (approx. 20% of total traffic) with clickstream logs, session lengths, and subscription/retention outcomes.
• Content Data: A complete archive of 62,801 articles with timestamps.
• Training Data: 17,776 parliamentary speeches used to train polarization classifiers.

The publisher's editorial process is key to identification: content delivery is primarily editorially curated with fixed weekly planning routines, and the platform does not use algorithmic headline testing or dynamic paywall adjustments during the observation window. This limits algorithmic amplification and ensures that supply-side variation is driven by editorial agenda-setting and exogenous news shocks rather than real-time user demand.

Measurement of Polarization

To address the limitations of single-label polarization measures, the authors develop a novel two-component framework:

1. Party-Referenced Measure (BERT): A hierarchical BERT classifier trained on parliamentary speech to detect rhetorical alignment with partisan discourse (Left/Right/Neutral). This captures the degree to which language aligns with party rhetoric.
2. Reader-Referenced Measure (LLM): A Large Language Model (Gemini 2.5 Pro) scores articles on four distinct dimensions using a structured rubric grounded in political science:
   • Affective Intensity: "Us-versus-them" emotional tone (1–5 scale).
   • Ideological Dimensions: Economic (Left–Right), Globalization (Globalist–Nationalist), and Environment (Pro-environment–Pro-growth), each on a 1–5 scale.

This dual approach allows the authors to distinguish between rhetorical alignment and the perceived stance/affective tone of the article.

Identification Strategy

To isolate causal effects from selection bias (where users self-select into polarizing content), the authors employ two complementary instrumental variables (IVs):

1. Bartik Instrument (Supply-Side): A shift-share design interacting users' stable pre-period topic preferences with weekly shifts in the publisher's supply of polarizing content within those topics. This exploits exogenous variation in editorial planning and news events to identify the effect of supply-driven exposure.
2. Election Instrument (Demand-Side): A difference-in-differences approach exploiting a federal election in the focal country. Users in the focal country (treatment) are compared to users in a linguistically similar neighboring country (control) where no election occurred. This exploits the exogenous increase in political salience to identify demand-side effects during high-salience windows.

The authors estimate two-stage least squares (2SLS) models with user and week fixed effects, controlling for sentiment, article quality, and platform activity.

3. Key Results

The Asymmetric Engagement-Commitment Trade-Off

The core finding is a "polarization trap": polarizing content reliably increases engagement but fails to convert to subscriptions, and actively damages commitment when political salience is high.

• Engagement (Bartik IV): Exogenous increases in affectively polarizing content raise time on site by approximately 11.6% (semi-elasticity). However, this engagement does not translate into subscription gains; the effect on subscriptions is statistically indistinguishable from zero.
• Commitment (Election IV): During the high-salience election window, the same affectively polarizing content significantly reduces subscription conversion probability (by ~1.6 percentage points, more than 2.5x the base rate) and accelerates churn.
• Ideological Dimensions: The engagement margin is narrow (driven mainly by affective and right-leaning BERT content), while the commitment penalty is broad. During the election, right-leaning ideological content (Nationalist, Economic-Right, Environmental-Right) also reduces subscriptions and increases churn, even when engagement effects are null.

Mechanism Analysis

The authors investigate why engagement does not convert to commitment:

• Nature of Engagement: Affective content increases the number of articles viewed but decreases time per article and does not increase active days. This suggests "low-commitment snacking" rather than deep engagement.
• Confirmation Bias vs. Balanced Consumption:
  • User-Side: Three pre-determined ideology proxies (residential location, pre-period reading lean via LLM and BERT) do not moderate the engagement or subscription effects. Ideologically congruent and incongruent users respond similarly.
  • Supply-Side: When the publisher's supply of content opposite a user's baseline ideology increases (exogenous shock), users increase their consumption of that counter-side content. This evidence supports balanced consumption over confirmation bias.
• Conversion Funnel: Polarizing content does not push users deeper into the monetization funnel; in fact, it reduces the probability of encountering subscription prompts (likely due to lower article depth) and does not moderate the conversion jump at the paywall threshold.

Economic Implications

The authors calculate that the implied subscription loss from one additional polarizing click (driven by the Election IV) exceeds marginal advertising revenue by 59 to 176 times. Even at the conservative end of the confidence interval, the revenue loss per click is economically significant, overturning the editorial heuristic that engagement metrics proxy publisher value.

4. Contributions and Significance

The paper makes three primary contributions:

1. Extending Polarization Literature: It moves beyond the established link between polarizing content and political behavior/attention to examine the subscription and retention margins critical to digital publisher economics. It demonstrates that the "polarization trap" is a first-order economic issue, not just a political one.
2. Adjudicating Demand Mechanisms: It provides evidence favoring balanced consumption over confirmation bias in the context of news demand. The lack of ideological moderation and the positive response to counter-side supply shocks suggest that readers do not merely seek reinforcement; they consume a mix of sources, particularly when supply shifts exogenously.
3. Methodological Innovation: It introduces a validated, article-level polarization framework that decomposes polarization into affective and multiple ideological dimensions using a combination of supervised BERT classifiers and LLM scorers. This allows for the operational testing of the affective-versus-ideological asymmetry behind the commitment penalty.

Significance for Stakeholders

• For Publishers: The findings suggest that optimizing for engagement metrics (clicks, time on site) via affectively polarizing content is a risky strategy that may undermine the trust and perceived value necessary for subscription conversion, particularly during periods of high political salience.
• For Regulators and Platforms: The results challenge the assumption that engagement metrics are a valid proxy for consumer welfare. Users may click on content that lowers their long-term willingness to pay, indicating a form of "revealed-preference regret." This has implications for algorithmic amplification on platforms that monetize both attention and commitment.

The paper concludes that while polarizing content captures attention, it does not convert to subscriptions and can actively erode commitment, creating a significant economic trade-off for digital news publishers.