pith. sign in

arxiv: 2605.18357 · v1 · pith:7UBX667Unew · submitted 2026-05-18 · 💰 econ.GN · q-fin.EC

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

Shunyao Yan , Klaus M. Miller This is my paper

Pith reviewed 2026-05-19 23:27 UTC · model grok-4.3

classification 💰 econ.GN q-fin.EC
keywords polarizing contentengagementsubscriptionsnews platformsinstrumental variablesaffective polarizationchurnbalanced consumption
0
0 comments X

The pith

Polarizing news content increases time on site but does not raise subscriptions and reduces them during elections.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

News platforms encounter a clear economic trade-off: content that boosts short-term user engagement through polarization does not translate into paid commitments. The study measures polarization at the article level using deep-learning and language models adapted to a multiparty context, then isolates causal effects with supply-side editorial variation and election-driven salience as instruments. Supply increases in polarizing material reliably lift engagement metrics yet leave subscriptions flat, and in high-salience election periods the same material lowers subscriptions and speeds up churn, with affective polarization producing the largest negative shifts. The data show no support for confirmation bias as the driver, since pre-set ideology measures do not change the patterns, while readers increase consumption of ideologically opposite content when the publisher supplies balanced coverage.

Core claim

Supply-driven rises in polarizing content raise engagement but leave subscriptions unchanged; during high-salience election windows the same content lowers subscriptions, accelerates churn, and shows the largest negative effects through affective polarization, while mechanisms favor balanced consumption over confirmation bias.

What carries the argument

Article-level polarization scores generated by deep-learning classifiers and large language models, identified through a Bartik instrument on supply-side editorial variation and an election instrument on demand-side political salience.

If this is right

  • Polarizing articles capture attention without converting viewers into subscribers.
  • Elevated political salience turns the engagement gain into a net loss for commitment and retention.
  • Affective polarization produces sharper drops in paid readership than other forms of divisiveness.
  • When publishers supply coverage of both sides, readers increase consumption of ideologically opposing material.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Platforms that optimize only for engagement metrics may see slower revenue growth from subscriptions over time.
  • Adjusting content mix during election cycles could protect long-term user bases even if short-term attention falls.
  • The pattern points to a broader tension in digital media between attention capture and durable revenue models.

Load-bearing premise

The two instrumental variables capture only exogenous supply shifts and salience changes without being correlated with unobserved demand shocks that would directly affect subscriptions or retention.

What would settle it

New data showing that polarizing content reduces subscriptions even outside election periods, or that pre-determined ideology proxies strongly moderate either engagement or subscription effects, would contradict the central pattern.

Figures

Figures reproduced from arXiv: 2605.18357 by Klaus M. Miller, Shunyao Yan.

Figure 1
Figure 1. Figure 1: Multi-dimensional Measurement of Political Polarization [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
read the original 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

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript studies how polarizing news content affects engagement (time on site) versus commitment (subscriptions and retention) on a major digital news platform. Polarization is measured via deep-learning classifiers and LLMs adapted to a multiparty setting. Causal identification relies on a Bartik instrument exploiting supply-side editorial variation and an election-period instrument exploiting demand-side salience. Core results: supply-driven polarizing content raises engagement but not subscriptions; during high-salience election windows the same content reduces subscriptions, accelerates churn, and shows the largest effects via affective polarization. Mechanism tests reject moderation by pre-determined ideology proxies (inconsistent with confirmation bias) but find support for balanced consumption when the publisher covers both sides of an issue.

Significance. If the identification holds, the paper documents a practically important asymmetric trade-off for digital publishers: polarizing content reliably captures attention yet fails to convert to paid commitment and can damage retention when political salience rises. The dual-IV design and machine-learning measurement of polarization in a multiparty context are strengths; the results speak directly to platform strategy and the economics of media polarization.

major comments (2)
  1. [§4 (Identification)] §4 (Identification): The headline claim that supply-driven polarizing content raises engagement but not subscriptions (and reduces subscriptions during elections) rests on the Bartik instrument satisfying the exclusion restriction. The manuscript must demonstrate that editorial polarization shifts are uncorrelated with unobserved demand shocks or anticipated retention trends; absent explicit pre-trend tests, placebo checks on lagged subscriptions, or correlations with other demand proxies, the engagement-subscription divergence is unidentified.
  2. [§5 (Main Results)] §5 (Main Results): The election-window subscription and churn estimates are load-bearing for the asymmetric trade-off conclusion, yet the presented results omit first-stage F-statistics, weak-instrument robust inference, and overidentification tests. Without these diagnostics it is impossible to assess whether the instruments isolate the intended variation or whether post-hoc sample choices affect the findings.
minor comments (2)
  1. [Abstract] The abstract refers to 'three pre-determined ideology proxies' without naming them; a short description or pointer to the data section would aid readability.
  2. [§3 (Measurement)] §3 (Measurement): Additional detail on classifier training, validation accuracy, and robustness to alternative polarization thresholds would help readers evaluate measurement error.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the identification strategy and strengthen the presentation of our results. We address each major point below and commit to incorporating the suggested diagnostics and robustness checks in the revised manuscript.

read point-by-point responses

  1. Referee: [§4 (Identification)] §4 (Identification): The headline claim that supply-driven polarizing content raises engagement but not subscriptions (and reduces subscriptions during elections) rests on the Bartik instrument satisfying the exclusion restriction. The manuscript must demonstrate that editorial polarization shifts are uncorrelated with unobserved demand shocks or anticipated retention trends; absent explicit pre-trend tests, placebo checks on lagged subscriptions, or correlations with other demand proxies, the engagement-subscription divergence is unidentified.

    Authors: We agree that more explicit evidence on the exclusion restriction would bolster confidence in the Bartik instrument. The instrument exploits within-publisher editorial shifts in polarization that are driven by supply-side decisions rather than contemporaneous demand. To directly respond to the concern, we will add pre-trend plots for subscriptions and retention, placebo regressions on lagged outcomes, and correlations between the instrument and observable demand proxies (e.g., search trends or competitor coverage) in the revision. These checks will be reported alongside the main results to document that the instrument is uncorrelated with anticipated retention trends. revision: yes

  2. Referee: [§5 (Main Results)] §5 (Main Results): The election-window subscription and churn estimates are load-bearing for the asymmetric trade-off conclusion, yet the presented results omit first-stage F-statistics, weak-instrument robust inference, and overidentification tests. Without these diagnostics it is impossible to assess whether the instruments isolate the intended variation or whether post-hoc sample choices affect the findings.

    Authors: We acknowledge that standard IV diagnostics are necessary for transparency, especially given the centrality of the election-window results. In the revised manuscript we will report first-stage F-statistics for both the Bartik and election instruments, implement weak-instrument robust inference (Anderson-Rubin confidence sets), and conduct overidentification tests for the periods where multiple instruments are available. These additions will be placed in the main results tables and appendix to allow readers to evaluate instrument strength and validity directly. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical identification

full rationale

The paper's derivation chain consists of measuring article polarization via external deep-learning classifiers, constructing a Bartik instrument from supply-side editorial variation and an election instrument from demand-side salience, then estimating IV effects on engagement and subscription outcomes. These steps rely on pre-determined ideology proxies and external instruments rather than any self-referential equations, fitted parameters renamed as predictions, or load-bearing self-citations. No step reduces by construction to its own inputs; the central claims remain independent of internal fitting loops and are self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review limited to abstract; no explicit free parameters, axioms, or invented entities are stated. Polarization measurement depends on trained classifiers whose training data and hyperparameters are unspecified.

pith-pipeline@v0.9.0 · 5739 in / 1099 out tokens · 33894 ms · 2026-05-19T23:27:33.790781+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 1 internal anchor

  1. [1]

    Econometrica , volume =

    Dufour, Jean-Marie , title =. Econometrica , volume =. 1997 , doi =

    work page 1997

  2. [2]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume =

    Benjamini, Yoav and Hochberg, Yosef , title =. Journal of the Royal Statistical Society: Series B (Methodological) , volume =

    work page

  3. [3]

    Milkman , year =

    Berger, Jonah and Katherine L. Milkman , year =. What Makes Online Content Viral? , journal =

    work page

  4. [4]

    Consuming Cross-Cutting Media Causes Learning and Moderates Attitudes: A Field Experiment with

    Broockman, David E and Kalla, Joshua L , journal=. Consuming Cross-Cutting Media Causes Learning and Moderates Attitudes: A Field Experiment with. 2025 , publisher=

    work page 2025

  5. [5]

    DellaVigna, Stefano and Ethan Kaplan , year =. The. The Quarterly Journal of Economics , volume =

    work page

  6. [6]

    Handbook of Media Economics , editor =

    DellaVigna, Stefano and. Handbook of Media Economics , editor =. 2015 , title =

    work page 2015

  7. [7]

    Paywalls and the Demand for News , journal =

    Chiou, Lesley and Tucker, Catherine , year =. Paywalls and the Demand for News , journal =

    work page

  8. [8]

    , year =

    Aral, Sinan and Dhillon, Paramveer S. , year =. Digital Paywall Design: Implications for Content Demand and Subscriptions , journal =

    work page

  9. [9]

    and Manchanda, Puneet , title =

    Pattabhiramaiah, Adithya and Sriram, S. and Manchanda, Puneet , title =. Journal of Marketing , year =

    work page

  10. [10]

    Shapiro , year =

    Gentzkow, Matthew and Jesse M. Shapiro , year =. Ideological Segregation Online and Offline , journal =

    work page

  11. [11]

    Shapiro and Matt Taddy , year =

    Gentzkow, Matthew and Jesse M. Shapiro and Matt Taddy , year =. Measuring Group Differences in High-Dimensional Choices: Method and Application to Congressional Speech , journal =

    work page

  12. [12]

    , year =

    Guess, Andrew M. , year =. American Journal of Political Science , volume =

    work page

  13. [13]

    and Ali Yurukoglu , year =

    Martin, Gregory J. and Ali Yurukoglu , year =. Bias in Cable News: Persuasion and Polarization , journal =

    work page

  14. [14]

    The Market for News , journal =

    Mullainathan, Sendhil and Andrei Shleifer , year =. The Market for News , journal =

    work page

  15. [15]

    , year =

    Puglisi, Riccardo and Snyder, Jr., James M. , year =. Empirical Studies of Media Bias , booktitle =

    work page

  16. [16]

    Snyder , year =

    Puglisi, Riccardo and James M. Snyder , year =. Newspaper Coverage of Political Scandals , journal =

    work page

  17. [17]

    and Mark I

    Rust, Roland T. and Mark I. Alpert , year =. An Audience Flow Model of Television Viewing Choice , journal =

    work page

  18. [18]

    Marketing Science , year =

    Schoenmueller, Verena and Netzer, Oded and Stahl, Florian , title =. Marketing Science , year =

    work page

  19. [19]

    Journal of Marketing Research , volume =

    Toubia, Olivier and Garud Iyengar and Ren. Journal of Marketing Research , volume =. 2019 , title =

    work page 2019

  20. [20]

    Sharif , year =

    Woolley, Kaitlin and Marissa A. Sharif , year =. Down a Rabbit Hole: How Prior Media Consumption Shapes Subsequent Media Consumption , journal =

    work page

  21. [21]

    From Feeds to Inboxes: A Comparative Study of Polarization in

    Yoganarasimhan, Hema and Iakovetskaia, Irina , journal=. From Feeds to Inboxes: A Comparative Study of Polarization in. 2024 , publisher=

    work page 2024

  22. [22]

    Journal of Marketing Research , volume=

    How does the adoption of ad blockers affect news consumption? , author=. Journal of Marketing Research , volume=. 2022 , publisher=

    work page 2022

  23. [23]

    Marketing Science , volume=

    Frontiers: The Impact of Ad-Blockers on Online Consumer Behavior , author=. Marketing Science , volume=. 2022 , publisher=

    work page 2022

  24. [24]

    1991 , publisher=

    Who Benefits from State and Local Economic Development Policies? , author=. 1991 , publisher=

    work page 1991

  25. [25]

    American Economic Review , year=

    Bartik Instruments: What, When, Why, and How , author=. American Economic Review , year=

    work page

  26. [26]

    Demand for Online News under Government Control: Evidence from

    Simonov, Andrey and Rao, Justin , journal=. Demand for Online News under Government Control: Evidence from. 2022 , publisher=

    work page 2022

  27. [27]

    Marketing Science , volume=

    The language that drives engagement: A systematic large-scale analysis of headline experiments , author=. Marketing Science , volume=. 2025 , publisher=

    work page 2025

  28. [28]

    Political Analysis , volume=

    Positioning political texts with large language models by asking and averaging , author=. Political Analysis , volume=. 2025 , publisher=

    work page 2025

  29. [29]

    Frontiers: The Persuasive Effect of

    Simonov, Andrey and Sacher, Szymon and Dub. Frontiers: The Persuasive Effect of. Marketing Science , volume=. 2022 , publisher=

    work page 2022

  30. [30]

    The Review of Economic Studies , volume=

    Quasi-experimental shift-share research designs , author=. The Review of Economic Studies , volume=. 2022 , publisher=

    work page 2022

  31. [31]

    Journal of Economic Perspectives , volume=

    A Practical Guide to Shift-Share Instruments , author=. Journal of Economic Perspectives , volume=

    work page

  32. [32]

    Digital News Consumption: Evidence from Smartphone Content in the 2024

    Aridor, Guy and Dekel, Tevel and Jimenez Duran, Rafael and Levy, Ro'ee and Song, Lena , journal=. Digital News Consumption: Evidence from Smartphone Content in the 2024

    work page 2024

  33. [33]

    Review of Economics and Statistics , volume=

    Cross-country trends in affective polarization , author=. Review of Economics and Statistics , volume=. 2024 , publisher=

    work page 2024

  34. [34]

    Public Opinion Quarterly , year=

    Iyengar, Shanto and Sood, Gaurav and Lelkes, Yphtach , title=. Public Opinion Quarterly , year=

    work page

  35. [35]

    and Wills, Julian A

    Brady, William J. and Wills, Julian A. and Jost, John T. and Tucker, Joshua A. and Van Bavel, Jay J. , title=. Proceedings of the National Academy of Sciences , year=

    work page

  36. [36]

    and van der Linden, Sander , title=

    Rathje, Steve and Van Bavel, Jay J. and van der Linden, Sander , title=. Proceedings of the National Academy of Sciences , year=

    work page

  37. [37]

    Robertson, Claire E. and Pr. Negativity drives online news consumption , journal=. 2023 , volume=

    work page 2023

  38. [38]

    American Economic Review , year=

    Gentzkow, Matthew , title=. American Economic Review , year=

    work page

  39. [39]

    2013 , publisher=

    Arceneaux, Kevin and Johnson, Martin , title=. 2013 , publisher=

    work page 2013

  40. [40]

    2025 , note=

    Braghieri, Luca and Eichmeyer, Sarah and Levy, Ro'ee and Mobius, Markus and Steinhardt, Jacob and Zhong, Ruiqi , title=. 2025 , note=

    work page 2025

  41. [41]

    , title=

    Wilbur, Kenneth C. , title=. Marketing Science , volume=

    work page

  42. [42]

    and Coate, Stephen , title=

    Anderson, Simon P. and Coate, Stephen , title=. The Review of Economic Studies , year=

    work page

  43. [43]

    The RAND Journal of Economics , year=

    Armstrong, Mark , title=. The RAND Journal of Economics , year=

    work page

  44. [44]

    , title=

    Chae, Inyoung and Ha, Jihyeon and Schweidel, David A. , title=. Marketing Science , year=

    work page

  45. [45]

    Outsourcing Tasks Online: Matching Supply and Demand on Peer-to-Peer Internet Platforms , journal =

    Cullen, Zo. Outsourcing Tasks Online: Matching Supply and Demand on Peer-to-Peer Internet Platforms , journal =

    work page

  46. [46]

    Marketing Science , volume =

    Barron, Kyle and Kung, Edward and Proserpio, Davide , title =. Marketing Science , volume =

    work page

  47. [47]

    and Ross Arguedas, Amy and Nielsen, Rasmus Kleis , title =

    Newman, Nic and Fletcher, Richard and Robertson, Craig T. and Ross Arguedas, Amy and Nielsen, Rasmus Kleis , title =. 2025 , address =

    work page 2025

  48. [48]

    European Journal of Political Research , year =

    Kriesi, Hanspeter and Grande, Edgar and Lachat, Romain and Dolezal, Martin and Bornschier, Simon and Frey, Timotheos , title =. European Journal of Political Research , year =

    work page

  49. [49]

    and Ensley, Michael J

    Carmines, Edward G. and Ensley, Michael J. and Wagner, Michael W. , title =. American Behavioral Scientist , year =

    work page

  50. [50]

    and McAfee, R

    Goldstein, Daniel G. and McAfee, R. Preston and Suri, Siddharth , title =. ACM Transactions on Economics and Computation , year =

    work page

  51. [51]

    Attention Spillovers from News to Ads: Evidence from an Eye-Tracking Experiment , journal =

    Simonov, Andrey and Valletti, Tommaso and Veiga, Andr. Attention Spillovers from News to Ads: Evidence from an Eye-Tracking Experiment , journal =. 2025 , volume =

    work page 2025

  52. [52]

    and Fiske, Donald W

    Campbell, Donald T. and Fiske, Donald W. , title =. Psychological Bulletin , year =

    work page

  53. [53]

    Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference

    Chiang, Wei-Lin and Zheng, Lianmin and Sheng, Ying and Angelopoulos, Anastasios Nikolas and Li, Tianle and Li, Dacheng and Zhang, Hao and Zhu, Banghua and Jordan, Michael and Gonzalez, Joseph E. and Stoica, Ion , title =. arXiv preprint arXiv:2403.04132 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  54. [54]

    Richard and Koch, Gary G

    Landis, J. Richard and Koch, Gary G. , title =. Biometrics , year =

    work page

  55. [55]

    Anderson, T. W. and Rubin, Herman , title =. The Annals of Mathematical Statistics , volume =. 1949 , doi =

    work page 1949

  56. [56]

    and Hansen, Christian B

    Conley, Timothy G. and Hansen, Christian B. and Rossi, Peter E. , title =. Review of Economics and Statistics , volume =. 2012 , doi =

    work page 2012

  57. [57]

    , title =

    Gentzkow, Matthew and Shapiro, Jesse M. , title =. Econometrica , volume =. 2010 , doi =

    work page 2010

  58. [58]

    , title =

    Iyengar, Shanto and Lelkes, Yphtach and Levendusky, Matthew and Malhotra, Neil and Westwood, Sean J. , title =. Annual Review of Political Science , volume =. 2019 , doi =

    work page 2019

  59. [59]

    A Robust Test for Weak Instruments , journal =

    Montiel Olea, Jos. A Robust Test for Weak Instruments , journal =. 2013 , doi =

    work page 2013

  60. [60]

    , title =

    Staiger, Douglas and Stock, James H. , title =. Econometrica , volume =. 1997 , doi =

    work page 1997

  61. [61]

    Journal of Econometrics , volume =

    Kleibergen, Frank and Paap, Richard , title =. Journal of Econometrics , volume =. 2006 , doi =

    work page 2006

  62. [62]

    Review of Economic Studies , year =

    Rambachan, Ashesh and Roth, Jonathan , title =. Review of Economic Studies , year =

    work page

  63. [63]

    Colin and Gelbach, Jonah B

    Cameron, A. Colin and Gelbach, Jonah B. and Miller, Douglas L. , title =. Review of Economics and Statistics , year =

    work page

  64. [64]

    2020 , month = jul, address =

    Piechota, Grzegorz , title =. 2020 , month = jul, address =

    work page 2020

  65. [65]

    2025 , month = dec, howpublished =

    The. 2025 , month = dec, howpublished =

    work page 2025