pith. sign in

arxiv: 2606.06812 · v1 · pith:SHTURG2Gnew · submitted 2026-06-05 · 💻 cs.CL

Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths

Farhan Samir , Jappun Dhillon , Meghna Ravikumar , Syed Ishtiaque Ahmed , Vered Shwartz This is my paper

Pith reviewed 2026-06-27 22:28 UTC · model grok-4.3

classification 💻 cs.CL
keywords media representationpolice-involved deathsnews narrativesperspective quantificationCanadian mediastate versus civilian voicescomputational content analysis
0
0 comments X

The pith

Canadian news on police-involved deaths quotes state officials nearly three times more often than civilians or family members.

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

The paper examines 4,000 Canadian news articles spanning 25 years to measure whose voices appear in coverage of policing-related deaths. It introduces a computational model called PerspectiveGap that tags and counts perspectives from state bureaucrats against those from civilians such as relatives, eyewitnesses, and community groups. The central finding is a consistent 3-to-1 imbalance favoring official accounts, although civilian perspectives have become more common in later years. State quotes tend to stay procedural while civilian ones carry stronger emotion. The authors present the model as reusable for tracking narrative patterns in other places.

Core claim

Reporting on police-involved deaths on average features perspectives from state bureaucrats at a rate nearly three times as much as perspectives from other members of the public, including relatives, community members, eyewitnesses, lawyers representing the family, or civil liberties groups. A considerable fraction of articles contain no points of view from civilian actors, though civilian representation has increased in recent years. Qualitatively, state bureaucrats' accounts tend to be clinical and procedural, while civilian discourse carries considerably more emotional valence.

What carries the argument

PerspectiveGap, a computational model that identifies and quantifies distinct perspectives from state bureaucrats versus civilian actors across news text.

If this is right

  • Civilian representation in these stories has risen over the 25-year span.
  • Many articles still include zero civilian perspectives.
  • State accounts remain more clinical and procedural than civilian ones.
  • The same model can be applied to news coverage in other countries or topics.

Where Pith is reading between the lines

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

  • The measured imbalance may shape how readers assign responsibility for these deaths.
  • Similar automated tracking could reveal whether the pattern holds for other high-stakes public events.
  • If the model generalizes, newsrooms could use it as an internal audit tool for source diversity.

Load-bearing premise

The PerspectiveGap model correctly distinguishes and counts state-bureaucrat perspectives from civilian ones in the collected articles.

What would settle it

Re-coding a random sample of the 4,000 articles by hand and finding that the model assigns the wrong perspective category to more than a small fraction of quotes, or that the overall 3-to-1 ratio disappears under that manual count.

Figures

Figures reproduced from arXiv: 2606.06812 by Farhan Samir, Jappun Dhillon, Meghna Ravikumar, Syed Ishtiaque Ahmed, Vered Shwartz.

Figure 1
Figure 1. Figure 1: We measure the volume of perspectives from state [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Confusion matrix between the labels. Baselines. To contextualize the performance of PerspectiveGap, we compare the model with two baselines. We assess GPT-4o’s capacity to perform this task. We used the training set for prompt￾tuning; see the full prompt in our repository. We also compare against a random classification baseline, where one of the labels is picked uniformly at random. Performance. We report… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of predicted vs. ground-truth refer [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Outlet and year distributions for the large-scale [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Median proportion of articles devoted to perspec [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Key terms identified for passages tagged as [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

We perform the largest known computational analysis of Canadian news narratives about police-involved deaths, spanning 4,000 articles from the last quarter-century. We develop a novel computational model, PerspectiveGap, grounded in prior sociological work on media representation of policing. We find that reporting on police-involved deaths on average features perspectives from state bureaucrats at a rate nearly three times as much as perspectives from other members of the public, including relatives, community members, eyewitnesses, lawyers representing the family, or civil liberties groups. A considerable fraction of articles contain no points of view from civilian actors, though civilian representation has increased in recent years. Qualitatively, we find that state bureaucrats' accounts of these deaths tend to be clinical and procedural, while civilian discourse carries considerably more emotional valence. The PerspectiveGap framework developed here can be contextualized to other jurisdictions, offering a scalable approach for analyzing how media systems construct narratives around policing and accountability.

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

1 major / 1 minor

Summary. The paper performs a computational analysis of 4,000 Canadian news articles spanning 25 years on police-involved deaths. It introduces the PerspectiveGap model, grounded in prior sociological work, to classify and quantify perspectives from state bureaucrats versus civilian actors (relatives, community members, eyewitnesses, lawyers, civil liberties groups). The central claim is that state-bureaucrat perspectives appear at nearly three times the rate of civilian ones on average, with a considerable fraction of articles lacking civilian viewpoints (though this has increased recently) and qualitative differences in tone (clinical/procedural for bureaucrats vs. emotional for civilians). The framework is positioned as scalable to other jurisdictions.

Significance. If the PerspectiveGap classification is reliable, the work supplies a large-scale, longitudinal measurement of media representation dynamics around policing accountability, with potential to generalize across contexts. The dataset size and temporal scope are clear assets for empirical media studies. However, the absence of any reported validation for the core classifier directly limits the interpretability of the quantitative ratio and trend claims.

major comments (1)
  1. [Abstract] Abstract: the headline quantitative result (state-bureaucrat perspectives at nearly 3 imes the rate of civilian perspectives) is produced by PerspectiveGap, yet the abstract supplies no information on model validation, held-out test sets, precision/recall for the state-vs-civilian distinction, or inter-annotator agreement on quote/entity labeling. Without these, it is impossible to assess whether the reported ratio is recoverable from the data or an artifact of systematic labeling error.
minor comments (1)
  1. [Abstract] Abstract: the statement that 'a considerable fraction of articles contain no points of view from civilian actors' is left unquantified; adding even a single percentage or count would improve precision.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address the major comment on the abstract below and agree that additional information on model validation is warranted there to support interpretability of the core claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline quantitative result (state-bureaucrat perspectives at nearly 3 times the rate of civilian perspectives) is produced by PerspectiveGap, yet the abstract supplies no information on model validation, held-out test sets, precision/recall for the state-vs-civilian distinction, or inter-annotator agreement on quote/entity labeling. Without these, it is impossible to assess whether the reported ratio is recoverable from the data or an artifact of systematic labeling error.

    Authors: We agree that the abstract, as a high-level summary of the headline result, should reference key validation details for PerspectiveGap to enable readers to evaluate the reliability of the reported ratios and trends. The full manuscript describes the model's grounding in sociological literature on media representation, the quote/entity labeling process, and evaluation procedures including inter-annotator agreement. We will revise the abstract to incorporate a concise statement on these elements, such as the inter-annotator agreement score and classification performance metrics on held-out data, without altering the overall length or focus. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical measurement via externally grounded model

full rationale

The paper describes an empirical study that applies a novel but externally grounded computational model (PerspectiveGap) to 4000 news articles to quantify perspective representation. No equations, derivations, fitted parameters renamed as predictions, or self-referential steps are present in the abstract or described methodology. The central claim is produced by running the model on data rather than by construction from the model's definition or self-citations. The model is stated to be grounded in prior sociological work, with no indication that the load-bearing justification reduces to the authors' own prior unverified results. This is a standard empirical measurement paper whose quantitative outputs are independent of any internal definitional loop.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated beyond the existence of the PerspectiveGap model itself.

pith-pipeline@v0.9.1-grok · 5707 in / 1044 out tokens · 13880 ms · 2026-06-27T22:28:03.712337+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

50 extracted references · 5 canonical work pages

  1. [1]

    Arnav Arora, Srishti Yadav, Maria Antoniak, Serge Belongie, and Isabelle Augen- stein. 2025. Multi-modal framing analysis of news.arXiv preprint arXiv:2503.20960 (2025)

    arXiv 2025

  2. [2]

    Bob Becken. 2023. ‘Defund the Police’ Calls in Canada Began in 2020. Today, Budgets Continue to Climb.CBC Radio(12 Feb. 2023). https://www.cbc.ca/radio/ day6/defund-police-2023-budgets-grow-1.6741711 Posted: Feb 12, 2023. Last Updated: Feb 12, 2023

    2023

  3. [3]

    Dallas Card, Amber Boydstun, Justin H Gross, Philip Resnik, and Noah A Smith

  4. [4]

    The media frames corpus: Annotations of frames across issues. InProceed- ings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 438–444

  5. [5]

    CBC News. 2009. Police clear officer in shooting of homeless man.CBC News (15 Dec. 2009). https://www.cbc.ca/news/canada/british-columbia/police-clear- officer-in-shooting-of-homeless-man-1.833040 Posted 1:10 PM EST

    2009

  6. [6]

    CBC News. 2010. Hold inquiry into Yukon jail death: lawyer.CBC News(19 May 2010). https://www.cbc.ca/news/canada/north/hold-inquiry-into-yukon-jail- death-lawyer-1.942769 Posted 1:06 PM EDT

    2010

  7. [7]

    CBC News. 2017. Abdirahman Abdi’s death not due to beating, lawyer claims. CBC News(26 Oct. 2017). https://www.cbc.ca/news/canada/ottawa/daniel- montsion-abdirahman-abdi-not-a-beating-1.4373493 Posted 4:33 PM EDT

    2017

  8. [8]

    2014.Racialization, crime, and criminal justice in Canada

    Wendy Chan and Dorothy Chunn. 2014.Racialization, crime, and criminal justice in Canada. University of Toronto Press

    2014

  9. [9]

    2022.The skin we’re in: A year of black resistance and power

    Desmond Cole. 2022.The skin we’re in: A year of black resistance and power. Anchor Canada

    2022

  10. [10]

    Andrew Crosby, Alexander McClelland, Tanya L Sharpe, Evelyn Maeder, Cather- ine Stinson, Kanika Samuels Wortley, Karyn Graham, Rina Faiyaz Khan, and Zhongmin Lin. 2025. Tracking (In) Justice: Documenting Fatal Encounters with Police in Canada.Canadian Journal of Law and Society/La Revue Canadienne Droit et Société40, 1 (2025), 23–47

    2025

  11. [11]

    KhudaBukhsh, Edson Severnini, and Daniel S

    Logan Crowl, Sujan Dutta, Ashiqur R. KhudaBukhsh, Edson Severnini, and Daniel S. Nagin. 2025. Measuring criticism of the police in the local news media using large language models.Proceedings of the National Academy of Sciences 122, 9 (March 2025), e2418821122. doi:10.1073/pnas.2418821122

    work page doi:10.1073/pnas.2418821122 2025

  12. [12]

    Justine Davidson. 2009. Man ‘looked right through us, ’ officer testifies. https:// www.whitehorsestar.com/News/man-looked-right-through-us-officer-testifies

    2009

  13. [13]

    2012.The public and its problems: An essay in political inquiry

    John Dewey and Melvin L Rogers. 2012.The public and its problems: An essay in political inquiry. Penn State Press

    2012

  14. [14]

    1994.An introduction to the bootstrap

    Bradley Efron and Robert J Tibshirani. 1994.An introduction to the bootstrap. Chapman and Hall/CRC

    1994

  15. [15]

    Robert M Entman and Andrew Rojecki. 1993. Freezing out the public: Elite and media framing of the US anti-nuclear movement. (1993)

    1993

  16. [16]

    Anjalie Field, Prateek Verma, Nay San, Jennifer L Eberhardt, and Dan Jurafsky

  17. [17]

    Developing speech processing pipelines for police accountability.arXiv preprint arXiv:2306.06086(2023)

    arXiv 2023

  18. [18]

    2013.Archaeology of knowledge

    Michel Foucault. 2013.Archaeology of knowledge. routledge

    2013

  19. [19]

    Jeremy Grimaldi. 2016. Crown appeals acquittal of York cop whose vehicle hit, killed pedestrian.Vaughan Citizen(4 Nov. 2016). https://www.yorkregion.com/ news/crime/crown-appeals-acquittal-of-york-cop-whose-vehicle-hit-killed- pedestrian/article_a7772050-8ca0-58f4-b95c-6963380eed3f.html

    2016

  20. [20]

    Stuart Hall, Chas Critcher, Brian Roberts, John Clarke, and Tony Jefferson. 2017. Policing the crisis: Mugging, the state and law and order. (2017)

    2017

  21. [21]

    Log Odds Ratio Informative Dirichlet Prior

    Dan Jurafsky and James H. Martin. 2024.Speech and Language Processing (3rd ed. draft). Stanford University. https://web.stanford.edu/~jurafsky/slp3/22.pdf Section 22.5.1: “Log Odds Ratio Informative Dirichlet Prior”, Chapter 22: “Lexicons for Sentiment, Affect, and Connotation”

    2024

  22. [22]

    2025.Copaganda: How police and the media manipulate our news

    Alec Karakatsanis. 2025.Copaganda: How police and the media manipulate our news. The New Press

    2025

  23. [23]

    Amanda Kelly. 2014. Shooting victim Alain Magloire described as a caring person. Global News(4 Feb. 2014). Posted 3:58 pm. Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths WebSci ’26, May 26–29, 2026, Braunschweig, Germany

    2014

  24. [24]

    Larry Kusch. 2013. Inquest to probe death of man, 26, shot by police in West End in 2008.Winnipeg Free Press(9 Aug. 2013). https://www.winnipegfreepress.com/local/2013/08/09/inquest-to-probe- death-of-man-26-shot-by-police-in-west-end-in-2008 Posted 1:00 AM CDT

    2013

  25. [25]

    2022.Police Use of Force: Understanding its Impact on Indigenous and Black Community Members in Ontario

    Erick Laming. 2022.Police Use of Force: Understanding its Impact on Indigenous and Black Community Members in Ontario. Ph. D. Dissertation

    2022

  26. [26]

    2017.Public opinion

    Walter Lippmann. 2017.Public opinion. Routledge

    2017

  27. [27]

    Jacques Marcoux and Katie Nicholson. 2018. Deadly force.CBC News(2018)

    2018

  28. [28]

    2025.Policing Black lives: State violence in Canada from slavery to the present

    Robyn Maynard. 2025.Policing Black lives: State violence in Canada from slavery to the present. Duke University Press

    2025

  29. [29]

    Alyce McGovern and Murray Lee. 2010. ‘Cop [ying] it sweet’: Police media units and the making of news.Australian & New Zealand Journal of Criminology43, 3 (2010), 444–464

    2010

  30. [30]

    Saif M Mohammad. 2025. NRC VAD Lexicon v2: Norms for valence, arousal, and dominance for over 55k English terms.arXiv preprint arXiv:2503.23547(2025)

    arXiv 2025

  31. [31]

    Kevin Nielsen. 2025. Ontario’s police watchdog clears officers in fatal shooting of Hamilton man.Global News(6 June 2025). https://globalnews.ca/news/11216626/ erixon-kabera-siu-update/ Posted 12:40 pm

    arXiv 2025

  32. [32]

    Nora O’Malley. 2022. Police watchdog clears Tofino RCMP officer in shooting death of Julian Jones. https://vancouverislandfreedaily.com/2022/11/25/police- watchdog-clears-tofino-rcmp-officer-in-shooting-death-of-tla-o-qui-aht- man/

    2022

  33. [33]

    Elinor Ostrom. 1996. Crossing the great divide: Coproduction, synergy, and development.World development24, 6 (1996), 1073–1087

    1996

  34. [34]

    Shon Otmazgin, Arie Cattan, and Yoav Goldberg. 2022. F-coref: Fast, Accurate and Easy to Use Coreference Resolution. InProceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing: System Demon- strations, Wray Buntine and Maria Lia...

    work page doi:10.18653/v1/2022.aacl-demo.6 2022

  35. [35]

    2014.Black Males’ Perceptions of and Experiences with the Police in Toronto

    Akwasi Owusu-Bempah. 2014.Black Males’ Perceptions of and Experiences with the Police in Toronto. Ph.D. University of Toronto (Canada), Canada – Ontario, CA. https://www.proquest.com/dissertations-theses/black-males-perceptions- experiences-with-police/docview/1668381846/se-2?accountid=14656 ISBN: 978- 1-321-65095-2 Publication Title: ProQuest Dissertatio...

    arXiv 2014

  36. [36]

    Kate Puddister and Danielle McNabb. 2021. When the police break the law: The investigation, prosecution and sentencing of Ontario police officers.Canadian Journal of Law and Society/La revue canadienne droit et société36, 3 (2021), 381– 404

    2021

  37. [37]

    Jim Rankin and Patty Winsa. 2012. Known to Police: Toronto Police Stop and Document Black and Brown People Far More Than Whites.Toronto Star(9 March 2012). https://www.thestar.com/news/insight/2012/03/09/known_to_ police_toronto_police_stop_and_document_black_and_brown_people_far_ more_than_whites.html

    2012

  38. [38]

    Eugenia H Rho, Maggie Harrington, Yuyang Zhong, Reid Pryzant, Nicholas P Camp, Dan Jurafsky, and Jennifer L Eberhardt. 2023. Escalated police stops of Black men are linguistically and psychologically distinct in their earliest moments. Proceedings of the National Academy of Sciences120, 23 (2023), e2216162120

    2023

  39. [39]

    1999.What are journalists for?Yale University Press

    Jay Rosen. 1999.What are journalists for?Yale University Press

    1999

  40. [40]

    Steve Russell. 2022. Why did Chadd die? 15 months after a ‘struggle’ with off-duty Toronto cops, his mom needs answers. https://www.thestar.com/news/gta/why- did-chadd-die-15-months-after-a-struggle-with-off-duty-toronto- cops/article_0500dd66-1636-5889-ade2-8b881c5c8cad.html

    2022

  41. [41]

    Ted Rutland, Lee-Anne Goodman, and Vinita Srivastava. 2023. Canadian cities continue to over-invest in policing. doi:10.64628/AAM.p6frg4d7w

    work page doi:10.64628/aam.p6frg4d7w 2023

  42. [42]

    2021.The Thin Blue Line in Black and White: Symbolic Construc- tions of Police Use of Deadly Force in Canadian Newspapers

    Danielle Sutton. 2021.The Thin Blue Line in Black and White: Symbolic Construc- tions of Police Use of Deadly Force in Canadian Newspapers. Ph. D. Dissertation. University of Guelph

    2021

  43. [43]

    Michael Tutton. 2010. N.S. RCMP Call Off Disciplinary Hearing Af- ter Fatal Shooting.The Globe and Mail(11 march 2010). https: //www.theglobeandmail.com/news/national/ns-rcmp-call-off-disciplinary- hearing-after-fatal-shooting/article573039/ The Canadian Press

    2010

  44. [44]

    Jonny Wakefield. 2024. ASIRT clears Edmonton police officers who killed by- stander while firing on robbery suspect. https://edmontonjournal.com/news/ crime/asirt-clears-edmonton-police-bystander-shooting

    2024

  45. [45]

    Kevin Walby and Babatunde Alabi. 2022. Examining Press Conference and Press Release Accounts of Canadian Police Shootings.Canadian Journal of Criminology and Criminal Justice64, 1 (Jan. 2022), 30–52. doi:10.3138/cjccj.2021-0021

    work page doi:10.3138/cjccj.2021-0021 2022

  46. [46]

    Miriam Wanner, Sophia Hager, and Anjalie Field. 2025. Does Local News Stay Local?: Online Content Shifts in Sinclair-Acquired Stations.arXiv preprint arXiv:2510.07060(2025)

    Pith/arXiv arXiv 2025

  47. [47]

    Kailey White, Forrest Stuart, and Shannon L Morrissey. 2021. Whose lives matter? Race, space, and the devaluation of homicide victims in minority communities. Sociology of Race and Ethnicity7, 3 (2021), 333–349

    2021

  48. [48]

    2023.Deadly discrimination: Police-involved deaths and mental health disability in Canada

    Robin Whitehead. 2023.Deadly discrimination: Police-involved deaths and mental health disability in Canada. Ph. D. Dissertation. Université d’Ottawa/University of Ottawa

    2023

  49. [49]

    Janyce M. Wiebe. 1994. Tracking point of view in narrative.Comput. Linguist. 20, 2 (June 1994), 233–287

    1994

  50. [50]

    Caleb Ziems and Diyi Yang. 2021. To Protect and To Serve? Analyzing Entity- Centric Framing of Police Violence. InFindings of the Association for Computa- tional Linguistics: EMNLP 2021. Association for Computational Linguistics, Punta Cana, Dominican Republic, 957–976. doi:10.18653/v1/2021.findings-emnlp.82

    work page doi:10.18653/v1/2021.findings-emnlp.82 2021