About

Dean Knox is an Assistant Professor of Operations, Information and Decisions at the Wharton School, University of Pennsylvania. He is a computational social scientist developing new methods for the study of complex and high-dimensional data. His research interests include policing, speech analysis, ethnic politics, and political communication. His work has been published or is forthcoming in prominent journals such as Science, the Journal of the American Statistical Association, the Proceedings of the National Academy of Sciences, and the American Political Science Review. His research has received recognition through awards including the Gosnell Prize for excellence in political methodology, the John T. Williams dissertation prize, and the best poster award by the Society for Political Methodology.

Research topics

• Computer Science
• Machine Learning
• Sociology
• Political Science
• Psychology
• Mathematics
• Econometrics
• Computer Security
• Artificial Intelligence
• Law
• Criminology
• Statistics
• Gender studies
• Medicine
• Developmental psychology
• Geography
• Data science
• Communication
• Social psychology
• Engineering
• Economics

Selected publications

• A framework for studying causal effects of speech style: application to US presidential campaigns

  Journal of the Royal Statistical Society Series A (Statistics in Society) · 2025-05-16 · 2 citations

  articleOpen access

  Abstract Numerous disciplines hypothesize about the causal effects of speech, which influences others not only through which words are spoken, but also how they are spoken. Yet applied research focuses almost exclusively on the textual component of speech—ignoring audiovisual components or reducing them to coarse summary statistics. We show that text-only analyses are biased, except in implausible scenarios where (a) non-textual speech elements are irrelevant to listeners, or (b) speakers’ vocal style does not change with the words spoken. Even analyses including audiovisual summary measures are biased unless (c) these measures satisfy a ‘sufficient reduction’ condition, fully capturing the non-textual mechanisms through which speech operates. To demonstrate, we develop a formal causal framework that accounts for the unstructured and multimodal nature of speech. We present an application to US presidential campaign speeches, using the framework to clarify implicit assumptions in prior work. We then demonstrate two designs that permit valid hypothesis tests and causal effect estimates under more plausible conditions: (1) a naturalistic experiment exploiting subtle variation in campaign speech ‘catchphrases’ with near-identical wording, identified with automated phrase-clustering methods; and (2) an audio conjoint experiment with nearly 1,000 recordings manipulating specific vocal mechanisms, produced with professional voice actors and audio-editing software.

  PublisherOA PDFDOI

• New data fill long-standing gaps in the study of policing

  Science · 2025-03-27

  letter1st authorCorresponding

  Data show discrimination, but analysis must be more policy relevant.

  PublisherDOI

• Short-term exposure to filter-bubble recommendation systems has limited polarization effects: Naturalistic experiments on YouTube

  Proceedings of the National Academy of Sciences · 2025-02-18 · 10 citations

  articleOpen accessCorresponding

  An enormous body of literature argues that recommendation algorithms drive political polarization by creating "filter bubbles" and "rabbit holes." Using four experiments with nearly 9,000 participants, we show that manipulating algorithmic recommendations to create these conditions has limited effects on opinions. Our experiments employ a custom-built video platform with a naturalistic, YouTube-like interface presenting real YouTube videos and recommendations. We experimentally manipulate YouTube's actual recommendation algorithm to simulate filter bubbles and rabbit holes by presenting ideologically balanced and slanted choices. Our design allows us to intervene in a feedback loop that has confounded the study of algorithmic polarization-the complex interplay between supply of recommendations and user demand for content-to examine downstream effects on policy attitudes. We use over 130,000 experimentally manipulated recommendations and 31,000 platform interactions to estimate how recommendation algorithms alter users' media consumption decisions and, indirectly, their political attitudes. Our results cast doubt on widely circulating theories of algorithmic polarization by showing that even heavy-handed (although short-term) perturbations of real-world recommendations have limited causal effects on policy attitudes. Given our inability to detect consistent evidence for algorithmic effects, we argue the burden of proof for claims about algorithm-induced polarization has shifted. Our methodology, which captures and modifies the output of real-world recommendation algorithms, offers a path forward for future investigations of black-box artificial intelligence systems. Our findings reveal practical limits to effect sizes that are feasibly detectable in academic experiments.

  PublisherDOI

• Administrative Records Mask Racially Biased Policing—CORRIGENDUM

  American Political Science Review · 2025-09-29

  article1st authorCorresponding
  PublisherDOI

• Political Diversity in U.S. Police Agencies *

  SSRN Electronic Journal · 2025-01-01 · 4 citations

  preprintOpen access
  PublisherDOI

• Political diversity in U.S. police agencies

  American Journal of Political Science · 2025-02-14 · 11 citations

  articleOpen access

  Partisans are divided on policing policy, which may affect officer behavior. We merge rosters from 99 of the 100 largest local U.S. agencies-over one third of local law enforcement agents nationwide-with voter files to study police partisanship. Police skew more Republican than their jurisdictions, with notable exceptions. Using fine-grained data in Chicago and Houston, we compare behavior of Democratic and Republican officers facing common circumstances. We find minimal partisan differences after correcting for multiple comparisons. But consistent with prior work, we find Black and Hispanic officers make fewer stops and arrests in Chicago, and Black officers use force less often in both cities. Comparing same-race partisans, we find White Democrats make more violent crime arrests than White Republicans in Chicago. Our results suggest that despite Republicans' preference for more punitive law enforcement policy and their overrepresentation in policing, partisan divisions often do not translate into detectable differences in on-the-ground enforcement.

  PublisherOA PDFDOI

• How to Use Causal Inference to Study Use of Force

  CHANCE · 2024-10-01

  articleCorresponding
  PublisherDOI

• Assessing the Reliability of Probabilistic US Presidential Election Forecasts May Take Decades

  2024-08-26 · 4 citations

  preprintOpen access

  Probabilistic election forecasts dominate public debate, drive obsessive media discussion, and influence campaign strategy. But in recent presidential elections, apparent predictive failures and growing evidence of harm have led to increasing criticism of forecasts and horse-race campaign coverage. Regardless of their underlying ability to predict the future, we show that society simply lacks sufficient data to evaluate forecasts empirically. Presidential elections are rare events, meaning there is little evidence to support claims of forecasting prowess. Moreover, we show that the seemingly large number of state-level results provide little additional leverage for assessment, because determining winners requires the weighted aggregation of individual state winners and because of substantial within-year correlation. We demonstrate that scientists and voters are decades to millennia away from assessing whether probabilistic forecasting provides reliable insights into election outcomes. Forecasters' claims of superior performance and scientific rigor should be tempered to match the limited available empirical evidence.

  PublisherOA PDFDOI

• An Automated Approach to Causal Inference in Discrete Settings

  Figshare · 2023-01-01

  datasetOpen access

  Applied research conditions often make it impossible to point-identify causal estimands without untenable assumptions. <i>Partial identification</i>—bounds on the range of possible solutions—is a principled alternative, but the difficulty of deriving bounds in idiosyncratic settings has restricted its application. We present a general, automated numerical approach to causal inference in discrete settings. We show causal questions with discrete data reduce to polynomial programming problems, then present an algorithm to automatically bound causal effects using efficient dual relaxation and spatial branch-and-bound techniques. The user declares an estimand, states assumptions, and provides data—however incomplete or mismeasured. The algorithm then searches over admissible data-generating processes and outputs the most precise possible range consistent with available information—that is, <i>sharp</i> bounds—including a point-identified solution if one exists. Because this search can be computationally intensive, our procedure reports and continually refines non-sharp ranges guaranteed to contain the truth at all times, even when the algorithm is not run to completion. Moreover, it offers an <i>ε</i>-sharpness guarantee, characterizing the worst-case looseness of the incomplete bounds. These techniques are implemented in our Python package, autobounds. Analytically validated simulations show the method accommodates classic obstacles—including confounding, selection, measurement error, noncompliance, and nonresponse. Supplementary materials for this article are available online.

  PublisherDOI

• An Automated Approach to Causal Inference in Discrete Settings

  CrimRxiv · 2023-09-22 · 9 citations

  preprintOpen access
  PublisherOA PDFDOI

Frequent coauthors

• Jonathan Mummolo

  12 shared

• Samuel A. Mehr

  10 shared

• Christopher Lucas

  Prince of Wales Hospital

  8 shared

• Jan Simson

  Munich Center for Machine Learning

  7 shared

• Daniel Ketter

  Missouri State University

  7 shared

• Manvir Singh

  University of California, Davis

  7 shared

• Matthew A. Baum

  6 shared

• Adam J. Berinsky

  Massachusetts Institute of Technology

  5 shared

Awards & honors

• Gosnell Prize for excellence in political methodology
• John T. Williams dissertation prize
• Best poster award by the Society for Political Methodology