A six-panel comic and an interactive page about one of the first lessons in research methods: if your data only contains the people something happened to, you can describe them, but you can't say how much anything raised their risk. Potato salad and sharks stand in for the serious version.
Smith et al. replicated our analysis of false confession wrongful conviction risk, then extended it to ask whether that risk rises with interrogation duration, reporting roughly 10% at six hours and 25% at twelve. Changing only which false confession study supplies one of the two duration distributions moves the twelve-hour estimate from 26.3% to 72.6%. A comparison drawn from both groups of a single study never reaches 10% at any duration we evaluated.
Across 99 experimental conditions, 41% of innocent participants signed a false confession. More strikingly, 37% did so in control conditions where the experimenter typically did little more than accuse them. Our new paper asks what behavior these experiments are actually measuring, and how far their results can be carried into real interrogations, courts, and policy.
Nine states banned deceptive interrogation tactics for juveniles between 2022 and 2024, raising concerns among some that the tradeoff would be fewer cases solved. Across 510,582 juvenile-involved incidents in NIBRS (2021-2024), arrest clearances did not decline after adoption --- estimates are null to modestly positive across every method --- and prosecution declinations, the one measure that fell, fell for adults in the same states too.
For two decades, interrogation research has graded techniques with a single ratio: true confessions divided by false confessions. That ratio cannot tell the difference between a method that is better at separating guilty from innocent suspects and one that simply makes confessing harder for everyone. An interactive dashboard demonstrates the problem.
How accurate are Americans' perceptions of police use of force?
Where do Americans most support calling off police pursuits---and what factors drive those differences?
A Bayesian analysis of seven years of daily data finds that patrol staffing levels are the most consistently significant factor affecting police response times across all priority levels.
Using structural topic modeling on a corpus of 29,035 news headlines spanning 1990-2021, this analysis identifies eight latent media narratives about police use of force and examines their association with public opinion.
A structural topic model analysis of public comments from a policing commission listening session reveals four distinct themes---including one of appreciation for officers that the local news coverage missed.
A Bayesian structural time series analysis finds a 388% increase in police resignations causally attributable to the post-George Floyd period, with probabilistic forecasts indicating continued increases.
A case study showing how pracademics used data analysis and research synthesis to drive a police department's internal COVID-19 infection rate below city, county, and state levels.
Using data from the Comparative Agendas Project, this analysis applies Punctuated Equilibrium Theory to explain the episodic, non-incremental nature of federal crime control legislation and criminal justice assistance funding from 1947 to 2017.
Text mining analysis of US senators' tweets in the two months following George Floyd's death reveals distinct differences in policing rhetoric between Republican and Democrat senators.