Beyond its shock value, I found police use "fuck" to serve various functions, from expressing stress to asserting control. Profanity, while often cast as unprofessional, can be a nuanced tool in law enforcement. Table 1 gives 50 whole fucks for your viewing pleasure.

Another demonstration of how the internal controls of academia fail and thereby make external control inevitable. Just remember that every correction is an over correction, and externally imposed corrections will invariably feel worse than those brought about thoughtfully from

We published that result a year before it was cool.

Just this week, saw multiple examples of grad students use LLMs to generate reference lists by uploading PDFs and saying “give me an APA 7 ref list.” Great example of how the *belief* that AI will bring us efficiency runs into the *reality* of an inefficient process. Grad

The Salt Lake valley is a wonder.

If you’re not putting up preprints of your work, you’re making a decision to forego the potential for many more readers. If you want readers, if you want impact, if you want better science…why leave it all behind big publisher paywalls?

Tough break for my Philly friends, but at least you still have the Eagles right?

Very worthwhile efforts here to support open criminology. @CrimConsortium now has individual memberships for those without the institutional support yet. Fee is hard to beat, but for those able I strongly suggest giving the very small amount to be a paid supporter.

Appearance matters for how the public views police, and AI-generated images open new possibilities for reproducible, transparent research on legitimacy and trust. 🔗

New preprint up @CrimRxiv. We test whether AI-generated police officer profiles can provide a valid, scalable way to study public perceptions of officers. Led by @UUtah's Chandler Robinson, and alongside Matt Logan and Pete Blair, we varied officer race, gender, and body type

Yes. “We propose fundamental shifts: prioritizing precise description of concrete phenomena, subjecting theories to severe tests of risky predictions about explanatory mechanisms, generating more high-resolution data, and developing formal computational models of theories.”

Paper: “Through Thick and Thin: Comparing Traditional Qualitative Analysis and NLP Using Narrative Data from Police Officers,” Justice Quarterly. DOI:

Bottom line: If you need rapid, high-level thematic extraction at scale, and you want to relate themes to covariates, STM works. If you need thicker description, latent meaning, and legally/policy-relevant categories, you still need humans.

What STM adds that hand coding can’t (easily): document-level topic proportions (“theta”) that can be modeled like outcomes. Demo: White officers emphasized “Officers are human” more; Black officers emphasized “Missing information” more; the third contrast was small/not sig.

Cost/benefit: the qualitative pass took ~12 hours of expert time; STM took ~1–2 hours (experienced users; short texts). Trade-off is clear: STM is fast and reproducible; hand coding is slower but yields interpretive richness and additional categories.

Where they diverge: the qualitative team also surfaced an additional top-level theme—ignorance of rules (policy, training, Graham v. Connor). They also recovered more contextual nuance (e.g., how media clips shape perceptions). STM missed some of that depth.

Result: convergence on three high-level themes— • Lack of experience: the public hasn’t lived these encounters • Missing information: judgments made without full context • Officers are human: split-second decisions under stress

Both, blinded to each other’s work, analyze 715 officers’ open-ended responses to: “What is the #1 thing the public doesn’t understand about police use of force?” One team does thematic coding; the other fits an STM. No cross-talk until the end.

New in Justice Quarterly: two great qualitative researchers (Logan Somers, Natalie Todak) team up with two over-indulged quants (@smourtgos and me). Same text corpus; different tools. How does a machine learning approach (STM) stack up against traditional qualitative analysis?

A problem in policing researchers too. The initially least popular, but long term very successful, requirement in my policing pro-seminar is requiring students to spend time with two different officers in two different roles in two different agencies. Might bump it to three next