Skip to content
EN

Back to the catalog

andrewpwheeler.com
rss2English

Andrew Wheeler

andrewpwheeler.com · English

Crime Analysis and Crime Mapping

rss2 audio content media slash wfw atom dc sy

Open the feed

https://andrewpwheeler.com/feed

Last post
Sep 12, 2026
Posts in 24 h · 7 days · 30 days
0 · 0 · 3
Our last check
Answering
Served from
United States
Text score at discovery
5,644
Format
rss2
Features in the feed
content, media, slash, wfw, atom, dc, sy

Posts

What our queue read from this feed. Open one to read it here, or go to the site that published it.

  1. Faking synthetic control estimates
    Sep 12, 2026 · original
    AI disclosure – this post was created via Claude Code using Claude Opus 5 on xhigh reasoning effort. I gave it the idea, the data, the estimator and my prior blog posts as style examples, and it wrote the code and the draft. I will always disclose when I use AI to heavily write any content on this blog. (I use it for minor copy editing all the time.) A few years back I wrote some notes on the Hogan/Kaplan back and forth . Short version, Tom Hogan used synthetic control to say Larry Krasner’s de-prosecution caused extra homicides in Philadelphia (Hogan, 2022), and Jacob Kaplan, JJ Naddeo and Tom Scott said the result was an artifact of a very short pre-period and a few other specification choices (Kaplan et al., 2026). I thought KNS had the better of the substantive argument. But the thing that has stuck in my head for three years is not the substance, it is that Hogan never released his
  2. Join CrimConsortium, ASEBP submissions open, VerusCite privacy, NACA Course
    Sep 2, 2026 · original
    I have minor updates for several activities. First, the American Society of Evidence-Based Policing is accepting submissions for its conference in spring 2027 at UPenn: If you are a vendor interested in promoting some of your work at the conference, also feel free to get in touch. I serve on the committee that works with vendors and the committee that reviews proposals. Second, I have updated several of the pages for Crim Consortium (roughly the nonprofit behind CrimRXiv). For individuals, you should consider becoming a member to support the work Scott (and to a lesser extent myself) is doing on this. (We do not make money to be clear, just keeping the lights on.) If you are a program head for a department (or involved in some other organization), you should consider becoming an institutional member. You get your own prioritized page (which is worth it just for the SEO for your group). C
  3. Crime De-Coder and VerusCite Updates
    Aug 27, 2026 · original
    If you do not follow me on social media ( X , LinkedIn ), you likely have missed the different work I have been up to. So this is a remix of some of that, posted here for folks to follow along. Crime De-Coder Blog On the Crime De-Coder Blog : Notes on Crime Trends in the US Architecture Notes on VerusCite Filtering Secrets from Coding Agents with a Hook I often debate whether to post things here or on the Crime De-Coder blog. So if you like my posts here, you should probably also follow Crime De-Coder. As you can see, it is a mix of recent technical and crime-analysis-related posts. For the crime trends work, also check out my web app , where you can filter and compare cities relative to national trends (based on the Real-Time Crime Index data). Note that I will be going to the Council on CJ working group on crime trends, on the causes of the homicide decline, in a few weeks, so let me k
  4. License Plate Reader Searches Should Require a Warrant
    Aug 12, 2026 · original
    So while I work with police departments regularly, I think it is critically important that technology be used reasonably. While this may be off-putting to some of my clients, I worked with the Institute for Justice as an expert witness in their trial Schmidt v City of Norfolk . (Any opinions herein are my own and not those of IJ, to be clear.) The gist of that case was whether searches of historically cached ALPR data (automated-license-plate-reader) constituted an illegal search. 1 The judge ruled against plaintiffs in that case. Here is a quote from the judgment: Consistent with Plaintiffs’ claims in this case and controlling precedent involving mass surveillance in public spaces, ALPR surveillance could become too intrusive and run afoul of [constitutional privacy standards] at some point. But when? While a definitive answer to that question is elusive, what is readily apparent to thi
  5. VerusCite: checking academic articles for hallucinations
    Aug 5, 2026 · original
    I have a new app out, VerusCite . With the recent rise in popularity of GenAI tools like ChatGPT and Claude, this has also come along with academics writing slop articles. One of the ways to check that slop is via looking at the articles citations. LLMs have some predictable failure modes in writing papers whole cloth – they tend to get details like complicated author lists wrong, or swap out incorrect journal titles. VerusCite is a tool for editors and reviewers to use to verify citations in a fast and cheap application. It costs $2 to review a paper (and you get two free reviews on sign-up ). If you want to see the output of a single example though, check out https://veruscite-data.com/share/E1-rn3TBwksENO_3hlmsRD0IG-EnXd6vfQg-upQZWQM In addition to hallucinations, I have made many parts of the application just useful to editors in general. Many papers have minor errors in their biblio
  6. Notes on document processing with LLMs
    Jul 23, 2026 · original
    One of the recent papers that came across my X feed was a researcher digitizing a corpus of old Sears catalogs to estimate inflation more accurately: The technical skills to do this work are closely related to many of the projects I am working on at Gainwell – large-scale document processing using LLMs. So if you are a PhD student and your thesis involves work like this, I would likely want to hire you for a six-figure job as a data scientist at Gainwell. I wanted to have a quick blog post on some of the tools to process documents. So this researcher processed 180 Sears catalogs that are typically well over 1000 pages, so they likely dropped well over $10k on this project. Like I said in the X post, for this type of volume processing, you shouldn’t be using the main models. You should first consider the cheaper models, like the flash-lite models from Gemini (not even the flash for this),
  7. Podcast with Jeff Asher, AI blog posts
    Jul 17, 2026 · original
    Just a quick update on a few things. I was interviewed on Jeff Asher’s Jeffalytics Podcast, mostly about AI and the current landscape in criminal justice: Using Technology to Improve Crime Analysis with Andy Wheeler by Jeff Asher I don’t always link to my recent Crime De-Coder posts (follow the Crime De-Coder RSS feed or follow me on Crime De-Coder LinkedIn to stay up to date). But for a few recent technical ones: Filtering Secrets from Coding Agents with a Hook (written by Marc Olson, employee #1 at Crime De-Coder) Using Agentic Coding Tools for Crime Analysis Using local LLM models vs APIs AI is not going away, as so folks looking to learn about the tech, you should buy my book, Large Language Models for Mortals: A Practical Guide for Analysts with Python . If you are an analyst and have questions always feel free to send me an email. Happy to expand those blog posts to material that i
  8. AI writing is better than no writing
    Jun 23, 2026 · original
    AI disclosure – this post was entirely written by myself. I know AI writing is still pretty cringey – so I get that people are quite opposed to it. For people like me though (academics promoting their work, more technical oriented) I would like to proffer a slight defense of (even cringey) AI writing. Having an LLM tool help you write a blog post is better than not writing at all. I have come to the personal opinion I just want you to disclose when you use AI. I am starting to get peer review requests for academic papers that are clearly LLM written, and they are not obviously worse than the typical (mostly horrid) way academics write papers (they may actually be better to be honest). Blog and social media posts I think are strictly worse to my personal tastes when using LLM writing (across many dimensions, for now anyway ). But it is better to write something than nothing if you have so
  9. Notes on Valuing the Cost of Crime
    Jun 22, 2026 · original
    AI disclosure – I used AI to write this blog post. I figure having an AI blog post is better than not writing it at all. I will always disclose though if I use AI to heavily write any content on this blog. (I use it for minor copy editing all the time.) For the tech details, I used gemini flash 3.5 with medium reasoning in the Antigravity IDE, using the same advice I said in this blog post. (Minor preference to Claude Code for writing blog posts for those who care.) It is the outline of the thread I did on X (which I wrote entirely by hand). Using this approach, e.g. I give a detailed outline and prior examples, Pangram says this is only lightly AI assisted . Notes on Valuing the Cost of Crime We often hear eye-popping figures about the “cost of crime.” For example, that a single aggravated assault costs society $100,000, or that a statistical life is worth $10 million. But if you look u
  10. How long to conduct your experiment: Talk at ASEBP
    May 12, 2026 · original
    Upcoming at the American Society of Evidence Based Policing Conference, I have a talk Thursday morning (9:45-10:00), How long to conduct your experiment . The talk goes over some of the simple metrics I have created to help plan how long to conduct your intervention. Such as how long to evaluate your hot spots intervention, or purchase to increase arrest rates, etc. I have prepared a ton of different resources. The main one is a web-based application (a WASM-based app with R as the backend) where you can enter your inputs and generate a graph showing how precise your parameter estimates are: The help page includes citations and additional materials, but here is a brief rundown: I have the math details in this github repo , see the methodology.pdf . It also includes notes on how I used different LLM tools to produce the webpage and the method materials. Each of the applications allows you

Discovered by the rss-feed-index crawler, which checks each feed at most once a month.

Same record as JSON: https://api.agentalog.com/api/feeds/fd_andrewpwheeler_com_48ecb435fd302119. More from this site: andrewpwheeler.com in the Feeds tab.