No one ever said “time your research for public salience” — but Scott Mourtgos () and I are happy to share a working paper that provides the first large-scale causal evaluation of Flock () ALPRs.
Almost a year ago, we started to work with Flock to get access to detailed implementation data: when agencies put up their first Flock camera, when the next camera came online, and the next, etc. We combine those deployment records with crime data across 216 U.S. law enforcement agencies to study what happens after fixed ALPR networks come online.
I suspect our results won’t make all of the loudest voices in this debate happy at every turn. But I think they make sense, and I’m eager to hear what others think after they read the study.
A few of the main findings:
Motor vehicle theft (MVT) falls about 11% after Flock deployment. MVT also declines relative to other property crime within the same agencies, which gives us more confidence that we’re seeing something specific to vehicle theft. The pattern is consistent with deterrence.
MVT arrest clearance rises 15.9% but that result needs an important caveat: clearance starts rising shortly before the cameras officially come online, so we’re less confident about how much of that increase can be causally attributed to the cameras themselves.
And among stolen vehicles that are ultimately recorded as recovered, the median recovery lag is about 7 hours shorter. But when we look at all stolen vehicles—including those never recorded as recovered—we don’t find evidence of a meaningful overall improvement in recovery time.
This technology obviously stirs a lot of emotions, and there’s no reason to think our initial results will—or even should—settle the privacy debate. Evidence that a technology “works” does not by itself tell us whether its benefits justify its privacy, governance, and data-sharing costs.
That said, it’s good to have some empirical ground to anchor those debates to.
Credit as well to Flock for making the data available. It is never easy for private companies to take a chance on the scientific method, but I think they showed themselves to be truly interested in the process and the answers.
Interested? Read the working paper here:
Post
Conversation
Post your reply
I'm glad you did this study. I wouldn't have expected the technology to have much effect on clearance rates because the images would rarely be sufficient to ID the occupants. I would expect FLOCK cameras in parking lots to aid recovery more than on streets
We may not need IDs of occupants to help at the margins of clearance, but we don’t test that directly here either.
Very timely!
Are you able to estimate some kind of net effect on incarceration or on total arrests (given the reduced baseline from deterrence)? I know you'd have to make some assumptions.
Good question but not reachable directly here. On the table going forward, but would take a different data linking approach.
Thanks for putting together this (much-needed) analysis! Had a couple questions:
1. Is there any potential for negative spillover effects here (i.e. ALPRs simply move motor vehicle theft into neighboring non-surveilled communities), and if so, is there a good way to quantify
So you didn't even look at how miss use and miss identification is hurting people?