Blog Ipsa Loquitur

Published on under Questionable Pilot Programs

Ars Technica’s Cyrus Farivar filed a Public Records Request with the city of Oakland, California for the Oakland Police Department’s license plate database. The OPD apparently has 4.6 million geotagged photos of license plates of cars in and around Oakland. There are plenty of police departments around the country doing this sort of thing.

I haven’t read an awful lot about these kinds of programs, and I’m not of any particularly strong opinion about the constitutionality (or lack thereof) of them. But Farivar does some excellent reporting on Oakland’s data:

Specialized [License Plate Reader] cameras mounted in fixed locations or on police cars typically scan passing license plates using optical character recognition technology, checking each plate against a “hot list” of stolen or wanted vehicles. […] Some cities have even mounted such cameras at their city borders, monitoring who comes in and out, including the wealthy city of Piedmont, California, which is totally surrounded by Oakland.

LPR collection began in Oakland back in 2006, and an early OPD analysis showed that the overwhelming majority of the data collected was not a “hit.” In April 2008, the OPD reported to the city council that after using just four LPR units for 16 months, it had read 793,273 plates and had 2,012 hits—a “hit rate” of 0.2 percent. In other words, nearly all of the data collected by an LPR system concerns people not currently under suspicion.

Despite this, in that same report, then-OPD Deputy Chief Dave Kozicki (who has since retired) dubbed the LPR setup an “overwhelming success.” Today, OPD’s LPR hit rate has fallen slightly, to just 0.16 percent.”

So wait. For every stolen car they found, the police were scanning nearly 400 cars? Like I said, I’m not sure this is necessarily unconstitutional, but it does seem wildly inefficient. It doesn’t seem even remotely successful, let alone “overwhelmingly” so. Though honestly, out of all the things cops could be spending their (our) money on, I’m not exactly going to complain about this kind of silliness.

Ars isn’t just spitballing about the accuracy here. In January, the Electronic Frontier Foundation looked at just eight days of Oakland’s license plate data, and concluded that Oakland’s license plate camera program was a pointless dragnet that ignored the actual locations of crimes. The police scans weren’t entirely random, however; when the EFF overlaid census data on the license plate data, they found that the dragnet was focused on areas with predominantly black and hispanic populations. Maybe that helps explain the poor hit rate.

That’s pretty lousy, but to give some small amount of credit where credit is due; the OPD’s scans didn’t appear to correspond to the locations of mosques, which is more than you can say about New York.