The body-cam number that reversed

In 2012 a small California police department ran the first randomized trial of body-worn cameras. Over 988 shifts in Rialto, dice decided which officers filmed their shifts and which did not; use-of-force incidents split 17 on the no-camera shifts against 8 on the camera shifts, and citizen complaints dropped from 24 the year before to 3. The 60% figure that fell out of that contrast became the single most-cited number in the case for putting cameras on cops.

The trouble is what its own authors found next. Barak Ariel and his Cambridge colleagues took the design to ten forces at once — 2.2 million officer-hours — and the civilizing effect vanished: no average change in use of force, and, more awkwardly, more assaults on the officers wearing the cameras. Force rose in some sites and fell in others, which they called puzzling and disturbing. The mechanism they had proposed for Rialto (an officer, knowing the tape is rolling, de-escalates) turned into its own failure mode when officers could choose when to hit record.

What survived replication was the quieter half: complaints against police fall sharply almost everywhere cameras go. The headline half — that filming makes officers gentler — rests on a within-study count of 17 versus 8, a difference the study's own catalogers note fell short of statistical significance between the randomized arms. A single small trial became national policy; the larger evidence base that followed says the strongest claim it launched is the one least supported.