Predicting when a crime is about to take place on CCTV

Are experienced CCTV operators better than naive participants at judging from an unfolding scene on CCTV whether or not a crime is about to be committed? The short answer is no, they aren't.

Presented with 24 real-life 15-second CCTV clips, and asked to predict which half ended just before a crime was about to be committed (examples included violence and vandalism) and which half were innocuous, 12 experienced CCTV operators managed just 55.5 per cent accuracy – no better than if they'd just been guessing. Twelve naive controls achieved an accuracy of just 46.5 per cent – no worse, in terms of statistical significance, than the CCTV operators.

Another purpose of the research was to find out if certain viewing tactics lead to more accurate predictions of criminality. To do this, Dawn Grant and David Williams of the University of Hertfordshire recorded the eye movements of the CCTV operators and control participants as they watched the brief clips. Also, for a subset of the clips, they asked the participants to talk through their thought processes regarding what was taking place.

The key to successful predictions seemed to be to pay attention to the social context. Specifically, when participants spent more time focused on the face or head of single individuals not engaged in any social interaction, or looking at the bodies of those in a social interaction, they tended to more accurately predict whether a crime was about to occur. Grant and Williams think this might be because the former allowed the participants to notice when a lone person in the scene was staring at other people, which could betray their plans to commit a criminal act. Meanwhile, viewing the bodies, rather than faces, of those in a social interaction, might have allowed the participants to notice aggressive body language and the spatial proximity of people in a group.

This speculation was backed up by the participants' spoken accounts of how they were appraising the scenes. For example, participants who made accurate comments about which people in a scene belonged to which social group tended to also make accurate predictions about when a crime was about to occur. Accurate predictions also tended to be preceded by comments about body language and the social proximity of people in the CCTV footage.

'For certain types of crime, it may be that understanding the social context and the relations between those in the CCTV image is the first step towards obtaining reliable indicators of criminal intent,' the researchers concluded. Of course, whether we actually want CCTV operators, accurate or otherwise, watching our movements and forecasting crimes is another question altogether.

Further reading

Grant, D., & Williams, D. (2010). The importance of perceiving social contexts when predicting crime and antisocial behaviour in CCTV images. Legal and Criminological Psychology DOI: 10.1348/135532510X512665