Estimating the effect of Europe’s largest street party on crime in London

Matt Ashby
matthew.ashby@ucl.ac.uk

UCL Department of Security and Crime Science

“The level of knife-related violence, aggression towards emergency workers, and on-the-street debauchery … is a fundamental betrayal of Notting Hill Carnival’s heritage and origins” Rakib Ehsan in CapX

“A two-day festival at which a million folk are having a wild time should not be judged by the villainy of a few” – Brendan O’Neill in The Spectator

“I dread this event every year, it gets more violent each time. How many more officers have to get injured or innocent members of the public killed?” – The Police Federation

“It’s a UK success story. I hope a few more of the naysayers experience it, before hastily jumping on the bandwagon of criticism!” – Leroy Logan in The Guardian

Working out how much crime is attributable to Carnival

We can’t do a randomised controlled trial …

… so we try to get as close as possible

… using a difference-in-difference model with two-way fixed effects

Step 1: create a grid of small cells covering London

Step 1: create a grid of small cells covering London

Step 2: count crimes of each type in each cell on each day

… June–October 2013–2023

… 15,462 grid cells × 1,323 days = 20,456,000 cell–day observations per crime type

Step 3: estimate the baseline daily frequency of crime in each grid cell

to account for population, socio-economic factors, land use, bars, shops, etc.

Step 4: estimate the London-wide frequency of crime on each day

to account for long-term trends, seasonal effects, public holidays, weather, etc.

After we take all that into account

… we’re left with only variation specific to each grid cell on a particular day

… and if Carnival had an effect on crime

  • the difference in crime levels between Carnival and non-Carnival days in grid cells closer to Carnival
  • would be bigger than the difference between Carnival and non-Carnival days in grid cells further away from Carnival