Speed Limit Reduction and Traffic Crashes: Empirical Evidence from São Paulo Ciro Biderman UCL FSP/USP Event November, 2017 FSP, São Paulo, Brazil
Motivation 1.25 million people die annually in traffic crashes. If nothing is done this number may go up to 1.9 millions in 2020. One person dies every 30 seconds from a car accident. This is the main cause of death for young people (15-29) and the second cause for child (5-14).
Motivation In 2011 São Paulo City signed the UN goal to reduce traffic fatalities by half (from 12/100K to 6/100K). Speed limit reduction is considered the single most effective way to reduce crashes at a negligible cost. However there are few robust evidence showing that this is indeed the case.
Traditional Approach: Physics Speed Limit for arterial roads suggested by OMS: 50km per hour or lower Velocidade do impacto [km/h] Fonte: OECD/ECMT Transport Research Centre, Speed Management report, 2006
Evidence in the literature Archer et al, 2008; Khan et al, 2001; TRB, 1998; McCarthy, 2001: correlations between reduction of speed and safety. Ashenfelter and Greenstone (2004): increasing speed limits from 55 mph to 65 mph increase average speeds by 3,5% and fatality rates by 35% (difference in difference). Van Benthem (2011): increasing the speed limit in 10mph causes 9-15% more accidents and 34-60% more fatalities (quasi experiment).
Descriptive Evidence: Before-After
Descriptive Evidence: Comparing Groups Hospital Admissions Due to Road Crashes (per 100,000) São Paulo State
Econometric Strategy Before and after may confound many other factors happening through time. Comparing with other cities may also be misleading since São Paulo City was doing many changes in transportation policy. Comparing roads that reduced speed versus roads that did not would be comparing apples with oranges. Crashes in roads that reduce the speed limit drop by 30% while it was reduced by 10% in other roads.
Econometric Strategy Since the speed limit reduction happens in time we can compare the segment of roads that reduced the speed limit earlier with the segment of roads that reduced the speed limit later. This is a variation on the difference in difference approach. Controlling for fixed effects by road section we can see if the reduction in speed limit changed the trend in traffic crashes.
Econometric Strategy
M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 km reduced Econometric Strategy Speed limit reductions kilometres of roads affected by month 280 240 200 160 120 80 40 0 2014 2015
Econometric Strategy Linear: crashes i,t = 0 + δ i + 1.intervention i,t + 'controls i,t + ϵ i,t Poison: E(crashes i,t X) = exp( 0 + δ i + 1.intervention i,t + 'controls i,t + ϵ i,t )
Results
Robustness Checks and Extensions
Placebo Test
Duration Analysis
Policy Consequences The reduction in speed limits generated at least (very lower bound) 10% reduction in crashes. 20% is still a lower bound The correlation between the treatment and the time trend is.95 The cost of the intervention is very low if you consider just painting the signs but the main cost is on enforcement. For any reasonable value of life the return is huge; just saving in hospitalization compensate the direct investment. Why all cities in the world are not reducing speed limits in urban areas?
politics
The Political Economy of Transportation There are social gains in transport policy that are clear cut but difficult to implement from the political perspective. Since crashes are a rare event people do not understand their social advantage. Tickets seem to be too much a punishment. It is similar to substituting an alternating plate system to a congestion tool.
The Political Economy of Transportation The good news: once it is implemented the political cost is gone. São Paulo is the exception that confirms the rule: the speed limit was changed in just one road after a hard political campaign against the speed limit reduction. The bad political equilibrium is not stable. Autonomous cars may induce a 100% enforcement at no cost.