Tuesday, March 4, 2014

The speed of a vehicle and pedestrian mortality

The over four thousand pedestrians killed ever year from motor vehicle collisions have motivated the following graphic.


Naturally, we expect the likelihood of mortality to increase as the vehicle travels faster and physics suggests that it will be nonlinear.   But like anything else, we want to make decisions based on accurate information and understanding of risk.

These estimates originate from research using 1980s and earlier data more likely to report serious injuries.  From the abstract of a literature review published in 2011.
Without exceptions, papers written before 2000 were based on direct analyses of data that had a large bias towards severe and fatal injuries. The consequence was to overestimate the fatality risks. We also found more recent research based on less biased data or adjusted for bias. While still showing a steep increase of risk with impact speed, these later papers provided substantially lower risk estimates than had been previously reported.
These authors produced the following table in their 2009 paper containing the following table.

Rosen and Sander (2009)
Clearly, estimates that correct for the bias or use less biased data are much lower than those suggested by the popular graphic.  Mind you, it's still the case that the likelihood of mortality rapidly increases as vehicle speed increases and that serious injuries are important too.

From DOT HS 809 021 October 1999
Note: Given the 2011 literature review, the table is potentially biased given that the paper is written so close to the year-2000-threshold.  The mortality estimates roughly match later estimates, however, suggesting that the table is based on less biased data.   
Broadly speaking, as a person who advocates transportation networks with strong walking and bicycling options, relying on bad or biased estimates when better options are readily available to make our points is a terrible strategy.  Besides being ethically questionable, it (1) makes advocates look naive, (2) sets expectations too high, and (3) leads to poor decision-making.  For example, suppose we believe the graphic such that a pedestrian struck by a vehicle traveling at 40 mph is almost certainly going to die.  One might reasonably conclude that there is no point in traffic calming a 50 mph arterial since there is nothing to be gained until you get below 40 mph.  However,  that's not the case based on the more robust estimates.

EDIT:

We can see the bias in this graph by Rosen.  You can see how the risk curves are dramatically shifted left when using the biased data.

No comments:

Post a Comment