This week, Walker wrote an excellent post comparing crash statistics and the structural responses to incidents by mode in the wake of the Amtrak derailment.

It’s a great read with great questions, and we are reminded once again how safe we’ve made rail and air travel relative to other modes. But one thing sticks out to me – how we frame safety through statistics. Like any data, statistics can be misleading in how they’re presented. Sometimes it’s intentional (like denigrating light rail transit because it’s unsafe for pedestrians). Other times, misleading data might be more benign, simply out of convenience of available datasets. So, how should we talk about transportation safety in the fairest, most honest way possible?

Incidents Per Passenger Mile or Per Passenger Trip?

Ultimately, what really matters is the big picture: how many people die or are injured in transportation-related incidents per inhabitant. It tells us how our land-use/transportation combination is doing at keeping us safe. In that regard, we know we’re doing very poorly (even if fatality rates vary greatly by state, road classification, etc):

It’s more common for DOTs to report safety as per mile (or km) driven. This is fine when comparing between countries, and it gives an idea of how good or bad our drivers and roads are by normalizing total distance traveled.

However, when comparing different modes to one another, it’s common to see another layer added: per passenger mile. The logic is that some modes carry more passengers per vehicles than others, and so it’s a fair comparison to normalize for the bus carrying 40 people that hits a bicyclist vs a car carrying 1 driver who does the same. Those 40 people are each going somewhere, traveling some miles, so the risk rate (to both the occupant and non-occupants, ie pedestrians and cyclists) should be equally compared.

But a per-passenger mile metric fails the sniff test despite its broad-usage. Look at the general formula and see if you can tell why:

Cheat sheet answer: increasing the denominator automatically makes the safety rating better without reducing the number of fatalities. Imagine you’re only comparing non-occupant safety (pedestrians, cyclists, etc) between modes as the linked O’Toole post above does.

Three Ways of Becoming Data

I’ll give a visual example of why passenger-miles of the vehicle striking the non-occupant favor motor vehicles over transit. Let’s say I’m a pedestrian walking along Lake Street in the LynLake area of Minneapolis. A person making a 1 mile trip in their car hits me halfway through its journey and I die:

Now, let’s rewind this (very sad) story, and a different person making a 2 mile trip by car hits me 1.5 miles into their journey:

Once again, let’s think bigger, and a person drives into LynLake from Richfield, hitting me 5.5 miles into their 6 mile desired journey:

Here’s the thing. As a non-occupant, it doesn’t matter how far that car traveled or intended to travel. All that matters is that I’m dead. Another statistic. All that matters is that driver walked out their front door with the intention of making the trip from A to B. They made a modal choice. They could have biked, or walked, taken transit, or driven. Yes, some modes would have taken more time, but we’re not talking about accessibility vs mobility right now, we’re talking about safety.

Big Picture Data

Now, multiply this by the millions of trips taken every day. Some are short, some are long, some are on buses, some on light rail, many in cars. Those modes have different numbers of passengers in each vehicle. But what matters is that each vehicle has a number of people making individual trips.

We should evaluate fatality or injury rates based on passenger-trips instead of passenger miles. Otherwise we could just convince everyone to drive 100 miles (compared to the average 9.7 mile car trip length in 2009) for every trip, hold pedestrian fatalities steady, and look like we’re all of a sudden ten times safer than we currently are. It’s misleading in favor of automobiles. Put another way, it doesn’t matter if my grocery store is 1 block by foot or 4 miles by car, but only how many times out of 100 I die or am injured making that trip.

Using numbers taken from a mixed-bag of sources*, here is what that looks like:

Thoughts:

• Occupant death rates for transit of all modes is clearly much lower than driving, whether taken as per-mile or per-trip.
• Car drivers kill far fewer non-occupants per mile driven or trip taken. No doubt about it.
• LRT is especially dangerous for pedestrians, no matter how you slice it.
• The total safety gaps between cars and transit widen (or, narrow for LRT) when passenger-trips are used as the metric.
• Random stat: 2/3 of heavy rail and 1/3 of LRT non-occupant deaths are suicides – a lot of people intentionally walk or jump in front of trains (there’s no hard data to support it in those reports, but my gut feel is suicides make up a very small share of non-occupant car crashes).
• Intercity rail (ex. Amtrak) and air travel are excluded here, but fatality rates so low they don’t warrant a comparison.

A similar trend is observed when comparing injuries by mode:

Cars look roughly the same as every other mode per passenger mile, but are clearly more dangerous than transit. The ratio of occupant to non-occupant injuries for transit is much closer to passenger cars than deaths. The most likely factor is that trains and buses are heavy and difficult to stop quickly and therefore kill at higher rates than a car-on-pedestrian or bike crash.

Takeaway

Our land-use and transportation priorities advocate for mobility over accessibility, and as a society we seem to be willing to let drivers and vehicle occupants take their lives into their own hands to achieve longer distances traveled. This is clearly a losing strategy from a safety perspective – vehicle occupant death and injury rates per trip are higher in cars than any transit mode, by a fairly wide margin too.

However, pedestrians are more likely to die per passenger trip or mile by a light rail vehicle than the vast number of cars on the road. Urbanists need to be aware of this – we shouldn’t accept the high number of pedestrian fatalities from car crashes, nor should we for transit. Transit deserves more grade separation in heavily-trafficked pedestrian areas and better station design to help reduce total fatalities.

But from a metrics standpoint, we should start moving toward a per-trip method rather than per-passenger mile. This informs our land-use decisions from a safety standpoint much better than per-passenger mile.

*Sources:

2011 Transit Passenger Miles, Fatalities, & Injuries by Mode taken from NTD data
2011 Occupant Fatalities/Injuries, Non-Occupant Fatalities/Injuries by Vehicle Involved, Miles Traveled by Vehicle Type
Vehicle Occupancy & Trips Taken for Passenger-Trips and Passenger Mile Calculations

This post is adapted from my personal blog, the Fremont Avenue Experience