Prius fatal accidents

This weekend, I went through the National Highway Traffic Safety Administration (NHTSA) Fatal Accident Reporting System (FARS) data to count every fatal accident that involved a Prius and every single vehicle Prius and Pedestrian accident:


I did not count bodies, just the accidents. Accidents that had more than one vehicle and a pedestrian were treated as just a vehicle accident. Only single vehicle, Prius accidents with a pedestrian were counted as pedestrian accidents. What is interesting is the 10% ratio of all accidents to single car, pedestrian accidents matches the 11% of all fatalities are pedestrians reported by the NHTSA. Please don't read anything into this relative ratios, just sometimes numbers seem to align.

The next step is to take the annual Prius sales, a "fleet loss rate," and the 15,000 miles per year per vehicle to calculate the fatality rate per 100 million miles. By summing all years of Prius activity, we'll have a large enough sample set, already evident by the total number of fatal accidents, to meet even the most skeptical statistical requirement.

Bob Wilson

 

He'll only be pissed that his broadcasts weren't required to be the noise ... Hummmm, perhaps we should do some 'noise detection' tests:

• Prius in EV mode
• engine noise
• Rush noise

Bob Wilson

 

The NHTSA told us in 2008 that they did not have 'enough hybrid data' to analyze the relative, hybrid electric risk. But now we have the 2007 FARS accident data and annual Prius sales and by accumulating the number of Prius and fleet mileage, we have a statistically valid sample set of Prius miles:


I'm attaching the excel spreadsheet for critical review but the approach I used:

• 5% fleet loss per year - 'pulled out of thin air,' this means one of every 20 Prius go to the salvage yard each year, 10% every two years. I suspect the actual rate is lower but I've yet to find a credible, fleet loss rate.
• one half of current year vehicles plus all existing vehicles - we assume that at the end of the year, one half of the vehicles sold in that year would have been in full use plus all of the previously sold Prius minus the fleet loss. This is the active fleet used to calculate the mileage.
• 15,000 miles/vehicle/year - using the EPA estimated miles per year
• 2000 Prius miles were ignored - a partial year, the mileage contribution is insignificant. I stopped looking with 2001 FARS data because the format seemed to change from 2000. I believe the 2001-2007 data is sufficient for a statistical analysis.

The NHTSA publishes a fatality rate of "deaths per 100 million miles." By my conservative (aka., under reporting fleet miles,) we are looking at 40 times the 100 million miles. This meets anyone's definition of a minimum sample set with better than a 90% confidence level.

One caution, we can't used the fatal accident rate for the fatalities nor can we ignore the multi-vehicle, fatal pedestrian accidents. There is also a question of who gets tagged with a fatality if one car strikes the pedestrian and a second car hits the pedestrian.

I'm looking for a car with similar characteristics, say a Scion or Echo/Yaris, to do a differential analysis. This would give us a way to compare similar engine and hybrid vehicles to see if there is any, statistically different fatal accident rate.

Bob Wilson

 

It is my habit in excel to count by starting with "0" or "1" in the top row of a column and then in the subsequent rows either increment or pass on the value. This causes the 'flat part.' In reality, the non-incrementing values could be set to a blank value but then plotting the line would have a lot of 'data gaps.'

Bob Wilson

 

Yes but first I have to handle the accidents. There are fields in the accident data that let me count the fatalities just as there are fields that let me count and then identify which vehicles were involved in the accident.

Now many fatal pedestrian accidents involve more than one car. The first impact can throw the pedestrian into the path of another car and either impact can kill the pedestrian. But understand I can't go after 'blame'. I'm only interested in:

• all fatalities per 100 million miles traveled
• pedestrian fatalities per 100 million miles traveled

The reason is the NHTSA publishes a lot of safety analysis:

• "In 2007, the fatality rate per 100 million vehicle miles of travel (VMT) fell to a historic low of 1.37." - Traffic Safety Facts 2007 Data, DOT HS 810 993.
• "Vehicle occupants accounted for 74 percent and motorcycle riders accounted for 13 percent of traffic fatalities in 2007. The remaining 13 percent were pedestrians, pedalcyclists, and other nonoccupants." - Ibid.
• "In 2007, 70,000 pedestrians were injured and 4,654 were killed in traffic crashes in the United States, representing 3 percent of all the people injured in traffic crashes and 11 percent of all traffic fatalities." - Ibid.

So I've been trying to figure out how to 'divide up the dead.' Let us say three vehicles are involved in an accident with one fatality. I can't assign that one death to all three vehicles as this would multiply the fatality rate per mile traveled. But what I can do is divide the fatality by the number of vehicles involved and assign to each vehicle that 'fraction of a death.'

Using this approach and the total number of Prius miles, I can come up with a credible:

• deaths per 100 million Prius miles
• pedestrian deaths per 100 million Prius miles

Now if I can find similar 1.6 L, vehicles whose sales records start about the time the Prius began and parallel the number of sales, I can do a differential analysis. I can calculate the numbers for both vehicles to test these questions:

• How does the Prius fatality rate compare to similar engine sized vehicles from 2000-2007?
• How does the Prius pedestrian fatality rate compare?

Bob Wilson