Crunching numbers: A closer look at Portland’s road fatality rates

1996-2015 Portland traffic fatalities
Portland traffic fatalities, 1996-2015

The past few weeks have been especially bad in terms of road fatalities in Portland. Within nine days between July 30th and August 8th we had four fatalities, which prompted me to run the numbers- so by the time you’re reading this, they have gone up.

For the year to date as of August 9, we’ve had 28 fatalities. I took PBOT’s fatality data and crunched some numbers:

Portland traffic fatalities, 1999-2015 with trendline

In the past 20 years our fatalities have ranged between 20 and 59 per year; the 59 in 1996 is an outlier (2.5 standard deviations away, in fact), and the following two years are high. From here on I’ll begin analysis at 1999, as the previous three years skew the trend terribly. So, since 1999 there has only been two years at or above 40: 2002 at 40 and 2003 at 47; the average fatalities are about 33 per year.

As of 8/8 we are 221 days into the year. That means we’re averaging nearly one fatality a week. To put it another way, we’re 60% of the way through 2016 and have 76% of last year’s fatalities; we also have 100% of the fatalities of 2014. If the current trend holds, based on the days remaining we’ll end up with about 46 fatalities, the second-deadliest since 1999.


Instead of assuming the rate is constant through the year and instead scale it by the rate of fatalities we had last year, we’re on track for 52 fatalities (last year, by 8/8, we had 20 of 37 fatalities).

While the trendline (linear regression) points down, it’s notable that we are above average in both 2015 and most likely again in 2016.

Histogram of fatalities, 1999-2015

Statistical footnote: I’m assuming a normal distribution of the data and generally using basic linear approximations. If you disagree, copy the spreadsheet and show your calculations. I can’t really make it easier to accommodate reanalysis than that. You’ll see that σ at 9 (and 6.3 post-1998) is fairly reasonable for the datapoints given. The R2 value of 0.06 for the post-1998 linear regression is surprisingly strong indeterminate.

Crunching these numbers is a grim task and it’s not our intent to stoke fear in anyone. But if we truly want to tackle Vision Zero we have to look in the mirror and be accountable for what we see.

Edit: on August 19, we had our 30th fatality. The lower bound of my estimate goes from 46.4 to 47.5.

– Ted Timmons, @tedder42

Exit mobile version