Finding 16 000 unsafe curves
There are many reasons why traffic accidents occur. And a lot of them has to do with the driver. Drunk driving, speeding or drivers that aren’t attentive enough to name a few.
We wanted to know if it was possible that the actual construction of roads could be the cause traffic accidents?
This project started with a report — a report written by an expert that claimed that a horrible traffic accident could have been caused by how the road was constructed — or at least that the road was part of the problem.
SVT’s investigative weekly show Uppdrag Granskning had been working on a story about a horrible accident in the northern parts of Sweden. Six young men died when their minivan drove into a curve and skid over to the other side of the road. The minivan collided with an iron ore truck. Only one person survived the accident.
Uppdrag granskning had talked to the expert about his findings — he claimed that it’s possible to calculate whether a car driving in a curve will stay on the road or not. The formula he used is also used by Trafikverket, the Swedish Transport Administration, when constructing new roads making sure that those roads makes it easier for drivers to stay on the road. (VGU 2015:090 p 41)
The formula calculates the friction needed to keep the car safely on the road using the speed limit on the road, the radius of the curve and the cross slope of the road.
As a start Uppdrag Granskning wanted to see if there were other deadly accidents that followed the same pattern, finding problems with how the road was constructed at the location of the accident.
They turned to us, a new team at SVT — a team where journalists and system developers work together with advanced data analysis for stories.
Once we started looking at the story — and the findings from the expert, we also got the idea to check the roads, instead of the accidents.
Getting the data
Every state road in Sweden is measured by a measuring car that drives around and collects all kind of data about the road: the radius of the curves, how uneven the surface is, how much (or how little) the road leans to give a few examples. The data is collected meter by meter, but it’s stored with an average of 20 meters. This data is stored in a database that is publicly available through an API. (PMSV3)
There are about 100 000 km of state roads in Sweden — and we realized that we needed to make a selection of roads to look at. We wanted to focus on rural roads, so we decided not to include roads without physical lane separation. We also didn’t want to look at roads in a town or village. Finally, we needed to look at roads where there had been one or more accidents.
Our selection became roads in Sweden where there had been at least five accidents in the last five years. Once we had those roads, we decided only to look at segments where the speed limit was 70 km/h or more — and where cars would meet, i.e. there was no separation of the road in the middle.
We started to download data for the 950 roads that were our selected roads. In all we looked at 26 000 km of roads and we did this in 100-meter segments. For each segment we first determined whether it was a curve or not using another formula found among guidelines for constructing roads. (VGU 2015:086, s 106) Once we concluded that the segment was a curve, we calculated the formula for friction for each segment.
Finding problems
Looking at the analysis we found that we had a problem with roundabouts. A roundabout is by definition a curve, but the road will not have a proper cross slope since there are roads connected to the roundabout. We realized that we need to take out all roundabouts to get better data. The data has no information about roundabouts or any other crossings, so we needed to use OpenStreetMap to locate the roundabouts. Unfortunately, the data from the measuring car is listed as meters from the start point of the road, not with geographical coordinates. In order to match the two datasets, we needed to convert meters from starting point to lat and long coordinates.
We then discovered the same kind of problem with crossroads. The measuring car couldn’t handle passing by a crossroad and the data was often wrong. Using the same method, we took out the segments close to a crossroad as well. (An additional problem here was roads passing over another.)
What we found
When examining all the segments we found that 13 percent of the segments defined as curves had problems with frictions. But 17 percent of accidents happen in these problematic curves. We also calculated the same thing for curves as a whole, for roads — and we divided the segments in different speed limits. We got the same kind of overrepresentation any way we did our calculations.
We spent a lot of time calculating whether we could find a statistical significance for our findings, but we didn’t. There are just too few accidents and too many roads to do that. And there are a lot of other variables that affect accidents, variables that we don’t have access to. We still find it interesting that we find a larger percentage of accidents in curves that are problematic than curves that are correct — and this we find no matter how we do our calculations.
In order to translate this so that reporters could use our analysis for their reports we redid our definition of the roads — now dividing them into curves and straights. Doing this we could say that we had found 16 000 curves that were not properly constructed.
Adapting our findings for reporters
Working with reporters has its challenges — and in order to make it as simple as possible we decided to create PDF-files with all the roads, one pdf for each local editorial office. An excel-file would have been very hard for them to interpret. Even though, we still had to spend a lot of time with reporters helping them find a good example road and explaining what language they could use. We wanted to avoid talking about “dangerous” roads — or “deadly” roads — instead we settled for calling the project “Un-safe roads”.
When confronting the agency responsible for road safety in Sweden they claimed that they had to focus on the roads where most people travel. The roads we had looked on was not a priority. They also said that just looking at the curves with friction problems was not something that was efficient, accidents have many reasons and they said that our findings was not significant.
However, the Minister for Infrastructure, Tomas Eneroth was interviewed on the nine o’clock news saying that this was something that needed to be corrected.
We also found that the stories coming from the rural areas of Sweden told numerous tales about roads that were problematic and were locals witnessed a lot of accidents happening, especially in the winter.
This project demonstrates how stories are made stronger through the joint contribution of highly skilled developers and journalists. This story could simply not have been told without this combination of skill sets.
Rickard Andersson
Helena Bengtsson
Fredrik Stålnacke
