The most dangerous cars on the road

When we don't grasp the fundamentals of data literacy, we can make mistakes which, in turn, can be dangerously misleading for our audience. Here's an example from the South African media that illustrates this.

In South Africa, traffic safety is a major public health issue. Road accidents are one of the top ten causes of death in the country, responsible for similar numbers of deaths as as coronary heart disease. The World Bank's data library records traffic fatalities as a ratio of the total population, using a measure of "deaths per 100 000 people" so that comparisons can be drawn between countries.

This is the context for road deaths in South Africa from the World Bank (https://bit.ly/road_deaths_SA)

It shows that while things have somewhat improved over the last 20 years, road fatalities are much higher than in countries in Latin America, for example.

One of the big challenges is that these numbers are not accurate. It's acknowledged that they are probably an underestimation of the total number of deaths due to poor data collection and lack of data sharing between the police and hospitals, for example. In early 2022, the Road Traffic Management Corporation released a report which analysed the data available.

The report included data about the types of vehicles involved in accidents, based on make and model, with the caveat that this data was available in only 67% of accident records. In other words, for one in three fatal accidents on South African roads, we don't even know what cars were involved.

How the report was received

The report received widespread coverage in South African media, with stories published on all major news outlets and prime time discussion on talk radio. Most of the coverage focussed on this table in the report.

The table is organised to show that, in the limited data sample available, Volkswagen Polos were involved in more fatal accidents (2 668) than any other vehicle. Most organisations published stories which interpretted this to say VW Polo drivers are more dangerous than drivers of other vehicles.

Samples of headlines covering the report launch

Why these reports were dangerously wrong

These reports weren't just wrong, they were dangerously wrong. The data table above makes it clear that there are more VW Polos on the road than any other vehicle by a considerable amount (over a quarter of a million more than Toyota Corollas), which automatically makes it more likely that an accident will involve a Polo than any other vehicle.

In fact, since only 16.7% of crashes recorded involved a VW Polo, you could argue that they are statistically safer drivers.

The best way to appraise this data is to use a relative measure, in this case "fatal accidents per 1 000 cars". To work this out in the table below, we have taken the earlier data and created a new column using the formula =sum((C2/B2)*1000) in cell D2, then rearranged the order based on this new calculation.

What we see is that not only are VW Polos not the most likely vehicle to be involved in an accident, but that given the incomplete data there's not a huge difference between many of the cars measured here. The report itself makes this point, yet somehow it was missed in most of the media coverage.

The two approaches tell very different stories. If we tell the story that VW drivers are the worst, it provides our audience with confirmation bias if they already think that - after all, they've probably also seen more bad behaviour from VW drivers, simply because there are so many of them.

Worse, however, also provides them with someone to blame for traffic accidents - when what the data shows is that there is no particular group of drivers based on the car they drive that can be considered better or worse than others, especially as there are other factors to consider too: high end luxury cars may have safety features that mean they are involved in accidents due to bad driving which do not result in as many fatalities. If anything, the report shows that bad driving is endemic to all social groups - which was definitely not the story South Africans were told.

Data literacy matters because without it, a serious issue can be turned into a jokey headline that dangerously misleads your audience.