This blog is a product of my passion for data visualization. The data shown here are sourced from other websites, but all statistical operations on these data and the resulting graphics are original.
I take requests and am available for freelance work. If you have a suggestion for a graphic or need support on a project, email me.
To learn more about me, visit my LinkedIn profile, and send an invitation to connect.
Thanks for visiting!
There are many ways to evaluate whether someone is a good driver. One could review traffic violations, DUI convictions, collisions, etc. Of course, someone may be a bad driver and simply get lucky and end up with a clean record. However, by reviewing the records of many people, we can get a sense of how safe drivers may be at the city or state level. This is what Allstate has done – using claims data from a two-year period, it calculated the average time between collisions for drivers in 200 of the largest cities in the US.
I’ve mapped these data – the average time between collisions – by city, and then derived population-weighted averages for each state where data were available: Σ(time · population)/Σ(population). Geographically, the region that immediately stands out is the northeast. Insufficient data were collected in MA, so nothing was published on that state, and most of the other New England states did not have cities on the list. Regardless, the five states with the shortest time between collisions were RI (5.4), MD (5.4), PA (6.2), CT (6.7), NJ (6.8), and NY (7.4), all of which are northeast states. The greenish states – those with longer average times between collisions – show a slight clustering in the middle of the country, but there are few data points (cities) here, so it is difficult to say if the trend is significant.
One assumption, based on the map, might be that drivers in larger cities might have more frequent collisions. Graphing the data, however, shows that this is only partly true. The trend holds for very large cities, with populations greater than about 1M, but is not seen for the vast majority of the data. If the relationship were true, population and years between collisions should be inversely related (i.e., show a strong negative correlation). At best, there is a statistically significant, weak, negative correlation. (Although there were several ties, I ran a Spearman rank correlation test, which provided a rho of -0.16 and p-value of 0.02. This result was similar to the Pearson coefficient of -0.20 and p-value of 0.004.)
Although the absence of this negative correlation should be clear from the lack of structure in the scatter plot, I have further illustrated it by showing the medians in the right-hand graph. The medians are calculated on all cities in population bins of 100k. For example, the first bar shows the median value (average years between collisions) for all cities with populations between 100k and 200k. The last bar includes all cities with populations greater than 1M. As you can see, the bars show little decrease in height for populations below 800k – this cutoff includes 92% of the cities.
Note that Allstate reported that the national average is 10 years, though this is not the average of the published data. The unweighted average of the published cities is 9.3, and the population-weighted average is 8.7. This likely suggests that smaller cities (those not on the list of 200) do tend to average longer times between collisions.