Have you wondered if your city has worse drivers than other parts of the country? Do you think drivers in your city use their phone too often while driving? Is speeding more common where you drive than in other places? We answer these questions and many more in this article by analyzing detailed driving data collected over a four-month period in 2016 from drivers across the US.
The DriveWell program from Cambridge Mobile Telematics measures driving quality using data gathered from sensors on a smartphone, applying machine learning and signal processing algorithms on sensor data to score driving quality. Working with our partners, we have developed a scoring algorithm that captures a driver’s level of safety. The basic components of the DriveWell Score include harsh acceleration, hard braking, harsh cornering, excessive speeding, and phone distraction.
- The first three components measure abrupt changes in car motion, indicating aggressiveness, poor anticipation, or crash avoidance.
- Unsafe speeding is flagged when the driving speed is considerably higher (by a dynamic threshold) than the speed of the ambient traffic or the posted speed limit.
- Phone distraction measures how often the driver is using their phone while the vehicle is in motion.[ref]DriveWell’s phone distraction score penalizes only significant handheld distractions, such as picking up a phone and talking, using any app by moving the phone and tapping on it, and does not penalize mounted use, hands-free use, and so on. The goal is to identify and dissuade significant distractions that likely take the driver’s eye off the road, rather than penalize things like hands-free calling, which are problems because of their cognitive impact, but still not as harmful as the behaviors that take both the mind and the eye off the road.[/ref]
The DriveWell program is being used by a large (and growing) number of users around the country to improve their driving using applications developed by Cambridge Mobile Telematics and our partners. We analyzed a subset of data collected from the app — the dataset includes driving data from over 15,000 drivers who drove a total of about 18 million miles (29 million km) between early April and mid-July, 2016.
As of 2016, there are 297 cities in the US with population of over 100,000 — we have driving data from every single one of these cities. The dataset contains a sufficient number of miles, drives, and distinct drivers in every city to obtain statistically meaningful results. We included only drives that started or ended within 30 miles (50 kilometers) of the city center in each city’s assessment.
And now for the results!
- The safest driving city? Billings, Montana, with an out-of-the-world score of 98.3 (out of 100)! Users in Billings seem to hardly use their phone while driving, and are a role model for the rest of us. These users have outperformed the other cities in every measured category (except speeding, where they tied with users in Sioux Falls, South Dakota — only 1.3 minutes of significant speeding per 100 miles).
- The city with the riskiest drivers? Jersey City, New Jersey, with an average score of only 58.9 across its drivers, “beating” Pembroke Pines, Florida! They speed a lot compared to almost every other city (nearly 25 minutes of speeding per 100 miles). On the positive side, they don’t drive using their phone as much as some other cities.
- The riskiest city for phone distractions? Odessa, Texas, where users on average were distracted by their phone for over 15 minutes per 100 miles of driving. Be extra careful if you drive there! Fontana, California, was close behind (13.7 minutes per 100 miles). Users in Billings, Provo (Utah), and Anchorage (Alaska) had the best phone distraction scores (no more than 1 minute of significant phone distraction per 100 miles).
- Which cities have the worst hard braking and harsh acceleration behaviors? Prior work in the insurance industry has shown that these factors are correlated with higher crash probabilities. Among our users in this analysis, cities in Southern California have the riskiest behavior on these metrics — Long Beach, Pasadena, Riverside, and so on. Amongst our users, the worst 13 cities on hard braking and harsh acceleration are all in (Southern) California (well, Sunnyvale in the Bay Area snuck in at 11th worst on braking and 13th worst on acceleration, but the rest are in SoCal). The best hard braking behavior after Billings are in Las Cruces (New Mexico), Sioux Falls, Brownsville (Texas), Kansas City, and Des Moines.
- The safest driving state? Take a bow, Montana! And also Alaska and Wyoming. (These are all states with low traffic and low population densities, but we must commend these drivers for being careful about using their phone and for not speeding excessively.) If these states were in a class at a university, they would score an A+ (the average score exceeds 90).
- The riskiest driving state? Rhode Island, New Jersey, and Connecticut all have substantial room for improvement — the average for our drivers in these states is under 70.
- How does the “average” American driver perform? In the table of cities, the nationwide average is tied for #153 with Evansville, Indiana (and just ahead of Tulsa, Oklahoma ). It’s tied with #33 South Dakota, a little worse than Virginia, in the table of states. The nationwide average DriveWell score is 78, with about 9.3 minutes of speeding and 3.7 minutes of significant phone distraction per 100 miles of driving.
- Is there a correlation between safety (the DriveWell score) and population? No, not really, at least for the overall city population. Spearman’s rank correlation between the CMT ranking and population is a mere -0.04 (r(295) = -0.04, p = 0.473). Pearson’s correlation between the DriveWell score and population was only -0.14 (r(295) = -0.14, p = 0.0194), which is small as well.
- Well, what about population density? Yes, there’s a strong correlation. Using population-density-by-state data to measure the correlation between our by-state scores and the population density, we found:
– Rank correlation between the DriveWell score and population density: -0.78 (r(48) = -0.78, p = 1.84e-11)
– Pearson’s correlation between DriveWell score and population density: -0.71 (r(48) = -0.71, p = 7.13e-9). The negative correlations indicate that the safety is strongly negatively correlated with population density even on as coarse a granularity as an entire state. This finding — that driving risk is correlated with population density — is generally known to insurance actuaries.
- Is this the first such study? Yes, to our knowledge this is the first study that uses smartphone telematics data. That said, it is important to note that recently Allstate published a “2016 Best Drivers” report covering 200 cities with a population over 100,000 using data from their insurance claims (that report also shows the number of “braking events” per 1000 miles for each city). We found that the Spearman rank correlation between Allstate’s ranking, obtained by ranking the collision rate, and the DriveWell rankings, is 0.44 (r(198) = 0.44, p = 7.65e-11) — a moderate correlation. That is encouraging and suggests that our score has some predictive power. We wouldn’t expect a stronger correlation because the set of users is very different, and they likely have different dispositions toward using a safe-driving app. Moreover, users in the DriveWell program do improve with time, and that effect has been captured in the observed performance of our users.Which brings us to…
- What does this analysis say about driving improvements with time? Not much. This analysis comprises nearly four months of driving data. As noted in our first blog article published a few days ago, driving behavior for most users in the DriveWell program improves significantly with time. That is an important and valuable effect, which we will analyze in an upcoming article. So please stay tuned for that!
Check out the two tables below to learn where your city (first table) and state (second table) place in terms of driving safety. Note that five states appear only in the state ranking because they have no city with 100,000 people.
The first table below shows for each city the average DriveWell Score (out of 100) as well as subscores (out of 100) for harsh acceleration, hard braking, and harsh cornering. It also shows the number of minutes spent speeding above a safety threshold (which is higher than the speed limit) and the number of minutes of phone distraction, both over 100 miles of driving. Click on each column to sort the data by that column. The second table below shows the same data by state. Below that is a scatterplot curve of each state’s score — which you can use to assign your own letter grades to each state!
Aggregate drivewell score by state (“Drivewell score” column from the state table above)