I am involved in question of why our company accident rate is rising.
We have internal data, miles driven, familiarity with route, number of tickets handled. We even have demographics data, population density and so on.
What I am sure is available, but I haven’t found is traffic density in areas.
It seems to me that the risk of a crash is not just the miles you drive, the state of mind you are in, your training and experience but also the number of crash opportunities.
Just looking at gross population trends for Atlanta, there has been a 25 percent increase in population in 10 years. I am unable to translate that into a usable metric of “crash opportunities” I don’t know how it is correlated if at all.
If anyone knows how to grab this data, or if there is an insurance company generated metric, I would be interested.
It seems to me that the risk of a crash is not just the miles you drive, the state of mind you are in, your training and experience but also the number of crash opportunities.
You have calculated against “experience”? Seems to me that would be the number one factor. I am seeing it elsewhere with unemployment so low.
I am involved in question of why our company accident rate is rising.
Total accidents or “at fault” (driver error) accidents? Have older vehicles been replaced by the company and those are getting into accidents? Too many variables could change outside anyone’s control.
* get a data source providing #licenseddrivers per ZIP (best)
* or get a data source of #oversixteen per ZIP (better)
* or just plain population per ZIP (meh)
* or get state data reflecting number of vehicles registered per ZIP
* use those as a proxy of how much driving activity is in that ZIP
* extract the "appointment data" that presumably has fields reflecting
* techidentifier
* servicelocation street / city / state / zip
* accidentinvolved (yes, no)
* accidenttype (traffic, stationary)
* finalfault (companydriver, otherparty)
By joining the appointment / incident data with the population by ZIP data, it should be possible to compare relative employee driver risks across these demographic deltas.
Metro techs are likely to have higher accident rates on a per-appointment basis and per-mile basis since they are around more things to hit or be hit by, both when driving or parked on-job. However, for two techs working in similarly dense or sparse areas, any statistically significant delta in the accident rate is likely due to driver-specific behavior that can be targeted for training / discipline.
You have calculated against “experience”? Seems to me that would be the number one factor. I am seeing it elsewhere with unemployment so low.
While experience is a factor, it is probably not a factor in the way you imagine. In the driver population we are looking at the outlier young ages are about 35 with most ages centered around 45 to 50. The years of experience in this job is typically 10 years.
However, we get a lot of voluntary transfers so a technician may only have a couple of years in a given area.
Worse, the job tends to gravitate work tickets the new areas of construction. So, even if a driver has 10 years of experience, that driver will spend a great deal of time in new and constantly changing situations.
As such miles driven is a lousy proxy. What is needed is some sort of metric that includes time driven, (a great deal of time for technicians in metro areas is spent starting stopping, patrolling slowly down right of ways and stuck in traffic. Rural technicians handle less tickets but drive faster and further between tickets and have many less crash opportunities per day.
However, even this is confounding as many areas that were rural 10 years ago are now metropolitan.
However, right now the low hanging fruit seems to be poor training regimen and would be the easiest answer. On the other hand, a large and intense training program for a few thousand technicians can run into millions of dollar per year. Of course one technician severely injured per year can be just as expensive.
By joining the appointment / incident data with the population by ZIP data, it should be possible to compare relative employee driver risks across these demographic deltas.
Metro techs are likely to have higher accident rates on a per-appointment basis and per-mile basis since they are around more things to hit or be hit by, both when driving or parked on-job. However, for two techs working in similarly dense or sparse areas, any statistically significant delta in the accident rate is likely due to driver-specific behavior that can be targeted for training / discipline.
WTH,
This is about what the company is doing, however it is probably not dynamic, in other words the data has been updated in a decade. (The company has been focused on losing 50 billion dollars. They were successful in that endeavor)
This really is sad as the company has, or had, a complete big data department. I remembered that after I posted. I will attempt to contact them next week.
I like your traffic density thought. If you have time of day/day of week, I’m guessing you would find a correlation with rush hour. If your data is limited to a local geographic area, it’s probably close enough to assume every day has a very similar rush hour (maybe starting a bit early on Friday). If your data is nationwide, you could maybe find something approximate for highway driving based on direction of travel (towards nearest city for morning rush hour, away from nearest city for evening rush hour), plus distance from nearest city, size of city.
The U.S. Census Bureau has free data that you could use to convert a street address to a latitude/longitude. The U.S. Geological Society has a free data set called the Place Names Database that gives you latitude/longitude of every city in the USA. IIRC, the Census Bureau has population numbers for cities. That should give you enough data to investigate the rush hour hypothesis.
You might find a GIS (“Geographic Information Systems”) consultant and see if they if they can help you and the cost. ArcInfo is the 500 pound gorilla in that industry, but they charge accordingly. You can probably find someone much cheaper.
I’m not sure if they are company vehicles or employee vehicles, but you might consider putting telematic devices in the cars. I’m not sure if it would be a popular idea. I think Progressive Insurance may have coverage that is priced based on telematics. That may be a way to both improve safety and incorporate telematics without being an evil boss.
We have internal data, miles driven, familiarity with route, number of tickets handled. We even have demographics data, population density and so on. What I am sure is available, but I haven’t found is traffic density in areas.
The first thing I would look at is the drivers. Are they younger? Do you have your techs doing more service calls now, leading them to speed between appointments?
Or perhaps your company is just feeling the same thing that the rest of the country is feeling:
Probably somewhere in the traffic density data, but what intersections/areas have the most accidents and how often do your drivers go through those areas.
How often are your drivers taking a call (even over a bluetooth connection) or worse, texting (even over a bluetooth connection) and driving?