Waze Data Can Help Predict Car Crashes and Cut Response Time | Hacker News Reader

Waze Data Can Help Predict Car Crashes and Cut Response Time

wired.com

71 pointsnsoonhui20 comments

“Here’s the thing about car crashes: They are blessedly pretty rare. In the US, nine people are injured in motor vehicle crashes for every 100 million miles traveled in cars, according to data from the National Highway Traffic Safety Administration.”

There are 6 million accidents every year in the US alone. 90 people die every day.

I suppose it’s just a matter of which statistic you choose. However, in my view accidents are not rare.

The best statistic will account for exposure. Also look at fatalities per 100 million miles traveled over the past few decades: incredible reductions!

“The number of deaths per passenger-mile on commercial airlines in the United States between 2000 and 2010 was about 0.2 deaths per 10 billion passenger-miles.[1][2] For driving, the rate was 150 per 10 billion vehicle-miles for 2000 : 750 times higher per mile than for flying in a commercial airplane”

https://en.wikipedia.org/wiki/Transportation_safety_in_the_U...

Commercial airlines are generally used for long distance trips so they'll naturally rack up more miles between deaths. You'd see the same thing if you compared freight haulers like Swift, Schneider and Walmart to companies that do a lot of local delivery.

Commercial airlines are also flown by a professional crew instead of the public making the utility of the comparison limited at best.

Commercial Aviation is the equivalent of US buses which have a dramatically better safety record. School buses for example are about 70x as safe as driving your kids to school.

The vastly more deadly general aviation compares more closely with the automotive fatality rates. GA has 16 fatal accidents per million hours, that’s horrific.

And the first number is primarily because of 911. It would be closer to 0.0 if not.

>best statistic

It depends on what you're trying to do. On the other hand, there are a lot of injuries and fatalities in an absolute sense (or relative to most other types of accidents) so it's certainly worthwhile to do things that reduce those numbers.

On the other hand, the accident rate is really quite low already. So any technology or policies have to deal with an already pretty low base rate--which, as you say, has already been reduced significantly over time. There isn't much in the way of low-hanging fruit.

It's slightly unfair, but I read this as Waze deciding to route a bunch of cars down a narrow residential street to divert around a congested highway then going "whoa, that's gonna be chaos! Watch the sparks fly!"

More seriously, I hope they're using this accident data - particularly involving pedestrians - to avoid making routing decisions that increase their risk.

Cross-referenced with time of day and weather conditions (rainy and foggy in an area where it's typically neither) I would think Google (Waze's parent company) could estimate that the chances for accident-based delays are elevated and recommend a sooner start to a commute. Maybe cross with Facebook's social data (users who were at a bar last night and for how long) combined with known start/end points for commutes and historical times that people make those commutes, the ability to guestimate the chances for an accident could be increased even more. Yes, it's very big-brother but if it saves me 20 minutes to 2 hours waiting in traffic...

Your insurer would definitely be very interested in such data aggregation.

You could just use location data (either from Google, or from signal triangulation) to see if their owners were out or at home last night. Add data from the accelerometer and you'd see how drunk they were.

Yeah, I'm glad there are laws against that kind of data collection. Or are the laws just there to keep me fooled?

Interestingly, Facebook probably can already tell who's in a relationship with whom and how far they are in it: how often do they talk to each other on the messaging apps, how often they check each other's profiles (FB, IG), were both their phones connected to the same WiFi network last night after their third date?

I'd love to know how many folks are under active guidance when they are party in a crash. All of the mapping services make ridiculous unsafe decisions. They are not liability free.

Even if they had that data, how do they apply it usefully?

Let's assume that, even when the odds of a crash are elevated, they're still relatively low. Perhaps the chances of a wreck along your route jump from 1% to 10%.

If a wreck happens, there's also a lot of variability to how much delay it could add. If it is minor, it could be cleared or moved to the shoulder in 20 minutes. If there is a fatality or a big truck flips over, it could close the whole road for hours.

You also don't know how critical it is to the user to arrive on time. Maybe tomorrow the first thing on the agenda is to do an important presentation to management, but the next day the first thing on the agenda is to do some code reviews and catch up on email. So do you need 99% confidence that you'll arrive on time or would 80% confidence be good enough? The answer to that question will have a big effect on how early you have to tell them to leave. (I guess you could just explicitly ask the user, though, which might be nice.)

With Waze directing streams of phone-watching drivers onto roads they are unfamiliar with and which are unsuited for the traffic, I can also predict there will be crashes.

It's at least possible that putting drivers on to roads they're less familiar with could decrease the amounts of fatal accidents.

For example - http://roadsafetygb.org.uk/news/end-of-the-road-for-white-li...

Even if that is valid in its limited domain - and I doubt that it is, given that no consideration seems to have been given to the dangers of increasing variance in driver behavior - there's no reason to believe it applies where roads gain a lot more traffic than they are suited for (and especially when most of the increase are technically lost, in that they, personally, do not know where they are or what they will encounter next.)

Other than having the new and shiny veneer of technology making political and management buy in easier to obtain (which is valuable) how is this any more effective asking the EMTs and tow truck drivers where the problem areas are?

Waze can't even get the traffic conditions right ON TIME. Every time I've ever used the app it tells me either way too late (as in that accident up ahead doesn't even exist and hasn't for two hours) or not at all. I'm curious as to how they honestly think Waze can be helpful when it's this slow.

Reply on news.ycombinator.com