Crashmapper: Visualizing Years of Traffic Collisions in NYC | Hacker News Reader

Crashmapper: Visualizing Years of Traffic Collisions in NYC

blog.accursedware.com

61 pointstalos21 comments

In Seattle a few intersections have been singled out as the most dangerous for cyclists because of the number of collisions involving cyclists there. Considering they are also the highest-traffic intersections for bicycles, it doesn't actually say anything about how dangerous they are per cyclist-trip.

Crashmapper is useless without knowing how many vehicles pass each intersection per day.

Useless for figuring out how absolutely safe an intersection is by specific mode of transit at a specific point in time, sure.

There are other uses though, you just have to make sure to take into account the unknown denominator. That's why being able to compare layers and run through time is so important.

Right. For example, if your goal is to reduce total number of crashes, (assuming you are in a position to meaningfully impact this,) the ability to identify "hotspots" is extremely valuable.

However, if you want to use it to know where to focus efforts to save the most lives it is useful. It's the correct representation for some uses, just not for all uses.

Although even in that case it's important to know both pieces of information. Past some level of safety, intersections likely hit a floor where no amount of further investment will actually make them safer.

> Past some level of safety, intersections likely hit a floor where no amount of further investment will actually make them safer.

That point is probably called "driverless cars."

And "driverless cycles" probably also.

That data is easy to sample if necessary, whereas the crash data is not.

I learned that pretty much every corner has crash potential. :)

We ran into a similar problem at my last job. I mapped crash locations in our game to the map in which they took place. When viewed in aggregate it ended up turning into more of a map of where people tend to congregate in the game then where problem spots actually exist.

Yep, without the denominator of how many people were using a specific mode, you can't really determine some kind of safety ratio.

In some ways, the map functions as that denominator. That's still pretty interesting. Also, it's pretty clear looking at the correlation between injuries and auto occupants how much more dangerous it is to get in a car than on your feet. :)

We did something similar and actually ended up finding a confusing hotspot. Turned out to be a ledge where people could clip through a fence and fall to their deaths.

That's not to imply it wasn't helpful. Just that once we knocked out all of the low-hanging fruit the map turned into an indication of where people played. Remaining crashes weren't related to location or content.

The first pass actually had a nice grid pattern on the map that was where ambient audio emitters happened to overlap. Something about the transition between two ambients emitters was falling over.

This is a good example where some color would really help.

There are points with the same color where their collision count is different by a factor of 10.

Comment on style: The transparency of the orange heat spots should not fall below 30% or so, even at the center. Having the heat spots opaque makes the map difficult to read.

Huh. I worked at a place that had access to data similar to this, but statewide and in a different state. I wanted to put something like this together, but i was told the data was very sensitive and that I couldn't get access to it, since knowing that some intersections are really dangerous opens up government agencies to lawsuits for not fixing them.

So its interesting to see this here.

I like how on error he redirects to the application to serve custom urls giving a dual meaning to crash mapping.

I've done something similar for Oslo, Norway; http://trafikk.quai.io/ (warning: Norwegian text)

Great start!

It would be wonderful if you could enter a time span, rather than seeing one month at a time.

Maybe filter by daylight/nighttime hours?

It's worth pointing out this historical context mentioned in the OP's Github:

> Council Member Jessica Lappin got into an animated discussion with Petito over traffic crash data. When Lappin asked why NYPD is releasing data in PDF form — and only after the council adopted legislation forcing the department to do so — Petito replied that the department is "concerned with the integrity of the data itself." Petito said NYPD believes data released on a spreadsheet could be manipulated by people who want "to make a point of some sort." An incredulous Lappin assured Petito that the public only wants to analyze the data to improve safety, not use it for "evil."

This is a persistent problem in public records law: officials who are asked to provide public data in computer parsable formats but instead, deliver it as PDFs or worse, as paper printouts.

The reason given by the NYPD is so common a trope that it's hard to say that they're just technically slow. When you consider that the NYPD has taken strong criticism for not investigating fatal accidents seriously, it's reasonable to suspect the NYPD of actively obfuscating when they use the "oh but they'll alter the spreadsheets!"

Digitally-signed spreadsheets are a thing. If concern about altered versions was the real issue, there are mitigations that don't remove the utility of having a consumable data format. Its not merely "reasonable to suspect" that refusal to give usable formats is motivated by desire to make it more difficult to use the data rather than concern about alterations, it is almost unreasonable to believe anything else.

for a second i thought this was mapping ethernet packet collisions and i thought it was cool but couldn't figure out why we would need a heatmap

Reply on news.ycombinator.com