Dear Diego,

thanks for the link. I did not remember that post. It is good to see that you get a nearly identical distribution of times of crashes as this data set gets. An interesting question would be: what are the differences, e.g. most accidents happen at crossings, but are the statistically significant differences between NYC and Atlanta in terms of where accidents happen.

It is obvious from your post, and also what we would expect, that larger crossings with more people passing have a higher accident rate. I thought about combining this with TravelDirections, similar to Bernat's post or Sander's. One would have to look at paths between any (or as many as possible) pairs of points and see which roads/crossings are taken most frequently. Of course, observed/measured traffic flow data might even be better.

Thanks,

Marco

PS: Are you aware of any other cities in the US that have similar data sets?

Dear Bernat,

thanks a lot - in fact the stunning part of the visualisations comes from your post and from Michael Trott's post. You have a really impressive way of creating spectacular graphics. I was merely trying to mimic that.

Yes, the dataset structure is nice and elegant. Your graphic clearly shows the difference between weekdays and weekends. There are also the hourly peaks which appear to be an artefact of the recording of accidents. People/police appear to be using full (or half) hours when they record a crash. I saw that effect, too, and tried to get the curve relatively smooth by choosing the bins accordingly.

Also you are quite right when you say that the data is very tragic. On that website there are lots of other data sets, some of which are really interesting. I wish there was more of that here in the UK.

I am also thinking in the direction that Conrad Wolfram was pointing at. What if more data is made publicly available? Will that further democracy and decision making? I'd love to see what can be learnt from combining different data sets. Is the total more than the sum of its parts? I guess that many of the individual entities that provided data will not be looking at all the other data sets. The Wolfram Language should be really convenient to achieve just that.

For example, I don't think that they have lots of roundabouts there, but it would be interesting to see how accidents/crashes relate to different types of road crossings for example. Or the Tree Census Data; one could also use netatmo data for example.

I think that the datasets on that website are quite amazing.

Thanks a lot for leaving a message!

Marco