Hacker News Reader

Comment by nhashem | Hacker News Reader

Am I missing something here?

The article basically uses crime as a proxy for "social population density." I'm pretty sure you could also use "restaurant reservations" as a similar proxy, but then I guess you wouldn't get to use words like 'prostitution' in the title of your blog post.

Then, Uber cab riders go to these areas that are densely social. I'm guessing they probably take other modes of transportation too.

Then, something about certain crimes being more prevalent on certain days of the week, with some pretty huge leaps of faiths made in the reasoning and no actual testable data to back it up.

I don't like to post too often on HN if I'm just going to stand there and drink some haterade, but this just seemed like such a sad attempt to put together something for pageviews that I couldn't help myself.

View on HN

I think the post was a decent attempt to do something similar to OKcupid's insightful blogs.

The difference is that okcupid had data in their database that is deep, and the team has a bunch of MIT math guys, if I recall correctly.

I'm not sure about Uber, but I'm guessing to write this post they had to go out looking for this information, rather than infer it from their primary data. Still a decent post, but realize the objective here was to market Uber, and the rest makes sense.

> Am I missing something here?

I believe so: the importance of open data.

This finding is a perfect example of the fascinating insights you can get when you combine big, seemingly disparate datasets.

I'm not sure where you would go to get an equivalent dataset for "restaurant reservations".

In the sense that it uses readily-available crime data to make inferences about possibly correlated phenomena, this analysis is another iteration on Adrian Holovaty's original mashup:

http://www.holovaty.com/writing/chicagocrime.org-tribute/

Is Uber's transportation data available too ? I doubt that.

Some of it is: http://www.infochimps.com/datasets/uber-anonymized-gps-logs

25000 trips, anyway.

Thanks !

The whole article comes across very "nerdy and scientific" but indeed isn't at all. Just the difference between correlation and dependence are already something they should look at: http://en.wikipedia.org/wiki/Correlation_and_dependence

That's because one of the side-effects of big data is the new definition "data scientist" = Ph.D. <graduate|dropout> in any field, no matter how unrelated to statistics or data mining.

To be fair, they were pretty clear that their data was a correlation, not a causation. But the stuff about prostitution on second Wednesdays is pretty crazily unsubstantiated, and controversial.

I think his goal was to see if one could use an activity proxy to predict where most rides would occur. Crime data is a fairly natural proxy because it's readily available in an easy-to-use format.

It doesn't seem all that useful — it's fluffy — but it's amusing nonetheless and I don't regret reading it.

I'm curious what leaps of faith do you see in his reasoning? Crimes "caught" versus actual criminal activity seems like the obvious one, but I don't think there's really anything in the piece that makes that distinction necessary. (Police activity might actually be a better proxy for activity than criminal activity.)

Reported crimes probably correlate more strongly to police presence, so the root question may be, Are police officers more heavily patrolling areas where people also need rides, and how do they know?

That's exactly what I was thinking about. Here in San Francisco, the police were recently cracking down in North Beach (mostly around the strip clubs on Broadway). It seemed to be a pretty clear causal relationship in that case — more people meant more police meant more tickets and arrests.

The plural of anecdote is data, right?

Reply on news.ycombinator.com