40% of NYC Taxi Trips Are Uniquely Identified by Census Tracts and Hour | Hacker News Reader

40% of NYC Taxi Trips Are Uniquely Identified by Census Tracts and Hour

gist.github.com

13 pointslil_tee11 comments

Very smart inquiry...I had been somewhat of skeptic initially that the potential danger to privacy outweighed the value of making the data as transparent as possible, but that's just a guess out of my preconceived notions of taxi use, which were already inadequate as it still blows my mind how many taxi trips there are on an average day.

I think an argument can still be made that even if the OP is right about the quantity that can be uniquely identified -- keeping the coordinate data still outweighs the real-life privacy risk, that is, the small number of people who want to hire a private investigator/specialist to analyze this data to catch a specific person would find it much faster to track the person the way that PI's normally do so. But the rebuttal can't simply be, "uniquely identifiable trips are probably so rare as to be inconsequential"

"it turns out that if you know the census tracts for pickups and drop offs, plus pickup times truncated to the nearest hour, then you can uniquely identify 40% of NYC taxi trips"

Hmmm... but if you already have those pieces of information (start tract, end tract, start hour) what would you want to get from the data? How much someone paid? How much they tipped? Whether they paid with cash or card?

Can anyone see an obvious nefarious use for this data?

So, you think your partner is cheating on you with their ex. You know they got a taxi from your apartment last Sunday at 6pm, you look up the data, and sure enough, the taxi went to the ex's address, and not home as your partner claimed.

Or, you arrive at work late, claiming that you stopped off at a client's before heading in to work, but your employer can now verify that you actually got the taxi from your home address.

I'm assuming in these examples that you just need to know pick-up OR drop-off - if you need both, then I agree with you that it's not much of a concern.

"if you need both, then I agree with you that it's not much of a concern"

You need both pick-up and drop-off to get a unique row.

Now, I think about your examples, though, perhaps the absence of a particular row could show you didn't do something.

I find this kind of analysis to be really awesome and I'd love to learn how to do even a more basic version of it. Does anyone have some resources they can point me to?

I'm actually working on a small project that has a much, much smaller dataset than the NYC Taxi data but some similar attributes (geographic coordinates mainly). I'd love to produce something like this with what I find (assuming I can find anything interesting).

I wonder how much one loses if the first and the last couple of miles are fuzzed over. Such data would still be quite useful.

> uniquely identified by birthday, gender, and ZIP code

This is not correct; you need to say "full birthday" which includes the year, otherwise the statement is nonsense.

I'd still bet that month/day, gender, and zip code are enough to uniquely identify a very large number of people.

I don't see how that would work unless you're talking about ZIP+4 and not just a regular ZIP code. It would work for very sparsely-populated areas and extremely small towns, but most residents will share a ZIP code with thousands of others.

I assumed he meant day, month, and year.

> Note the distinction between birthday and birthdate

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

Reply on news.ycombinator.com