One interesting phenomenon is that the % of riders paying cash consistently increases every December. Purely speculative guess, but maybe holiday tourists account for a larger % of all taxi riders in December, when many New Yorkers are on vacation, and tourists are more likely to pay cash?

This video is processing – it'll appear automatically when it's done.

One issue is that my weather dataset contains only daily totals. With minute-by-minute weather data, we could measure more precisely how taxi/Uber activity changes based on whether it is literally raining at a given moment

This video is processing – it'll appear automatically when it's done.

New York of course has many ways to get around that don’t involve cars: subways, buses, ferries, bike lanes, walking, and perhaps others. Just about all of these options are cheaper than taxis, so although I don’t have any data, I’d guess that taxi riders are on average more affluent than the median NYC resident. So really when we use taxi traffic as a proxy for neighborhood popularity, we’re implicitly measuring popularity among those who elect to ride in taxis. I think the official word for this is “gentrification”

This video is processing – it'll appear automatically when it's done.

Personal aside: I took a memorably terrible 92 minute trip from Williamsburg to LaGuardia on Friday, March 6, 2015. Here’s the Uber receipt:

Shockingly, I made the flight, but only because it was delayed by an hour and I sprinted to the gate. Sure enough, the taxi dataset shows that on the afternoon/evening of March 6, 2015 the median trip from Williamsburg to LaGuardia took 59 minutes; half an hour longer than usual. And trips from Midtown to LaGuardia took a median of 63 minutes, up from an expectation of 35 minutes.

The culprit was traffic on the Grand Central Parkway—Robert Moses strikes again!

This video is processing – it'll appear automatically when it's done.

An NYC-based consultant who flies 25 weeks a year once told me, “if you don’t miss at least 5 flights a year, you’re spending too much time at the airport”, so if he wants to miss 5 out of 50 flights, 90th percentile travel time seems like a reasonable scenario to plan for

This video is processing – it'll appear automatically when it's done.

Even less if you include livery cabs and black cars, which seem to be particularly popular among corporate clients that provide consistent demand. The FiveThirtyEight GitHub repo has some additional for-hire vehicle data, but I did not integrate it into my dataset

This video is processing – it'll appear automatically when it's done.