[rusbus]

Uber is pretty great, and I enjoy using it, but many of these statistics are pretty weak and out of context. As others have mentioned, using "income" to describe "revenue" is inaccurate. You can't compare making $90k driving an Uber to $90k at a desk job with a completely different set of expenses.

While the statistics used to determine the DUI rate in SF and Taxi crime in Chicago are beyond the levels of my knowledge and I can't verify their correctness, the calculations certainly seem a bit suspect. For example, as a pair city when doing a discontinuity test, they chose San Francisco, a city which notably, has Uber. The post doesn't mention a confidence interval or P value in a clear explicit way (outside of the table, with unexplained column headers).[1]

In the infographic, they claim these statistics as facts without mention of any underlying confidence intervals. I'm pro Uber, but this post seems to contain a lot of fuzzy math. It'd be interesting if someone with a statistics background could confirm or deny these suspicions...

[1] http://blog.uber.com/DUIratesdecline

[shakethemonkey]

For taxis, medallions are usually leased to drivers (few could afford one). This takes much of their revenue -- perhaps half or more surrendered to a medallion holding company just for the privilege of driving a taxi.

New York City presentation suggests a driver with $90k revenue has $41k expenses. [1]

[1] http://www.nyc.gov/html/tlc/downloads/pdf/presentation.pdf

[YokoZar]

They did give a P value, it's the final column labelled Pr (>|t|). The findings are significant.

They also gave the standard error, which you can use to compute your own confidence interval (95% would be about 2x, or ranging from no effect to a 20% reduction)