A blog post on Medium got a lot of attention recently for claiming that Uber drivers are now faster than ambulances in New York City. The numbers seemed pretty damning for the ambulances: the median Uber rider in Manhattan can expect to get a ride in 2.42 minutes. In the outer boroughs, the median wait is still just 3.1 minutes. The median ambulance wait? About twice as long at 6.1 minutes.
We wondered, how does the Bay Area stack up? Newsweek obtained some figures from Uber and mapped them.
According to Newsweek, the median wait time for an Uber in San Francisco is just 3.2 minutes. Los Angeles is not far off with 3.8 minutes. If that sounds like a hard number to beat, that's because it is. The target response time for San Francisco ambulances is 10 minutes for 90% of calls. In January, 2015, the average SF ambulance took 7.49 minutes to arrive(pdf).
But wait--you might say--aren't you comparing averages and medians? Yes, and that's a problem. But that's just where the problems start when you try to compare these two services.
Recall that a "median" is the halfway point in a set of numbers. In other words, half of Uber rides in SF are less than 3.2 minutes, and half are more. How much more? That, we don't know.
This isn't just a matter of semantics. It's crucially important in a life or death situation. If we know that half of all Uber rides take more than 3.2 minutes, we might hope that a lot of those are very close to 3.2 minutes. Maybe they are. But maybe a lot are closer to 10 minutes. Maybe 30? An hour? The median doesn't tell us that information. It could be anything.
The worry is that the set of Uber waits might be "skewed", meaning the first half might be very close to 3.1 minutes, but the last half might be very far from 3.1 minutes. If you plotted all those rides and put time on the X axis with number of rides on the Y, a skewed curve might look like this:
Importantly, we simply don't know what this curve actually looks like and Uber has no responsibility to provide it.
Local
This is one reason why emergency services use the 90th percentile as the target: it lets you know what the wait time is for the vast majority of rides in a way that a "median" cannot.
The conclusion Medium writer Minqi Jiang came to was that Uber should get into the ambulance game--that Uber drivers with defibrillators and CPR training could act like first-responders and help take the load off of city ambulance services.
I would say that the numbers alone can't justify a call like that. We simply don't know nearly enough. But one thing we do know is that San Francisco's ambulance response has been in pretty bad shape recently.
The city has struggled for years to hit its 10-minute target for ambulance waits. In January, SF ambulances missed the goal target by 2.49 minutes. This was actually an improvement over recent months. In August of last year, the 90th percentile was 14.63 minutes.
A San Francisco Chronicle investigation last fall uncovered residents that have had to wait 30 minutes or more for an ambulance.
Fire Chief Joanne Hayes-White and Mayor Ed Lee said the problem stemmed from shortages in staffing and ambulances. The mayor convened an "Ambulance Working Group" to recommend solutions.
In February, the working group released its final memo, recommending the the fire department hire 26 new EMTs and build a new $40 million ambulance facility. An earlier recommendation to acquire 19 new ambulances and replace many older ones has already been carried out. One recommendation this blog is very excited about is the creation of a "performance data dashboard" that would publish online data for ambulance response times and other "critical public safety measures."