CURA’s analysis of Richmond’s bus network redesign is real bad

The following originally appear on Twitter as an epic thread. It’s been adapted and lightly edited for this website.

A new report is out on @GRTCTransit‘s year-old system redesign, and it shows benefits that starkly contradict a prior report by @Wilder_Planning‘s CURA. The CURA report is so flawed that the authors should immediately retract it. Let’s take a look.

First, the topline: the new system increases the number of jobs the average resident can reach in 45 minutes by 6%, and the number of jobs the average low-income and minority resident can reach by 10%. Better on average, in particular for disadvantaged populations.

So how did CURA find that access for disadvantaged populations decrease by 22%. Well, besides using methodology that doesn’t make sense and is not generally accepted in the field, they used the methodology incorrectly! 😲

CURA took all the bus stops (red) and calculated all the areas within a half a mile (blue) and considered loss of access for areas over half a mile (white). As you can see, some of the areas considered over half a mile from bus stops are bus stops themselves! 🤔

Obviously, this makes no sense, but sometimes data spits out weird results. That’s why you need to do a sanity check and, if you find something wrong, fix your analysis. Unfortunately, CURA didn’t, and just published, and when this was brought to their attention they did nothing.

What’s the result of this lack of simple checking? Take a look at Creighton Court. It is two large parcels. The parcel on top, in yellow, has 0.2% of it more than 1/4 mile from a bus stop. So CURA excludes the entire parcel from those having <1/4 mile to a bus stop.

Now you would think, “Well, if 0.2% is outside 1/4 mile, you should count 99.8% within a quarter mile, so of the 356 dwellings, include 355, especially since the 0.2% is grass.” CURA disagrees, and excludes the whole parcel, so all 356 units “lost access” in their report.

“Well, using this methodology is bad, but at least it affected both systems equally, right?” you might ask. Wrong again. You see, CURA noticed that some parcels, especially housing courts, were being excluded, so they manually included them…only for the old system!

[w]e were concerned that many multi-family and public housing courts were not completely contained in the polygons due to their size even though they overlapped with the coverage polygons, and made an effort to accurately include them in both old and new polygons.

Yes, the CURA report manually fudged the numbers while analyzing the old system but not the new one! At best it’s sloppy; at worst, dishonest. Note that when trying to replicate the results (CURA vs QGIS), the new network is the same, but the old is different, due to fudging.

Don’t be scared by the math, but the formula CURA used isn’t right either. The first one, from a peer-reviewed paper, has seven variables, while CURA’s, below, has five. They just eliminated two variables! One is activity density, the other is speed. Yes, speed disappeared!

It doesn’t take a transit expert to realize that the faster a bus goes, the better the service is. While the peer-reviewed experts incorporated speed as a key element, CURA just lops it off. 🤷‍♀️ It’s not even hard to figure out: distance/time. The schedules are all online.

CURA messes up the frequency factor too! It’s easy to look up, but they instead pulled the data from a source that isn’t updated, so they got a ton of weird results, including saying a training platform with no regular service gets a bus every 7 minutes!

Want more sloppy data handling? CURA comingled weekday and weekend times when calculating how often a bus came. This shows two buses coming one minute apart, and then they just averaged it, producing a wait time that is half of reality. This mistake is EVERYWHERE.

CURA also just assumed all buses ran 12 hours a day, even though some run only at rush hour and some run 18 hours a day, making the most frequent and useful routes seem worse and the commuter ones (mostly in rich areas) seem better.

As a result, for peak-only services like Routes 26, 27, 28, 29, and 64, which actually operate about 6 hours per weekday, the hours of operation have been overstated by more than double. And for routes that run later in the evening, like The Pulse, 1, 2, 3, 4A, 4B, 5, 12, 13, 14, the hours of operation have been understated by a third: those long-span routes run 18-19 hours per weekday.

For the distance factor instead of looking at how long a route is (more things to go to), CURA looked at how far a stop was from other stops (long trip). Thus Short Pump had some of the best service, while downtown had the worst. They multiplied when they should have divided!

CURA then decided to grade more positively a stop if it has more routes serving it. So a stop with two routes that run once a day is better than one with one route that runs every 15 minutes. A quick thought experiment and you should easily spot the error. CURA didn’t.

That’s right: of the 7 variables in the equation CURA took from peer-reviewed researchers, they used 5 incorrectly and didn’t use the other 2 at all. There’s not a single variable CURA used correctly. Not one! How can we trust any of this? Garbage in, garbage out.

I could go on, as there’s lots more, but it’s clear that CURA has no idea what it’s doing here. They do really good work in a lot of fields, but their expertise is not in transit, and anything they put out in this field should clearly be ignored.

We all have specialties and weak spots, and when venturing into another field it’s good to speak to those experts before publishing. Unfortunately, it seems they didn’t do this, and when I and others brought up these concerns after publication, they were kind but didn’t change.

It’s really disappointing that such shoddy research is sullying such a respectable organization. If CURA has any honor, they’ll immediately retract their report. This doesn’t live up to their academic tradition.

It’s worth adding that @VCU put the @Wilder_Planning CURA report on their front page. I hope, in the vein of intellectual honesty, when this paper gets retracted they offer up a mea culpa that is similarly prominent, as it is to reporters. cc: @ByRobertoR @Suarez_CM /fin

VCU back on the bus (for the next three years)

Big, huge transit news out of VCU:

Virginia Commonwealth University signed a three-year paid agreement with the GRTC Transit System on Tuesday to fund unlimited transportation access on Pulse Bus Rapid Transit, local and express routes for all VCU, VCU Health System and Virginia Premier students and employees effective Aug. 1.

Unlike basically every other mid-sized city in America, Richmond is right smack in the middle of an unprecedented increase in bus ridership—and VCU’s unlimited rides program has been a huge part of that growth. At the moment, VCU-adjacent folks make up 12% of GRTC’s total ridership (you can check out ridership trends by route for the whole system in this PDF). V-C-U! Go Rams, go!

VCU’s original pilot program with GRTC, which was set to end on July 31st, cost the University $1.2 million. With way more VCU folks riding buses all over town than originally expected, the effective cost per ride for the University went way down. Ultimately, this starts to become an equity issue with some rides costing less than other rides. That’s why it’s good to see that VCU will increase their financial support for GRTC immediately, and then continue to increase it each year, presumably paralleling projected ridership increases.

Under the new agreement, VCU will pay GRTC $1.42 million for services in the first year, $1.57 million for the second year, and $1.65 million for the third year to cover the cost of ridership for students and employees on local routes and the Pulse and to maintain 10-minute headways for the Pulse.

Additionally, the University has shown its commitment to encouraging people to get on the bus by eliminating their Campus Connector.

In an effort to eliminate redundant services and contribute to the cost of the new partnership, VCU will eliminate its Campus Connector transportation service, effective July 1.

Heck, why not even go a step further and eliminate the M Lot Route as suggested by @_smithnicholas_?

Now, how can the Commonwealth of Virginia create the same deal for their Richmond-based employees? Or what about SunTrust? If Richmond’s other major employers decided to “get on the bus” so to speak, we’d see some real mode shift take place in the city.

The fundamental principle is this: Mid-sized cities can quickly increase public transportation ridership by spending money making things fast, frequent, and reliable. It works! Take it from Richmond!