Last week after I posted about Capital Bikeshare's Reverse Rider Rewards experiment, Mr. T in DC tweeted this in response:

@bikeshare needs to better incentivie uphill trips to Ward 1!
There's some anecdotal evidence that suggests that people are riding CaBi bikes downhill to get to wherever they're going, and then taking the bus or Metro back uphill, rather than trying to huff it and puff it on one of the heavy red bikes. It's something that was discussed last year on Kojo. In theory, it makes perfect sense.

I dug into the data and found that it's happening to an extent - more people are riding CaBi bikes down hills than back up them; but it's not happening at quite the exaggerated rate that some people seem to suggest.

The truth is that the majority of Capital Bikeshare trips start and end at about the same elevation. I looked at 327,680 trips taken from January through March of 2012 in the District (unfortunately I had to exclude Arlington due to issues with the elevation data), and the median elevation change was only 2.9 feet downhill. That's it!

In fact, more than three-quarters of all Bikeshare trips had elevation changes of less than 60 feet. Still, there were more trips that went 60 or more feet downhill than there were that went uphill. Why 60 feet? Because that's roughly one standard deviation from the mean; and it seemed more reasonable than picking an arbitrary number. The results are basically the same whether you're talking about registered or casual users.

(click to enlarge)

Regardless of where you draw the cutoff, even at a generous 30 feet up or down, the fact remains that a majority of CaBi trips start and end at roughly the same elevation.

(click to enlarge)

A histogram of all CaBi trips from the first quarter of this year shows what seems approximately to be a normal distribution with a slight skew to the left (downhill trips). The median, nonetheless, is squarely in the middle.

(click to enlarge)

There is one additional thing worth noting. Of course while we can calculate the elevation where the trip starts and ends, it ignores any elevation changes along the route. For example, someone traveling from Woodley Park to Capitol Hill via the National Mall will experience greater changes in elevation than this analysis suggests. Unfortunately, it's very difficult to calculate exact elevation changes without knowing actual routes. For now, we have to leave this as-is. 

What's the implication of this? I think it suggests that some people are riding CaBi in only the downhill direction, and that could be one reason the system becomes unbalanced. but the problem isn't as widespread as some may have feared. The reason why stations become unbalanced is certainly more complex than the topology of the city.


    On June 12, 2012 Chris said...

    Good article. The only thing better than an article that draws a conclusion from actual data is one that draws a conclusion from actual data AND uses graphs to display the data. Nice job!

    On June 12, 2012 Anonymous said...

    If Capital Bikeshare is designed to support commuters unbalanced bike distribution is actually a good thing (as long as there are enough parking spots) since the bikes will be where the commuters are. Of course this make it difficult for the non-commuters running errands during the day. To solve this problem designate some bike as "local" and can only be parked at their home station(s). The rates could be reduced for "local" bikes

    On June 14, 2012 Anonymous said...

    There's nothing wrong with using trucks or vans to re-distribute bikes throughout the day, provided this is included in the monthly membership dues.

    On June 14, 2012 James said...

    Rob - good insights. If you were interested in following this up with some more refined analysis that actually tracks changes in elevation along paths (rather than just start/end points) then I'd invite you to look into the git we posted about this week on GGW. Using OTP, you can definitely pull out some better elevation change information that will highlight any valleys or hills over which people might be riding. For example, when we ran those planned bike trips, we could have pulled out the total elevation gained/lost. I like what you're doing with filtering non-members though; in retrospect we should have done that to get rid of more of the lost/leisure riders. Interesting stuff!

    On August 13, 2012 Lisa C. said...

    Hi, What's your elevation resource? Great stuff here btw.


    I used "Topography - Spot Elevations" from the DC Data Catalog.