EDITOR’S NOTE: This is the third in a four-part series of posts this week by statistician Mike Logsdon. Mike bikes across the Fremont Bridge every day, and put together this analysis of the bike counter data “mostly for the amusement of friends and coworkers.” Note that at the time Mike did this analysis, the 2014 Bike Month data was not yet available. So this post only looks at 2013.
The Cascade Bicycle Club promotes May of every year as Bike Month. Local businesses sign up teams of bicycle commuters for the commute challenge, and the Cascade Bicycle Club offers a variety of events, incentives, and aid to raise awareness and promote bicycle commutership. Obviously the benefits are numerous to everyone: Bicycle commuting boosts the health of the participants and takes cars off the roads. But how much do these efforts actually modify commute habits?
To make fair comparisons, we consider “residual” trips. Residual trips represent the difference between the observed number of crossings by bicycle and some expected number given the season and the weather. Intuitively, we expect lots of riders on a sunny day in June and few riders on a rainy day in January, so we need to account for that when making comparisons. A particular week may have seen unusually high rates of bicycle crossings, but perhaps that week experienced unusually pleasant weather. Fair comparisons adjust for the season and the weather, and we do so by fitting a regression model to calculate “expected counts”, then investigate the differences between observed counts and expected.
If commute rates during Bike Month consistently and emphatically exceed the expected rates, then we can conclude that the Cascade Club advocacy proved fruitful. As a demonstration, Figure 4 shows the daily observed counts, with a squiggly line showing the expected counts. The regression model is the same as that in the previous section, when it was used to estimate the effect of rain.
Well… see Figure 5. Bike Month 2013 started with a bang; people were excited to bike to work! The first few weeks of the month saw bridge crossings routinely between one thousand and fifteen hundred riders per day above expectation given weather alone. However, commuters seemed to lose interest and momentum by the end of the month, when Fremont bridge crossings returned to normal, or even possibly a bit below normal.
Returning to the regression model of the previous section, we can assess statistically whether the Cascade Bicycle Club’s advocacy encouraged bicycle commutes, at least across the Fremont Bridge. Adding an additional binary term to the model – indicating whether the day fell within Cascade Bike Month – yielded an estimated bike month effect of 675 riders per day, with 95% interval (430, 920). Although the detailed look revealed a late decline, total Bike Month bicycle trips across the Fremont bridge significantly exceeded what would have been expected given season, temperature, and precipitation alone.
In the next and final part of the series, we will look at whether the number of people biking over the bridge is increasing.
Comments
9 responses to “A statistical analysis of biking on the Fremont Bridge, Part 3: Bike Month”
[…] ← A statistical analysis of biking on the Fremont Bridge, Part 1: Overview A statistical analysis of biking on the Fremont Bridge, Part 3: Bike Month → […]
Mike — I just have to say that this analysis series is absolutely fantastic. Thank you!
Yes! Thanks, these are fun and you have a great writing style (clear explanations but still short and to the point).
I think the most interesting test is if bike month changes permanent behavior – I know it was the little push I needed to show me biking in the rain was better than taking the bus! Now the only reason I don’t ride is if post-work activities keep me downtown late into the evening with others who won’t be riding home or would involve leaving my bike locked for many hours on the street.
That’s a much tougher question. Part 4 that goes up today will look at the long term trend, although it’s impossible just from this dataset to decompose time trends into changes caused by altering commute rates and changes caused by population growth
I know it did for me. The first time I biked to work in Seattle was during bike to work month 7 years ago. After that I was hooked. I’ve been doing it rain or shine ever since.
This is absolutely fantastic stuff, Mike. I am curious if weather or other effects influenced this count. Early May of 2013 set records, if some websites are to be believed, whereas I recall we eventually complained that summer did not really start until July 4th last year. I have a terrible memory so I probably got this wrong, but if you look at this graphic (http://fs.weatherspark.com.s3.amazonaws.com/production/reports/history/year/000/029/735/2013/temperature_temperature_f.png) , it would support that the “weather effect” you previously documented may have confounded the “Cascade Bike Club” effect you look at here. Perhaps a pioneering grad student at UW wants to run a regression analysis or something to disaggregate some of this.
I suspect Cascade may be wisely trying to associate themselves with an already existing phenomena, that Seattlites are mad to get out and about after the rainy winter and biking to work more is one consequence. (Perhaps a pro-walking group should promote walks around greenlake in May?) It does give the phenomena a handle to raise awareness, however, and may even boost turnout a titch, but I doubt the P value approaches .05 :)
This analysis already adjusted for the weather. Eagerness to get outside after the cold, dark winter could certainly contribute to the early Bike Month surge
Interesting Read