I posted previously about the 2010 bicycle commuting statistics released by the census bureau, which caused one commenter to imply that more dense cities have more bicycle commuting. So I thought I would plug in some surface area numbers, draw some graphs and check some correlations.
The results, if you are like me and want to just skip straight to them, is that I could generally agree with that statement, which surprised me a bit because it seems overly simplistic, but looking at the top 25 US cities by population, sorted by density, it does look like the denser cities have higher rates of bicycle commuting. However, to get there, I kind of had to throw out Seattle and New York. I also added a column for current LAB BFC status, with 5=platinum, 4=gold, 3=silver, 2=bronze, 1=honorable mention, 0=not ranked or I'm not sure). Wasn't sure where I was going with that, actually, but it seemed sort of interesting, and now looking at this chart, I'm thinking Philadelphia really needs to get to work on a BFC application.
(the data in this chart comes from the 2010 statistics on bicycle commuting and Wikipedia's surface area figure for the corresponding city as nearly as I could make it out)
| City | Pop | Workers | Cmtr % | # Bike Cmtr | Land Surface Area | Pop Density | Worker Density | BFC Status |
| New York City | 8,184,899 | 3,615,588 | 0.8% | 27,917 | 304.8 | 26853.34 | 11862.17 | 3 |
| San Francisco | 805,463 | 437,814 | 3.5% | 15,208 | 46.87 | 17185.04 | 9341.03 | 4 |
| Washington, DC | 604,453 | 296,717 | 3.1% | 9,288 | 40.1 | 15073.64 | 7399.43 | 3 |
| Boston | 621,383 | 309,620 | 1.4% | 4,369 | 48.4 | 12830.54 | 6393.14 | 3 |
| Chicago | 2,698,831 | 1,168,318 | 1.3% | 15,096 | 227.2 | 11878.66 | 5142.24 | 3 |
| Philadelphia | 1,528,306 | 583,734 | 1.8% | 10,503 | 135.1 | 11312.41 | 4320.75 | 0 |
| Los Angeles | 3,797,144 | 1,706,116 | 0.9% | 16,101 | 468.7 | 8101.96 | 3640.34 | 0 |
| Seattle | 610,710 | 339,160 | 3.6% | 12,306 | 83.9 | 7281.63 | 4043.88 | 4 |
| Baltimore | 620,583 | 256,622 | 0.7% | 1,788 | 92.1 | 6738.14 | 2786.34 | 0 |
| San Jose | 949,197 | 426,136 | 0.6% | 2,708 | 176.5 | 5376.97 | 2413.96 | 0 |
| Detroit | 711,910 | 196,706 | 0.3% | 651 | 138.8 | 5129.03 | 1417.19 | 0 |
| San Diego | 1,311,886 | 620,939 | 1.0% | 6,390 | 325.2 | 4034.21 | 1909.47 | 0 |
| Denver | 604,414 | 296,453 | 2.2% | 6,514 | 153.3 | 3942.69 | 1933.81 | 3 |
| Columbus | 789,939 | 379,334 | 0.7% | 2,498 | 210.3 | 3756.25 | 1803.78 | 0 |
| Houston | 2,107,208 | 961,240 | 0.5% | 4,393 | 579.4 | 3636.88 | 1659.03 | 0 |
| Dallas | 1,202,797 | 543,348 | 0.2% | 820 | 342.5 | 3511.82 | 1586.42 | 0 |
| San Antonio | 1,334,359 | 591,725 | 0.2% | 1,159 | 407.6 | 3273.70 | 1451.73 | 0 |
| Austin | 795,518 | 412,291 | 1.0% | 4,242 | 264.9 | 3003.09 | 1556.40 | 3 |
| Phoenix | 1,449,481 | 620,072 | 0.6% | 3,576 | 516.7 | 2805.27 | 1200.06 | 1 |
| El Paso | 652,113 | 260,318 | 0.1% | 217 | 249.1 | 2618.09 | 1045.12 | 0 |
| Fort Worth | 744,114 | 330,652 | 0.1% | 473 | 292.5 | 2543.98 | 1130.43 | 0 |
| Charlotte | 734,418 | 344,436 | 0.2% | 835 | 297.7 | 2466.97 | 1156.99 | 0 |
| Indianapolis | 824,199 | 366,017 | 0.5% | 1,935 | 365.1 | 2257.46 | 1002.51 | 0 |
| Memphis | 647,870 | 262,033 | 0.1% | 153 | 302.3 | 2143.14 | 866.80 | 0 |
| Jacksonville | 823,316 | 375,579 | 0.2% | 843 | 767.0 | 1073.42 | 489.67 | 0 |
I added the calculated worker density because I thought that might make a difference, in the sense that it should give a better indication of how spread out the people who are commuting to work by bicycle are, and it did make a small difference. New York and Seattle are the outliers though. Actually, Denver is sticking up there about 1.2% or 3000 commuters higher that you would expect, too, but maybe hasn't gone all the way yet, whatever "all the way" may be. Here's my first graph with New York and Seattle included. I did use the spreadsheet's correlation function to calculate a .51 from these numbers, to show you what we get including NYC and STL.
Looking at this graph, I conclude something else besides just population density is going on in New York, and in Seattle going the other direction, to impact bicycle commuting. Also that Denver may be catching whatever it is that Seattle has got. I will go ahead and assume that New York is in transition and also is so big that it's Different, and also that Seattle is being special in its own way (Portland didn't quite make the top 25 by population, or else it would have been up off the top in its own 6% zone saying hey look at me in my tight pants with my excellent coffee and free-range fair trade organic statistics). Removing those two, here's the resulting graph, also switching to workers rather than overall population, which helped a little (correlation now at .87) still with Denver holding out. The two superstars up there that look like they are just about right where they should be for their density are San Francisco and Washington DC.
If you go all the way and take out Denver, it looks even nicer with a .93 but let's not get too carried away. I included the data above because if you actually have read this far and disagree with the overall idea or approach, it is very probable that you know a lot more about this than I do and can straighten me out, so please do. Regarding the original commenter's suggestion about DFW and size and density, there's some different terms that result in different surface areas and densities, from city, to metro, metropolitan area, CSA and MSA, and probably others. The 2010 census bicycle commuting numbers only relate to the city itself, though, and not the larger or different ways of looking at our complex metro landscapes, so I used surface area corresponding to the population number given. For example one metropolitan area boundary for some Phoenix-related statistics (mentioned in this post) has it almost the same size as the Netherlands, which is interesting but not used to compile the city-specific bicycle commuting statistics given above. And anyway looking at some of the other MSA boundaries, it's getting harder and harder to tell where one starts and the next stops, but at least on Wikipedia, DFW is not the largest or most populous or least dense..
And my main conclusion out of all of this is that I am feeling a little cooped up from this cold and need to get back out on the road really soon.
And my main conclusion out of all of this is that I am feeling a little cooped up from this cold and need to get back out on the road really soon.
That is nice work. Interesting.
ReplyDeleteThough I cheated and gave you the DFW metro area (almost all of which has people), not all such are created equal. For example, Seattle's metro includes Everett and Tacoma; both comparative cycling deserts. Seattle's also includes two major mountain passes. If you take Seattle proper, most of that cycling is coming from a small portion of the city. Like about a quarter of the area. You get to the hills around Northgate and cycling is, shall we say, rather more rarified.
ReplyDeleteYou should, however, EXPECT Seattle to be an outlier. After all, it is my home town.
It is too bad you can't do it, but I bet the correlation would be stronger if you were able to plot cycling rate versus when the neighborhood was built. You can also add a bump for major student populations (which Seattle has).
It gets addicting when those numbers start to talk, eh?
I think NYC would not be an outlier, if you included peds and transit, along with bikes as people that commute by means other than private motor vehicle. That'd still leave Seattle. At one level, you could even count ME as a Seattle bike commuter...
ReplyDeleteAwesome post! The Portland bit is hilarious.
ReplyDeleteI was inspired to add Portland and our densest suburb, Beaverton, to the scatterplot.
I actually see this as a challenge for Portland: for our above-average biking (and transit) infrastructure to be equitably accessible, we need to start building more densely. We're *half* as dense as Seattle.
Beaverton, our densest suburb, is denser, and it's actually about as much of a leftward outlier on this scatterplot as Denver or Seattle. It's only 90,000 people, though, so obviously a very different situation.
Mark, thanks!
ReplyDeleteSteve, approaches like the one I took in this post surely are limited since the actual commuting picture in large cities is complex, being tightly related to the city's history of public transit, New York for example, as well as how spread out the whole metro area is, U.S. western suburbian metroplexes being one species, eastern multi-city megalopolises being another, northern European cities an entirely different other. Even mode itself is complex, I don't know how you would account for me when I used to commute every day by carpool in the morning and bicycle home at night. Each place has its own context, which ends up meaning for us that successful bicycle encouragement strategies from one city need to be translated, or may not translate at all, to another city. Some places are so dense with people already that they end up with a whole other set of complexifying challenges. Still, what works in one place may be valuable to talk about in another, to explore options. And I don't think Dallas counts as a suburb of Seattle.
Michael Portland Afoot, thanks! There's also a school of thought that Phoenix would be much better off in several ways by building denser. In this era of zombie subdivisions sitting empty 60 miles from the city center and being reclaimed by slow natural processes, proposals for how cities could build better, denser, and more affordably may get slightly more attention, although here we still build and tear down block by block, development by development, one get-rich-quick opportunity at a time, it seems like. I am of the school that there will be a time within a hundred years if not much sooner where the economics of relying on petroleum-based energy for moving around our cities will become untenable as supplies run out, and alternatives will increase in share, if we survive the change that is. I don't know how that will turn out, although it seems that cities with a mix of density and a variety of local sources of essential resources like food and water will be stronger in that era than spread-out cities with few local essential resources.
Bear in mind that picking the most populous 25 cities (as defined by municipal boundaries) gets you places like Indianapolis and Jacksonville that include nearly their entire counties (including still-rural land) within the "city". Picking the central cities of the most populous metropolitan areas might get you a better selection (and would then also include Minneapolis-St Paul, which has one of the highest bike commuting rates).
ReplyDeleteExpansive municipal boundaries would also explain Denver's position -- I think the city proper includes both the new Denver International Airport and the old Stapleton Airport territory, which must lower the density numbers a bit. The city also includes suburban and undeveloped areas to the east.
ReplyDeleteafiler, good points, I found the different approaches to defining, delineating, and counting the population of cities interesting and challenging. On the one hand, the distinct historical and geographical circumstances of each place may mean it makes more sense to count and define the "City" in a distinctly local way, while on the other hand, as our major cities more and more just melt into one another to become one vast metropolitan expanse the lines may become more and more arbitrary and merely traditional. "Phoenix" is an interesting example to me, since I am relatively familiar with its vast contiguous similarity.
ReplyDeleteJohn, you asked for a response. My initial reaction to your figures is that they're all so low. You have nowhere with a high cycling rate in your data and you are left with no choice but to plot figures which are to some extent noise. After all, what is the error margin on the measurement of a 0.1% commuting rate ?
ReplyDeleteAlso, I have my doubts that this is the only correlation that could be found. These cities vary in more ways than just their densities. e.g. in student population, average distances, quality of infrastructure.
However, Michael from Portland adding in his city shows something really interesting. This city is an outlier with low density and high cycling rate compared with the rest of your figures.
As you know, I found absolutely no correlation at all, between population density and cycling rate when we look at the international picture. Rather, we're left with a correlation between the quality of cycling provision and the cycling rate, and density becomes difficult to spot as an incentive to cycle.
I suspect this is also what you're seeing. There may well be a small effect due to density, but this is important only where infrastructure is bad, cycling rates are incredibly low and where cyclists are under a huge amount of pressure on the roads.
Portland's better than average (for the US - poor by Dutch standards) infrastructure is enough to make it jump right out of the top of the scatter graph because infrastructure is a far more powerful incentive to cycle than is density.
If you added Dutch cities to your graph you'd end up with something that looked like my plots, showing a strong relationship with having Dutch infrastructure and very little correlation with anything else.
Thank you David. To your concern about the low percentages yielding only graph noise, I would also add that the error bars on the census commuting data are almost laughably large when you look at them. To cite a local example, the published census number for female bicycle commuters in Scottsdale is 0 with a 305 margin for error (which may just be a typo). On the other hand, this is the best data I know of for nationwide bicycle commuting statistics in the US, and it does have the advantage of being gathered with the same methodology at a particular time. I'm not sure what the error bars are the data you used, or why you picked those particular cities rather than some direct sample like "Top 100" or something.
ReplyDeleteMy thought after looking at the US example is that if one could gather reliable data on cycling and population density ranging from the Sahara Desert to Copenhagen, and then do some extremely difficult calculations that no one would understand to correct for economic and other significant variables, more dense places would have more cycling, all things considered. But I'm not sure that could even be done.
I am sure that comparing the bottom with the top and the big with the small without correcting for those things and other variables will also yield just graph noise. Comparing New York and Assen and Groningen on the same graph may be overlooking a few other variables which also add noise to the picture. Looking just at cities in China over the last twenty years would show very strange statistics on cycling vs. density unless you understand the economic boom there which has resulted in streets which were filled with bikes in the 1980s when I was there to now be packed with bumper to bumper cars. More than a simple bivariate analysis is required to get some useful understanding of what's really going on, in my opinion.
New Yorkers have very low automobile usage but also very low cycling rates, what's going on? Well, for starters, they have an fantastic Metro / subway system integrated with ferries and regional transportation, and walk a lot. But what does that mean for a worldwide look at cycling vs. density, and how to correct or compensate for it? That's probably a question better suited to statisticians, demographers, and transportation experts than to a bicycle blogger like me.
In this TED talk, Alex Steffen shows a graph which boldly asserts that the higher the density of city, the lower the overall emissions, worldwide. With Phoenix in the low/bad zone by the way. While cycling is my thing, in the end I am just as pleased that denser cities have lower overall emissions because this shows that regardless of historical, economic, political, and random noise, people in denser cities get from A to B, however they do, in ways which lower emissions. When I start to feel that higher density yields higher cycling in the future, I will remind myself of the difficulties of actually demonstrating that statistically, leave that exercise to more qualified people, and instead just conclude that denser cities have lower emissions, which seems like a more reliable bivariate way to look at how people in very different situations get around.