Monday, May 28, 2012

Transit Types and Modal Share: More Numbers

I've put together three additional charts, correlating per capita highways, heavy rail and light rail with commuting transit share for an expanded list of thirty American cities, in the hope that these statistics might add to and shed light on the results from the previous post.  Although the first two charts are probably in line with most expectations, the third, for light rail, may come as somewhat of a surprise.

First, here are highway lane miles per capita plotted against commuting transit modal share, using highway and population data from the Federal Highway Administration (h/t Walkable DFW):

As might be expected, there is a correlation between decreasing freeway lane miles and increasing transit modal share, but it is not strong.  Eliminate the three outliers of Kansas City, St. Louis and New York, and even that modest correlation is cut in half.

By contrast, the correlation between transit share and kilometers of heavy rail is much more robust.  For this chart, only urban heavy rail systems, such as subways, were included. Regional commuter rail, such as New York's Long Island Railroad or San Diego's Coaster line, is excluded to keep the comparison consistent:


Eliminating the cities with no heavy rail from the sample reduces the correlation to an R2 value of .17.

Last, and perhaps most surprising, are the light rail numbers.  Regardless of whether cities with no light rail are included, there is a statistically insignificant (but consistently negative) correlation between light rail kilometers per capita and transit share.  The United States seems to be flush with cities with substantial light rail networks and low transit modal share.  The exceptions are the cities which also have heavy rail networks (SF, Philadelphia and Boston).  Three other heavy hitters have no light rail, although all had systems in the past (NYC, DC and Chicago):


Two cities with no or essentially no light rail at all, Milwaukee and Detroit, surpass numerous other cities with extensive networks.  Portland's extensive system, more than five times larger than Seattle's on a per capita basis, has only earned it 60 percent of that city's modal share (arguably, Seattle should be given an even lower value on this chart, as the King County branch of its light rail system opened more than halfway through 2009, the year measured by the ACS in computing modal share data).

Are there any further points to glean here?  Certainly no correlation shown here approaches the correlation of urban population density to modal share that I provided in the previous post, which remains by far the strongest correlate of transit use. The role of bus networks is obviously huge, but I wasn't able to find a single metric useful in making cross comparisons that was available for all bus operators (routes per capita is a possibility, but there is no way of knowing the length of these routes in most cases.  I might add it in later anyways, if I can locate the data for it). Commuter rail would introduce more complexities.

Chart data are below.  Numbers in the last three columns show miles and kilometers per capita (per 1,000 or 100,000).  Note that modal share data are for cities, rather than the greater urban areas from which population figures used to compute per capita highway and rail figures are drawn, but are more closely tied to the types of urban transit systems I have included.  You are welcome to reuse this information, but no guarantee of accuracy is offered.  Contact me if you'd like the excel file.

Sources:
DOT: Highway Statistics 2007
2009 ACS via The Transport Politic
Wikipedia (for light and heavy rail statistics)

Related posts: 
Chris Bradford: The association between density and mode of commute. Chris finds that standard density is weakly predictive of transit modal share, but weighted, or perceived, density is strongly correlated with it.
Laurence Aurbach: Fun with Density and Transit Statistics.

22 comments:

  1. Too many confounding variables here. Old cities like Detroit have pre-car-dependency residential and office development which provide built-in reasons to take the bus that nobody in Portland or Dallas really has, for instance.

    I think there might be some data to mine here based on development of the city core in, say, 1950 versus bus transit share - but I can't think of a way to get at that off the top of my head.

    ReplyDelete
    Replies
    1. I agree, and I think that the charts in the earlier post help to get at those variables (of residential and job concentration). That residential density and job concentration correlate so strongly with transit use suggests that these "built-in" reasons are very significant.

      That said, Detroit is very much a city of the auto age. 75% of its population growth, to the 1950 maximum, occurred post-1910. For contrast, only 40% of NYC's growth was after 1900, and 25% of Philadelphia's. It is denser than Portland, but only very slightly.

      Delete
    2. 1910 - 1950 was the "unsubsidized auto age" where a typical family eventually got a car but didn't use it for most commutes. 1950-2000 is the real problem, and why Sunbelt cities are so much harder to serve than Detroit.

      Delete
  2. There is also a selection bias here, with some places having light rail because they couldn't swing a heavy rail system, and in the ability to gain local support for a heavy rail system, high public transport mode share would be a substantial help.

    One wonders what the correlation would be for total passenger miles by rail of either type.

    ReplyDelete
    Replies
    1. That's a good point, and a great idea for another chart -- I will try to put it together. It is possible that the viability of heavy rail (politically, financially, etc) is largely dependent on density, so that the heavy rail figures are simply an echo of the density correlation. Atlanta was the sole city in the sample with very low population density to have built a heavy rail system (Cleveland built its system in the 50s, when population density was much higher than today).

      Delete
    2. The distinction light and heavy in formal lexicons is regarding capacity, so there'd have to be a multiplier. Offhand, I'd multiply one light rail mile as 0.5 of a heavy rail mile, not based on any deep information on average capacity relationships nationwide, but because if the factor is too small a fraction, the light rail impact disappears in the noise.

      And 0.5 is the maximum entropy estimate for "smaller than one, bigger than 0".

      Delete
    3. I can't speak for other cities but St. Louis' 'light' rail system is indistinguishable from a heavy rail system, all running on dedicated tracks, with some of it underground.

      The real difference between light and heavy rail is that heavy rail is in dense cities where rail makes some economic sense while light rail is in dispersed cities like St. Louis or Dallas where the economics are utterly insane.

      I did a back of the envelope calculation for the total costs per passenger mile of the St. Louis/BiState Transit system (including buses and rail) and including depreciation on the fixed assets the real cost per mile equals the TCO for an Economy car: car, taxes, gas, insurance, etc.

      I have no reason to believe that St. Louis is a particularly bad system. In my opinion "Light" rail in most cities is simply the Lunatic phase of transit insanity.

      PS: I like and use the St. Louis system - its clean and efficient. Just economically insane.

      Delete
  3. I'm confused. How does San Francisco have a heavy rail ratio of 5.28? The MUNI metro consists of light rail and buses. BART and Caltrain could be considered "heavy rail" but both are commuter systems. Or are you counting the portion of BART that runs within the city? Even so, that's only 12km.

    Also, what's the split for Boston? Green Line vs the others?

    "Heavy" vs "light" rail is kind of a weird distinction to make, since most of these systems were designed and built long before that terminology came into use in the 70s.

    ReplyDelete
    Replies
    1. BART is quite definitely urban heavy rail. These must be either Metro areas or Urbanized areas, as you'll note that the entry in the table is San Francisco - Oakland.

      Delete
    2. I counted the entire BART system (167 km), even the portions outside SF proper, since the context was the greater urban area, not just the city itself. The same was done for DC's metro system. "Pruning" the system to match political boundaries seems artificial, and I was aiming at a general figure showing the amount of high-capacity rapid transit rail line laid down relative to overall population.

      Caltrain is a close call, but I omitted it since it's more regional in scope (going beyond the population area covered), and really is a dedicated commuter line, rather than a mixed-purpose line like BART. MARC is not included for DC, for similar reasons. I'll admit the distinction is unscientific, but I had to draw the line somewhere. Long-winded answer, but the basic idea is "urban heavy rail with rapid transit function that mostly serves the metro area rather than outlying suburban communities," and to apply that as consistently as possible. In that sense, BART, the DC Metro, Boston's T and the NY subway are are all recognizably related.

      For Boston, the Red, Orange and Blue lines are counted as "heavy rail," while the Green line is under "light rail." That's not my distinction, though -- it's from the American Public Transit Association, which also uses the heavy rail/light rail distinction, via Wikipedia.

      Delete
    3. And in response to Bruce, as you can see from the FHA data, urbanized areas were used. This is not necessarily a perfect match with the transit systems covered, but it's reasonably close.

      Delete
    4. Well, I do think of BART being more of a regional commuter line, but it's not a big deal either way. Counting BART leads to SF/Oak being an outlier on that chart, and omitting it would probably strengthen the correlation slightly.

      Delete
    5. Yeah, the SF number may be a bit high, but at the same time the NY number is a bit low, since the population figure includes a large portion of northern NJ, which isn't served by the NY subway (I did include PATH, but omitted the Hudson-Bergen LR since it does not enter the city at all -- not that it would have made a difference). The urbanized area population figure for NYC seems to be more generous, geographically and proportion-wise, relative to some other cities. Adjusting both of these would probably leave the correlation in a similar place, but then, starting to tinker with the urbanized area figures or with the length of individual systems really opens a can of worms.

      Delete
    6. I do prefer urbanized area for these things.

      If the figures were trips or passenger miles, then the difference for BART would not be a big deal, as most of BART patronage is within the SF-Oakland Urbanized Area, but track miles would be inflated, with the BART to Livermore and the BART that will be running to San Jose.

      Caltrain is primarily regional intercity between the SF-Oakland urbanized area and the San Jose urbanized area, so if regional rail systems are excluded, leaving out Caltrain makes sense.

      Delete
    7. One data point: at a trivia night in New York, I asked people to name the only commuter rail network in the US that's all-electrified, and most teams answered BART. Its outer ends definitely look very commuter rail-like.

      However, one key distinction is frequency. Every BART branch gets at least 15/20 service, better than any commuter rail branch in the US. A few branches get 30/60 service and even fewer get 30/30. (Here the first number is midday frequency, not peak frequency, and the second is evening or weekend frequency, whichever is worse.)

      Delete
  4. Great stuff. It will take me a while to digest these posts, but I also wonder about downtown parking density as another variable. I find transit use to be a multiplicity of incentives.

    ReplyDelete
    Replies
    1. Downtown parking is a huge variable, for sure. It is frustratingly difficult to calculate, though. Even Norman Garrick and Chris McCahill, in their study of Northeastern cities, were only able to add up surface and above-ground parking -- below-ground garages are very difficult to measure without on-site inspection, and citywide statistics aren't usually available. It would be doable, but a lot of work, to calculate surface parking ratios and use that as a baseline for comparison.

      Delete
    2. Edmonton and Calgary have no freeways to downtown, and extremely high transit share. That's not the same as having limited parking, but the effect is similar.

      Delete
    3. I also would love to see what the correlation is with parking. However, calculating existing parking would likely be extremely difficult.

      It might not be that hard to run a regression of transit use vs present required minimum parking standards (perhaps an average of residential and commercial). I would hazard a guess that these policy standards say a lot about how fully a city embraces transit.

      Delete
  5. Control for poverty first, and for weighted density second. Then see what the scatterplots look like.

    --Nathanael

    ReplyDelete
  6. Would be really nice to see that someday in the future, all of those back garages could function as storefronts.

    ReplyDelete
    Replies
    1. Wrong blog post =_=

      Delete