Wednesday, June 26th, 2013

Where the Density Is

I’m sure many will not like or agree with Wendell Cox’s latest New Geography piece on transit and density. But if you’ll put aside preconceived notions for a moment, he has some very interesting charts about density in the United States. He measured density at 10,000 people per sq mi and 25,000 per sq mi at the zip code level and mapped to metropolitan areas. Here’s the breakdown at 10,000:

I’m sure someone will take this opportunity to ding pie charts, but you can easily convert that to a bar chart if you want. Speaking of bar charts, here’s one of total population at both 10K and 25K:

New York has 88% of America’s total population living at 25K/sq. mi. or higher.

And here’s a bonus infographic series. Nathan Yau over at FlowingData posted this graphic of grocery story geography in America. Click to enlarge.

Topics: Demographic Analysis

35 Responses to “Where the Density Is”

  1. I feel dirty writing this, but I (for the first time) didn’t really disagree with Wendell Cox. Trying to force more transit into areas that weren’t built for it just ain’t gonna happen within current development policy. His analysis of density in LA and San Jose makes that point well.

    The elephant in his argument’s room is that it is ILLEGAL to build housing the market demands in expensive, dense areas. Zoning, parking minimums, approvals, environmental reviews, etc mean that supply is nowhere near the demand, and the price of housing in LA, San Diego, San Jose, NYC, etc is proof. The fact that residential supply can’t rise to meet demand isn’t proof that transit doesn’t work.

  2. John Morris says:

    This is my central objection to Indianapolis’s transit plans. Density, isn’t everything- other factors like mixing land use and basic pedestrian infrastructure also effect transit, but it’s pretty important.

    Indy needs to fix these issues first, which is why development, line by line linked with introduction of transit oriented zoning is probably the best plan.

    Forcing transit where it can’t work bleeds the system of resources that could be used to provide better service on viable lines.

  3. Eric Fischer says:

    10,000 people per square mile is a remarkably low bar to consider dense. That’s two people in a house on a 50-foot lot, and yeah, nothing about houses on 50-foot lots particularly encourages transit use.

  4. John Morris says:

    Cox is right about design. I googled Garden Grove California, because of my conversation with Chris Barnett about The Crystal Cathedral. 9,500 people per square mile passes as close to dense by American standards & there is even a core grid of main streets. A closer look, however shows not only a proliferation of cul-de-sacs, but the general absence traffic lights or even sidewalks, making the walk from any transit stop to one’s home almost suicide.

    What’s also clear is that the low hanging fruit in transit use isn’t so much retrofitting car oriented suburbs, but infill development. The classic post industrial donut around many downtowns is ripe for new transit oriented construction.

  5. Chris Barnett says:

    Eric, 10,000 is the top theoretical density of many “streetcar suburban” neighborhoods in US cities away from the coasts, using 2 people per dwelling… except that they were built in the day when household size was closer to 4.

    8 units per acre is close to 5,000 units/sq mi. That means a 40×75 lot, plus 10′ more in the back and 30 in the front for utility easements, alley, street, and sidewalks. The buildable footprint (not lot coverage) with a detached garage would be about 1,500 sf, more if there is a party wall or zero setback on one side.

    This allows enough for large street and yard trees, landscaping, some on-site stormwater infiltration, narrow cross-streets, a wide arterial every few blocks and a neighborhood school and park every few blocks.

    Such a neighborhood would be considered walkable and transit friendly…the busiest buslines in Indianapolis serve neighborhoods of slightly lower density. It would have enough people and incomes to attract business to neighborhood nodes.

    In reality, outside the coasts I think this represents the kind of idealized urban setting many suburbanites would find comfortable…the kind of setting cities must sell as competitors with suburbs for residents and business.

  6. John Morris says:

    Yes, but as places like Garden Grove show, one can have a neighborhood with a density of close to 10,000 with cul-de-sacs and few sidewalks and still not have a walkable place or much transit use.

  7. Chris Barnett says:

    I don’t know, John…it might be relatively inexpensive to retrofit places like Garden Grove. As Cox showed, that is far more the norm in the US.

    Buy out one house (or take easements) at the ends of cul-de-sacs to create bike-ped through-passages and neighborhood greenspace amenities (including rain gardens for stormwater processing).

  8. John Morris says:

    Yes, in theory that’s true but ask yourself what kind of person buys a house on a street without a sidewalk, with a large garage, in a neighborhood with few sidewalks? There is often a pretty big cultural barrier.

    Also, while in theory a place like that has space for lots of little parks, the reality is that space is needed for parking lots.

    I’m not saying one can’t gradually make adapt some of these places, Garden Grove has a good main street grid, but it’s much harder than doing infill construction in the former industrial donut zones.

    Almost all of NYC’s new high density residential has been built in places that had few residents before. Even there strong political barriers are thrown up by existing residents.

  9. Chris Barnett says:

    Suburban neighborhood streets really aren’t that dangerous, and can easily perform as “shared space” when there aren’t sidewalks. I’m not talking about busy arterials, but about the “winding ways” and “manor courts” where only a few dozen people live. If bike-ped cross-connections were created that still prevented through car-traffic, I think those developments would hit suburbanites’ sweet spots better.

  10. Eric Fischer says:

    The meaningful numbers that could have come out of this analysis are that, at the block group level of aggregation, car commute share drops below 50% at 40,000 people per square mile (which is to say, basically, two-flat buildings on 25-foot lots), and, for most of the range, each additional resident per square mile means a .0011% drop in the car commute share.

    I guess Cox has made his point that, from the transportation perspective, forcing someone who wants to live on a gigantic lot to live in the streetcar suburb format is pointless, since it will make no difference in their mode choice. But it still seems ludicrous to refer to 10,000 per square mile as “high density.”

  11. Alon Levy says:

    Three thoughts:

    1. Why use zip code data when there’s more granular census tract data? On the census tract level, Greater New York isn’t 43% of the 10,000/mi^2-and-denser US, but more like 33%. (Actually, 33% of the two-thirds of states that I have data for, but those two-thirds are cherry-picked to have the highest average population per county because I was grabbing data manually by county, so all dense major cities and their suburbs are included. I may be missing some density in Texas, though.)

    2. More granular data, i.e. census tracts, is especially useful for measuring the possibility of rapid transit, which is very spiky. If you try to look at the density of Vancouver even on the census tract level you’ll miss a few residential towers near SkyTrain; if the data is only as granular as American zip codes, you’ll miss nearly all of them.

    3. The problem with these thresholds even on a more granular scale is that they don’t measure everything that’s going on with transit choice. Elizabeth, New Jersey has the same population density as Zurich. Switching to a threshold of 5,000/km^2 because that’s what Canada makes available, Richmond, BC has 16% of its population living in census tracts denser than this, versus 21% of Union County, NJ. But Elizabeth has nothing like the Canada Line or the ridership it generates.

  12. Alon Levy says:

    Eric, I didn’t see your comment when I posted mine, but re “each additional resident per square mile means a .0011% drop in the car commute share,” you’re overinterpreting a weak correlation. I don’t know if the original mistake is due to you (in which case it’s a subtle mistake) or Cox (in which case it may well be deliberate), but when you have a linear correlation that’s not very strong, the regression coefficient is going to be low, and this leads to fallacies of the form “you’ll need to enormously change the independent variable to get the desired change in the dependent variable.” Here is what I said almost two years ago about freeways and congestion, between which there’s even less correlation than between density and transit use:

    [Link] Freeways are not a very strong correlate [of congestion]. The regression coefficient is -233: increasing the number of freeway miles per thousand people by 1 (the range is 0.13-1.4 [in the 101 biggest US metros], with few large metros above 1 or below 0.35) reduces the congestion cost per capita by $233 per year, also uninspiring.

    The regression number alone can be used as a dishonest trick when arguing on the Internet. If we overinterpret weak correlations, we can declare that the only way to decrease congestion is to build an unrealistic number of freeways, and thus declare the problem unsolvable. Of course, for most cities we can find other cities of comparable size with much less congestion and without enormous amounts of asphalt – this is why the correlation is so weak. But a good hack should not bother himself with such caveats to talking points.

  13. I don’t think using zip codes is a totally unreasonably analysis. Urban zip codes are fairly small. I think there’s value in looking at density over some reasonable piece of terrain. Looking at a Census block might pick up pockets of density, but do these really correlate with urban function? If you picked just the building I live in (an apartment building in a converted mill with 250 units), it would probably look spectacularly dense, but I’m not sure exactly what that translates into. Census block is probably useful for some things, but I think zip code level type analysis is also a useful way to look at it.

  14. Eric Fischer says:

    Alon, this is my own conclusion ( from Census 2000 Summary File 3, not using anything of Cox’s. Yes, of course, there are many other factors that determine mode choice, even though there is a trend with density.

    Zip codes are a terrible aggregation unit. Aaron, you know how big 46220 is in Indianapolis, including not just the heart of Broad Ripple but also huge-lot areas off Binford Boulevard? If I had my way, things would be tallied to block faces, not the interior of blocks, but blocks are a lot closer to being meaningful than zip codes are, and arguably better than census tracts too.

  15. Alon Levy says:

    I don’t think it’s totally unreasonable either, it’s just less reasonable than more granular census tracts. None of these is on the same level of obviously ridiculous granularity as looking at the footprint of just one building. A 5,000-person census tract is a couple of blocks, enough to support a few stores and almost enough to support a supermarket. The bigger residential TOD projects around SkyTrain are about that large or slightly larger – that’s why even on that scale you don’t see all of them. On the level of a zip code with 30,000 people, you could see Metrotown but the rest of Burnaby and East Vancouver would look indistinguishable from uniform medium-low density, which doesn’t capture the urban layout here faithfully. On a slightly more granular level than a census tract you’d see the form here pretty well: the areas along the Expo Line have small islands of high density surrounded by low density, whereas the neighborhood I live in has fairly uniform medium-high density.

  16. Eric Fischer says:

    I think I’m going to claim that the best aggregation unit to choose is the one that provides the best correlation for the thing that you’re trying to explain.

    I am surprised that Congressional District (Summary Level 500) density/mode seems to correlate the best of any of the national categories, with R²=0.628. Then comes Census Tract (Summary Level 140, R²=0.461) and Block Group (Summary Level 150, R²=0.399). Metropolitan Statistical Area is almost as good (Summary Level 390, R²=0.398). Zip Codes are far down the list (Summary Level 871, R²=0.144).

  17. Alon Levy says:

    How much of that correlation persists if you remove New York from the sample as an outlier? In Gary Barnes’ weighted density paper the correlation ex-New York is still significant but much lower than the correlation with New York

  18. Eric Fischer says:

    Leaving out New York State lowers it quite a lot. R²=0.325 for congressional districts, 0.271 for tracts, 0.228 for block groups, 0.388 for MSAs, and 0.056 for ZCTAs.

  19. Eric, I think zip codes can be great neighborhood proxies in many cases. In Chicago, for example, there’s a great match of zip code to neighborhood. I suspect that the average density of 46220 tells you more about how Broad Ripple functions than a few select census blocks, even if it does extend pretty far east. (The commercial core of Broad Ripple survives almost entirely on auto “commuters” and is not really a traditional neighborhood commercial node). Similarly 46204 and downtown.

    However, I posted the charts less because of the transit correlation matter, but because of what they show about density in the US. Whatever aggregation level you choose, the general thrust of the conclusion would be the same. The vast majority of high density areas in the US are in New York. And a comparative handful of metros in America contain almost all areas of even modest density.

    But to consider the transit-density connection, I don’t think you would pick the geographic entity with highest correlation measure if determining whether they are correlated is what you want to do. Nor would you pick the lowest one. Otherwise we would just be pulling numbers that best supported our own desired outcome.

    I happen to think zip codes are ok for density measures because I want would want to know if there’s a legitimate dense neighborhood, not just a pocket here or there. When I’ve looked at census tract maps, they don’t appear to be functional entities. Zip codes aren’t perfect, but in big, dense cities they at least cover functional neighborhoods. And in those dense cities (i.e., not Indianapolis), zip codes can be reasonably compact. It depends on your application. For your metro style maps, more granular is probably better since you’re providing a sort of mosaic view.

  20. Eric Fischer says:

    That makes sense. Neither zip codes nor census tracts is really a perfect fit, but either of them works if you just want to categorize the densest ones, and people know their zip codes but not their census tracts. It’s Cox, whose goal is supposedly to understand the correlates of transit ridership, who really needs to care about what unit gives the best correlation.

  21. Alon Levy says:

    But a lot of modern TOD is exactly a pocket here and there. Land values rapidly drop with distance from the train station or whatever else the anchor is. The densest census tract in Metro Vancouver is a small tract near Joyce-Collingwood: the area there is your standard SkyTrain TOD, with a cluster of towers over a low-rise base, and this isn’t the same picture as the triple-deckers in Hyde Park or Somerville even though on the level of a 20,000-person area the density is about the same.

    Absent such extreme TOD, the census tract level is still a unit at which you can analyze issues of walkability. For example, Greater College Hill is just four census tracts (see map here): two correspond almost perfectly to Fox Point and Wayland, and two are the two halves of College Hill proper. And Fox Point is a meaningful neighborhood for small trips. When you adjoin it with the other three census tracts, you get an area that when I lived in Providence I rarely left except for trips to other metro areas.

    Incidentally, on the census tract level, Greater New York’s percentage of 25,000/mi^2-or-denser people in the two thirds of America I checked drops to 65%. A majority, but not an 88% one.

  22. John Morris says:

    Anyway, with it’s flaws, zip code & census tract data is far more granular and useful than MSA. (LA MSA counted as denser than NY)

    I think most developers etc.. know pretty well what matters is how density is distributed near the actual line first. A Forest Hills layout- of apartment buildings anchoring the line is much better than a uniform spread of housing units.

  23. John Morris says:

    @ Chris Barnett

    Garden Grove has no sidewalks on many significant streets. Usually, not the main ones, but often everything else. Many two way through streets with no sidewalks. Overall they pop up and disappear randomly- which puts pedestrians in a dangerous spot.

    The massive parking around every business on the the main drags indicates how few people walk to the store.

  24. Chris Barnett says:

    John, in the grand scheme, putting sidewalks where none exist is not “redevelopment”. It’s fine-tuning.

    I have a close friend who lives in a nice suburban subdivision outside Indy. There are sidewalks on both sides of the street through the whole development. But when you get to the gatehouse where the development joins the (former mile square edge country road) “arterial”, the sidewalks end. You can wind through the development for a couple of miles on sidewalks. You just can’t walk anywhere else.

    When the arterials at both ends are (inevitably) rebuilt, they should (and likely will) be rebuilt with sidewalks or multi-use paths. Fine tuning, not redevelopment.

  25. John Morris says:

    Call it whatever you want. Often the actual costs are small, but the change in culture is important. In the case of Garden Grove one is talking about an overall absence of sidewalks along most streets in a pretty dense area.

    In this case, the total lack of sidewalks makes denser mixed use development along any of the main roads difficult since people and businesses demand the parking.

    All I’m saying is that Cox does make good points about culture that density per square mile numbers don’t show.

    For this reason, the sweet spot in creating TOD is often in areas along or near existing lines with few current residents like the post industrial donuts.

  26. Chris Barnett says:

    My comments have been more about reinforcing existing density and making it a more-livable residential product suited to today’s tastes.

  27. John Morris says:

    First of all, there is no one place that fits “today’s tastes” and regulation ensures that we can never know in many areas what real demand might be.

    Sad to admit but bitter experience shows how hard it can be politically to make mild changes in existing zoning or layouts. Aaron has stated how much opposition pops up in Indy neighborhoods to even small multi family construction.

    Garden Grove while fairly dense seems to have little or no non single family development. It also seems to have never had many sidewalks- and google indicates that few people have even suggested adding them. In that context, the chances are changing that will not be an easy process.

    Amazingly, attitudes like this prevail in large parts of Brooklyn and Queens. Developers wisely have decided that it’s easier to fill the former industrial and business districts with high density mixed use than to add it to existing lower density neighborhoods.

  28. Alon Levy says:

    Don’t say “developers wisely have decided.” Bloomberg has decided, not the developers: he’s the one responsible to the upzoning. And this is because his political incentives are to upzone now. In reality the city’s fast running out of non-residential neighborhoods to redevelop, to the point that it has to deck over railyards at high cost to get more buildable land that’s not too far from Midtown. But the moment of reckoning is going to be after he leaves office, so he doesn’t care.

    Conversely, the biggest chunk of redevelopable non-residential land in the region is in Secaucus, but there’s little TOD there because none of the political leaders in the area is interested. To say nothing of the fact that NJ Transit still thinks of Secaucus as a transfer station rather than as a place people might walk to to get to Manhattan.

    Long run, there’s no way around right-to-build laws. Either put urban growth boundaries everywhere or don’t, it doesn’t matter, but put laws restricting the ability of neighborhoods to turn into housing cartels.

  29. John Morris says:

    I guess that depends on what one calls development. Certainly there are still plenty of underdeveloped property along major transit routes, often in areas like Corona with huge housing shortages and illegal rental markets.

    In spite of housing shortages, political opposition shows up to almost all multifamily construction in the outer boroughs.

    Placing a few new apartment buildings along an existing bus route is still a pretty hard sell.

    Given the current political environment new investments in transit upgrades outside of existing routes seem like a waste.

  30. John Morris says:

    “put laws restricting the ability of neighborhoods to turn into housing cartels.”

    That’s what has happened in NY. They are political cartels.

  31. John Morris says:

    I somewhat disagree that Secaucus represents the only big underdeveloped non residential site in the region. It could be one of the bigger, largely empty sites. (Never got off there)

    IMHO, most of the outlying “downtowns” along the commuter lines offer big opportunities for mixed use constriction- Yonkers, New Rochelle, Norwalk, Stamford and of course Newark. A few, like Patterson have special problems like flooding.

    Sorry, I forgot to mention downtown Jamacia, Queens as a large underdeveloped area with good transit access.

    You correctly identify the core problem which is that suburban commuters feel entitled to these sites as park and ride lots.

  32. John Morris says:

    Oh, what about The Metropark station in North Jersey? It screams out TOD opportunity.

  33. Alon Levy says:

    You’re right that Jamaica, Newark, Paterson, Yonkers, and Stamford are big TOD opportunity. I mention Secaucus because unlike the rest, it has practically no high-value development – it’s just warehouses, of the kind that would’ve been replaced by lofts thirty years ago if they were in Manhattan.

  34. John Morris says:

    Patterson has Huuge flooding issues.

  35. Nathanael says:

    The density chart is right, of course…

    ….but it partly reverses cause and effect. I mean, the primary cause of density is geographical constraints (so, Hawaii will be dense, Wichita won’t). But after that, train service has a densifying character to it.

The Urban State of Mind: Meditations on the City is the first Urbanophile e-book, featuring provocative essays on the key issues facing our cities, including innovation, talent attraction and brain drain, global soft power, sustainability, economic development, and localism. Included are 28 carefully curated essays out of nearly 1,200 posts in the first seven years of the Urbanophile, plus 9 original pieces. It's great for anyone who cares about our cities.

About the Urbanophile


Aaron M. Renn is an opinion-leading urban analyst, consultant, speaker, and writer on a mission to help America’s cities thrive and find sustainable success in the 21st century.

Full Bio


Please email before connecting with me on LinkedIn if we don't already know each other.



Copyright © 2006-2014 Urbanophile, LLC, All Rights Reserved - Click here for copyright information and disclosures