Austin Contrarian

Chronicling downtown's decline

CityLab had an interesting piece a few weeks ago on Sanborn fire insurance maps. These maps were created by Daniel A. Sanborn beginning in the 1800s to help insurance companies map fire risks. The maps were exquisitely detailed catalogs of existing structures, and included the outlines of each building and outbuilding, the location of windows and doors, street names, street and sidewalk widths, building composition and a host of other information, and updated every few years. They were a hand-crafted, pre-digital GIS database. 

There were Sanborn maps for Austin. Here is the east side of the 500 block of Congress from the 1956 Sanborn map. Note the density of businesses. I count 24 separate buildings in this one block. 

 

 

This map portended the future character of downtown development, though. By 1956, the southeastern corner of the block had been converted to an "open deck garage" and drive-thru bank, the eventual fate of many downtown properties.

There’s an interesting quirk to these maps. To save money, Sanborn did not produce totally new maps between 1934 and 1956. Instead, map makers (called "pasters" or "correctors") carefully cut out overlays to paste over the existing maps. The pasters would visit subscribers' offices to paste the slips on top of the old maps. But the paper they used was not thick enough to completely blot out the underlying buildings. They inadvertently created a record of how the building stock changed between 1934 and 1956. 

They tell a story of a drastic reduction in building stock during the Great Depression and early post-war years.

For example, here is the 900 block of Lavaca/Guadalupe, which today is the site of two drive-thru banks. The map documents the destruction of four buildings (numbered 1-4), and their replacement with parking.

A close-up of the southeastern corner clearly shows the ghostly image of a demolished building:

The period was especially hard on Rio Grande. Take the 400 block of Rio Grande, just north of today's 360 condominiums. Nine buildings on this block -- excluding small outbuildings and sheds -- were demolished during the Depression or post-war years and replaced with used car lots.

The damage was not limited to the west side of downtown. The 1400 block of San Jacinto and Brazos, smack next to the Capitol, was almost completely leveled. Depending on how you interpret the map, either eight or nine principal structures were knocked down, turning this into a mostly-vacant block.

(The subsequent development of this block shows that there are worse terminal states than being a parking lot. There are a lot of complaints about all of the new parking plinths downtown, but this is how we used to build parking plinths: impregnable walls with small offices perched, castle-wise, on top.) 

Even where there was not wholesale, block-level destruction, there was a general thinning out of buildings. Take the four blocks meeting at 8th and Lavaca: 

Every block experienced at least one demolition.

This is not to suggest that all of downtown's parking lots date to the Depression and early post-war years. During the 1960s and 1970s, structures continued to be demolished and replaced with parking lots at a disheartening rate. Judging from aerial photos, the early 1980s were the apex of the surface parking lot in downtown Austin. 

But the Sanborn maps document a clear collapse in demand for downtown space, whether for office, retail, industry, or warehousing, before the mid-1950s.  The interesting question is why?

One cause can be ruled out right away: the collapse was not caused by the construction of I-35 along downtown's eastern edge, because I-35 did not open through downtown Austin until 1962. Downtown property owners no doubt cheered the freeway's development, hoping that the new highway would reverse the collapse in demand.

Nor can the suburbanization of office and retail be credibly blamed, because this phenomenon didn't take off until the 1950s. Suburban retail either had not yet been built or was just coming on line by 1956. For example, the Twin Oaks shopping center at Oltorf and South Congress was completed in 1953. The Lamar Plaza shopping center on South Lamar, just one mile south of Town Lake, was completed in 1958.  There is a lag between the loss of tenants and the decision to demolish the building. Demolition is an of despair, a decision an owner makes once he's convinced that the rental value of a building has dropped below the cost of maintaining and paying taxes on the building. Incipient suburbanization would not have lead to such a widespread loss in confidence in downtown rents.

So what's the answer?

Probably the Great Depression. The Great Depression caused a monumental collapse in demand for office and retail space, not just in downtown Austin but in downtowns across the country. Historian Robert Fogelson chronicled the collapse in his excellent book Downtown. For example, by the mid 1930s, the vacancy rate in downtown Detroit was running between 35 and 40 percent. With rents falling and increasingly hard to collect, property owners began demolishing buildings and replacing them with parking lots. Between 1932 and 1940 about 92 million cubic feet of commercial space, close to 10 percent of the total, was demolished in the Chicago Loop, most of it replaced by parking lots and one- or two-story garages. By the early 1940s in Los Angeles, roughly 25 percent of the buildable land was used to store autos -- large chunks of the business district were leveled for more than nine hundred parking lots and garages, with space for more than sixty-five thousand cars. Fogelson explained the phenomenon:

[H]itherto these buildings had been torn down to make room for taller, more up-to-date buildings, which, it was assumed, would make more money. Now, in the depths of the worst depression in the nation’s history, the owners were facing a unique situation, in which, said one real estate appraiser in 1934, 'there is no demand.' Hence the demolition of buildings like the Temple Theater [in Detroit] and their replacement by parking lots or one- and two-story garages, which were commonly referred to as “taxpayers.” The “taxpayers” were as much a legacy of the depression as the “Hoovervilles,” bread lines, soup kitchens, and dance marathons. They symbolized downtown in the 1930s as much as skyscrapers, department stores, and high-rise hotels had in the 1920s.

Downtown Austin likely experienced the trend afflicting other American downtowns. The Great Depression caused the collapse in demand for downtown buildings, resulting in the widespread conversion of buildings to parking lots.

Austin still bears some of those scars today. For example, here is the northwest corner of 9th and Lavaca today: 

This is now the site of a drive-thru bank but mostly remains a parking lot, just as it was in 1956. This should be thought of as a scar of the Great Depression. Change in the built environment usually happens very slowly; damage can take a long time to heal.

 

The Price You Pay: Austin's Affordability X-Factor

Leadership Austin has organized a breakfast panel discussion for tomorrow morning at the Long Center entitled, "The Price You Pay: Austin's Affordability X-Factor." Francie Ferguson, Brian Kelsey and I will be on the panel, moderated by KXAN's Robert Hadlock.

Brian (who has his own firm, Civic Analytics) has posted a blog entry organizing his thoughts for the panel tomorrow. It's chock full of interesting statistics. I don't want to steal his thunder, but I can't help passing this one along:

In 2000, 14% of households in Austin (city) had incomes of at least $100,000. In 2012, it was up to 24%. In 2000, households with incomes of less than $15,000 outnumbered households with incomes of $100,000 or more. In 2012, $100,000+ households outnumbered < $15,000 households by almost two to one. This is a remarkable shift in the city’s demographics in a very short amount of time and can go a long way toward explaining why housing is so expensive, especially in centrally located neighborhoods.

I don't think I'm mischaracterizing him when I say that Brian believes that affordability is mainly an economic development problem (with "economic development" meant in its broadest sense). Brian mainly advocates policies that increase human capital (let's give our young people better schools/education) and income (let's draw better paying jobs).  Or put more simply (perhaps too simply), he thinks the affordability problem is mainly about making people wealthier. He's "skeptical that we can 'build our way' to greater affordability in Austin, especially without an equally ambitious strategy for human capital development–i.e. investing in people so they can afford to live and raise families here."

There's actually a bit of daylight between our views on this, of course. My own view is that unless you allow housing supply to respond elastically to demand, higher wages will be offset (more or less) by higher housing costs. Freezing the housing supply while raising wages will attract higher-skilled workers and push out lower-skilled workers, turning the city's real estate market into a giant sorting machine. See, e.g., San Jose. You'll get a housing market that is increasingly occupied by households making over $100,000 and devoid of households making less than $15,000.

Honestly, there's almost no limit to how expensive you can make a really nice city with really wealthy people if you put your mind to it. We complain about high housing costs, but they're nothing like San Jose's.

If you want rising wages to be broadly shared by the people who already live here, you need to make sure the gains aren't eaten up by rising housing prices. The only way to do that, for the vast majority of people who will never get a subsidized housing unit, is to allow housing supply to respond elastically to rising demand.

New York City is much cheaper than such a dense place ought to be

Charlie Gardner finds an association between metropolitan area weighted density and housing affordability:

Income is more strongly correlated with weighted density than total population, although not dramatically so. However, median home values were even more strongly correlated with weighted density.  The result is that, for cities of equivalent size, the city with the higher weighted density will generally be less affordable in relative terms, even if incomes are higher (for instance, Sacramento is almost twice as dense as similarly-sized and lower-income Kansas City, but is only two-thirds as affordable).

I believe high home prices are generally caused by housing being too scarce rather than housing being built too close together, so I didn't expect weighted density to have much bearing on home affordability. In fact, it doesn't: the association between weighted density and affordability is weak, with weighted density explaining only about 17% of the variation in affordability (the ratio of median home price to median income*) among metropolitan areas. Whatever the culprits for unaffordability, they're mostly something other than density. (Being located in California is a better predictor of high relative housing cost, for example, than having high density. The variation in affordability explained by weighted density drops to 10.5% when California cities are excluded.)

Below is a scatterplot pairing the weighted density of each MSA with the ratio of its median home price to median income. The trend line slopes upward, which means that housing tends to get less affordable as the metropolitan area gets denser. As the scatterplot should make clear, though, weighted density doesn't really have much predictive value, particularly for MSAs with weighted densities less than 5,000-6,000 pssm. (Contrast this with having "San" or "Santa" in the city name; this seems to be a very good predictor of high unaffordability.)

 (click to enlarge). 

One thing that jumps out from the chart is just how much of an outlier the New York City metropolitan area is. It's an outlier in density, of course - it's almost three times as dense as LA and San Francisco. But it's also an outlier in its distance below the trend line.  The New York City metropolitan area is much, much more affordable than the trend line predicts. The trend line predicts that the median home in the New York MSA should cost a bit more than 15 times the median income; in reality, the ratio is under 11. Moreover, the slope of the trend line is itself skewed downward by New York City: if you fitted a trend line to the data ignoring NYC, it would predict a home value-to-income ratio of arond 20 for a city with 31,000 ppsm. 

New York City has horrible land-use regulations. Ed Glaeser et al. estimated a few years ago that regulatory constraints are responsible for at least 50% of the overall unit value in the average owned apartment in Manhattan. Vast swaths of the City have been mummified by historic preservation districts. The City has casually downzoned large chunks of Brooklyn. Many of New York's suburbs are notoriously NIMBY.

Still, it ain't California.

As California's coastal cities prove, a city (with sufficient amenities) can make itself completely unaffordable at any given weighted density. While California has many natural amenities, they're no greater than, say, Honolulu's. Honolulu, of course, is squeezed into a small sliver of land on an island between volcanos and the ocean. (Note that roughly half of New York's metropolitan population is also situated on three islands.) Santa Cruz and San Luis Obispo don't have Honolulu's (or New York's) geographic constraints. What they do have is a nasty set of land-use regulations that are, evidently, just as effective as volacanos, rivers and oceans  in depressing the supply of new housing below the market-clearing level.

*Charlie uses the reciprocal; i.e., median income divided by median home value.  That yields a slightly stronger correlation with weighted density, but the resulting trend line is not as steep.

(NB. Corvalis, another outlier, is not particularly expensive; it just has a very low median income. I assume this means that college students are an unusually high percentage of the population.)

Old Urbanist on weighted density (and other stats)

Charlie Gardner at Old Urbanist has done the tedious but very useful work of collecting a bunch of metropolitan area statistics, running pairwise regressions, and then publishing the correlation coefficients in an elegant little chart. If you are looking for the correlation between, say, 2010 metropolitan area population  and 2010 weighted density (spoiler: it's 0.793.), or the correlation between weighted density and median home value (0.582.), there you go.

He's also got some provocate thoughts on the relationship between weighted density, productivity and affordability.

Census density data back online

I noted the other day that the Census spreadsheet with density data for all U.S. metro areas had some errors and had been taken down. A corrected version has now been re-posted.

Again, do check out the nifty graphing feature that allows comparisons between cities. Here, for example, is a graph comparing the weighted densities of  Portland and Austin at various distances from their repective city halls:

As you can see, Portland is much denser at the very center, Austin is much denser in the 1-2 mile band, and then Portland is generally much denser between the 2-mile and 13 mile marks. (The cities are pretty close between 5-7 miles).

Portland's MSA population-weighted density is roughly a third denser than Austin's. The chart above gives part of the reason -- Portland is denser across most distances. But this chart in conjunction with the last provides the full explanation:

Portland has a huge bulge in population between the 4-mile and 12-mile marks, where it is consistently more dense than Austin. So it's not just that Portland has a large swath that is denser than Austin -- that swath also happens to be where the bulk of the people live.

The 50 densest American metropolitan areas, by weighted density

As reported by the Census Bureau in the spreadsheet here:

A few comments:

1. The Census Bureau went all in on the weighted density metric. I say that because it obviously invested a lot of resources in the weighted density calculations. For example, in order to allow meaningful comparisons between 2000 and 2010 weighted densities, it retabulated the 2000 census tract data using the 2010 census tract boundaries.  In order to calculate the weighted densities at various distances from downtowns, it identified the location of the city hall in the principal city of each metropolitan area and the population centroids of tens of thousands of census tracts. This was not an idle undertaking.

2. I've been pushing weighted density for a few years now, so I'm glad to see the Census Bureau buying into it. I'm also hopeful that since this data is now out there, in readily accessible form, and from an authoritative source, it will actually be used. When researchers investigate relationships between a particular characteristic of an MSA and density, perhaps they'll now use weighted density rather than the useless standard density. (For an example of researchers using standard density, see this paper on urban scaling, as discussed by Wendell Cox.)

3. Most MSAs saw a decline in weighted density. This does not mean most MSAs lost population. On the contrary, all but a handful gained population. Nor does it mean that population fled the central cities for the suburbs. The Bureau reports that, on average, the largest metro areas experienced double-digit percentage growth within 2 miles of their largest city's city hall, and those in the second largest metro area size category also experienced growth, although at a slower rate. Chicago, which saw a sharp drop in weighted density, gained over 48,000 downtown residents. In fact, a metropolitan area can gain population everywhere and still see a drop in weighted density.

4. I previously calculated the weighted densities of the four large Texas MSAs. My 2000 calculations don't match the Bureau's 2000 calculations because, as noted above, it retabulated the 2000 census data using 2010 census tract boundaries. I can't do that. I note that apparently I got the Houston MSA 2010 weighted density wrong; it is now denser than the Dallas MSA rather than significantly less dense. I apologize to all 6,086,895 +/- 2,485 residents of the Houston MSA. Note that El Paso is Texas's densest MSA.

5. Austin barely made the list at number 48. Atlanta did not -- its 2010 MSA weighted density was only 2,173 ppsm.

Austin's weighted density, 2000-2010

One of the counter-intuitive things about weighted density is, well, that it's weighted. To calculate the weighted density of, say, a metropolitan statistical area, you first calculate the density of each census tract and then assign it a weight equal to its share of the total MSA population. A census tract that has 1% of the total population is assigned a "weight" of 0.01, a census tract that has 2% of the total population is assigned a "weight" of 0.02, and so forth. The idea is that a tract with twice the population should have twice as much impact on the final average, because a person chosen at random is twice as likely to live there. (Weighted density is precisely the expected density of the census tract in which a random residents lives.)

This can produce seemingly weird results. One that surprises people is that a city can add people everywhere (and grow denser everywhere) and yet still experience a decline in weighted density.

Austin provides a nice example. The Austin MSA experienced rapid growth during the aughts, growing by 37.3 percent. But its weighted density (as calculated by the Census Bureau) dropped by 3.9%, from 3,259 ppsm to 3,132 ppsm.  

Here is a graph of the change in weighted density by distance from City Hall in downtown Austin. (I made the chart using data taken directly from the Census Bureau's Excel spreadsheet.)

Austin experienced a sharp increase in density in the center of town (0-1 miles), a steady increase in density between the 7-mile and 23-mile points, and an outright (noticeable) decline in density in only three tracts. But even these declines were relatively small.

You might be tempted to infer from this chart that Austin's weighted density rose, but you do not, in fact, have enough information to make that inference. You need to know where the growth took place. So consider this chart, which shows the population within the concentric bands at one-mile intervals from City Hall:

This chart shows that while the part of town nearest City Hall experienced healthy growth, almost all of the growth in the MSA population occurred between the 7-mile and 23-mile marks. Now, if we go back to the weighted-density graph, and draw in a line (brown) to mark the 2000 MSA weighted density, it's easy to see what happened:

Almost all of the MSA population growth occurred in the area between the black lines. But this entire area, in 2000, had below-average density, because it lies below the brown line denoting the 2000 MSA weighted density. And although the healthy population growth during the aughts pushed up its density some, it generally did not push it up to the 2000 average. Because this area now comprised a greater percentage of the total population, it received a larger weight and pulled down the average. This merely reflects the fact that a 2010 resident was more likely to live in the low-density 7-23 mile band than a 2000 resident. The MSA's weighted density accordingly declined even though the MSA added population or retained population almost everywhere.

The density profiles of Austin and Houston

The Census report I linked in the last post has just partial tables and graphs. The Bureau has a backup spreadsheet online that contains data on each metro area. It includes a nifty feature that allows you to compare the population-weighted density profiles of two cities, or even the 2000 and 2010 profiles of the same city.

For example, here is a comparison of the density profiles of the Austin and Houston metropolitan areas, based on 2010 data. Basically, the Bureau calculated the weighted density of all census tracts within 0 miles of the city halls, the weighted density of all census tracts within 1 mile of the city halls, etc., and then plotted the points.

Basically, Houston is denser in the center, Austin is denser at the 1-mile mark (because of UT and West Campus, I'm sure), Houston is slightly denser at the 2-mile and 3-mile marks, and Houston is mostly much denser between  the 4-mile and 27-mile marks.

Update: The Census Bureau has removed the Excel spreadsheet from the link. I can only speculate that it has something to do with the fact that the distance data looked wrong for some cities -- e.g., it showed weighted densities under 1,000 ppsm for Portland, OR between 4 and 14 miles from city hall, which is highly unlikely given the Portland's MSA weighted density of over 4,000 ppsm. Hopefully, they're cleaning up that data.

The Census Bureau embraces weighted density

The Census Bureau has just issued a fascinating report (warning: 130 MB pdf) on patterns of metropolitan change between 2000 and 2010. It's loaded with all sorts of interesting statistics. But I was particularly interested in the fact that it embraces population-weighted density as a relevant metropolitan-area metric. The Bureau's justification for the measure more or less matches what I've argued for a few years:

Overall densities of CBSAs [core based statistical areas] can be heavily affected by the size of the geographic units for which they are calculated. Metropolitan and micropolitan statistical areas are delimited using counties as their basic building blocks, and counties vary greatly across the country in terms of their geographic size. With this in mind, one way of measuring actual residential density is to examine the ratio of population to land area at the scale of the census tract, which—of all the geographic units for which decennial census data are tabulated—is typically the closest in scale to urban and suburban neighborhoods. To gain perspective on the densities at which people live, in addition to overall density, the population-weighted density for 2010 was computed for the United States, inside and outside CBSAs, and CBSA size categories. Details on how these measures were calculated are presented in the “Population-Weighted Density” text box.

Table 3.1 shows how overall and population weighted density measures varied by the population size of CBSAs in 2010. The densities at which most people reside are revealed as much higher when the population-weighted density measure is used. While the overall U.S. density stood at 87 people per square mile, population-weighted density shows that people actually lived at an average of 5,369 people per square mile. Density was more intense and the difference between the two density measures was even larger when looking inside CBSAs, in metro areas, and in metro areas of 5.0 million or more, where people were living at an average of over 13,000 people per square mile. Like overall density, population-weighted density decreased with decreasing CBSA size but only dropped to an average of 617 people per square mile for micro areas with populations less than 50,000. The population-weighted density approach reveals that the areas with people living at the highest density levels—metro areas with 5,000 or more people per square mile—were clustered mainly in California and along the corridor stretching from Boston to Washington. Other very dense metro areas included Chicago, Honolulu, Laredo, Las Vegas, Miami, Milwaukee, and San Juan. Low-density metro areas, on the other hand—those with fewer than 1,000 people per square mile— were generally clustered in the South (Figure 3.1).

The Bureau also included a neat graph comparing the weighted densities of the New York and Los Angeles metro areas at various distances from NYC and LA city halls, which the Bureau used as the central points of the metropolitan areas. The graph illustrates nicely why standard density can be such a misleading description.

 The standard density of the New York City metro area is only slightly higher than that of the Los Angeles metropolitan area (2,826 ppsm vs. 2,646 ppsm). But, as we see, within 17 miles of City Hall, New York is denser -- in most cases, multiple times denser -- than Los Angeles. At around the 17 miles mark, the weighted population density of Los Angeles creeps past New York's, and generally stays higher (although not a lot higher) throughout the rest of the metropolitan area. 

The standard density statistic obscures the sharp difference between these density profiles.  But weighted density does not: just knowing that the weighted density of the New York City metropolitan area is 31,251 ppsm and the weighted density of the Los Angeles area is 12,113 ppsm allows you to predict density profiles that look something like the above.

Standard population density is a useless statistic for metropolitan areas

Standard population density -- total population divided by total land area -- is a useless statistic when applied to metropolitan statistical areas (MSAs).

For starters, it does not tell you anything you really want to know. The population density of a metropolitan area is typical of no particular place within the metropolitan area. Most of any MSA is vacant land because MSAs are made up of counties, not cities; they often include not just rural areas, but entire rural counties. Most of the population is concentrated in a relatively small portion of the land at a relatively high density. But all of that vacant land is added to the denominator, which causes the average density to be much lower than the density of the built-up urban areas -- the density at which most people live -- and much higher than the density of most of the land, which is sparsely settled.

That makes it an uninteresting statistic. It's like averaging the temperature outside on a sweltering summer day with the temperature inside a nicely air-conditioned house. You get a number, but it's not a number that tells you much about either the outdoor or indoor temperature.  

Likewise, we might be interested in the density of the built-up, settled area within a metropolitan area, and we might be interested in the percentage of the metropolitan area that has been developed, but combining the two statistics -- which is what standard population density essentially does -- erases the useful information.

A second problem with the statistic is that it does not permit reliable comparisons between cities. Standard population density is too sensitive to how boundaries are drawn. Consider San Antonio and Austin. The San Antonio MSA had a population in 2010 of 2,142,508 spread over a land area of 7,313 square miles for a standard density of 293 ppsm. The Austin MSA, by contrast, had a 2010 population of 1,716,289 spread over 4,220 square miles for a standard density of 406 ppsm.

So the Austin MSA appears to be much denser than San Antonio. But in what sense is that true? If we look merely at the urbanized areas (the contiguous, built-up urban core -- what you would circle as the "city" in a nighttime satellite photo), San Antonio's urbanized area is a bit denser than Austin's, using either weighted or standard density.  San Antonio's urbanized area contains a larger share of the metropolitan population (82%) than Austin's (79%), so the San Antonio MSA is also more centralized than the Austin MSA.

I submit that there's no meaningful sense in which the Austin metropolitan area is denser than the San Antonio metropolitan area. It's just that the San Antonio MSA was assigned eight counties while the Austin MSA was assigned five, as a result of which San Antonio has more rural counties.  The San Antonio MSA includes sparsely-populated Bandera County (standard density: 26 ppsm), which contains less than 1% of the MSA population. If Bandera County were excluded from San Antonio's MSA, its standard density would jump by more than 10%, from 293 ppsm to 325 ppsm. If rural Atascosa and Medina Counties were also excluded, then the San Antonio MSA's standard density would jump to 511 ppsm, much higher than Austin's.

Atascosa, Bandera and Medina Counties together account for only 5% of the San Antonio MSA population. But they determine whether the San Antonio MSA is 28% less dense than the Austin MSA or 26% more dense. Any statistic that fluctuates so wildly depending on whether such a small and insignificant share of the population is included is not a reliable measure of anything.

Finally, standard population density is practically useless for measuring change within a MSA over time. The reason, again, is that the boundaries are based on county (i.e., political) boundaries, which are static over time. If a metropolitan area is not assigned new counties between one census and the next, then the percentage change in standard density is exactly equal to the percentage change in population (because the denominator -- land area -- does not change). On the other hand, if it is assigned a new county, the MSA inevitably sees a dramatic increase in land area and a plunge in density . . . even if every single county in the MSA adds population (and thus density).

I see even serious urban economists like Ed Glaeser using metropolitan area densities from time to time in regressions. I wish they'd stop. Better to use weighted density or stick to urbanized areas.