» Over the decades, cities change size, but they gain and lose population in varying ways: Some in-town, some on greenfield land. How does that impact our understanding of population change?
Every few months, the U.S. Census releases new data on population change, chronicling the rise and fall of America’s cities, counties, and regions as they grow and shrink. The data are fascinating, bringing us useful insights about migration flows and economic shifts. They also point to fundamental changes in the places Americans live: Houston over Chicago, Phoenix over Philadelphia, and so on. And they produce breathless news reports that emphasize that the fastest-growing places are 15 cities you’ve never heard of.
Yet as data are released and evaluated, the trends as described by the levels of information presented by the Census often fail to directly represent underlying facts about how cities are changing–or they at least do not do so adequately. Comparing the changes in population size in the Birmingham and Buffalo regions, for example, explains very little about the health of their respective center cities. Comparing how the cities of Houston and New York have grown overall tells us little about how their in-town neighborhoods have held up over time.
This post delves into the question of how to measure population growth in urban environments by examining frequently used measures of demographic change and comparing them to alternatives. It is geared toward a discussion of demography rather than transportation, but its implications are important for how we think about cities and their component parts, including transportation. Indeed, as I’ll delve into in this article, the question of what cities are growing and what cities aren’t is at the core of some of the most pressing debates in today’s urban planning–so understanding how a place’s population change is occurring is essential.
Levels of data reporting
The U.S. Census collects data about people, either as a full sample (on decennial years) or using sample-based estimates (through the American Community Survey). These data are aggregated by the Bureau to different geographical levels. Depending on the sample used and the year collected, they are aggregated to the block, block group, tract, place (city), county, metropolitan statistical area (region), and other geographies, which are then used by analysts to make conclusions about the way in which the country’s population is changing.
The typical way for demographers to make comparisons between centers of population is to use regional (MSA) data and to track their changes, as MSAs provide a broad view of a place’s demographics and offer insight into how that place is evolving, regardless of political boundaries. The justification for this approach is based on the fact that boundaries in different places work differently.
For example, the city of Philadelphia is the major center city of its region (and jobs center), and it happens to share borders with Philadelphia County, the major county in the region. On the other hand, while Boston is the largest city in its region, the city of Cambridge next door has a large share of the region’s jobs and many of its dense, urban neighborhoods. Meanwhile, while Boston is in Suffolk County, there are other municipalities also in Suffolk County, and Cambridge is not in Suffolk County. As a result, comparing trends in Philadelphia with Boston as cities or Philadelphia with Suffolk as counties with one another at the national level could result in inappropriate conclusions.
Yet there are significant tradeoffs in using a regional level of analysis, as well. Indeed, every level of analysis has advantages and disadvantages, as summarized by the following table.
|Tract, block group, block||
How, then, can we compare cities across the country with one another? If no Census geography is problem-free, are cross-regional comparisons useless?
The answer is to first determine what it is, exactly, that we are trying to ask. If the question is, for example, which areas of the country are growing most quickly, looking at regional demographics make sense (rather than, say, emphasizing large percentage growth in tiny cities). If the question is whether different political approaches are affecting growth differently, then examining population in cities may be effective.
But if the question is something more complicated, such as how similar neighborhoods in different places around the country are acting, these Census geographies often cease to be relevant. This is particularly important for understanding transportation trends, since the way people move around is often directly related to the physical characteristics of the places people live and work. But cities or regions as a whole rarely respond to this issue because cities are often too big and inadequately uniform.
Components of urban population change
More than just pondering the rather simple (but important!) question of what metropolitan regions are growing or declining most quickly, I wanted to get a better sense of how specific parts of regions changed over time. I wanted to be able to answer questions more relevant to urban transportation patterns, like how downtowns in various cities grow or shrink over time. Downtowns are almost uniformly the places in urban regions with the highest transit, walking, and biking mode shares; their health is indicative of whether a region is moving in the right direction on that front. Similarly, how much are already-developed areas of cities changing over time? This is particularly relevant for understanding the pace of infill development, to determine whether cities are adapting to become denser places, or whether they are focusing on suburban growth instead.
To conduct this analysis, I moved beyond the standard Census geographies and to create more appropriate and nationally comparable methods. Taking as a base the 100 largest U.S. cities in 1960 (many of which, though not all of which, are the same as today’s 100 largest cities), I compared changes in not only (a) the overall population within city boundaries, but also (b) the areas of those cities that were already built up in 1960 (with a density of at least 4,000 people per square mile*), and (c) the areas within 1.5 and 3 miles of city hall, irrespective of whether those areas are within the relevant city or not.
The following four maps illustrate how these geographies look for four representative cities–Las Vegas, Indianapolis, Houston, and New York City. The cities have changed dramatically between 1960 and 2014. Las Vegas, Indianapolis, and Houston increased the area within their city boundaries dramatically through annexation or, in the case of Indianapolis, a merger with the surrounding county. New York City, on the other hand, has the same boundaries, with the exception of some landfill such as at Battery Park City.
The areas that were built up in 1960 also differ considerably between the cities. Whereas most of 1960 Indianapolis had neighborhoods of densities of more than 4,000 people per square mile, less than half of Las Vegas did and Houston’s density was arrayed along corridors emanating from downtown. Finally, while the areas within 3 miles of Houston city hall are entirely within the city, those within 3 miles of Las Vegas and New York city halls include suburban jurisdictions (including parts of New Jersey, in the case of New York City).
When examining just a comparison between changes in population in the city as a whole and those in the neighborhoods that were already built up in 1960, some remarkable trends become apparent.
As the following interactive graph shows (mouse over the graph to get more information; not all cities are shown in the X-axis), very few cities saw significant overall growth between 1960 and 2014 in neighborhoods that were already built up. Houston and San Antonio, which each gained hundreds of thousands of people overall during that period, also each lost more than 100,000 people in their already-built up areas. So did Indianapolis, Columbus, Louisville, and Memphis. What’s surprising is that these are cities often acclaimed for their dramatic growth over the past few decades. Yet their growth has been premised largely on annexation–suburbanization–even as their already-built up cores have declined.
In fact, the average of the 100 largest cities grew by 48 percent overall. Yet the average city also lost 28 percent of its residents within its neighborhoods that were built up in 1960.
Some cities did expand through infill quite dramatically, and Los Angeles is a true outlier on this front, gaining almost 1,000,000 people in areas that were already at least partially built up. Other coastal cities had similar but less dramatic trends, like San Diego, San Jose, Long Beach, Miami, San Francisco, Seattle, Arlington (VA) and Oakland. San Francisco is often singled out as a place where growth is not moving fast enough, yet this chart illustrates that the city is at least as willing to accept infill growth as most others.
Of course, change from 1960 is just one criterion to measure change in urban environment. The following graph illustrates how the built-up neighborhoods in 1960 fared over the next few decades. In some cases, they rebounded from significant declines in the 60s and 70s; in others, their populations have continued to fall. (Note: this paragraph and the following graph added in a post-publishing update.)
Looking at areas within 1.5 and 3 miles of city halls produces equally interesting results. While these areas often overlap with the areas that were built up in 1960, they do not match directly as many downtowns had few residents in 1960 (look at the map of downtown Manhattan above, for example), and they often include communities outside of the city itself.
When looking at these neighborhoods, as shown in the following chart, the overall trend is negative: The preponderance of U.S. cities has lost a significant number of people within 1.5 miles of city hall and between 1.5 and 3 miles of city hall, with Philadelphia, Baltimore, New Orleans, and St. Louis leading the way.
There are some clear exceptions, however: San Jose, Los Angeles, Las Vegas, Miami, Long Beach, Honolulu, Arlington, and Austin each grew dramatically within these core areas. New York City and Chicago grew dramatically (in fact, more than any other cities) very close to their city halls–but they also lost a significant number of people between 1.5 miles and 3 miles of downtown.
It is worth pointing out that these trends have changed over time and that choosing a starting point in 1960 was an arbitrary choice based on the availability of Census tract-level data for that year.
As the following graph illustrates, the population of the neighborhoods within 1.5 miles of city hall in the 100 largest cities has changed dramatically over time, and while the change from 1960 to 2014 is a key indicator, it is not all meaningful. Indeed, it is interesting to point out that of the areas within 1.5 miles of city hall of the 100 largest cities, only 5 grew between 1960 and 1970, but 6 grew between 1970 and 1980, 35 between 1980 and 1990, 51 between 1990 and 2000, and 53 between 2000 and 2014. In other words, while the central areas of most large cities are still less populated than they were in 1960, many have recovered a significant share of their population in the intervening years.
Diverging paths to growth
Of the 15 largest U.S. cities in 2014, 13 grew between 1960 and 2014. Yet of those 15, only 7 grew in the areas within 1.5 miles of city hall, and only 5 grew in their respective neighborhoods that were already built up in 1960. Even New York City, whose growth has been outpaced by just a few cities, has increased in population only in areas that were underdeveloped before from the perspective of residential occupancy, such as on Staten Island, in the financial district, and on land that was reclaimed from the rivers.
Los Angeles is an outlier, seeing stellar growth both overall and through infill development during this period.
Seen alternatively, examine the following chart illustrating how the 100 largest cities in 1960 have changed over time when separated by region (by chance, 1960’s 100 largest cities are roughly evenly distributed between the Midwest, Northeast, South, and West).
What’s intriguing is that though large cities in the South and West grew spectacularly over this period (the median city grew by more than 50 percent overall even as the median cities in the Midwest and Northeast lost people during that period), their infill development–particularly in the South–stalled out. Indeed, the median city in the South, much as in the Midwest, saw its 1960 developed areas decline in population by more than 40 percent; the median city in those two regions also experienced a decline in population in the 1.5 miles closest to city hall of more than 50 percent.
From the perspective of these alternative measurements, cities in the Northeast actually appear closer in trends to those in the West than those in the Midwest and South.
Implications for discussing population data
The U.S. has gained more than 140 million people since 1960, and the growth of its largest cities has at least to some degree corresponded to that; the total population of the 100 largest cities in 1960 grew from 47.5 million then to 57.4 million in 2014. Yet this growth has come largely through annexation and not through infill development or construction downtown, as I’ve noted above. This gives some clue as to why the country’s residents continue to rely on personal automobiles to get around. Overall, this paints a worrying picture about the renaissance that many cities appear to be going through; is it simply a blip on the overall continued suburbanization of the country?
Yet by evaluating growth from a different perspective using the indicators I presented above (surely there are other measures that would also be helpful) suggests that we need a more nuanced look at population growth. It is simultaneously true that Chicago lost the country’s second-largest number of inhabitants between 1960 and 2014 and that the same city gained the second-largest number of inhabitants living within 1.5 miles of city hall. It is simultaneously true that San Antonio gained almost 800,000 people in the city as a whole even as it lost more than 100,000 people in the areas that were already developed in 1960.
Why take these alternative measures of city growth so seriously? They should help us question whether the cities that grew fastest from 1960 to 2014 were Las Vegas, San Jose, and Austin or, alternatively, Los Angeles, Long Beach, and Miami or perhaps New York City, Chicago, and Honolulu. Each tells a different but useful tale about demographic change.
These measures might help us to understand, for example, how it is possible for half-vacant neighborhoods to exist just blocks from central Houston, which is otherwise booming. Or it may help us to understand why that city’s transit ridership has increased by just 5 percent since 1996 even though the city has grown by more than 25 percent since then. And they might help us get a better idea of what cities are truly regenerating their inner-city neighborhoods versus those that are simply gobbling up suburban growth to feed into their growing population counts.
Much of the rhetoric in the urban planning discourse over the past few years has focused on the lack of adequate housing construction in many of our cities’ most-desired communities. This is, no doubt, one of the most pressing issues facing places where supply has not kept up with demand and, as a result, rents have risen out of control.
Cities like San Francisco, notably, are frequently cited for inappropriately preventing growth, whereas places like Houston have been lauded for their unbelievable growth. Yet the data presented above should make us question this argument, or at least make us evaluate whether it conclusions correspond to actual demographic change. How is it that “growing” Houston lost almost 120,000 people in its neighborhoods that were already developed in 1960, while “growth-inhibiting” San Francisco gained more than 70,000 in its similar communities? Context matters.
* I used this density measure because it approximates mid-level suburban density levels.
Image at top: Construction in Los Angeles, from Flickr user fliegender (cc)
30 replies on “Reorienting our discussion of city growth”
Very interesting analysis. Will take a lot more digging into. First reaction: California cities seem to have done well at encouraging infill/downtown growth. San Diego, Long Beach, and Sacramento all pop out. Even Fresno, where the downtown core rivals that of the rust belt in regards to abandoned buildings and empty lots has apparently managed to grow more so than most cities. The question is, is this downtown growth in the California cities a result of policies or preferences?
I’m not sure this isn’t some kind of artifact. Certainly I can think of numerous examples of infill in LA, but it’s such a huge outlier that I have to imagine the result comes from how you define “1960 built-up area”.
Even in 1960, Los Angeles was a constellation of cities – Long Beach, Santa Monica, San Bernardino, Santa Ana, etc. were all far-flung corners of a vast “built up area” with millions of acres of farm fields in between, perfect for sprawl development. Leapfrog development, where developers built large master-planned communities like Reseda in the middle of nowhere, fueled by freeway connections, only enlarged the built up area without being contiguous.
Bravo Yonah! This is in your Top 3 most insightful articles. I will never look at U.S. city population change data the same way again.
I really dislike the use of city hall as the nominal center of the city. In New York, it’s in neither the primary CBD nor close to the geographic center of the city; this is why that 538 analysis of airport connectors gets New York so wrong – it misses entirely which subway lines the AirTrains and LGA buses feed, and where most travelers’ destinations are. In Vancouver, city hall is again not in the primary CBD, although it is geographically more central. In Tel Aviv, the area around city hall is centrally located and is a major entertainment and retail center but not much of a job center. These are quibbles, but when your radius is 1.5 miles, they matter.
I also dislike using fixed distances, in your case 1.5 and 3 miles. The reason is that in small cities, these radii are relatively more central than in large cities. For example, in Chicago, 1.5 miles gets you to the booming neighborhoods just outside the Loop, like the Near North Side. In New York, a larger city, you’re still mostly within the CBD, so there’s less of a residential boom, and it gets even smaller if you check 1.5 miles from Grand Central rather than from Civic Center (in the 1990s, Lower Manhattan had something like 33% residential growth). In smaller cities, 1.5 miles is beyond the radius of early gentrification.
These are fair points! I don’t think that the alternative measurements I’m suggesting are universally useful, much in the same way that I don’t think the standard Census ones are. The point is that we need to be careful about which geographies we’re using while conducting this analysis.
In terms of New York specifically, I evaluated your alternative suggestion: Grand Central Terminal at a distance of 1.5 miles. Here’s a comparison:
___________1960 pop______2014 pop______Change
What’s interesting is that the change is similar nominally, but downtown has certainly grown more quickly percentage-wise. But they’ve both grown more than all other downtowns in the U.S.
I agree that the 1.5- and 3-mile buffers from any single point will always be a problematic approach in itself–especially when you throw very special New York into the mix. That’s one reason why I think a measure look at already-built areas is useful.
It might be useful to use census tracts but not by population already existing in 1960 but rather where street networks existed at the time. Prof Garrick at UCONN found that in small towns in California you could predict collisions by looking at pre-1950 street networks versus post. Older street networks were safer. Of course this takes a lot more research but it’s likely that it would be a better starting point than the somewhat arbitrary distance buffers.
At RA one of my last projects was the “Are We There Yet?” paper in which we developed opportunity areas as a geography. They were places that met a certain block size and population/employment density threshold. If you looked at contiguous areas that had block sizes smaller than a few acres as the norm, it’s likely that you could compare central city growth to more suburban growth over time with out the hangover of distance measures. It might also inform why transit ridership is dropping because these older streetcar suburb created street networks are still the basic building block of transit/bike/walking today.
Remarkable research! well done and thanks for sharing.
Regarding the first interactive graph showing the change in population between 1960 and 2014: Is there anyway to understand how the data for NYC is organized by geographies -ie: what constitutes part of the 1960’s built up areas, and what is not?
Click on the map of New York in the post above. The areas that are hatched had residential densities of at least 4,000 people per square mile in 1960.
I assume you used current city hall locations? I’d be interested to know how population has changed relative to the location of 1960 city hall’s. Many cities moved their city hall’s as part of failed renewal plans. I’m specifically thinking of Dallas where there is much more growth within 1.5 miles of the old city hall than within 1.5 miles of the robocop version of city hall we have now. Austin would show more growth too, I imagine.
Based on your comments, I looked at Dallas with an alternative location right in the middle of the financial district. Here’s the data:
___________1960 pop_____2014 pop____Change
The problem with this argument is that the housing market does not stop at neighborhood or municipal boundaries. True, Houston’s growth has been mostly outward, unlike SF or the Bay Area. This is bad in terms of urbanism, but it works equally well in terms of increasing the housing supply, which is what determines housing prices.
Totally agree with your points–housing costs are related in large part to regional supply. I would suggest these data should make us question, though, the perception that places like San Francisco are uniquely hostile to new population/housing.
Interesting analysis. Some of this is shrinking household size. An existing housing stock simply has fewer children.
It would be interesting, if it isn’t too hard, to see the same thing done with households.
Look at Boston. It has fewer residents than in 1960, and little demolition — but massive housing scarcity.
Very good point, and this can hide some of growth pressures “shrinking” areas are experiencing.
Going from a SAH mom, 3 kids, and husband who works to a DINK couple looks like a massive population decrease. But you’ve ~doubled rush hour traffic.
This is fascinating but I would say using 1960 as a baseline is problematic. I was shocked by the low growth numbers for New York and the conclusion that most of the “growth” happened in non-built-up areas when New York has so few of those to begin with. Then I realized that starting in 1960 catches the massive urban flight that happened in eastern industrial cities in the ’60s and ’70s. Staten Island, which was a destination for white flight, looks like it’s the growth center for the city because all but the most recent growth in Manhattan and Brooklyn is rebound. If you switched your starting point to 1980 I’d be willing to bet your conclusions would be turned on their heads.
Interesting and reasonable point. I have added a graph to the article (“Population change, top 20 cities, for neighborhoods that were developed in 1960”) that should help address this question. It shows how the population of these built-up communities has changed over time. In many cases, the declines of the 60s and 70s have been made up (such as in New York); in other cities, the decline continues.
1950 or 1960 could be the the Census starting date for this article, due to the interrelated impacts of:
• 1946-63, streetcar systems were sabotaged and run down
• 1946-63 was America’s suburban home & auto-ownership boom due to the G.I. Bill
• For 35 months in 1951-53 U.S. manufacturing peaked reducing unemployment to 2.5-3.7%
• 1956 Interstate Highway System construction started
• 1958 the Commercial Jet Age started, leading to suburban airport construction that decimated long distance train patronage from CBD to CBD
• 1960s, White Flight amps up, depopulating older cities that could not annex land
I prefer 1960, because Interstate Highway and International Airport project completions did not cross a “Threshold of Major Impact on American lifestyles” until shortly after 1960.
Is there something wrong with the data for Tacoma? It’s shown as having lost 154,000 people from pre-war areas but appears to be mostly prewar development and has a total population of 203,000.
There is indeed something very odd about the Tacoma data. Of the 100 cities, Tacoma had by far the largest drop in its built-up areas from 1960 to 1970: -58%. I double-checked my figures and saw no discrepancy in the data, so I’m not sure what might be happening here. Any local knowledge of Tacoma?
Tacoma, south of Seattle, is only just recently beginning its own urban revival. Perhaps, from my own knowledge of my region’s history, Tacoma saw greater migration north, considering Seattle had some of the largest involvement in the war effort on the West Coast–Boeing, etc. My guess would also be that, again, until recently, there was no concentrated settlement in Tacoma, and the downtown consists of a few commercial buildings that a small number of people would have commuted to for work, and therefore I would assume that growth simply moved outward. It’s a fairly spread-out city. But developers are now breaking ground on the infamous “mixed-use” sites becoming so popular in Seattle.
Nice work, but there are definitely some shortcomings to your analysis. Taking New York City as an example, for instance, I noticed that much of the “greenfield” development was actually “brownfield redevelopment”—residential conversion of former business, maritime, and industrial districts, such as the Financial District, Battery Park City, the Meatpacking District, and the Williamsburg/Greenpoint/Long Island City waterfront (to be fair, ~350,000 residents were added in then-rural parts of Staten Island, too). When you include these areas, NYC actually gained ~100,000 residents in built-up areas, rather than losing residents. Thus, cities that saw a lot of industrial-to-residential conversion will appear to have sprawled a good deal according to this analysis.
Maybe a similar measure, such as the envelope (i.e., the convex hull) of built-up census tracts in 1960, would be more appropriate?
I absolutely agree! There are many different ways to think about how to measure growth, and if there is any point to this article, it is that no one measure is all-encompassing.
I think that the measure you propose is interesting, though outside of New York, the truth is that there are few cities with gaps between residential areas that would be filled in with the envelope you suggest.
I would also emphasize that one of the benefits of the analysis I’m proposing, rather than one encompassing brownfield land, is that it is in some ways a measure of neighborhood acceptance of additional residents.
Very interesting analysis and congrats on assembling and presenting a complicated dataset in such a clear way!
A few suggestions on improving some of the cross-city insight, especially since your core question is about development patterns, not necessarily residential population counts.
1. As Larry has suggested, a household (really a housing unit) based analysis would eliminate regional variability in household size change, which is highly related to immigration and poverty level.
2. If there is a way to add commercial density into this somehow, that would also address some of the downplaying of some of the revitalized cities (e.g. Washington). The Economic Census might be a possible data source here. After all, the baseline year is a time when the downtown of a city of any size was highly non-residential – I’m not sure there were more than a handful of residences within NYC or LA’s city halls. (this is also a reason why crime rates appear to be higher in urban areas because the daytime population is much higher than the nighttime residential population that is the denominator of the crime rate).
3. Ditto on Alon’s comment on the distance from city hall being arbitrary. Something like the radius that encompasses 50% (or 30% or whatever) of the center city’s 1960 housing units would be more useful, and potentially address my concern in #2 as well. A more dynamic measure (but more complicated) that could define the dense urban core, like the inflection point of density (max of 1st derivative of density by radius) could also get closer to what you’re looking to demonstrate, but would be less standard across cities.
Should you be interested in taking any of these next steps, happy to help.
Adding to Pete Gruett’s point on the starting date of your analysis – using 1960 as a reference point means you catch a lot of the loss of urban residential areas to interstate construction, which are areas permanently lost to population growth. It might be interesting to see what the population change is in areas that are still available to residential growth. Plotting the change in area within cities that have >4000 per sq mile would also be an interesting view on that loss of residential areas which has contributed to the suburbanization of the “urban” populations.
Yonah – it’s great to see these graphics and maps analyzing urban population change at a more granular level. Well done! I used to work at the Census Bureau, and we published some work similar to this (though not a lot, admittedly).
Check out these two papers:
Population Trends in Incorporated Places: 2000 to 2013
Patterns of Metropolitan and Micropolitan Population Change: 2000 to 2010
There’s some ‘proximity to city hall’ and annexation analysis in them that you might find interesting.
I disagree a bit with MSA’s not “abiding by “arbitrary” political boundaries.”
They do, because they are built up out of counties and county equivalents, which are vastly different in different jurisdictions and vary by as much as 3 orders of magnitude in geographical size.
For the most part, this doesn’t cause _too_ much discrepancy, but can in the Western states, California in particular. San Bernadino (and the MSA that includes it) is the worst offender here.
I would propose instead comparing the census-defined urbanized areas.
I would agree when it comes to the largest counties in California which contain large areas of desert. San Bernadino, the largest county,is bigger than many small countries but has one large city and many small towns.
[…] a more detailed discussion of the advantages and disadvantages of various levels of analyses, see this article from Transport […]
Its a demographic bulge and social changes. Falling crime, slowly improving schools, and better race relations among millenials mean suburbs are not the default options for 30 somethings with options the way they were in the 70’s. The same time, people are having smaller families and bigger homes (sqft/resident).
On the other hand, myopia, bias, sclerosis and entrenched interests in government and connected organizations/people, block the kind of massive and inclusive job creating, and equity building redevelopment that Singapore embarked on during the 1960’s-1990’s.
It might be interesting to see what the population change is in areas that are still available to residential growth. I agree with your this quote
” The U.S. has gained more than 140 million people since 1960, and the growth of its largest cities has at least to some degree corresponded to that; the total population of the 100 largest cities in 1960 grew from 47.5 million then to 57.4 million in 2014.”
it works equally well in terms of increasing the housing supply and it is determines housing prices.