How important is proximity to jobs for the poor?

More jobs are close at hand in cities.  And on average the poor live closer to jobs than the non-poor.

One of the most enduring explanations for urban poverty is the “spatial mismatch hypothesis” promulgated by John Kain in the 1960s.  Briefly, the hypothesis holds that as jobs have increasingly suburbanized, job opportunities are moving further and further away from the inner city neighborhoods that house most of the poor. In theory, the fact that jobs are becoming more remote may make them more difficult to get, especially for the unemployed. How important is proximity to getting and keeping a job?

A new Brookings Institution report from Elizabeth Kneebone and Natalie Holmes, The Growing Distance Between People and Jobs  sheds some light on this old question.  Their data show that between 2000 and 2012, jobs generally decentralized in U.S. metropolitan areas, with the result that on average, people live further from jobs than they did a decade ago.  Put another way:  there are fewer jobs within the average commute distance of the typical metropolitan resident.

While job access has diminished for most Americans, the report notes that the declines in job access have been somewhat greater for the poor and for racial and ethnic minorities than for non-poor and white metropolitan residents.  This, in the report’s view, has exacerbated the spatial mismatch between the poor and jobs.

The Kneebone/Holmes findings emphasize the change in job access over time.  As jobs decentralized, the average American had about 7 percent fewer jobs within a typical commuting radius in 2012 than in 2000.  But its illuminating to look at the level of job access.  Certain patterns emerge:

People who live in large metropolitan areas have access to many, many more jobs, than do residents of smaller metropolitan areas.  The typical New Yorker is has just shy of a million  jobs within commuting distance; the typical Memphian, only 150,000.  This is what economists are talking about when they describe “thick” urban labor markets.

Dig deeper, and it turns out that within metropolitan areas, cities have much better job access than suburbs.  We’ve taken the Brookings data for 2012 and computed the relative job accessibility of cities compared to their to suburbs for each of the nation’s 50 largest metro areas.  For example, an average city resident in Charlotte has about 320,000 jobs within typical commuting distance.  The average suburban resident in the Charlotte metro has just 70,000.  (Metro level data are shown in the table below).  This means that a Charlotte city resident has about 4.6 times as many jobs within commuting distance of her home than does her suburban counterpart.  For the typical large metro area, city residents have about 2.4 times as many jobs within commuting distances as their suburban neighbors.  This pattern of higher job accessibility in cities holds for every large metro area in the country–save one:  Las Vegas.

At first this may seem counter-intuitive, but consider:  even though jobs may have been decentralizing, central locations are often better able to access jobs in any part of the region.  Its also the case that despite decentralization, job density–the number of jobs per square mile–still tends to be noticeably higher in urban centers than on the fringe.  Its also interesting to note that the difference in job accessibility between cities and suburbs (+140 percent) dwarfs the average decline in job accessibility (-7%) over the past decade.  While aggregate job accessibility may have decreased slightly, individuals have wide opportunity to influence their access to jobs in every metropolitan area based on whether they choose to live in cities or suburbs.

Perhaps even more surprisingly, on average the poor and ethnic minorities generally are closer to jobs than their white and non-poor counterparts.  We can do the same computation to compare relative job accessibility within each metro area for poor and non-poor populations, and to compare job accessibility for blacks and whites.  Despite job decentralization, and the fact that poorer neighborhoods often themselves support fewer local businesses and jobs, the poor residents of the typical large metropolitan area have about 20 percent more jobs within typical commuting distance than do their non-poor counterparts.  The black residents of large U.S. metropolitan areas are have on average about 50 percent more jobs within typical commuting distance than their white counterparts in the same metropolitan area.  Again, this pattern holds for virtually all large metropolitan areas.  Data showing relative job accessibility for poor and non-poor persons and black and white persons by metropolitan area are shown in the two right hand columns of the table above.

Of course, a pure distance-based measure of job accessibility may not fully reflect the transportation accessibility to particular jobs–especially for poor persons who are disproportionately more likely to not have access to automobiles for commute trips.  But the data show that city residents have strikingly better access to a large number of jobs, and other forms of transportation–transit, cycling and walking–generally work better in cities.  The density and proximity of jobs in cities, plus the availability of transit is one reason why poor persons disproportionately concentrate in cities, according to research by Ed Glaeser and his colleagues.

The very much higher level of physical job accessibility in cities, and the relative proximity that poor people and black Americans enjoy to employment opportunities is a signal that physical employment mismatch is at best only a partial explanation for persistent urban poverty.  Other important barriers, particularly a lack of education, concentrated poverty, and continued discrimination are also important factors.

We’re deeply appreciative of our friends at Brookings undertaking this analysis, and making their methodology and findings accessible and transparent.  The metro-by-metro data they present add a new dimension to our understanding of urban land use and evolving labor markets.  While we strongly encourage everyone to explore this data, we offer an observation. In measuring job accessibility, Kneebone and Holmes chose to use separate and locally customized estimates of local commute distance.  For example, the average intra-metropolitan commute (according to data from the LEHD program) in Houston is 12.2 miles, while in New Orleans it is 6.2 miles.  This means that a big part of the difference in measured job accessibility between these two metropolitan areas reflects the fact the typical commute shed for Houston cover a far larger area than for New Orleans.  While this may be an accurate reflection of typical commuting behavior in each cities, it makes direct comparisons between different metropolitan areas problematic.

On the Road Again

The last few months have witnessed a notable rebound in vehicle miles traveled. The U.S. Department of Transportation reports that for the year ended December, 2014, American’s drove 3.015 trillion miles, up about 1.7 percent from the previous year–the first noticeable increase in driving in more than a decade. The upward trend has led the highway lobby to excitedly claim that “demand on the roadway system is returning to historical trends of increased freight traffic and more overall use of passenger vehicles.”

But is that really the case? What is behind the increase in driving–and do the last few months of data really signal a return to a period of increasing driving?

It’s very clear what’s behind the surge in VMT. The big news of the energy market last year was the collapse in oil prices. Oil that had averaged roughly $100 a barrel for more than five years suddenly dropped to less than $50, taking gasoline prices down with it. According to the Energy Information Administration, the average price of a gallon of regular gas fell from $3.26 in January 2014, to $1.99 in February 2015.

The data for the last calendar year show that the rebound in driving roughly corresponds to the last quarter of the year–the time when gas prices dropped the most. The blue line shows the price of a gallon of gas in dollars, by week, and the orange line shows the percentage change in vehicle miles traveled compared to the same month one year earlier.

It would be extremely surprising if the lower price of gasoline didn’t prompt Americans to drive more. It’s quite clear that the run up in gas prices since 2004 was a major factor in reducing the amount of automobile travel. Academic studies suggest that the short-run elasticity (price responsiveness) of driving is about -0.1 to -0.2, meaning that a 10 percent increase (decrease) in fuel prices will result in a 1-2 percent decrease (increase) in miles driven.

As our colleague Clark Williams-Derry at the Sightline Institute has pointed out, the increase in driving is about what you’d expect given the historic relationship of gas prices and driving. The increase in driving is likely to lead to more traffic congestion and more crashes–externalities of automobile use that have been easing in recent years.

Lower gas prices also influence vehicle purchasing patterns. Sales of light trucks increased about 3.6 percent, year-over-year from December 2012 to December 2013. Truck sales accelerated briskly in 2014, recording a 12.4% increase between December 2013 to December 2014, to a total of more than 9 million light trucks.

The net effect was to slow the rate of improvement in the average fuel economy of newly purchased cars. The University of Michigan computes the sales-weighted average fuel economy of newly purchased cars on a monthly basis. Over the past seven years, Americans have been purchasing progressively more fuel efficient vehicles–average fuel economy of newly purchased cars has increased from 20.8 miles per gallon in model year 2007 to 25.3 miles per gallon last year.

In the last quarter of 2014, the year-over-year rate of improvement in vehicle fuel efficiency declined sharply. For the first three quarters of the year, new cars sold in each month averaged about 0.7 more miles per gallon than cars sold in the same month in 2013. For the last quarter of 2014 (when gas prices were dropping precipitously) the fuel economy of new cars averaged only about 0.3 more miles per gallon that cars sold in the same month in the previous year.

There’s a lot of emphasis put on the raw mileage number–3 trillion miles–but a better way of thinking about whether we’re driving more or less as individuals is to adjust these gross mileage numbers by population. Nearly half of the 1.7 percent increase in miles driven is due simply to having more people. Adjusted for population, the increase isn’t an impressive, about 0.9%. Per capita driving is still well below its 2005 peak– we’re driving about as much as we did in 1998.


You can see the underlying data for this chart here.

The big question going forward is whether this uptick is the harbinger of a reversal in the decade-long decline in driving, or whether it’s just a temporary blip. The evidence so far is simply too fragmentary to draw hard conclusions. But the recent rebound in gas prices–up 18% to $2.36 per gallon, according to the Energy Information Administration– may mean that the boost to driving that was provided by cheap gas is already ebbing.

The Cappuccino Congestion Index


City Observatory, April 1. 2015

A new City Observatory analysis reveals a new and dangerous threat to the nation’s economic productivity: costly and growing coffee congestion.

Yes, there’s another black fluid that’s even more important than oil to the functioning of the U.S. economy: coffee. Because an estimated 100 million of us American workers can’t begin a productive work day without an early morning jolt of caffeine, and because one-third of these coffee drinkers regularly consume espresso drinks, lattes and cappuccinos, there is significant and growing congestion in coffee lines around the country. That’s costing us a lot of money. Consider these facts:

  • Delays waiting in line at the coffee shop for your daily latte, cappuccino or mocha cost U.S. consumers $4 billion every year in lost time;
  • The typical coffee drinker loses more time waiting in line at Starbucks than in traffic congestion;
  • Delays in getting your coffee are likely to increase because our coffee delivery infrastructure isn’t increasing as fast as coffee consumption.

Access to caffeine is provided by the nation’s growing corps of baristas and coffee bars. The largest of these, Starbucks, operates some 12,000 locations in the U.S. alone. Any delay in getting this vital beverage is going to impact a worker’s start time–and perhaps their day’s productivity. It’s true that sometimes, you can walk right up and get the triple espresso you need. Other times, however, you have to wait behind a phalanx ordering double, no-whip mochas with a pump of three different syrups, or an orange-mocha frappuccino. These delays in the coffee line are costly.

To figure out exactly how costly, we’ve applied the “travel time index” created by the Texas Transportation Institute to measure the economic impact of this delay on American coffee drinkers. For more than three decades TTI has used this index to calculate the dollar cost of traffic delays–here we use the same technique to figure the value of “coffee delays.”

The travel time index is the difference in time required for a rush hour commute compared to the same trip in non-congested conditions. According to Inrix, the travel tracking firm, the travel time index for the United States in July 2014 (the latest month for which they’ve released this data) was 7.6, meaning that a commute trip that took 20 minutes in off-peak times would take an additional 91 seconds at the peak hour.

We constructed data on the relationship between customer volume and average service times for a series of Portland area coffee shops.  We used the 95th percentile time of 15 seconds as our estimate of “free flow” ordering conditions—how long it takes to enter the shop and place an order.  In our data-gathering, as the shop became more crowded, customers had to queue up. The time to place orders rose from an average of 30 to 40 seconds, to two to three minutes in “congested” conditions. The following chart shows our estimate of the relationship between customer volume and average wait times.


Following the TTI methodology, we treat any additional time that customers have to spend waiting to place their order beyond what would be required in free flow times (i.e. more than 15 seconds) as delay attributable to coffee congestion.

Based on our observations and of typical coffee shops and other data, we were able to estimate the approximate flow of customers over the course of a day. We regard a typical coffee shop as one that has about 650 transactions daily. While most transactions are for a single consumer, some are for two or more consumers, so we use a consumer per transaction factor of 1.2. This means the typical coffee shop provides beverages (and other items) for about 750 consumers. We estimate the distribution of customers per hour over the course of the day based on overall patterns of hourly traffic, with the busiest times in the morning, and volume tapering off in the afternoon.

We then apply our speed/volume relationship (chart above) to our estimates of hourly volume to estimate the amount of delay experienced by customers in each hour.  When you scale these estimates up to reflect the millions of Americans waiting in line for their needed caffeine each day, the total value of time lost to cappucino congestion costs consumers more than $4 billion annually. (Math below).


This is—of course—our April First commentary, and savvy readers will recognize it is tongue in cheek, but only partly so.  (The data are real, by the way!) The real April Fools Joke here is the application of this same tortured thinking to a description and a diagnosis of the nation’s traffic problems.

The Texas Transportation Institute’s  best estimate is that travel delays cost the average American between one and two minutes on their typical commute trip. While its possible–as we’ve done here–to apply a wage rate to that time and multiply by the total number of Americans to get an impressively large total, its not clear that the few odd minutes here and there have real value. This is why for years, we and others have debunked the TTI report. (The clumping of reported average commute times in the American Community Survey around values ending in “0” and “5” shows Americans don’t have that precise a sense of their average travel time anyhow.)

The “billions and billions” argument used by TTI to describe the cost of traffic congestion is a rhetorical device to generate alarm. The trouble is, when applied to transportation planning it leads to some misleading conclusions. Advocates argue regularly that the “costs of congestion” justify spending added billions in scarce public resources on expanding highways, supposedly to reduce time lost to congestion. There’s just no evidence this works–induced demand from new capacity causes traffic to expand and travel times to continue to lag:  Los Angeles just spent a whopping billion dollars to widen Interstate 405, with no measurable impact on congestion or traffic delays.

No one would expect to Starbucks to build enough locations—and hire enough baristas—so that everyone could enjoy the 15 second order times that you can experience when there’s a lull. Consumers are smart enough to understand that if you want a coffee the same time as everyone else, you’re probably going to have to queue up for a few minutes.

But strangely, when it comes to highways, we don’t recognize the trivially small scale of the expected time savings (a minute or two per person) and we don’t consider a kind of careful cost-benefit analysis that would tell us that very few transportation projects actually generate the kinds of sustained travel time savings that would make them economically worthwhile.

Ponder that as you wait in line for your cappuccino.  We’ll be just ahead of you ordering a double-espresso macchiato (and holding a stopwatch).

Want to know more?

Here’s the math:  We estimate that a peak times (around 10am) the typical Starbucks makes about 100 transactions, representing about 120 customers.  The average wait time is about two and one-half minutes–of which about two minutes and 15 second represents delay, compared to free flow conditions.  We make a similar computation for each hour of the day (customers are fewer and delays shorter at other hours).  Collectively customers at an typical store experience about 21 person hours of delay per day (that’s an average of a little over 90 seconds per customer).  We monetize the value of this delay at $15 per hour, and multiply it by 365 days and 12,000 Starbucks stores.  Since Starbucks represents about 35 percent of all coffee shops in the US, we scale this up to get a total value of time lost to coffee service delays of slightly more than $4 billion.

Misleading Medians & the McMansion Mirage

A story published by the Washington Post’s Wonkblog last week made the headline claim that “The McMansion is back, and bigger than ever.”  The article says that new homes are an average of 1,000 feet larger than in 1982, and that the “death of the McMansion” has been highly exaggerated, as have claims that development is shifting to smaller, more urban and more walkable development. The Wonkblog article echoes an 2014 post in CityLab –“The Increasingly Bloated American Dream”–which claimed that “American homes are getting bigger and bigger.”

While the data seem to superficially support this argument, a closer reading shows that the apparent surge in McMansions is actually a bit of a statistical mirage. These analysts have overlooked a key limitation of the reported data. It’s actually the case that American homes are only getting bigger if one believes that people living in multi-family housing either aren’t Americans or don’t have homes.

If instead of looking at the median, we look at the actual number of houses built, a different story emerges. As with all single-family housing, the market for big houses remains depressed—housing starts of 4,000 square feet or more are down 59 percent from the peak and are lower now than they were in 2001.  Homebuilders built 137,000 of these huge homes in 2006, but only 56,000 in 2013, according to the Census Bureau.

The only reason these big houses have increased as a share of total new housing is because the market for affordable, smaller single family homes has done even worse. The smaller yet still catastrophic decline in McMansions is hardly evidence of a growing, or even a continuing consumer love-affair with big houses.

Medians are funny measures—they’re highly dependent on the composition of the population being measured. If the housing market were so bad that only Bill Gates had the wherewithal to build a house, the “median” new home would balloon to 66,000 square feet (the size of his Lake Washington mansion). While that’s an extreme example, that’s the kind of thing that has happened to the U.S. housing market since the bubble days of last decade.

When the housing market collapsed, the bottom fell out. The big decline has been in smaller houses. The apparent popularity of the McMansion is a statistical artifact of the misleading median in a still very depressed housing sector. If anything, the rising median size of new homes is more a testament to the continued growth of income inequality in the U.S., coupled with tougher (i.e. more realistic) lending standards by banks.

This becomes apparent when you look at the actual number of new houses built in the U.S. The growth in the share of new single family homes is not due to some burgeoning increase in the demand for McMansions—rather, it represented the bottom falling out of market for single-family homes. Since the housing bubble peaked in 2007, single-family housing construction is down 66 percent. The construction of 4,000 square foot and larger homes—the McMansions—is down 59 percent. Smaller single-family homes under 1,800 square feet are down 75 percent. Meanwhile, the number of multi-family homes constructed has been increasing steadily, and is now back to pre-recession levels. Multi-family housing now makes up 40 percent of new home starts, up from 20 percent a decade ago. If we recalculated the median new home size including both multi- and single-family homes, the increase in the McMansion share would look much smaller.

We’re far from having what by historical standards would be considered a “healthy” housing market. Total housing constructed over the past five years is lower than any five-year period in the past 50 years. Does anyone believe that if the single-family housing market boomed back to 1.5 million housing starts, that the demand would come proportionately from McMansions? Of course not: the only way to get unit growth in single family housing is by getting households of more modest means back into homeownership—if that ever happens. They will be buying smaller houses.

Unlike the old days of NINJA (no income, no job or assets) lending, where even those with poor credit could qualify for loans, today’s credit standards are much higher. The other key factor has been the demise of the trade-up market. Because most people buy their new homes in significant part with the accumulated appreciation on their existing home, the decline in home values meant that very few middle-income households were in any position to trade-up in the real estate market.

There’s another problem with this median measure: it only looks at single-family housing, not all housing. The one bright spot in the housing market is not in single-family homes, but in multi-family units. By excluding the smaller multi-family homes, this automatically biases the median measure upward.

So in large measure, the only healthy segment of the single-family market is for those with very high incomes. Even here, “health” is a relative thing. Compared to the peak of the housing bubble years, sales of McMansions were lower in 2013 than any year since 2001.

If anything, the growth of the median size of new houses is evidence of the continued and growing impact of income inequality. With growth in incomes occurring mostly among those with the highest incomes, it figures that to the extent there is demand for housing, it’s coming disproportionately from those in the highest income brackets who can afford larger homes, and who qualify for credit.

An accurate measure of the popularity of McMansions would look at the extent to which high-income households are buying large new houses. We don’t have a good annual public data series on wealth by household, but a number of private firms estimate the number of high-net-worth households that form the market for these very large single-family homes. The Spectrem Group has estimated the number of U.S. households with net financial worth of $5 million or more (exclusive of the value of their principal home). By their reckoning there are about 1.24 million such households in the U.S. The number fluctuates from year to year, chiefly due to changes in financial markets.

We can get a good contemporaneous gauge of the popularity of McMansions by dividing the number of new 4,000 plus square foot homes sold by the number of households with a net worth of $5 million or more: call it the McMansion/Multi-Millionaire ratio. (There’s no universally accepted definition of McMansion, but since the Census Bureau reports the number of newly completed single-family homes of 4,000 square feet or larger, most researchers take this as a proxy for these over-sized homes.)

The McMansion to Multi-Millionaire ratio started at about 12.5 in 2001 (the oldest year in the current Census home size series)—meaning that the market built 12 new 4,000 square foot-plus homes for every 1,000 households with a net worth of $5 million or more. The ratio fluctuated over the following few years, and was at 12.0 in 2006—the height of the housing bubble. The ratio declined sharply thereafter as housing and financial markets crashed.

Even though the number of high-net-worth households has been increasing briskly in recent years (it’s now at a new high), the rebound in McMansions has been tepid (still down 59 percent from the peak, as noted earlier). The result is that the McMansion/Multi-Millionaire ratio is still at 4.5–very near its lowest point. Relative to the number of high-net-worth households, we’re building only about a third as many McMansions as we did 5 or 10 years ago. These data suggest that even among the top one or two percent, there’s a much-reduced interest in super-large houses.

There are a couple of key lessons here for thinking about the state of the U.S. housing market. Don’t be fooled by the misleading median, and don’t overlook the big rebound in multi-family housing.

Twenty-somethings are choosing cities. Really.

Over at 538, Ben Casselman offers up a provocative, contrarian article “Think Millennials prefer cities?  Think Again.” He claims that newly released census data show that, contrary to the “all the hipsters are moving to cities” meme, millennials–like previous generations–are actually migrating towards the suburbs.

This is a case where we think the usually reliable 538 gets it wrong.

Here’s the key problem:  Casselman’s data looks only the subset of migration between suburbs and cities in metropolitan areas–that is he only counts people who move from the suburbs of a metro to the principal city of a metro (and vice versa). He ignores the people that move to city centers from the city centers of other metropolitan areas, from non-metropolitan areas and from abroad. So Casselman’s tabulation only looks at whether people are moving from Scarsdale or Bethesda to Brooklyn (or vice versa).  A young adult moving from the central city of another metro (like Washington DC or Portland to Brooklyn) or from a rural area or another country doesn’t count in this tabulation. As it turns out, this makes a big difference.

To get a more comprehensive picture of migration, we’ve pulled together data showing all the twenty-something migrants to principal cities and all the migrants to suburbs, and classified them by place of origin (where they lived in the previous year).  The top panel of our table shows the complete data on moves to cities and suburbs; the bottom panel presents an addenda showing only the data on city-suburb moves that 538 used.  (Like Casselman, we’ve excluded city-to-city and suburb-to-suburb moves within a metropolitan area, and non-metro to non-metro moves).

Movers to Principal Cities and to Suburbs, 2013-14

     Age Group
Destination of Move Origin of Move 20-24 25-29
To principal city From own metro suburb 436 302
From other metro suburb 118 124
From other metro principal city 277 331
From non-metro area 107 82
From abroad 104 106
All moves to principal cities 1,042 945
To suburb From own metro principal city 479 390
From other metro principal city 242 138
From other metro suburb 136 193
From non-metro area 109 74
From abroad 56 85
All moves to suburbs 1,022 880
Net migration suburb to principal city 20 65
Addenda:  City-Suburb/Suburb-City Moves Only (538 Analysis)
To principal city From own metro suburb 436 302
From other metro suburb 118 124
Suburb-to-city moves 554 426
To suburb From own principal city 479 390
From other metro principal city 242 138
City-to-suburb moves 721 528
Net migration suburb to principal city -167 -102

Source:  Current Population Survey, 2014.  Table 16.  Metropolitan Mobility, by Sex, Age, Race and Hispanic Origin, Relationship to Householder, Educational Attainment, Marital Status, Nativity, Tenure, and Poverty Status:  2013 to 2014.  Numbers in Thousands.

These data show that there was actually a net inflow of about 85,000 20 to 29 year-olds into principal cities in 2014, in contrast to Casselman’s data showing a net outflow of more than one quarter million. The difference stems from the fact that young adults moving into a metropolitan area from some other metro, or a non-metro area, or from abroad, were much more likely to live in the principal city than young adults moving within a metropolitan area.

Let’s focus for a moment on 25 to 29 year-olds moving into principal cities. It turns out that more of them come from principal cities in other metropolitan areas (331,000) than move to the principal city from suburbs in the same metropolitan area (302,000). So inter-metropolitan moves are actually more important to this demographic shift than are within metro moves.  Also notice that principal city residents moving to a different metro are about two and a half times as likely to move to a principal city in that new metro as they are to move to a suburb in the new metro–331,000 residents of principal cities in other metros moved to principal cities in a new metro; only 138,000 residents of principal cities in other metros moved to suburbs in a new metro.  Among migrants from other metropolitan areas, only those who previously lived in suburbs were more likely to move to the suburb in a new metro (and this group was far less numerous). Movers from non-metro areas and from abroad were more likely to move to the principal city than to its suburbs.

The 538 result is skewed by the fact that principal city residents are much more likely to move, period, than are suburban residents.  Among 25 to 29 year olds living in principal cities in 2013, about 9.7% moved compared to only about 6.2% of 25 to 29 year-olds living in suburbs.  Fewer suburban residents move to cities simply because fewer suburban residents move anywhere.

By looking at only a subset of movers, 538 misses several important sources of migration of young adults to central cities.  This is important because central cities often serve as a kind of port-of-entry or “Ellis Island” in metropolitan areas.  New migrants to a region from other metropolitan areas other states and other countries seem disproportionately to settle, at least initially, in central city locations.  Its also the case that better educated workers are more likely to make longer moves, and move between states and to different metropolitan areas.  As a result, city centers are disproportionately attracting well-educated young adults.  Our data show that between 2000 and 2012, the number of 25 to 34 year-olds increased twice as fast within 3 miles of the center of the central business district in the 51 largest metropolitan areas as it did outside that circle.

There’s an important technical limitation to using municipal boundaries of the largest city to separate metros into “city” and “suburb.”  Principal cities vary widely in how much of a metro area they cover.  Some like Boston and Miami are a small fraction of the urban area; others municipalities like Jacksonville and San Antonio, encompass vast swaths of low density development.  At City Observatory, we strongly prefer using radius-based measures for making metropolitan comparisons. Unfortunately, CPS migration data aren’t available at the finer geographic detail needed to perform this kind of analysis.

It must also be said that 538 has set up a bit of a straw man:  the point is not that all millennials want to live or are living in cities. The point is that preferences have demonstrably changed in favor of cities. The migration patterns of young adults today are very different from those we observed just a decade or two ago. Looking at aggregate population data–not just year-to-year moves–we noted that the probability that a 25 to 34 year old lived in a close-in urban neighborhood, relative to all metro residents quadrupled from 1990 to 2010.

In our view, Ben Casselman glosses over the really critical point about changing migration patterns:

Millennials are moving to the suburbs at a much lower rate than past generations did at the same age. In the mid-1990s, people ages 25 to 29 were twice as likely to move from the city to the suburbs as vice versa. Today, they’re only about a quarter more likely.

That’s a big change.  And where people move in their 20s is important because the probability of migration falls precipitously with age:  a 35 year-old is roughly half as likely to move as a 25 year-old, and that probability declines steadily with age. If principal cities are doing a better job of attracting people in their 20s, it has major ramifications for future city population and economic growth.  City population change is highly sensitive to relatively minor changes in the probability and duration of city residence of young adults:  even if they move to the suburbs as they age, the growing proportion and longer tenure of young adults in cities has a measurable and continuing impact on city demographics.

Young adults are highly mobile:  they’re voting with their feet for the kinds of metropolitan areas and neighborhoods they want to live in.  When you look at the entire sample of movers to cities and suburbs–and don’t arbitrarily narrow the analysis–the data show that young adults, especially the most well-educated, are increasingly choosing cities.


Has the Tide Turned?

Last month, City Observatory released a new report—Surging City Center Job Growth—chronicling a widespread rebound in city center jobs. For the first time in decades, job growth in city centers around the country has surpassed the rate of job growth in peripheral areas.

In an article called “Fool for the City,” Jacob Anbinder of The Week responded to recent media reports about the return of city centers, commenting that the issue may have been over-hyped in the media. You might think that a publication that bills itself as “All you need to know about everything that matters”  might be a little bit more reticent in accusing others of hyperbole. Just the same, let’s take a minute to address the points raised in this article, and allay Mr. Anbinder’s fears.

As we described in our report, what’s remarkable about this trend is how it runs counter to the decades-long pattern of job decentralization. In analyzing data reaching back to the 1940s, we showed the steady ebbing of the relative economic importance of city centers. The message here isn’t a new era of urban triumphalism, so much as it is the end of a long period of unabated decentralization.

As a reminder, here are the key data points from our report:

We were careful in our work to flag the importance of the industrial composition of employment change through the business cycle on the observed patterns of job losses and gains. There’s no question that cities benefited from the strength of centralized industries, like professional services and finance, relative to the weakness of more decentralized industries like manufacturing, construction, and distribution. (In addition, contrary to the implication of the article that city center results were driven by government employment, our data excluded public administration employment.) But even after controlling for these industry variations, we showed that city centers had recorded a significant gain in their competitive position vis-a-vis suburbs.

The Week points out that for the entire nine-year period under consideration, many city centers were still below their 2001 level of employment. There is no question that the earlier 2002-07 period was a continuation of the historical trend, and that nearly all cities were then losing share of total metro employment. The key point in our study is that during the last four years for which we have data, the trend is quite different. (It is worth noting that 2002-07 was the height of the housing bubble and the peak of ex-urban development and job decentralization.) It’s hardly remarkable that most city centers didn’t grow fast enough in the four years coinciding with a very weak national economy to offset the relative decline they endured in the previous five.

Our report was quite clear that this pattern of city center revival isn’t universal. In the 2002-07 period, seven of 41 metropolitan areas outperformed their peripheries; in the 2007-11 period, 21 outperformed their peripheries. (21 of 41 is not an overwhelming majority; it is, however, a much bigger group than the 7 cities that saw this pattern in the previous time period.) As in politics, all economic geography is local: some city centers continue to follow the historic pattern of having growth that lags well behind their expanding suburban peripheries. We noted that job decentralization is still the order of the day in places like sprawling Houston and Kansas City.

It’s not surprising to find that the nascent urban comeback is happening faster in some places than in others: centralized employment did well in New York and San Francisco in the earlier 2002-07 period, when nearly all other city centers were lagging well-behind their peripheries in job growth. More analysis is needed to discern what sets of factors—industry mix, local policies, population movement—are at work in each metropolitan area.

There’s undoubtedly a lot to be learned from a closer, city-by-city and industry-by-industry examination of the data. This is the reason we published our report, and also why we’ve made our data available for others to download and analyze. Moreover, the underlying source of data, the Census Bureau’s Local Employment and Housing Dynamics series, is a powerful, yet under-used source of insight into the economic processes at work in our nation’s cities. We hope others will mine this data to generate an even richer picture of the changing geography of urban employment.

As we stressed in our report, four years of data drawn from a particularly turbulent time in our economic history is hardly the final word. We’re eager to see more recent data—the 2012 to 2014 data should be released by the Census Bureau later this year. When they are, we’ll be better able to judge whether the changes we’ve recorded in the past few years are a cross-current or a true turning of the tide.

What does it mean to be a “Smart City?”

The growing appreciation of the importance of cities, especially by leaders in business and science, is much appreciated and long overdue.  Many have embraced the Smart City banner.  But it seems each observer defines “city” in the image of their own profession.  CEOs of IT firms say that cities are “a system of systems” and visualize the city as an increasing and dense flow of information to be optimized.  Physicists have modeled cities and observed relationships between city scale and activity, treating city residents as atoms and describing cities as conforming to “laws.”

In part, these metaphors reflect reality.  In their function, cities have information flows and physical systems.  However, it is something more than its information flows and physical systems, and its citizens need to be viewed as something other than mindless atoms.

The prescriptions that flow from partial and incomplete metaphors for understanding cities can lead us in the wrong direction if we are not careful.  The painful lessons of seven decades of highway building in U.S. cities is a case in point.  Epitomized by the master builder, Robert Moses, we took an engineering view of cities, one in which we needed to optimize our cities to facilitate the flow of automobiles.  The massive investments in freeways (and the re-writing of laws and culture on the use of the right of way) in a narrow way made cities safe for much greater and faster travel–but at the same time they produced massive sprawl, decentralization and longer journeys, and eviscerated many previously robust city neighborhoods.

If we’re really to understand and appreciate cities, especially smart cities, our focus has to be elsewhere:  it has to be on people.  Cities are about people, and particularly about the way they bring people together.  We are a social species, and cities serve to create the physical venues for interaction that generate innovation, art, culture, and economic activity.

What does it mean for a city to be smart?

The most fundamental way a city can be smart is to have highly skilled, well-educated residents.  We know that this matters decisively for city success.  We can explain fully 60% of the variation of economic performance across large U.S. metropolitan areas by knowing what fraction of the adult population has attained a four-year college degree.  There’s strong evidence that the positive effects of greater education are  social–it spills over to all residents, regardless of their individual education.

Educational attainment is a powerful proxy measure of city economic success because having a smart population and workforce is essential to generating the new ideas that cause people and businesses to prosper.

So building a smart city isn’t really about using technology to optimize the efficiency of the city’s physical sub-systems.  There’s no evidence that the relative efficiency of water delivery, power supply, or transportation across cities has anywhere near as strong an effect on their success over time as does education.

It is in this process of creating new ideas that cities excel.  They are R&D facilities and incubators, and not just of new businesses, but of art, music, culture, fashion trends, and all manner of social activity.  In the process Jane Jacobs so compelling described, by juxtaposing diverse people in close proximity, cities produce the serendipitous interactions that generate what she called new work.

We don’t have an exacting recipe for how this happens.  But we do know some of the elements that are essential.  They include density, diversity, design, discovery and democracy.

Density. The concentration of people in a particular place.  Cities, as Ed Glaeser puts it, are the absence of space between people.  The less space, the more people, and the greater the opportunities for interaction.  Cities are not formless blobs; what happens in the center–the nucleus–matters, because it is the place that provides key elements of identity and structure and connection for the remainder of the metropolitan area it anchors.

Diversity. The range of different types of people in a place.  We have abundant evidence that the diversity of the population– by age, race, national origin, political outlook,and other qualities– helps provide a fertile ground for combining and recombining ideas in novel ways.

Design.  We are becoming increasingly aware that how we populate and arrange the physical character of cities matters greatly.  The arrangement and aesthetic of buildings, public spaces, streetscapes and neighborhoods matters profoundly for whether people embrace cities or abandon them.  We have a growing appreciation for urban spaces that provide interesting variety and are oriented to walking and “hanging out.”

Discovery.  Cities are not machines; citizens are not atoms.  The city is an evolving organism, that is at once host to, and is constantly being reinvented by, its citizen inhabitants.  A part of the attraction of cities is their ability to inspire, incubate, and adapt to change.  Cities that work well stimulate the creativity of their inhabitants, and also present them all with new opportunities to learn, discover, and improve.

Democracy.  The “mayor as CEO” is a tantalizing analogy for both mayors and CEOs; CEOs are used to wielding unitary, executive authority over their organizations; many mayors wish they could do the same.  But cities are ultimately very decentralized, small “d” democratic entities.  Decision-making is highly devolved, and the opportunities for top-down implementation are typically limited.  Citizens have voice (through voting) and the opportunity to “exit” by moving, appropriately limiting unilateral edicts.  Cities also give rise to new ideas, and when they work well, city political systems are permeable to the changing needs and values of their citizens– this is when many important changes bubble up.

All of these attributes of cities are susceptible, at least in part to analysis or description using the constructs of “information flows” or “systems of systems.”  They may be augmented and improved by better or more widespread information technology. But it would be a mistake to assume that any of them are capable of being fully captured in these terms, no matter how tempting or familiar the analogy.

Ultimately, when we talk about smart cities, we should keep firmly in mind that they are fundamentally about people; they are about smart people, and creating the opportunity for people to interact.  If we continuously validate our plans against this key observation, we can do much to make cities smarter, and help them address important national and global challenges.

Who’s Vulnerable to Retail Retrenchment?

This week comes news that Target is laying off 1,700 workers at its Minneapolis headquarters, looking to become leaner and more efficient. It’s just the latest move in a shifting retail landscape in the United States.

Target is not just downsizing its headquarters, it’s shifting to smaller urban stores–Target Express. Other retailers like Walmart and Office Depot have have also been developing smaller stores. The days of big boxes and power centers seem to be giving way to to more urban-centered and smaller-footprint retailing, undermining the economics of larger-scale retailing. It’s estimated that there are over 1,200 dead or dying malls in the U.S. It appears that we’re way overbuilt for retail space. Finding productive uses for these disused spaces is now a major undertaking for communities around the nation.

Several factors seem to be driving the tectonic shifts in retailing. Part of the problem is that retail, like housing,was overbuilt during the bubble: commercial developers typically followed new housing development, and as the housing stock sprawled in the last decade, so too did the expansion of retail space.

Another important factor is the technological change in the form of growing e-commerce. More and more, we’re purchasing goods and services via the Internet and mobile devices. According to data compiled by Erik Brynjolfsson, e-commerce now accounts for about 30 percent of non-food, non-auto retailing, and is continuing to grow:


There’s a bit of irony here: big box stores only become economically feasible thanks to earlier technological advances, including universal product codes, computerized inventory management, real-time ordering, and global data networks. These same technologies now help enable smaller stores (tailoring inventory to localized demand) and empower consumers to order online at home and via pervasive mobile devices.

The shifting retail environment will have impacts on the transportation system as well. The latest transportation data show a decline in the number and length of shopping trips (which decreases transport intensity of retailing), but this is at least partially offset by more travel by commercial delivery vehicles (like UPS and Fedx). It’s an open question as to how this will play out: will these shifts encourage (more) fleets of smaller transit trucks, or will increasing e-commerce retail sales and smaller urban stores mean larger trucks on urban roads? (Regardless, the D.O.T. believes e-commerce will significantly impact our road infrastructure by 2045, and that despite the hopes of Jeff Bezos, drones may not help solve that any time soon.)

To judge who’s most likely to be affected by these trends, we compiled some metropolitan level data on the amount of retail space per capita. The data come from Co-Star, a private firm that tracks retail space leasing throughout the nation. (They helpfully make their market reports available here). These data are for 2007 and we’ve computed retail space per capita in each market by dividing total square footage by each metropolitan area’s 2007 population.

The national average is about 46 square feet of retail space per capita, with most metropolitan areas having between 40 and 55 square feet per capita. There are a number of outliers, however.

Milwaukee/Madison has the highest amount of retail space per capita, and many southern, sprawled metros rank higher on this metric as well. These are the places most likely to struggle with a dwindling appetite for retail space, and the economic consequences that follow, be it in fewer retail jobs, large swathes of unused space, or transportation costs. At the other end of the spectrum, some metropolitan areas have far more space-efficient retailing: Portland has just 30 square feet of retail space per capita, fully one-third less than the national average.

By global standards, the U.S. has much more space devoted to retailing than anyone else: comparable estimates for other countries include: 23 square feet per capita in the United Kingdom, 13 square feet per capita in Canada, and 6.5 square feet per capita in Australia. If the experience of these countries is any indication, it’s a good bet that there’s lots there’s still lots of room for downsizing in the U.S. retail sector. However, despite these trends, Miami apparently isn’t concerned.

Florida’s Biotech Bet

For more than a decade, one of the hottest trends in economic development has been pursuing biotechnology. Cities and states around the nation have made considerable investments in biotech research, ranging from California’s voter-approved $3 billion research program, to smaller efforts in cities around the country, including Indianapolis, St. Louis, and Phoenix.

One of the states that made the biggest bets on biotech was Florida, which in 2003 committed state funds to luring the Scripps Research Institute to building a new campus in Palm Beach County. The Scripps deal served as a template for subsidies to other life sciences research institutions opening similar research labs in other Florida cities. The total cost of the program is estimated to reach more than $800 million.

In a new article, Reuters has questioned whether Florida has gotten its money’s worth for the investments it has made in biotechnology. The biotech investments were originally sold based on the promise that they would lead to a flourishing new industry employing more than 44,000 people. But a decade later, there’s little evidence of progress.

We’ve long followed the biotechnology industry. In 2002, my colleague Heike Mayer and I undertook an extensive study of the clustering of the US biotech industry, published by the Brookings Institution–Signs of LIfe–which showed that the industry’s economic impact was tightly concentrated in just a few leading centers around the nation. While life sciences research was becoming slightly more widespread as more cities competed for National Institutes of Health (NIH) funding, all of the measures of commercialization–new firm startups, venture capital investment, and privately funded research and development partnerships– were becoming more concentrated in a few leading cities. Our analysis showed that three biotech leaders (Boston, San Francisco, and San Diego) had decisive competitive advantages in starting and growing new biotech firms that other cities would find difficult, if not impossible, to overcome.

The succeeding decade has confirmed our original analysis. The three leading centers are even more dominant today that they were a decade ago. In 2000-01, Boston, San Francisco, and San Diego accounted for about 54 percent of venture capital invested in biotechnology. In 2010-12, the three metros accounted for 60 percent of biotech venture capital. Data on venture capital flows come from the PriceWaterhouseCoopers Moneytree survey.

There’s probably no better indicator of the growth of biotechnology commercialization than the flow of venture capital funds to new startup companies. By this measure, the state of Florida’s position is essentially no different than it was a decade ago. While venture capital funding fluctuates from quarter to quarter, Florida’s share of national biotechnology venture capital funding is still less than 1 percent–in the same range that it was before its subsidies to Scripps and other research laboratories.

As it turns out, doing biomedical research doesn’t automatically lead to new companies and job creation. The hard and costly work of turning promising research ideas into marketable products happens in only a few places. The challenge in growing a commercial biotechnology hub is in overcoming the overwhelming competitive advantages that established clusters have in being places that have the financial, human, and institutional resources to succeed in this complicated and risky business. Despite the time and expense that Florida and other states have invested in biotech research, there’s almost no evidence that anyone has made anything more than marginal changes to the landscape of the U.S. biotech industry.