Comments and discussion.

Glaeser, Edward L. ; Gottlieb. Joshua D.

COMMENT BY ROBERT E. HALL The issues addressed in this paper by Edward Glaeser and Joshua Gottlieb will be coming into focus in the United States in the rest of this century as the country adapts to rising energy prices, including the appropriate carbon tax. The economic geography of the United States features extremely low residential densities made possible by universal automobile travel over good roads. People here live in free-standing houses surrounded by lawns, a style of living almost unknown in the rest of the world apart from our close cousins, Canada and Australia. On the other hand, the United States lags far behind many East Asian and European countries in residential broadband deployment because our dwellings are so far apart.

The economic logic of raising residential density seems powerful, but so far none of the policies discouraging higher density have changed. California puts heavy taxes on cars and gives heavy subsidies to mass transit, but almost nobody uses mass transit, because it does not pass near where they live. Sixty-passenger diesel buses rumble up and down Silicon Valley's El Camino, seldom carrying more than three passengers. And it is utterly unlawful to develop new land for housing or replace a single-family house with anything other than a new single-family house. Restrictions on land use condemn us to driving cars at a time when any reasonable society would shrink car travel substantially.

Glaeser and Gottlieb's interesting and wide-ranging paper touches on a number of the research issues underlying the response to higher energy prices and many other policies that involve economic geography. The authors take the standard economist's position that the primary case for government intervention is the correction of an externality. They find little support for the view that taxes and subsidies that vary by location alter the distribution of income or that they would be desirable if they did. They believe that an attack on income inequality through policies with direct effects on the income distribution would generally be better than the more roundabout approach of subsidizing the locations where the poor live. This conclusion is unlikely to be controversial at meetings of the Brookings Panel.

One might think that geographic policy faces the classic issue: Build on strength, or focus on improving the weak places? The paper takes the resolutely middle-of-the-road stand that policy should do neither. Policy should strive to do good for its own sake, not to alter the geographic distribution of economic activity.

The paper's concern with geographic policy gives it something of a European flavor. Such policy is not on the agenda in the United States. In Europe, government puts substantial resources into subsidizing poorer areas, such as eastern Germany and southern Italy. Poor Portugal lost its EU subsidy when the European Union incorporated some of the countries of eastern Europe. I am not aware of any important policy push toward regional subsidies in the United States. We let our airlines sink or swim on their own, more or less--three sank in the week before this Brookings Panel meeting. There is no chance that Alitalia would still be flying if it were a U.S. airline. We tend to take the same tough-love attitude toward poorer regions. The authors set forth a good case for tough love toward cities like Detroit: no need to subsidize new housing there; Detroit already has plenty of cheap housing.

The paper pushes the point that it is not enough to identify spatial externalities to rationalize government intervention. To justify an intervention that moves people, say, from Boston to Philadelphia, one needs to show that the social gain per person moved is greater in Philadelphia than the loss in Boston. I concur with the authors that the earlier evidence in favor of agglomeration externalities is fairly compelling. Much of the interesting new empirical work in the paper shows that the marginal effects of pro-agglomeration policies are roughly the same across cities. The authors find no good evidence that moving people from city to city generates any net agglomeration economies.

The paper confirms earlier estimates of positive agglomeration effects. Using historical population as an instrument, equation 2-5 of their table 2 provides an instrumental variables estimate of the structural response-([omega] - [alpha][gamma])/(1 - [alpha] + [alpha][gamma]) in their notation--of 0.089, a large effect by the standards of this literature. The authors back away from this finding, however, saying, "... historical instruments of this kind do not naturally solve the identification problem in a spatial model...." Their main concern is that using historical population as an instrument may be invalid, because it is correlated with current productivity differences. They quite properly dismiss the ordinary least squares results in table 2 because the identification assumption that current population is uncorrelated with productivity seems obviously false. One wonders why the pages of the Brookings Papers need to be cluttered with OLS results that the authors believe are invalid.

I am puzzled by some features of table 3. The basic idea is that the growth in income per capita has a structural coefficient of ([omega] - [alpha][gamma])/(1 - [alpha] + [alpha][gamma]) on population growth. Here co is the elasticity of agglomeration productivity with respect to population, 1 - [alpha] is the elasticity of output with respect to labor input, and [gamma] is the factor share of fixed capital. The estimated value in equation 3-2 is 0.004. Thus, if 1 - [alpha] = 0.66 and [alpha][gamma] = 0.04, values the authors suggest, then the implied value of the agglomeration coefficient co is 0.04, somewhat above the finding of the earlier literature. But the implied value of [gamma] is a staggering 0.12. What is the basis for making fixed capital so important? Without fixed capital, that is, with [alpha][gamma] = 0, the implied value of co is 0.003, which is contrary to the entire theme of the paper that agglomeration is important. The authors defend their implicit assumption about the share of fixed capital by identifying it with nonresidential buildings. But buildings, although fixed in space, are not a fixed quantity over time, but instead are producible. In studying spatial equilibrium, I believe that one should take a long-run view, so that the durability of buildings is not a good argument in favor of the assumption that they are fixed. If fixed capital is limited to land (which is not really fixed either when production competes with housing in a city), the usual view is that its share is only a few percentage points. Thus table 3 seems to undermine the main idea of the paper that agglomeration itself is an important fact.

The authors go on to test whether the agglomeration effect differs by city size. They approach this issue by measuring the extra effect of population growth in larger places. Table 3 reports the results. It goes without saying that regression 3-1 should have been removed from the table, as neither the authors nor any reader is interested in OLS results when the right-hand variables are plainly endogenous.

The third coefficient in equation 3-3 in table 3 finds that the wage-population growth slope was lower in 2000 in larger places. The t-statistic on this effect is 1.2, indicating moderately persuasive evidence of smaller agglomeration effects in areas with higher population. Nonetheless, the authors remain agnostic about the variation in agglomeration effects between small and large places. Given that the most important theme of the paper is that differences in marginal agglomeration effects are the primary rationalization for policies with spatial effects, I think this equation deserves a lot more attention. On its face, it contradicts the authors' general skepticism about disparities in marginal effects.

Table 4 investigates the relationship between city size and some of the adverse consequences of size. In this table the odd-numbered equations look for differences between large and small places, and the even-numbered equations look for differences based on centralization. Positive values of the coefficients in the second row indicate a nonlinearity that could be exploited by moving people from bigger to smaller cities. All of the size coefficients are essentially zero, suggesting the lack of any exploitable differences by city size. The coefficient in the fourth row measures the difference in the marginal effect of population in more centralized cities. Here the results are mixed. Congestion is more sensitive to population in more centralized cities, but pollution and murders are less sensitive. Sampling variation obscures any definite conclusion as well.

Although the authors are careful about endogenous fight-hand variables in most of the empirical work in the paper, they drop that concern in table 4, where all estimation is by OLS. For congestion and air quality, the direction of the bias seems obviously downward. If a city has a geography that results in naturally high congestion--for example, if it surrounds a bay, so that traffic concentrates on the edges of the bay and along the few bridges that cross it--congestion will be high and population low. This implies a negative correlation between the disturbance in the regression and the right-hand variable, population, and a consequent downward bias. The same argument applies to air pollution, as some cities are in basins that collect polluted air. Los Angeles is smaller than it would be if the air circulated more effectively. Because nothing rules out nonlinear effects from endogeneity bias, the results on nonlinearity in table 4 are less than conclusive.

The authors might make the same point about city-size policy as they do about income redistribution: direct policies are surely better for dealing with congestion, pollution, and crime. Notwithstanding the recent setback in New York City, congestion taxation is making steady advances around the world. Progress in controlling air pollution in developed countries has been astronomical, and even China is beginning to take the issue seriously. Crime rates for less serious crimes have proven remarkably responsive to simple changes in law enforcement: the United States and Western Europe have switched places over the past few decades, as burglary and mugging have declined here and exploded in Europe.

The first part of the section of the paper titled "U.S. Policies toward Places" deals with transportation and presents moderately persuasive evidence that canals, railroads, highways, and airports shape urban growth. The paper does not delve into evidence on the marginal effects of transport subsidies on different places. The authors conclude, "current spending does not appear to be targeting the high-income, high-density areas where the agglomeration effects are likely to be strongest." I don't see where the empirical work supports this conclusion. The authors' observation that it may be desirable to subsidize transportation in poor areas for its direct effect does not involve agglomeration effects.

The section of the paper on housing policy argues that policy has not made the mistake of trying to attract people to particular places by subsidizing housing there, but rather has made the huge mistake of constraining density and thus grossly failing to achieve the social optimum where land prices are maximal. As I noted at the beginning, these policies may need to relent in the face of high and rising fuel prices, especially when the appropriate carbon tax is included in fuel prices.

COMMENT BY

PAUL ROMER Economists do not quite know what to make of the yin and yang of cities. More than 150 years ago, Frederic Bastiat famously captured the invisible-hand yin:

 On coming to Paris for a visit, I said to myself: Here are a
 million human beings who would all die in a few days if supplies of
 all sorts did not flow into this great metropolis. It staggers the
 imagination to try to comprehend the vast multiplicity of objects
 that must pass through its gates tomorrow.... What, then, is the
 resourceful and secret power that governs the amazing regularity of
 such complicated movements? ... That power is an absolute
 principle, the principle of free exchange. (1)

So cities are the perfect illustration of the miracle of the market, right?

Not exactly. Turns out that economists can't capture what goes on in a city with the model of competitive equilibrium that is supposed to capture the invisible hand. Cities are dense with goods and services that are characterized by inherent nonconvexities and therefore cannot be provided competitively: water, sewerage, garbage collection, electric power, communications, roads, parks, police protection. More fundamentally, in a model of perfect competition, cities should not even exist. If production technologies exhibited the kind of convexity required to show that one can use the price system to achieve an efficient outcome with prices, there would be no reason to pile up so much economic activity on so little land.

Like their colleagues who have had to confront fundamental nonconvexities in international trade and economic growth, economists working on cities have invoked both the Marshallian extension of the competitive equilibrium model based on external increasing returns, and the more recent extension based on monopolistic competition. With these extensions, they can build models in which people are willing to pay high prices for the chance to be around lots of other people. But these extensions get one only partway toward a model that can capture the variety of outcomes in different cities. In the universe of interactions that take place in dense urban environments, the missing markets far outnumber the ones that are present. As a result, the nonmarket mechanisms that city governments use to control public health, crime, traffic, air pollution, noise, sight lines, visual clutter, and the activities permitted in any specific location make a world of difference to the quality of city life. Someone comparing life in Lagos today with life in Paris in 1845 might reasonably conclude that successful cities tell us more about some miracle of good governance than about the miracle of the market.

Getting the right perspective on cities would not matter much if cities themselves did not matter, but of course they do. Half of the people on earth now live in cities. Most of the other half live in grinding poverty. Despite the romance of the rural that infiltrates many discussions of development, it seems likely that this second half will escape from poverty only when most of them can find places to live and work in cities. So even before taking account of population growth, either the world will need a lot more cities, or the existing ones will have to get a lot bigger. It is a pressing priority to understand the appropriate roles for markets and governments in carrying out this expansion.

In a rich country like the United States, the structural transformation that moves most people into cities is largely complete, but the United States will experience significant population growth in the coming century, and so Americans, too, have to think about the processes that will determine where the growth in urban population will take place. Moreover, although the United States does not face the same challenge in reducing poverty that remains in much of the world, income inequality driven by the interaction between skill and technology will likely be a growing policy concern. The facts that Edward Glaeser and Joshua Gottlieb cite at the beginning of their paper suggest that there may be an important interaction between urbanization on the one side and skills and technology on the other. Moreover, whatever that connection was in the past, it may now be changing.

So the issues that Glaeser and Gottlieb touch on in this paper are among the most challenging for economists and the most important for policymakers. The common thread in my comments will be that economists can provide the most help to policymakers by focusing more on understanding the fundamental issues and less on trying to do their job of designing or advocating specific policies.

This paper has two conceptually distinct parts: a summary of key facts about urban areas in the United States, and an analysis of various U.S. federal policies toward cities and regions. The bulk of the paper is devoted to this second part. Here it seems to me the authors are overly constrained by the requirement (whether self-imposed or externally imposed) that they speak directly to the wisdom of specific federal policies. This part is rich with detail but can be boiled down to a simple summary message: In formulating policy toward cities, economists should focus on the yin and ignore the yang. The authors conclude that there is little evidence to support the expansion or continuation of any of the active government policies they consider. The only positive role they can find is for the feds to suppress an active policy (land use restrictions) implemented by some local governments. If one takes the "do no harm" view seriously, telling government to do nothing may be safe advice, but the contrast between Lagos and Paris suggests that policy errors of omission can be as harmful as those of commission.

There are at least three ways to interpret the conclusion that emerges from this paper--that no policy is good policy. The first is that market mechanisms are all it takes to achieve an efficient outcome, so there is no room for government policy in the development and operation of cities. The second is that government matters, but that in the United States policy has already been optimized, so that no further policies are needed (other than perhaps to drop some old bad ideas). The third is that government policy matters a lot and is far from being optimal, but the required policies are best implemented at the local level, and so the federal government, which seems to be the audience for the advice offered here, had best stay out of the way.

The first of these positions, that government services do not matter, seems intellectually indefensible to me. I suspect that the authors would agree, and if so, it might be worth saying so explicitly, because the paper is simply silent on this point. The second seems unlikely to me, but because I am not a specialist, I am not sure how much evidence one could marshal to undermine it. But as for the third position, if the quality of governance explains part of the difference between Paris in 1845 and Lagos today, might it not explain part of the difference between Pine Bluff, Arkansas, and Sioux Falls, South Dakota? Or between New York in 1970 and New York today?

My hunch is that there is, in fact, a lot of variation in the quality of governance between different cities in the United States, just as there is in the quality of management between firms. Economists might be no better at prescribing what managers of city and state governments should do than at prescribing what managers of firms should do, but it does not follow that the quality of governance is irrelevant. Nor does it follow that economists have nothing useful to say.

For example, if there is variation in the quality of city governance, then the mobility of people has added significance that is captured neither in the model nor in the verbal analysis. But allowing for mobility of people clearly cannot solve all the problems associated with bad city governance. People can move but most physical capital cannot. In principle, one might want to consider mechanisms that could do for the enormous amount of capital that can be trapped in a badly run city what a private equity takeover or a bankruptcy can do for capital trapped in a badly managed firm.

The paper is strongest when it develops or deepens a robust abstract insight. A good example is what the authors call the concept of spatial equilibrium, which is based on the observation that people can freely move between cities. As the authors observe, this fact has deep implications for how one looks at, for example, the difference in average income between Brownsville and Bridgeport. It does not make sense to try to help people living in Brownsville by making Brownsville more like Bridgeport. For someone living in Brownsville, if moving to Bridgeport will not make her better off, moving Bridgeport to her will not help either. This insight is important and not obvious. The paper makes a good start at driving this point home.

There is room to contribute other robust general insights like this one in a model that is more flexible than the one presented in the paper. To illustrate what this model might look like and the kind of results it might generate, consider two cities on two islands, one larger in area than the other. Workers consume land and a single produced good. Firms produce output using land and labor. To simplify the analysis of migration between the two islands, assume that all land is owned by absentee landlords.

Next, consider adding in two types of agglomeration effects. Suppose first that productivity in a city is an increasing function of its population, precisely as in the paper's model. In addition, assume that the utility that a consumer can achieve per unit of expenditure is also an increasing function of the number of people who live in the city. The assumption about productivity is easy to rationalize using a variety-in-production model. In just the same way, the assumption about additional utility from being around others can be captured in a variety-in-consumption model.

It seems clear that the following results should hold in this two-island model:

1. If there is any agglomeration effect in production or consumption, population density will be higher on the larger of the two islands.

2. As a special case, suppose that agglomeration effects are present in production but not in consumption. Assume as well that land is used only in production. Because of the agglomeration effect, total factor productivity will be higher on the larger island. However, wages and the marginal productivity of labor will be equal between the two cities. Firms producing on the larger island will face a higher cost of land and use less land per worker.

3. Continue to suppose that there are no agglomeration benefits in consumption, only in production. Now, in contrast to case 2, assume that land is used only in consumption. Then wages and labor productivity will be higher on the larger island. Land will be more expensive there, and land consumed per person will be lower.

4. Now reverse the assumption about agglomeration benefits. Suppose that there are no agglomeration benefits in production but there are agglomeration benefits in consumption. Continue to assume that land is used only in consumption. Then wages will be the same in the two cities, but housing will be more expensive on the larger island. Workers there will consume less land and (depending on the elasticity of demand for land) may have more or less income to spend on other goods. They will derive more utility from each unit of spending on produced goods.

Because the situation outlined in case 4 does not arise in the model considered in the paper, it is worth pausing briefly to evaluate its plausibility. One implication of this case is that because of the broader variety of consumption activities that the larger island offers, someone who is wealthy and who does not work might choose to live there even though the price of land is higher. This sounds like the behavior one observes in at least some cities, but under model 3, the one assumed in the paper in which the only benefits from density come via production, someone who did not work would never choose to live in the larger and denser city.

The richer class of models outlined here offers several insights that differ slightly from the analysis in the paper. First, in the section on basic facts about cities, the paper's discussion of agglomeration effects seems to suggest that these effects are present only if wages are higher in more dense or more populous locations. The broader class of models outlined here suggests that agglomeration effects are present any time population density is higher in one place than in another. In effect, once there are cities, some kind of agglomeration effect must be operating. Given this, all that the analysis of cross-sectional differences in wages can do is provide a window into the form that these agglomeration effects take. In cases 2 and 4, wages are the same across cities. In case 2, firms trade higher total factor productivity against higher costs of land. In case 4, consumers pay higher prices for housing to access better consumption opportunities.

A blended version of cases 2 and 4 is particularly interesting because of the key fact cited in the early section of the paper: people who move between cities in the United States do not experience an immediate increase in their wage on moving to a bigger city. Taking these cases and this fact at face value, one has to recognize that the apparent correlation between city size or density and wages may have nothing to do with agglomeration effects. Instead it may simply reflect differences in the characteristics of the workers in different cities--differences that are not adequately captured even with the standard individual variables in a wage regression.

Other basic facts cited in the early part of the paper also suggest that worker heterogeneity and positive assortment of worker types across cities are a crucial part of the observed variation across cities, one that may be becoming more firmly entrenched over time. Skilled cities are becoming even more skilled. Interestingly, they no longer seem to be the fastest-growing cities. It is as if they are limiting population growth and swapping lower-skilled for higher-skilled residents. In a development that may be related, in the 1980s and 1990s, incomes in different cities stopped converging.

All of these facts call for a model with at least two types of workers. To see what this might look like, suppress for the moment the questions addressed in cases 2 through 4 above. That is, set aside the issue of precisely where the agglomeration effects show up (in production or in consumption) and where a higher cost of land pinches as cities get more dense (on the production side or the consumption side). Instead, consider a very reduced form model in which utility is linear in a general-purpose form of consumption good that is produced from land, low-skilled labor, high-skilled labor, and roads. Roads are provided by the local government and are available at no charge to all city residents. Here roads are a stand-in for any of the many different services provided by local governments. In this simple model it is easiest to take the stock of roads as given by history and set aside the question of how inputs provided by the government are financed and produced.

In this reduced-form equilibrium, the stock of land, skilled and unskilled labor, and roads produce total output Y. Skilled and unskilled labor are paid returns that may be greater than or less than their marginal product in producing 11. Agglomeration effects will tend to make the private return to an additional worker less than the social return. By itself, unpriced access to a congestible good like a road will tend to make the private return to a worker who comes to this town greater than the social return.

Imagine that the positive agglomeration effects are larger for high-skilled than for low-skilled workers. (This, too, could easily be captured in a variety model where skilled workers are the inputs needed in the fixed-cost process that creates new goods.) Assume that the congestion costs from more users of roads are the same for all workers. If the congestion effects are large enough, the net social return to an additional low-skilled worker could be lower than the private return. At the same time, because the high-skilled workers generate agglomeration effects, the social return to an additional high-skilled worker is higher than the private return. If the stock of roads cannot be changed, cities in this kind of environment might maximize income for those who remain by limiting increases in the total population, particularly if in so doing they screen out low-skilled workers and allow in the high-skilled workers. Better still, they could induce the low-skilled workers already present to leave as high-skilled workers enter.

The point here is that there may be important links among (i) the clustering by ability that may be behind the higher wages observed in bigger and denser cities, (ii) the tendency for skilled cities to become more skilled yet, recently, to grow more slowly, and (iii) intentional growth-limiting policies. Because of these links, flexible theories that let us understand cities more completely and that have a chance at capturing all of the disparate trends outlined in the early part of the paper might offer a better basis for thinking seriously about the reasons behind growth limitations and what their aggregate effects might be.

The paper's discussion of growth limitations turns on the observation that the same home would sell for only a slightly higher price if it were moved to a town with regulations that increase the minimum lot size, resulting in a lower population density. The coefficient estimate that the authors report for the effect of density on housing prices suggests that cutting the average lot size in half and doubling the number of houses would reduce the price per house by only about 10 percent. This certainly does seem to suggest that towns are not maximizing the value of their land.

But a closer look shows that the regressions that find this surprisingly small effect of density on house price also show that, holding density constant, each additional regulatory barrier to new construction increases home prices by about 7 percent. (2) This suggests that the barriers are acting through channels other than density. Moreover, the descriptive data show that lower density is associated with a higher percentage of white residents, a lower percentage of residents who are foreign-born, and a higher percentage of residents with a B.A. degree. It therefore seems possible that lot size restrictions are part of a larger package of policies that may actually be adjusting the skill mix in ways that increase the number of high-skilled workers and decrease the number of low-skilled workers.

Without a full model, one cannot say whether decentralized decisions by individual cities about the mix of residents by skill type will be efficiency enhancing or not, even if they do increase the local value of land. Nor can one say what effects such policies might have on overall inequality. Because there are many possible models, and the evidence will probably not narrow them down to just one, any investigation of these issues will probably not lead to precise policy recommendations. Nevertheless, the facts from the first section of the paper about the evolving skill mix across cities, and the changing relationship between skill on the one side and population growth and convergence in income on the other, strike me as the most provocative part of the paper. I think the authors could usefully have devoted less attention to the analysis of specific policies and instead have given us a richer set of models that could help us better understand what these facts are telling us about cities, governments, and markets.

(1.) Frederic Bastiat, Economic Sophisms, edited and translated by Arthur Goddard (Irvington-on-Hudson, NY: Foundation for Economic Education, 1996), sect. I, ch. 18, para. 12. www.econlib.org/library/Bastiat/basSoph.html.

(2.) E. L. Glaeser and B. A. Ward, "The Causes and Consequences of Land Use Regulation: Evidence from Greater Boston," Journal of Urban Economics (forthcoming)

GENERAL DISCUSSION Lawrence Summers remarked that some of the paper's conclusions would be shocking to any local politician actually engaged in place-making policies. In particular, most would be surprised to hear that investment in urban renewal has no impact except to create capital gains for homeowners. Benjamin Friedman compared the notion that place-making policies have zero impact to the Malthusian model, which worked under a certain set of conditions but not when those conditions changed. Local politicians may want to know how to discriminate between circumstances in which place-making policies are effectual and those in which they are not. Edward Glaeser replied that his model differs from the Malthusian model in that local investment positively affects the welfare of the whole society, even if utility is equilibrated over space; one party does not gain at the expense of another. For example, if an investment in Baltimore makes Baltimore more productive, the spatial model predicts that utility outside Baltimore will eventually equal that in Baltimore, but the utility of all will be higher. Summers suggested that the authors defend the idea that utility is really the same in all areas, as this is highly counterintuitive to most people.

Robert Gordon observed that the pattern of growth and decline of American cities seems to suggest convergence across regions in terms of income per capita relative to the U.S. average, with a few exceptions. For example, metro-area relative income per capita in the former Confederate states has risen in the last forty years, while that in old Midwestern industrial cities such as Detroit and Cleveland has declined. Urban growth seems consistently driven by desirability of location only in coastal cities, such as San Francisco, Seattle, and the Northeastern cities. To explain low urban density in the United States, Gordon cited four factors that date back to the 1940s: the tax deductibility of mortgage interest, zoning regulations that preclude small suburban lot sizes, the starvation of mass transit, and the subsidization of interstate highways. On this point Lawrence Katz suggested that increasing urban density may not be optimal from a global health perspective, even though it may be optimal from a global pollution perspective.

Katz noted further that the paper focused on static, long-run equilibrium issues. He would have liked to see more discussion of how policies might reduce barriers to migration when people are hit with negative shocks or find better opportunities elsewhere.

Rebecca Blank observed that most place-making policies focus on creating spillovers among neighborhoods within a particular municipality, not on allocating investment among different regions of the country. She suggested that the paper could address more local, intracity concerns.

Gary Becker expressed surprise at the emphasis on the wage effects of migration. He noted that the impact on real wages of nominal wage versus land price changes is sensitive to how production and consumption processes are modeled. Thus, it is difficult to determine how much regional growth or decline is reflected in movement in wages and how much in changes in land prices.

Martin Baily observed that land use policies have a large impact on productivity growth. Changing industry composition is an important factor in productivity growth, and countries that heavily regulate land use make such change more difficult. In the case of declining areas, it seems clear that whatever equilibrium was once in place has changed, and it may not make sense to preserve places that have lost their economic base. Still, individuals and families in these places can incur major losses (if housing prices decline, for example), and one could make a case for helping them. One way might be to facilitate their relocation to a different city.

APPENDIX A

Proofs of Propositions

The core equations state that demand equals supply for the nontraded good (housing), which means that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [L.sub.H] is labor in the housing sector and [L.sub.p] is labor in the traded sector. The two wage equations imply that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Together these equations imply that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Solving for traded capital gives us that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], or

(A1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

The price solves [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] or

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], where [Q.sub.W,1], [Q.sub.W], [Q.sub.P,1], and [Q.sub.P] are constants.

Finally, to solve for overall population, we use [U.bar] = [[theta].sub.i] [N.sup.-[sigma].sub.i] (1 - [t.sub.i]) [W.sub.i][P.sup.-[beta].sub.i], which delivers the result that

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Equations 2 through 4 in the text then follow.

Proof of Proposition 1: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(A2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [N.sub.T] is aggregate population in all regions.

We use the formulation above where both wages and prices can be thought of as a function of population and exogenous parameters, but not of the tax rate or the amenity level. The derivative with respect to [N.sub.i] is [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], which can also be written as k[U.sub.i] - [[lambda].sub.N], where k is a constant and [U.sub.i] is total utility, which will be equal everywhere in a spatial equilibrium. As such, the decentralized equilibrium is location equilibrium, despite the existence of agglomeration economies and congestion effects in amenities.

As shown in equation 2, population is an increasing function of [[theta].sub.i](1 - [t.sub.i]), so any choice of [t.sub.i] that maximizes [[theta].sub.i](1 - [t.sub.i]) will also maximize population. Wages (equation 3) are a monotonically decreasing function of [[theta].sub.i](1 - [t.sub.i]), so maximizing wages per capita will end up minimizing utility. Equation 4 reveals that house prices are increasing in [[theta].sub.i](1 - [t.sub.i]) as long as [omegas](1 - [mu] + [mu][eta]) + [mu][eta](1 - [alpha]) - [alpha][gamma](1 - [mu]) > 0 and decreasing in [[theta].sub.i](1 - [t.sub.i]) if the reverse inequality holds.

Proof of Proposition 2: (a) Social welfare maximization can be written as maximizing the Lagrangian, where [H.sub.i] now stands for housing consumption per capita and T is total taxes:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is constant over space in the competitive equilibrium, then the competitive equilibrium is equivalent to the first three first-order conditions. If [U.sub.Y][[Y.sub.i], [H.sub.i], [[theta].sub.i]([N.sub.i])] is not constant over space, then the competitive equilibrium cannot be a social optimum.

The fourth condition can be rewritten as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

If [N.sub.i] [[theta]'.sub.i]([N.sub.i]) [U.sub.[theta]][[Y.sub.i]i, [H.sub.i], [[theta].sub.i]([N.sub.i])] + [U.sub.Y][[Y.sub.i], [H.sub.i], [[theta].sub.i]([N.sub.i])] [A'.sub.i] ([N.sub.i]) [F'.sub.i]([N.sub.i] - [L.sub.i]) is constant across space, then this condition will also be implied by the competitive equilibrium, since U[[Y.sub.i], [H.sub.i], [[theta].sub.i]([N.sub.i])] is constant across space and so is unearned income, which equals [A.sub.i] ([N.sub.i]) [F'.sub.i]([N.sub.i] - [L.sub.i]) - [Y.sub.i] [U.sub.H]/[U.sub.Y] [H.sub.i].

If [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is not constant, then the competitive equilibrium cannot satisfy this condition when U[[Y.sub.i], [H.sub.i], [[theta].suib.i]([N.sub.i])] is constant across space.

(b) Moving someone from area j to area i represents setting [N.sub.i] = [N.sub.i,C] + [epsilon] and [N.sub.j] = [N.sub.j,C] - [epsilon], where [N.sub.i,C] and [N.sub.j,C] are the competitive population allocations. To determine the value of this change, we take the derivative of social welfare with respect to [epsilon] and evaluate it at [epsilon] = 0. Total social welfare can be written as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where total unearned income is [[summation].sub.i] [N.sub.i][Z.sub.i] = [summation][[A.sub.i] ([N.sub.i])[F.sub.i]([N.sub.i] - [L.sub.i]) + [P.sub.i][G.sub.i] ([L.sub.i]) - [N.sub.i][W.sub.i] - T, or total revenue minus labor costs minus taxes.

If the marginal utility of income is constant across space, then this derivative equals [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

If the marginal utility of income is not constant across space, but if unearned income from an area goes entirely to the residents of that area, then the condition becomes

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Both of these are equivalent to the expression given in the text.

Proof of Proposition 3: We can write social welfare maximization as the maximization of a social planner's Lagrangian:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This yields a first-order condition for transportation spending of

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

In a competitive equilibrium, total social welfare can be written as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] again. The derivative of this with respect to transport spending is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

The added benefit of transport spending is positive if and only if

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Proof of Proposition 4: Social welfare can be written as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

First-order conditions are [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

For these conditions to be satisfied in a competitive equilibrium, where net unearned income is constant across space, [U.sub.Y]([Y.sup.k.sub.i], [[theta].sup.k.sub.i]) and

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] must be constant across space.

Starting with a competitive equilibrium where the marginal utility of income is constant across space, welfare can be improved by moving type k individuals from city j to city i if and only if

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

APPENDIX B

Data Description

Census aggregated data are taken from the compilation provided by the Inter-University Consortium for Political and Social Research, under record number 2896. This compilation includes Census county data from 1790 to 2000, including data from various issues of the Census's City and County Data Book.

To analyze the metropolitan-area level data, we aggregate the county data according to metropolitan statistical area (MSA) definitions released by the Office of Management and Budget. Each figure or table specifies the definition used for that particular application. We use different definitions for different purposes in order to be consistent with data from other sources used in a particular figure or table.

A word of caution is in order regarding some aggregate numbers computed from these data. In order to use a consistent set of MSA definitions for each purpose, we need median family income and median house value data at various MSA definitions; these medians are only presented under certain definitions. We therefore estimate the median by averaging the component counties' median values, weighting by families in the case of family income and by housing units in the case of house values. The resulting numbers are not equal to the true median for the metropolitan area, but they should be a close enough approximation for our purposes.

We obtained the Census Bureau's 5 percent Public Use Microdata Sample (PUMS) of the 2000 Census from the Integrated Public Use Microdata Series (IPUMS) service of the Minnesota Population Center. The sole geographical identifier included in the PUMS is a Public Use Microdata Area (PUMA), which IPUMS links to an MSA where appropriate. (In particular, IPUMS uses the 1999 MSA definitions, using primary rather than consolidated MSAs where applicable.) This identification is imperfect because the Census does not ensure that each PUMA is contained within a county, so PUMAs do not necessarily map onto MSAs. Nonetheless it is the best that can be done to link the Census microdata to other geographical data.

When running wage regressions on the individual-level data, we include only prime-age men (defined as those 25 to 55 years old) in order to avoid picking up differing labor force participation rates. We exclude anyone who reports not having a full-time job or whose annual earnings are below half of the annual minimum wage. We include dummy controls for each individual's age (grouped by decade: 20s, 30s, 40s, or 50s) and educational attainment (high school dropout, high school graduate, or college graduate), and in the repeated panel regressions (table 3) we allow the coefficient on each dummy variable to change across Census years.

When we use industry-level data in conjunction with Census industry categorization, it is necessary to match the different industry classification systems used in the different datasets. Census industry codes for manufacturing industries, on the 1990 basis, are matched to Standard Industrial Classification (SIC) codes using appendix A to Census Technical Paper No. 65, which is available online at www.census.gov/hhes/www/ioindex/ tp65_report.html. Since there is not an exact one-to-one relationship between Census industry codes and SIC codes, the concordance is necessarily imperfect; we select one SIC code if multiple ones are given, and we use data from the SIC code given even when informed that the Census industry code matches only part of the SIC category. A given Census industry code can be matched with a two-digit, three-digit, or four-digit SIC code, so our resulting dataset uses a mixture of levels of detail. These SIC data are in turn matched to exporting data from the International Trade Administration, available online at ita.doc.gov/td/industry/otea/industry_sector/tables.htm.

Table B1. Additional Regressions (a)

 Dependent variable

 Log
 Log wage Log
Independent variable wage residual population

Human capital predicted by 2.32 0.28 9.36
 industrial distribution (b) (0.33) (0.21) (2.06)
Constant 9.20 -0.17 10.0
 (0.10) (0.07) (0.6)
No. of observations 285 285 285
Adjusted [R.sup.2] 0.15 0.01 0.07
Human capital predicted by 2.39 -0.45 -0.55
 industrial distribution (b) (0.59) (0.37) (3.62)
Predicted human capital x -0.01 0.15 2.06
 above median subsample (0.10) (0.06) (0.62)
Constant 9.18 (0.03) 11.5
 (0.16) (0.10) (0.4)
No. of observations 285 285 290
Adjusted [R.sup.2] 0.15 0.02 0.10
Exporting predicted by 0.35 0.16 1.48
 industrial distribution (c) (0.24) (0.14) (1.43)
Constant 9.85 -0.11 12.7
 (0.05) (0.03) (0.30)
No. of observations 285 285 285
Adjusted [R.sup.2] 0.01 0.005 0.004

Exporting predicted by 0.27 -0.06 -5.18
 industrial distribution (c) (0.47) (0.28) (2.74)
Predicted exporting x 0.03 0.10 2.99
 above-median subsample (0.18) (0.11) (1.05)
Constant 9.86 -0.08 13.60
 (0.07) (0.04) (0.40)
No. of observations 285 285 285
Adjusted [R.sup.2] 0.01 0.01 0.03
Wage premium predicted 1.71 -0.44 -31.3
 by industrial distribution (d) (3.76) (2.23) (22.30)
Constant 9.92 -0.08 12.9
 (0.01) (0.01) (0.06)
No. of observations 285 285 285
Adjusted [R.sup.2] 0.0007 0.0001 0.007
Wage premium predicted -2.1 -3.16 -60.3
 by industrial distribution (d) (13.2) (7.83) (78.2)
Predicted wage premium x 4.3 3.03 32.3
 above-median subsample (14.0) (8.35) (83.4)
Constant 9.92 -0.082 12.9
 (0.01) (0.007) (0.1)
No. of observations 285 285 285
Adjusted [R.sup.2] 0.001 0.0006 0.007

 Dependent variable

 Population Income
 growth, growth,
Independent variable 1990s 1990s

Human capital predicted by 0.23 0.02
 industrial distribution (b) (0.20) (0.11)
Constant 0.06 0.41
 (0.06) (0.03)
No. of observations 285 285
Adjusted [R.sup.2] 0.005 0.0001
Human capital predicted by -0.12 0.29
 industrial distribution (b) (0.35) (0.20)
Predicted human capital x 0.07 -0.056
 above median subsample (0.06) (0.034)
Constant 0.16 0.34
 (0.10) (0.06)
No. of observations 285 285
Adjusted [R.sup.2] 0.01 0.01
Exporting predicted by -0.31 0.03
 industrial distribution (c) (0.13) (0.08)
Constant 0.19 0.41
 (0.03) (0.01)
No. of observations 285 285
Adjusted [R.sup.2] 0.02 0.0004

Exporting predicted by -0.38 0.18
 industrial distribution (c) (0.26) (0.15)
Predicted exporting x 0.03 -0.07
 above-median subsample (0.10) (0.05)
Constant 0.20 0.39
 (0.04) (0.02)
No. of observations 285 285
Adjusted [R.sup.2] 0.02 0.01
Wage premium predicted 0.5 -1.21
 by industrial distribution (d) (2.11) (1.18)
Constant 0.13 0.417
 (0.01) (0.003)
No. of observations 285 285
Adjusted [R.sup.2] 0.0002 0.004
Wage premium predicted 2.91 3.37
 by industrial distribution (d) (7.41) (4.12)
Predicted wage premium x -2.68 -5.09
 above-median subsample (7.89) (4.39)
Constant 0.13 0.418
 (0.01) (0.003)
No. of observations 285 285
Adjusted [R.sup.2] 0.0006 0.008

Source: Authors' regressions.

(a.) Units of observation are MSAs according to the 1999 definitions,
using primary rather than consolidated MSAs where applicable. Data
for the dependent variables are from the U.S. Census Bureau as
described in this appendix.

(b.) Calculated from the Census Public Use Microdata Sample (PUMS) as
described in this appendix. The percent of workers with a college
degree is first calculated for each industry code as defined by the
Census Bureau. These numbers are then averaged for each metropolitan
area, weighting by the distribution of industry employment in the
metropolitan area from the 1980 PUMS.

(c.) Calculated using data from the International Trade Administration
as described in this appendix as the value of an industry's exports
divided by the total value of its shipments. The industry-level export
fraction is averaged for each metropolitan area, weighting by the
distribution of industry employment in the metropolitan area from the
1980 PUMS.

(d.) Calculated from the PUMS as described in this appendix. A wage
premium is first calculated for each industry code, as defined by
the Census, as the industry fixed effect in a wage regression
containing controls for individual age, sex, and education. These
industry wage premiums are then averaged for each metropolitan
area, weighting by the distribution of industry employment in
the metropolitan area from the 1980 PUMS.

ACKNOWLEDGMENTS We are grateful to Robert Hall, Patrick Kline, Paul Romer, and Clifford Winston for insightful comments and suggestions, and to Charlie Redlick and Kristina Tobio for research assistance. Edward Glaeser thanks the Taubman Center for State and Local Government for financial support.

References

Acemoglu, Daron, and Joshua Angrist. 2001. "How Large Are Human-Capital Externalities? Evidence from Compulsory Schooling Laws." NBER Macroeconomics Annual 2000: 9-59.

Achenbach, Joel. 2004. The Grand Idea: George Washington's Potomac and the Race to the West. Simon and Schuster.

Alonso, William. 1964. Location and Land Use: Toward a General Theory of Land Rent. Harvard University Press.

Barro, Robert J., and Xavier Sala-i-Martin. 1991. "Convergence across States and Regions." BPEA, no. 1: 107-58.

Baum-Snow, Nathaniel. 2007. "Did Highways Cause Suburbanization?" Quarterly Journal of Economics 122, no. 2: 775-805.

Bergstrom, Theodore. 1986. "Soldiers of Fortune." In Equilibrium Analysis: Essays in Honor of Kenneth J. Arrow, edited by Walter P. Heller, Ross Star,

David J. Starrett, and Kenneth J. Arrow. Cambridge University Press.

Bernstein, Peter. 2005. Wedding of the Waters: The Erie Canal and the Making of a Great Nation. W. W. Norton.

Berry, Christopher R., and Edward L. Glaeser. 2005. "The Divergence of Human Capital Levels across Cities." Papers in Regional Science 84, no. 3: 407-44.

Bewley, Truman F. 1981. "A Critique of Tiebout's Theory of Local Public Expenditures." Econometrica 49, no. 3:713-40.

Black, Sandra E. 1999. "Do Better Schools Matter? Parental Valuation of Elementary Educations." Quarterly Journal of Economics 114, no. 2: 577-99.

Blanchard, Olivier J., and Lawrence F. Katz. 1992. "Regional Evolutions." BPEA, no. 1: 1-61.

Brueckner, Jan K. 1983. "Property Value Maximization and Public-Sector Efficiency." Journal of Urban Economics 14, no. 1: 1-15.

Bureau of Transportation Statistics. 1996. Airport Activity Statistics of Certificated Route Air Carriers 1995. Washington.

Busso, Matias, and Patrick Kline. 2008. "Do Local Economic Development Programs Work? Evidence from the Federal Empowerment Zone Program." Cowles Foundation Discussion Paper 1638. Cowles Foundation, Yale University.

Caudill, Harry. 1963. Night Comes to the Cumberlands: A Biography of a Depressed Area. Little, Brown.

Ciccone, Antonio, and Robert E. Hall. 1996. "Productivity and the Density of Economic Activity." American Economic Review 86, no. 1: 54-70.

Combes, Pierre-Philippe, Gilles Duranton, and Laurent Gobillon. 2008. "Spatial Wage Disparities: Sorting Matters!" Journal of Urban Economics 63, no. 2: 723-42.

Craig, Lee A., Raymond B. Palmquist, and Thomas Weiss. 1998. "Transportation Improvements and Land Values in the Antebellum United States: A Hedonic Approach." Journal of Real Estate Finance and Economics 16, no. 2: 173-89.

Duranton, Gilles, and Matthew Turner. 2007. "Urban Growth and Transportation." Working Paper 305. University of Toronto.

Ellison, Glenn, and Edward L. Glaeser. 1999. "The Geographic Concentration of Industry: Does Natural Advantage Explain Agglomeration?" American Economic Review 89, no. 2:311-16.

Erickson, Rodney A., and Paul M. Syms. 1986. "The Effects of Enterprise Zones on Local Property Markets." Regional Studies 20, no. 1: 1-14.

George, Henry. [1879] 1975. Progress and Poverty: An Inquiry into the Cause of Industrial Depressions and of Increase of Want with Increase of Wealth: The Remedy. New York: Robert Schalkenbach Foundation.

Gilbert, Neil, and Harry Specht. 1974. "Picking Winners--Federal Discretion and Local Experience as Bases for Planning Grant Allocation." Public Administration Review 34, no. 6: 565-74.

Glaeser, E. L. 1998. "Should Transfer Payments Be Indexed to Local Price Levels?" Regional Science and Urban Economics 28, no. 1: 1-20.

--. 2005. "Urban Colossus: Why Is New York America's Largest City?" Federal Reserve Bank of New York Economic Policy Review 11, no. 2: 7-24.

--. 2008. "The Economic Approach to Cities." Harvard Institute of Economic Research Discussion Paper 2149. Harvard University.

Glaeser, Edward L., and Joshua D. Gottlieb. 2006. "Urban Resurgence and the Consumer City." Urban Studies 43, no. 8: 1275-99.

--. 2008. "The Economics of Cities." Harvard University.

Glaeser, Edward L., and Joseph Gyourko. 2005. "Urban Decline and Durable Housing." Journal of Political Economy 113, no. 2: 345-75.

Glaeser, Edward L., Joseph Gyourko, and Raven E. Saks. 2005. "Why Have Housing Prices Gone Up?" American Economic Review 95, no. 2: 329-33.

Glaeser, Edward L., and Matthew Kahn. 2004. "Sprawl and Urban Growth." In Handbook of Regional and Urban Economics, vol. 4: Cities and Geography, edited by J. Vernon Henderson and Jacques Thisse. New York: North Holland.

--. 2008. "The Greenness of Cities: Carbon Dioxide Emissions and Urban Development." Working Paper 14238. Cambridge, Mass.: National Bureau of Economic Research.

Glaeser, Edward L., Hedi D. Kallal, Jose A. Scheinkman, and Andrei Shleifer. 1992. "Growth in Cities." Journal of Political Economy 100, no. 6:1126-52.

Glaeser, Edward L., and Janet E. Kohlhase. 2004. "Cities, Regions and the Decline of Transport Costs." Papers in Regional Science 83, no. 1: 197-228.

Glaeser, Edward L., and David C. Mare. 2001. "Cities and Skills." Journal of Labor Economics 19, no. 2:316-42.

Glaeser, Edward L., and Giacomo A. M. Ponzetto. 2007. "Did the Death of Distance Hurt Detroit and Help New York?" Working Paper 13710. Cambridge, Mass.: National Bureau of Economic Research.

Glaeser, Edward L., and Bruce Sacerdote. 1999. "Why Is There More Crime in Cities?" Journal of Political Economy 107, no, 2, part 2: 225-58.

Glaeser Edward L., and Albert Saiz. 2004. "The Rise of the Skilled City." Brookings-Wharton Papers on Urban Affairs, pp. 47-94.

Glaeser. Edward L., Albert Saiz, and Jed Kolko. 2001. "Consumer City." Journal of Economic Geography 1, no. 1: 27-50.

Glaeser. Edward L., and Raven Saks. 2006. "Corruption in America." Journal of Public Economics 90, no. 6-7: 1053-72.

Glaeser Edward L., and Jesse M. Shapiro. 2003. "Urban Growth in the 1990s: Is City Living Back?" Journal of Regional Science 43, no. 1: 139-65.

Glaeser Edward L., and Kristina Tobio. 2008. "The Rise of the Sunbelt." Southern Economic Journal 74, no. 3: 610-43.

Glaeser Edward L., and Bryce A. Ward. Forthcoming. "The Causes and Consequences of Land Use Regulation: Evidence from Greater Boston." Journal of Urban Economics.

Green, Richard K. 2007. "Airports and Economic Development." Real Estate Economics 35, no. 1:91-112.

Gyourko, Joseph E., Albert Saiz, and Anita A. Summers. 2007. "A New Measure of the Local Regulatory Environment for Housing Markets: The Wharton Residential Land Use Regulatory Index." Wharton Real Estate Center Working Paper 558. University of Pennsylvania.

Gyourko, Joseph, and Joseph Tracy. 1989. "The Importance of Local Fiscal Conditions in Analyzing Local Labor Markets." Journal of Political Economy 97, no. 5: 1208-31.

Haines, Michael R., and Robert A. Margo. 2006. "Railroads and Local Economic Development: The United States in the 1850s." Working Paper W12381. Cambridge, Mass.: National Bureau of Economic Research.

Isserman, Andrew, and Terance Rephann. 1995. "The Economic-Effects of the Appalachian-Regional-Commission--An Empirical-Assessment of 26 Years of Regional-Development Planning." Journal of the American Planning Association 61, no. 3: 345-64.

Kahn, Matthew E. 2003. "The Geography of US Pollution Intensive Trade: Evidence from 1958 to 1994." Regional Science and Urban Economics 33, no. 4: 383-400.

Katz, Lawrence, and Kenneth T. Rosen. 1987. "The Interjurisdictional Effects of Growth Controls on Housing Prices." Journal of Law and Economics 30, no. 1: 149-60.

Lucas, Robert E., Jr. 1988. "On the Mechanics of Economic-Development." Journal of Monetary Economics 22, no. 1: 3-42.

Marshall, Alfred. 1890. Principles of Economics. London: Macmillan.

Moretti, Enrico. 2004. "Estimating the Social Return to Higher Education: Evidence from Longitudinal and Repeated Cross-Sectional Data." Journal of Econometrics 121, no. 1-2: 175-212.

Papke, Leslie E. 1994. "Tax Policy and Urban-Development-Evidence from the Indiana Enterprise Zone Program." Journal of Public Economics 54, no. 1: 37-49.

Rauch, James E. 1993. "Productivity Gains from Geographic Concentration of Human Capital: Evidence from the Cities." Journal of Urban Economics 34, no. 3: 380-400.

Rappaport, Jordan, and Jeffrey D. Sachs. 2003. "The United States as a Coastal Nation." Journal of Economic Growth 8, no. 1: 5-46.

Roback, Jennifer. 1982. "Wages, Rents, and the Quality of Life." Journal of Political Economy 90, no. 6: 1257-78.

Romer, Paul M. 1986. "Increasing Returns and Long Run Growth." Journal of Political Economy 94, no. 5: 1002-37.

Rosen, Sherwin. 1979. "Wage-Based Indexes of Urban Quality of Life." In Current Issues in Urban Economics, edited by Peter Mieszkowski and Mahlon Straszheim. Johns Hopkins University Press.

Schwartz, Amy E., Ingrid G. Ellen, Ioan Voicu, and Michael H. Schill. 2005. "The External Effects of Place-Based Subsidized Housing." NYU Law and Economics Research Paper 05-05. New York University.

Schwarz, John E., and Thomas J. Volgy. 1988. "Experiments in Employment-A British Cure." Harvard Business Review (March/April): 104-12.

Shapiro, Jesse M. 2006. "Smart Cities: Quality of Life, Productivity, and the Growth Effects of Human Capital." Review of Economics and Statistics 88, no. 2: 324-35.

Simon, Curtis J., and Clark Nardinelli. 1996. "The Talk of the Town: Human Capital, Information, and the Growth of English Cities, 1861 to 1961." Explorations in Economic History 33, no. 3: 384-413.

Staples, John H. 1970. "Urban Renewal--Comparative Study of Twenty-Two Cities, 1950-1960." Western Political Quarterly 23, no. 2: 294-304.

Tym, Roger. 1984. "Finance and Affordability." In Low-Income Housing in the Developing World: The Role of Sites and Services and Settlements Upgrading, edited by Geoffrey K. Payne. Wiley.

U.S. Census Bureau. 2004. Federal Aid to States for Fiscal Year 2003. Washington.

Von Hoffman, Alexander. 2000. "A Study in Contradictions: The Origins and Legacy of the Housing Act of 1949." Housing Policy Debate 11, no. 2: 299-326.

Warner, John C. 1978. "Unfulfilled Long-Term Interest-Rate Expectations and Changes in Business Fixed Investment." American Economic Review 68, no. 3: 339-47.

EDWARD L. GLAESER

Harvard University

JOSHUA D. GOTTLIEB

Harvard University

(1.) Haines and Margo (2006); Duranton and Turner (2007).

(2.) Busso and Kline (2008).

(3.) Alonso (1964); Rosen (1979); Roback (1982).

(4.) Some places, such as Sioux Falls, South Dakota, and Charlotte, North Carolina, have GMP per capita far above their income per capita, presumably reflecting relatively high levels of physical capital.

(5.) This exercise was done using 2000 Census microdata. Our individual-level controls are age, race, educational attainment, and speaking a language other than English at home.

(6.) The NAR figures for inexpensive areas are considerably higher, since those data only look at recent home sales, which skew the sample toward newer, higher-quality homes. Using NAR data, it is the Rustbelt, not Texas, that has the cheapest homes: Decatur, Illinois, Saginaw, Michigan, and Youngstown, Ohio, are the three cheapest metropolitan areas according to the NAR, with median sales prices below $90,000.

(7.) Glaeser and Gottlieb (2008).

(8.) Including Roback (1982), Gyourko and Tracy (1989), and Black (1999),

(9.) Barro and Sala-i-Martin (1991).

(10.) Glaeser and Gottlieb (2008).

(11.) All of our metropolitan areas have at least 60 respondents to the happiness question over the period, and 95 percent have at least 100.

(12.) Blanchard and Katz (1992).

(13.) Bewley (1981).

(14.) For simplicity we assume that this is true for the rentier class as well.

(15.) The decentralized equilibrium is, however, not a Pareto optimum. There would still be gains from redistributing income across space, holding populations constant.

(16.) George (1879).

(17.) See, for example, Brueckner (1983).

(18.) This is a spatial version of the Bergstrom (1986) problem, where wages for the volunteer army equate total utility levels but not marginal utility levels. Glaeser (1998) examines this issue in the context of indexing transfers for local price levels.

(19.) This failure of the decentralized equilibrium is not an artifact of assuming a utilitarian social welfare planner. The decentralized equilibrium will not generally maximize any weighted average of individual utility levels. If the planner puts a larger weight on people in areas where the marginal utility of income is high, then the spatial equilibrium will generically be suboptimal, because the utility of those people is the same as the utility of people in areas where the marginal utility of income is low.

(20.) This argument for redistribution appears in many situations where the market equilibrium equates total utilities, rather than the marginal utility of income; the same argument can be used to justify transfers to high-amenity occupations that pay lower wages, at least if those amenities are not a complement to earnings. However, as Oliver Hart emphasized in his comment on Bergstrom (1986), this problem can be solved privately with cash lotteries that occur before choosing occupations or locations.

(21.) Ellison and Glaeser (1999).

(22.) Ellison and Glaeser (1999).

(23.) Ciccone and Hall (1996).

(24.) Glaeser and Mare (2001).

(25.) Combes, Duranton, and Gobillon (2008).

(26.) Ciccone and Hall (1996).

(27.) Glaeser and Tobio (2008).

(28.) The index was developed by Gyourko, Saiz, and Summers (2007) based on a survey of local zoning authorities nationwide.

(29.) Glaeser and Kahn (2004).

(30.) Glaeser and Kahn (2004).

(31.) Kahn (2003).

(32.) Glaeser and Sacerdote (1999).

(33.) Glaeser and Gottlieb (2006).

(34.) Achenbach (2004).

(35.) Bernstein (2005).

(36.) Glaeser (2005). An easier case can be made that Chicago's success depended on artificial waterways. Its growth was set off by a boom anticipating the completion of the Illinois and Michigan Canal, which connected the Great Lakes to the Mississippi. With that canal, the city of Chicago became the linchpin of a system of waterborne transportation that stretched from New York to New Orleans.

(37.) Haines and Margo (2006).

(38.) Craig, Palmquist, and Weiss (1998). The data were kindly provided by Robert Margo.

(39.) Glaeser and Saks (2006).

(40.) Warner (1978).

(41.) Baum-Snow (2007).

(42.) Duranton and Turner (2007).

(43.) Glaeser and Kohlhase (2004).

(44.) Glaeser and Ponzetto (2007).

(45.) Green (2007).

(46.) Glaeser, Saiz, and Kolko (2001).

(47.) Glaeser and Shapiro (2003).

(48.) The full environmental calculus is quite complex and involves the trade-off between more air pollution in densely settled areas and less pollution per square mile over a larger area.

(49.) Caudill (1963).

(50.) Isserman and Rephann (1995).

(51.) Isserman and Rephann (1995).

(52.) Papke (1994)

(53.) Tym (1984).

(54.) Schwarz and Volgy (1988)

(55.) Erickson and Syms (1986).

(56.) Papke (1994).

(57.) Busso and Kline (2008).

(58.) Von Hoffman (2000).

(59.) Von Hoffman (2000).

(60.) Schwartz and others (2005).

(61.) Glaeser, Gyourko, and Saks (2005).

(62.) Glaeser and Gyourko (2005).

(63.) Staples (1970).

(64.) Gilbert and Specht (1974, p. 565).

(65.) Glaeser and Ward (forthcoming); Glaeser, Gyourko, and Saks (2005).

(66.) Glaeser and Ward (forthcoming).

(67.) Glaeser and Ward (forthcoming).

(68.) Katz and Rosen (1987).

(69.) Brueckner (1983).

(70.) Why do communities fail to maximize land value? The Coase theorem, after all, suggests that side deals between property owners should lead to maximizing joint wealth. One answer is that property rights are murky and that the democratic process is not geared toward such side payments. In many cases the right of an owner to build is the outcome of a complicated regulatory process that cannot be predicted in advance. In other cases explicit legal impediments prevent such side deals. Since each new development creates a windfall for one owner and a host of inconveniences for everyone else, one can understand why democratic decisionmaking would lead to many restrictions on building.

(71.) Glaeser and Kahn (2008).

(72.) Glaeser (2008) provides a more detailed plan of this kind.

(73.) Lawrence Summers emphasized this point to us.

(74.) Marshall (1890, Book 2, ch. 10).

(75.) Lucas (1988); Romer (1986).

(76.) Rauch (1993).

(77.) Glaeser and others (1992).

(78.) Shapiro (2006).

(79.) Berry and Glaeser (2005).

(80.) Glaeser and Kahn (2004).

(81.) Following Rauch (1993).

(82.) As in Moretti (2004).

(83.) Simon and Nardinelli (1996).

(84.) Glaeser and Saiz (2004); Shapiro (2006).

(85.) Berry and Glaeser (2005).

(86.) Acemoglu and Angrist (2001).

(87.) Moretti (2004).

(88.) We also examined the share of the industry's goods that are exported and the average wage residual in the industry. Both of these variables were suggested by Lawrence Summers. We do not find a significant tendency of wages to be higher in areas surrounded by such industries, nor do we find any significant nonlinearities in the impact of these variables on wages or population growth.

(89.) Glaeser and others (1992).

Table 1. U.S. Metropolitan Statistical Areas Ranked on Selected
Indicators (a)

Indicator Top, five MSAs Average value

Gross metropolitan Bridgeport, CT $74,285
 product per Charlotte, NC $68,406
 capita, 2004 San Jose, CA $68,311
 Washington, DC $62,251
 Sioux Falls, SD $59,003

Median family San Jose, CA $81,053
 income, 2000 Bethesda, MD $81,000
 Bridgeport, CT $77,690
 Nassau-Suffolk, NY $76,595
 San Francisco, CA $75,416

Share of adult Boulder, CO 52.4%
 population with Bethesda, MD 50.2%
 a college degree, Ann Arbor, MI 48.1%
 2000 (b) San Francisco, CA 43.6%
 Cambridge, MA 43.6%

Median house San Jose, CA $441,900
 price, 2000 San Francisco, CA $440,500
 Santa Cruz, CA $377,500
 New York, NY $362,500
 Honolulu, HI $309,000

Indicator Bottom five MSAs Average value

Gross metropolitan Brownsville, TX $16,025
 product per McAllen, TX $16,149
 capita, 2004 Lake Havasu City, AZ $17,126
 Cumberland, MD $18,302
 Prescott, AZ $18,482

Median family McAllen, TX $26,009
 income, 2000 Brownsville, TX $27,853
 Laredo, TX $29,394
 El Paso, TX $33,410
 Las Cruces, NM $33,576

Share of adult Dalton, GA 9.0%
 population with Lake Havasu City, AZ 9.9%
 a college degree, El Centro, CA 10.3%
 2000 (b) Hanford, CA 10.4%
 Merced, CA 11.0%

Median house Odessa, TX $47,700
 price, 2000 McAllen, TX $52,400
 Brownsville, TX $53,000
 Danville, IL $56,000
 Pine Bluff, AR $56,200
Sources: Bureau of Economic Analysis; U.S. Census Bureau.

(a.) Units of observation are metropolitan statistical areas (MSAs)
under the 2006 definitions, using metropolitan divisions where
applicable. All dollars are current (2000 or 2004) dollars.

(b.) Among MSAs with at least 100,000 population.

Table 2. Regressions of Wages on Metropolitan-Area Population
and Density (a)

 Regression

Independent variable 2-1 2-2 2-3

Log of population in 2000 0.041 0.023
 (0.009) (0.010)
Log of population in 2000, 0.076
 below-median subsample (0.029)
Log of population in 2000, 0.038
 above-median subsample (0.012)
Log of population density, 0.029
 2000 (0.015)
Log of density in 2000,
 below-median subsample
Log of density in 2000,
 above-median subsample
No. of observations 1,591,140 1,591,140 1,591,140
No. of MSAs 283 283 283
[R.sup.2] 0.22 0.22 0.22

 Regression

Independent variable 2-4 2-5 (b)

Log of population in 2000 0.089
 (0.037)
Log of population in 2000, 0.057
 below-median subsample (0.029)
Log of population in 2000, 0.020
 above-median subsample (0.013)
Log of population density,
 2000
Log of density in 2000, 0.041
 below-median subsample (0.017)
Log of density in 2000, 0.027
 above-median subsample (0.020)
No. of observations 1,591,140 1,282,116
No. of MSAs 283 210
[R.sup.2] 0.22

Source: Authors' regressions.

(a.) The dependent variable is the logarithm of the individual
wage. The regression method is ordinary least squares except where
noted otherwise. Only fully employed men aged 25 to 55 are included
in the sample. All regressions include individual controls for age
and education. Individual-level wage data are from the U.S. Census
Public Use Microdata Sample, as described in appendix B.
Metropolitan-area covariates are from the U.S. Census Bureau as
described in appendix A. Units of observation are metropolitan
statistical areas (MSAs) presented under the 1999 definitions,
using primary rather than consolidated MSAs where applicable and
New England county metropolitan areas where applicable. Standard
errors (in parentheses) are clustered by MSA.

(b.) Instrumental variables regression using the logarithm of
population in 1850 to instrument for log of population in 2000.

Table 3. Instrumental Variables Regressions Testing for Agglomeration
Effects (a)

 Regression

Independent variable 3-1 (b) 3-2 3-3

Population growth in 1990s x 0.200 0.004 0.166
 year 2000 dummy (0.044) (0.098) (0.088)
Above-median population dummy x 0.006
 year 2000 dummy (0.017)
Population growth in 1990s x -0.165
 above-median population (0.133)
 dummy x year 2000 dummy
Above-median centralization
 dummy x year 2000 dummy
Population growth in 1990s x
 above-median centralization
 dummy x year 2000 dummy
Dummy for bottom quartile of
 MSA population growth, 1970-90
 x year 2000 dummy
Log of population in 2000 x
 dummy for bottom quartile of MSA
 population growth, 1970-90 x
 year 2000 dummy
No. of observations 2,950,850 2,950,850 2,950,850
No. of MSAs 287 287 287

 Regression

Independent variable 3-4 3-5

Population growth in 1990s x -0.020 -0.099
 year 2000 dummy (0.121) (0.141)
Above-median population dummy x
 year 2000 dummy
Population growth in 1990s x
 above-median population
 dummy x year 2000 dummy
Above-median centralization -0.038
 dummy x year 2000 dummy (0.022)
Population growth in 1990s x 0.216
 above-median centralization
 dummy x year 2000 dummy (0.163)
Dummy for bottom quartile of -0.057
 MSA population growth, 1970-90 (0.026)
 x year 2000 dummy
Log of population in 2000 x 0.398
 dummy for bottom quartile of MSA
 population growth, 1970-90 x (0.252)
 year 2000 dummy
No. of observations 2,490,733 2,950,850
No. of MSAs 229 287

Source: Authors' regressions.

(a.) The dependent variable is the logarithm of the individual wage
for employed men aged 25 to 55. Regressions are instrumental
variables regressions except where noted otherwise. Instruments are
mean January temperature, mean July temperature, and precipitation,
from the 1994 City and County Data Book. Individual data are from
the 1990 and 2000 Census Public Use Microdata Sample, as described
in appendix B. Population data are from the Census. Units of
observation are metropolitan statistical areas (MSAs) under the
1999 definitions, using primary rather than consolidated MSAs where
applicable and New England county metropolitan areas where
applicable. All regressions include individual controls for sex,
age, and education and MSA and year fixed effects. Standard errors
(in parentheses) are clustered by MSA.

(b.) Regression is by the ordinary least squares method.

(c.) Centralization is defined as the fraction of MSA employment
located within five miles of the central business district, from
Glaeser and Kahn (2004).

Table 4. Regressions of Urban Disamenity Measures on Metropolitan-Area
Population and Centralization Measures (a)

 Dependent variable

 Log of concentration
 Log of average of TSP-10
 communte (minutes) particulates (b)

Independent variable 4-1 4-2 4-3 4-4

Log of population in 2000 0.12 0.057 0.142 0.145
 (0.012) (0.016) (0.056) (0.079)
Log of population x -0.003 0.001
 population above (0.002) (0.012)
 median in 2000
Centralization (d) -0.874 2.065
 (0.420) (2.134)
Log of population in 0.052 -0.222
 2000 x centralization (0.033) (0.156)
Constant 1.65 2.531 1.945 2.284
 (0.137) (0.210) (0.657) (1.134)
No. of observations 318 248 40 37
Adjusted [R.sup.2] 0.48 0.56 0.36 0.52

 Dependent variable

 Log of murder rate Log of real wage (c)

Independent variable 4-5 4-6 4-7 4-8

Log of population in 2000 0.222 0.485 0.028 0.037
 (0.146) (0.234) (0.015) (0.024)
Log of population x 0.004 0.000
 population above (0.018) (0.003)
 median in 2000
Centralization (d) 4.305 0.974
 (5.476) (0.611)
Log of population in -0.309 -0.089
 2000 x centralization (0.442) (0.048)
Constant -1.371 -4.947 8.363 8.336
 (1.733) (3.025) (0.184) (0.330)
No. of observations 153 126 220 186
Adjusted [R.sup.2] 0.07 0.09 0.05 0.06

Source: Authors' regressions.

(a.) Population, income, and commute time data are from the U.S.
Census Bureau, as described in appendix B. Murder rate is from the
Federal Bureau of Investigation Uniform Crime Reports, and TSP-10
data are from the Environmental Protection Agency. Units of
observation are metropolitan statistical areas (MSAs) under the
1999 definitions, using primary rather than consolidated MSAs
where applicable and New England county metropolitan areas where
applicable. Numbers in parentheses are standard errors.

(b.) Total suspended particulates of less than 10 microns diameter,
a standard measure of atmospheric pollution.

(c.) Computed by adjusting the average nominal wage for the cost of
living index from the American Chamber of Commerce Research
Association.

(d.) Percent of employment within 5 miles of the central business
district, from Glaeser and Kahn (2004).

Table 5. Historical Regressions of Population and Income Growth
on Metropolitan-Area Transportation Measures and Controls (a)

 Dependent variable

 Population Population
 growth, growth,
 1850-60 1850-1900
Independent variable 5-1 5-2

Log of initial population -0.335 -0.629
 (0.011) (0.017)
Distance in miles to ocean or 0.04 0.037
 Gulf of Mexico (0.003) (0.005)
Dummy for county accessible by 0.144 0.203
 rail in 1850 (0.023) (0.036)
Congregationalists per capita 0.925 0.96
 in 1850 (0.181) (0.286)
Log of new miles of highway,
 1960-90
Dummy for top-50 airport

Percent of population with
 college degree in 1990
Constant 3.154 6.431
 (0.100) (0.158)
Unit of observation Counties Counties
No. of observations 1,517 1,517
Adjusted [R.sup.2] 0.53 0.57

 Dependent variable

 Population Income per
 growth, capita growth,
 1960-90 1960-90
Independent variable 5-3 5-4

Log of initial population -0.101 -0.038
 (0.023) (0.010)
Distance in miles to ocean or -0.021 -0.013
 Gulf of Mexico (0.004) (0.002)
Dummy for county accessible by
 rail in 1850
Congregationalists per capita
 in 1850
Log of new miles of highway, 0.111 0.039
 1960-90 (0.026) (0.009)
Dummy for top-50 airport

Percent of population with
 college degree in 1990
Constant 1.309 2.412
 (0.245) (0.109)
Unit of observation MSAS MSAS
No. of observations 20-5 205
Adjusted [R.sup.2] 0.15 0.22

 Dependent variable

 Population Income per
 growth, capita growth,
 1990-2000 1990-2000
Independent variable 5-5 5-6

Log of initial population -0.007 -0.02
 (0.007) (0.004)
Distance in miles to ocean or
 Gulf of Mexico
Dummy for county accessible by
 rail in 1850
Congregationalists per capita
 in 1850
Log of new miles of highway,
 1960-90
Dummy for top-50 airport 0.046 0.025
 (0.021) (0.011)
Percent of population with 0.337 0.097
 college degree in 1990 (0.088) (0.047)
Constant 0.136 0.65
 (0.086) (0.046)
Unit of observation MSAS MSAs
No. of observations 318 318
Adjusted [R.sup.2] 0.07 0.1

Source: Authors' regressions.

(a.) Metropolitan statistical areas (MSAS) are under the 1999
definitions, using primary rather than consolidated MSAS where
applicable and New England county metropolitan areas where
applicable. Population, income, and Congregationalists data are
from the U.S. Census Bureau, as described in appendix B. Data on
proximity to rail transportation are from Craig, Palmquist, and
Weiss (1998). Distance to ocean or Gulf is from Rappaport and Sachs
(2003) and is the distance from the closest county in the MSA.
Highway data are from Baum-Snow (2007). Airport data are from
Bureau of Transportation Statistics (1996). Standard errors are in
parentheses.

Table 6. Correlations of Federal Transportation Spending with
State Size (d)

 Correlation Correlation
 Correlation with log of with log of
 with log of population income
Spending measure population density per capita

Log of highway spending -0.54 -0.32 -0.18
 per capita, including
 trust fund (b)
Log of highway spending -0.48 -0.49 -0.39
 per capita, excluding
 trust fund
Log of air travel spending -0.34 -0.79 -0.45
 per capita
Log of railroad spending -0.50 -0.18 -0.14
 per capita
Log of public transit 0.11 0.46 0.53
 spending per capita


Source: Authors' calculations.

(a.) Each cell reports the correlation coefficient between the
indicated transportation spending measure and the indicated state
characteristic. Transportation spending data are from U.S. Census
Bureau (2004). Population data are from the U.S. Census Bureau as
described in appendix B. Land area is from Rappaport and Sachs
(2003).

(b.) Includes contributions by the states to the national
transportation trust fund.

Table 7. Regressions Estimating Impact of the Appalachian Regional
Commission (a)

 Dependent variable

 Growth in income
 Population growth per capita

Independent variable 1970-80 1970-2000 1970-80 1970-2000

Dummy for county in ARC 0.037 -0.002 0.004 -0.029
 coverage area (0.008) (0.020) (0.005) (0.008)
Log of initial population -0.018 -0.036
 (0.004) (0.010)
Log of initial income -0.323 -0.406
 per capita (0.011) (0.016)
Constant 0.299 0.637 3.418 5.172
 (0.038) (0.099) (0.082) (0.123)
Adjusted [R.sup.2] 0.051 0.015 0.512 0.420

Source: Authors' regressions.

(a.) Units of observation (N = 898) are counties. Income and
population data are from the U.S. Census Bureau, as described in
appendix B. Standard errors are in parentheses.

Table 8. Regressions Estimating Impact of Urban Renewal (a)

 Dependent variable

 Growth from 1960 to 1970

 In income
Independent variable In population per capita

Urban renewal spending 0.0022 0.0004
 per capita (dollars) (0.0014) (0.0006)
Dummy for Model Cities
 participant
Log of initial population -0.027
 (0.051)
Log of initial income -0.459
 per capita (0.152)
Constant 0.054 5.92
 (0.768) (1.17)
No. of observations 21 21
Adjusted [R.sup.2] 0.20 0.45

 Dependent variable

 Growth from 1970 to 2000

 In income
Independent variable In population per capita

Urban renewal spending
 per capita (dollars)
Dummy for Model Cities -0.051 0.023
 participant (0.063) (0.016)
Log of initial population -0.053
 (0.021)
Log of initial income -0.177
 per capita (0.035)
Constant 1.06 3.34
 (0.26) (0.28)
No. of observations 318 318
Adjusted [R.sup.2] 0.04 0.07

Source: Authors' regressions.

(a.) Units of observation are metropolitan statistical areas under
the 1999 definitions (primary rather than consolidated MSAs where
applicable, New England county metropolitan areas where
applicable). Income and population data are from the U.S. Census
Bureau, as described in appendix B. Urban renewal spending per
capita is from Staples (1970).

Table 9. Regressions of Wages on Metropolitan-Area Human Capital
and Industrial Structure (a)

 Regression

Independent variable 9-1 9-2

Percent of population with 0.814
 college degree (0.024)
Percent with college degree, 0.638
 below-median subsample (0.072)
Percent with college degree, 0.862
 above-median subsample (0.031)
Percent employment in top quartile
 of human capital
Percent employment in top quartile,
 below-median subsample
Percent employment in top quartile,
 above-median subsample
Herfindahl index for MSA

Herfindahl index for MSA and sector

No. of observations 3,789,707 3,789,707
No. of MSAs 285 285
Adjusted [R.sup.2] 0.31 0.31

 Regression

Independent variable 9-3 9-4

Percent of population with
 college degree
Percent with college degree,
 below-median subsample
Percent with college degree,
 above-median subsample
Percent employment in top quartile 0.671
 of human capital (0.027)
Percent employment in top quartile, 0.035
 below-median subsample (0.070)
Percent employment in top quartile, 0.946
 above-median subsample (0.034)
Herfindahl index for MSA

Herfindahl index for MSA and sector

No. of observations 3,789,707 3,789,707
No. of MSAs 285 285
Adjusted [R.sup.2] 0.31 0.31

 Regression

Independent variable 9-5 9-6 (b)

Percent of population with 0.792
 college degree (0.024)
Percent with college degree,
 below-median subsample
Percent with college degree,
 above-median subsample
Percent employment in top quartile
 of human capital
Percent employment in top quartile,
 below-median subsample
Percent employment in top quartile,
 above-median subsample
Herfindahl index for MSA -1.011
 (0.243)
Herfindahl index for MSA and sector -0.087
 (0.001)
No. of observations 3,789,707 3,789,707
No. of MSAs 285 285
Adjusted [R.sup.2] 0.31 0.32

Source: Authors' regressions.

(a.) The dependent variable is the logarithm of the individual
wage; the sample includes fully employed men aged 25 to 55 only.
Individual-level data are from the 1990 and 2000 Census Public Use
Microdata Sample, as described in appendix B. All regressions
include individual controls for sex, age, and education and
industry fixed effects (but not MSA fixed effects except where
noted otherwise). Data on industry concentration and human capital
by industry are calculated from the microdata. Metropolitan-area
education data are from the U.S. Census Bureau as described in
appendix B. City characteristics are at the level of MSAs under the
1999 definitions, using primary rather than consolidated MSAs where
applicable and New England county metropolitan areas where
applicable. Standard errors (in parentheses) are clustered by MSA
and industry.

(b.) Regression includes MSA fixed effects.