Economic growth in a cross-section of cities

JOURNALOF Monetary ECONOMICS ELSEVIER Journal of Monetary Economics 36 (1995) ll7-143 Economic growth in a cross-section of cities Edward L. Glaeser *'a'b'¢, Jos6 A. Scheinkman d, Andrei ShleifeP "Harvard University, Cambridge, MA 02138, USA b Hoover Institution, Stanford, CA 94305, USA c National Bureau of Economic Research, Cambridge, MA 02138, USA d University of Chicago, Chicago, IL 60637, USA (Received September 1993; final version received May 1995) Abstract We examine the relationship between urban characteristics in 1960 and urban growth between 1960 and 1990. Income and population growth move together, and both types of growth are (l) positively related to initial schooling, (2) negatively related to initial unemployment, and (3) negatively related to the initial share of employment in manufac- turing. Racial composition and segregation are uncorrelated with urban growth across all cities, but in cities with large nonwhite communities segregation is positively corre- lated with population growth. Government expenditures (except for sanitation) are uncorrelated with growth; government debt is positively correlated with later growth. Key words. Growth; Cities; Regions; Education J EL classification: 040; RI 1 1. Introduction Over the last 30 years, the growth experiences of United States cities have varied widely. The population of some cities grew significantly while other cities * Corresponding author. The title is partially borrowed from Barro (1991). Participants in the Harvard Urban Workshop, the Harvard Growth Workshop, the NBER Summer Institute, and one anonymous referee provided helpful comments. David Mare and Melissa McSherry provided excellent research assistance, and the NSF provided research suppport. 0304-3932/95/$09.50 © t995 Elsevier Science B.V. All rights reserved SSDI 030439329501206 4 118 E.L. Glaeser et al. / Journal of Monetary Economics 36 (1995) 117-143 virtually disappeared. Some dispersion of growth experiences can be explained by geographic factors, such as the movement of population west and south. But what are the economic forces that explain city growth over the last 30 years in a cross-section of the United States cities? In this paper, we empirically investigate this question. We examine how the growth experiences of 203 large U.S. cities relate to their location, initial population, initial income, past growth, output composition, unemployment, inequality, racial composition, segregation, size and nature of government, and the education of their labor force. The primary purpose of our analysis is descriptive: we want to understand which cities grew between 1960 and 1990. As a description, this analysis continues an extensive regional growth litera- ture, including the studies by Borts (1960), Kain and Neidercorn (1962), Mills (most recently, 1992), and others. Our focus on human capital as a determinant of city growth is particularly closely related to Chinitz (1962), who emphasized the connection between urban success and the transmission of enterpreneurial skills. Similar arguments have been made by Jacobs (1969) and Marshall (1890). In addition, our analysis aims to contribute to the recent studies of economic growth. Starting with Baumol (1986), Delong (1988), and Barro (1991), econo- mists have compared income growth experiences of different countries as a func- tion of their characteristics. These studies typically find weak evidence of convergence of incomes between countries, and stronger evidence that educa- tion, physical investment, political stability, and openness to trade contribute to growth. Other studies, including Barro and Sala-i-Martin (1991) and Blanchard and Katz (1992), look at growth experiences of U.S. states. Barro and Sala-i- Martin find much more income convergence between states than between countries. Blanchard and Katz show that employment growth in different states is very persistent, while unemployment is not.1 Looking at cities complements looking at countries and states in three ways. First, unlike countries, cities are completely open economies; there is tremen- dous movement of capital, labor, and ideas between cities. Cities are more specialized (and less arbitrary) economic units than states, and hence it may make more sense to study the movement of resources and convergence between cities than between states. National boundaries that bar factor mobility and national policies that encourage industrial diversification eliminate the gains from factor mobility. These forces complicate work on cross-national studies. Cities allow us to look at economic growth without these concerns. Second, many of the cross-sectional - particularly the cross-national - studies of growth argue that ideas are important for growth. Various versions of this 1More precisely, they show that unemployment rate deviations from long-run state average unemployment levels are not persistent. E.L. Glaeser et al. / Journal of Monetary Economics 36 (1995) 117-143 119 theory focus on the external benefits of physical capital or disembodied know- ledge (Romer, 1986), human capital (Lucas, 1988), particular industries such as manufacturing or high-technology industries (Porter, 1991), and various other spillovers. Glaeser, Kallal, Scheinkman, and Shleifer (1992) found evidence that cross-industry intellectual externalities were particularly important for urban growth. This paper complements the earlier one. We now focus on whole cities, rather than individual industries, and on the sources of city-wide externalities, such as those from human capital. Third, recent studies of economic growth across countries have focused on political and social, as well as economic determinants of growth. For example, several studies starting with Barro (1991) have shown that political instability is bad for growth, while Alesina and Rodrick (1994) and others have argued that inequality is bad for growth. DeLong and Shleifer (1993) have shown that limited government, as opposed to absolutist government, strengthened the growth of medieval cities. This paper will use the political and social character- istics of cities to provide further evidence on the importance of political and social factors for growth. Section 2 presents a formal framework for our empirical work. Section 3 describes our data. Section 4 then presents our results on the relationship between economic characteristics of cities and their growth. Section 5 focuses on social and political characteristics of cities. Section 6 concludes. 2. A framework This section provides a formal setting for our empirical work. 2 Cities will be treated as separate economies that share common pools of labor and capital. Differences in urban growth experiences cannot then come from savings rates or exogenous labor endowments. Because of our assumption of mobile labor and capital, cities differ only in the 'level of productivity' and the 'quality of life'. Total output in a city is given by Ai . , f ( L , . t ) = A,. ,L~., , (2.1) where Ai., represents the level of productivity in city i at time t, 3 Li., denotes population of city i at time t, f ( . ) is a common across cities Cobb-Douglas production function. The coefficient of this production function, a, is a nation- wide production parameter. 2 This framework is an extension of the framework used in Glaeser et al. (1992). 3 We interpret A~., broadly, to allow for the possibility that social, technological, and political forces all determine the overall productivity of a city. 120 E.L. Glaeser et aL / Journal of Monetary Economics 36 (1995) 117-143 The labor income of a potential migrant will be the marginal product of labor: Wi.t = aA i . tLT f 1 (2.2) Total utility equals wages multiplied by a quality of life index. We assume that this index is declining in the size of the city, or using a simple functional form: Quality of life L - ~ = Qi, t i.t , (2.3) where 6 > 0. Quality of life is meant to capture a wide range of factors including crime, housing prices, and traffic congestion. Total utility of a potential migrant to city i is Util ity = 6Ai.tQi,tL~,t '~- 1 (2.4) We initially assume free migration across cities. This assumption ensures con- stant utilities across space at a point in time, so each individual's utility level in each city must equal the reservation utility level at time t, which we denote _Ut. Thus, for each city: Ai ,+ l + (Q i . ,+ I )+ _ Lit+~ l og (~)=log(~) log \ Qi,t / (a 6 -1 ) log(~) . (2.5) We also assume that ( A,., + ~'] = , l og \ Ai,, ] X i.tfl + ei., + l , (2.6a) log \ Qi, t / X i,,O + (i,t + l , (2.6b) where X~,, is a vector of city characteristics at time t which determines the growth in both the quality of life in the city and the growth of city level productivity. Combining (2.5), (2.6a), and (2.6b) yields log L i t + l - 1 ( ~ ) 1+6- -a X' i ' ' ( f l+ O)+x i , ,+ l , (2.7) /'Wi t+l \ 1 log~J \ i., / -- 1 +6- -0" XZi.,((~ fl "4- 0"0 - - O) "4- (f)i.t+ l , (2.8) 4 Assuming wages are average output makes no qualitative difference for the calculations. E.L. Glaeser et al. / Journal of Monetary Economics 36 (1995) 117-143 121 where X, and e3~t are error terms uncorrelated with urban characteristics) The net result is that employment growth regressions can be interpreted as showing how city level variables (the X's) determine the sum of quality of life and productivity growth. The wage growth regressions can be thought of as showing a weighted average of the productivity growth and a - 1 times quality of life growth. One difficulty in interpreting wage growth regressions is that they may be reflecting population composition changes as well as compensation changes. While our model presumes homogeneous labor (like most growth models), heterogeneity of labor is a principal feature of urban growth. We will handle this problem by (1) discussing how our results might be interpreted in a hetero- geneous labor model and (2) examining movements of population subgroups to allow different urban characteristics to attract different people differently. Migration vs. convergence A negative correlation between initial wages and wage growth (convergence) might occur because (1) technology improves more slowly in advanced cities (real convergence) or (2) because the in-migration of labor to high-wage regions causes the wages in those regions to decline. For the second explanation to make sense, migration of labor must respond slowly to shocks in local labor demand. To examine this question, this subsection presents a model with migration costs and delayed labor supply responses to local shocks. We assume that the quality of life for potential migrants declines not only in the level of population but also in the growth rate of population. This decline might occur because the costs of migration are rising in the number of in- migrants. A negative connection between the quality of life and growth might also occur because it takes time to build certain public goods, or basic infra- structure, or housing. The residents of quickly growing cities may suffer in terms of quality of life until those cities are built up. Quality of life is now given by f L- \ -~ Quality of life = Q,.zL[.~' ~ L~.t . (2.3') In this case the correctly specified wage growth equation is 6 (Wi . t+l" ~ X'i,t((~l~ --~ ¢~2j~ ~- o0 -- O) 1 / l og \ Wi., ] 1 +61+32-a t~2(1--(7) log( Li't ~ - - 1 + 61 + 62 - a \ L i , , - l J + ~oi , ,+1. (2.8') 5Formally, Zi.f+t = (-Iog(_Ut+l/_Uf) + eia+t + ~i.,+l)/(l +f-a) Iog(_U,+ I/_U,) + 6ei.,+t + (a - 1)(i,+ t)/(1 + 6 - o'). 6 The definition of ~., + 1 in (2.8') has been changed appropriately. and (hi.,+1 =((1 --a)× 122 E.L. Glaeser et al. / Journal of Monetary Economics 36 (1995) 117-143 If lagged growth rates of population are omitted from the regressions and 62 # 0, then the coefficients on initial characteristics will be biased (and in particular, the coefficient on initial wages should be biased downwards since current wages are positively correlated with lagged growth). Including lagged population growth rates into a wage growth regression is a test of the existence of this bias. As long as the coefficient on the lagged population growth rate is zero, we can accept that 62 = 0. If the coefficient is nonzero, then we must concern ourselves with the possibility that convergence comes from in- migration slowly meeting labor demand and wages being driven down by in-migration. Controlling for lagged growth rates may have substantial costs. If the first version of our model is correct, and our measures of the X variables are imperfect, and the true X variables are correlated with lagged growth (perhaps because lagged growth came about because of lagged X variables), then control- ling for lagged growth rates will decrease the signal to noise ratio in the X variables and spuriously lower the coefficients on these variables. 3. Data description The analysis in this paper is based on a sample of 203 U.S. cities between 1960 and 1990, although we also use some information from 1950. The data were hand-collected from County and City Data Books (1950, 1960, 1970), the 1990 Census (earlier censi provided the basis for the Data Books - the data are absolutely comparable), and from Taeuber and Taeuber (1965). Sample selec- tion is primarily determined by our reliance on Taeuber and Taeuber for racial characteristics of cities: they looked at all United States cities which in 1960 had over 1,000 occupied housing units with a nonwhite head. The resulting sample of 203 cities includes all but one of the largest 100 cities in the United States, but oversamples Southern cities in the next 100. We have verified that none of the principal results of this paper change if we restrict attention to the largest 100 cities. Table 1 presents the means and correlations of the variables used in this study (Appendix I provides variable definitions). The average city population in our sample of 203 cities is 269,000 in 1960, growing to an average of 288,000 in 1990, which amounts to only 8.5% growth over the whole period. East St. Louis shrunk the most in this sample (70% over 30 years) whereas Las Vegas grew the most (139% over 30 years). In this sample, 40% of the cities are in the South, 27% are in the Central Region, 19% are in the Northeast, and the remaining 14% are in the West. In 1960, these cities have an aver- age of 25.5% of their activities in manufacturing, an unemployment rate of 5.4%, black population equal to 18% of total, and the average median years of schooling of the population above 25 years old in these cities is 10.8 years. E.L. Glaeser et al. / Journal of Monetary Economics 36 (1995) 117-143 Table I Means and standard deviations, city variables 123 Mean Std. dev. Minimum Maximum Changes Growth in log of population 1960 1990 0.085 0.396 - 0.691 1.389 Growth in log of population 1950-1960 0.037 0.054 - 0.065 0.309 1960 Startiny variables Log of population 11.864 0.931 10.168 15.867 Per capita income ($1000) 1.993 0.4088 1.150 3.872 Migrants per capita 17.154 8.588 4.2 42.6 Geographical South 0.397 0.490 0 I Central 0.270 0.445 0 1 Northeast 0.191 0.394 0 1 Race Segregation index 86.19 7.46 60.4 98.1 Weighted segregation index 1.589 1.090 0.145 5.054 % nonwhite 18.09 11.91 2.1 57.5 Labor force Unemployment rate 5.4 1.7 1.9 10.5 Manufacturing share of employment 0.255 O. 118 0.043 0.589 Education Median years of schooling 10.833 1.116 8.4 13.7 % of pop with 16 + yrs of school 8.9 4.533 2.5 38.7 % of pop with 12-15 yrs of school 33.781 6.463 17.4 49.5 % of pop with less than 5 yrs of school 8333 3.948 1.9 21.4 Income inequality % of pop earning under $3000 19.437 7.411 5.0 41.6 % of pop earning over $10,000 15.98 6.383 5.7 43.8 Government Per capita revenue 103.639 64.423 11.775 581.353 Property tax share of revenue 0.445 0.175 0.082 0.876 lntergov funds share of revenue 0.161 0.109 0.0002 0.516 Per capita expenditure 110.002 66.107 13.931 634.169 Police share of expenditure 0.121 0.043 0.039 0.240 Highway share of expenditure 0.133 0.070 0.014 0.428 Sanitation share of expenditure 0.143 0.083 0.026 0.534 Per capita debt 205.326 164.684 8.146 1679.698 However, as Table 1 illustrates, there is a great deal of variation in all of these variables. We use the segregation index from Taeuber and Taeuber (1965). This index is defined as the percentage of nonwhites who would have to move so that the T a b le 2 R a w c o rr e la ti o n s , c it y v a ri a b le s G R 6 0 9 0 G R 5 0 6 0 L N P O P P C IN C S T H C E N T N E S E G C h an g es ¢~ E" G ro w th i n l o g o f p o p u la ti o n 1 9 6 0 -1 9 9 0 1 0 .5 2 - 0 .0 8 - 0 .0 2 0 .2 6 - 0 .2 8 - 0 .3 2 0 .0 8 G ro w th i n l o g o f p o p u la ti o n 1 9 5 0 -1 9 6 0 0 .5 2 1 - 0 .0 2 0 .0 0 0 .2 3 - 0 .1 5 0 .3 5 0 .2 3 - 1 9 6 0 st a rt in g va ri ab le s L o g o f p o p u la ti o n - 0 .0 8 - 0 .0 2 1 O . 1 1 - 0 .1 3 0 .0 4 0 .0 1 - 0 .0 2 P e r c a p it a in c o m e (0 0 0 ) - 0 .0 2 0 .0 0 O . 1 1 1 - 0 .4 7 0 .0 7 0 .2 2 - 0 .4 0 G eo g ra p h ic al ~ .~ S o u th 0 .2 6 0 .2 3 - 0 .1 3 - 0 .4 7 1 - 0 .4 9 - 0 .3 9 0 .5 4 C e n tr a l - 0 .2 8 - 0 .1 5 0 .0 4 0 .0 7 - 0 .4 9 1 - 0 .3 0 0 .1 2 N o rt h e a s t - 0 .3 2 - 0 .3 5 0 .0 1 0 .2 2 - 0 .3 9 - 0 .3 0 1 - 0 .4 2 ,~ R a ce S e g re g a ti o n in d e x 0 .0 8 0 .2 3 - 0 .0 2 - 0 .4 0 0 .5 4 0 .1 2 - 0 .4 2 1 % n o n w h it e - 0 .0 4 - 0 .1 7 - 0 .0 2 - 0 .3 8 0 .5 1 - 0 .2 1 - 0 .1 6 0 .3 4 La b o r fo rc e ~ " U n e m p lo y m e n t r a te - 0 .2 5 - 0 .2 8 O .O 1 - 0 .3 3 - 0 .1 9 O .O l O .l 1 - 0 .1 3 M a n u fa c tu ri n g s h a re o f e m p lo y m e n t - 0 .4 5 - 0 .3 5 - 0 .0 2 - 0 .0 1 - 0 .4 8 0 .4 0 0 .3 2 - 0 .1 9 ,~ E d u ca ti on M e d ia n y e a rs o f s c h o o li n g 0 .3 9 0 .4 5 0 .0 1 0 .5 8 - 0 .0 8 0 .0 0 - 0 .2 7 - 0 .0 7 G ov er n m en t P e r c a p it a re v e n u e - 0 .0 7 - 0 .0 9 0 .2 6 0 .2 0 - 0 .1 4 - 0 .1 7 0 .3 4 - 0 .2 6 P e r c a p it a e x p e n d it u re - 0 .0 2 - 0 .0 5 0 .2 2 0 .1 4 - 0 .0 8 - 0 .2 0 0 .2 8 - 0 .2 1 P e r c a p it a d e b t 0 .2 2 0 .2 3 0 .2 3 - 0 .0 8 0 .2 1 - 0 .1 9 - 0 .0 5 0 .1 7 T a b le 2 ( c o n ti n u e d ) % N O N W U N E M M E R G M E D Y R S R E V E X P D E B T r~ C h a n g e s G ro w th in l o g o f p o p u la ti o n 1 9 6 0 -1 9 9 0 - 0 .0 4 - 0 .2 5 - 0 .4 5 0 .3 9 - 0 .0 7 0 ,2 2 - 0 .0 2 E" G ro w th in l o g o f p o p u la ti o n 1 9 5 0 -1 9 6 0 - 0 .1 7 - 0 .2 8 - 0 .3 5 0 .4 5 - 0 .0 9 0 ,2 3 - 0 .0 5 1 9 6 0 st a rt in g va ri ab le s L o g o f p o p u la ti o n - 0 .0 2 0