Employment subcenters in Chicago: past, present, and future.

McMillen, Daniel P.

Introduction and summary

Employment in large American metropolitan areas has become increasingly decentralized over time. However, employment is not distributed evenly throughout the suburban landscape. Firms congregate at highway interchanges, along rail lines, and in former satellite cities. An employment subcenter is a concentration of firms large enough to have significant effects on the overall spatial distribution of population, employment, and land prices. Large subcenters can look remarkably similar to a traditional central business district (CBD), with thousands of workers employed in a wide variety of industries. A polycentric city--a metropolitan area with a strong central business district and large subcenters--can potentially combine the advantages of the traditional centralized city and a more decentralized spatial form. Large subcenters offer agglomeration economies to firms, while potentially reducing commuting times for suburban workers. As traffic congestion increases in the suburbs, an important advantage of subcen ters over more scattered employment is they can potentially be served effectively with public transportation. As a result, the location and growth patterns of subcenters in major cities are of interest to policymakers.

In this article, I document the growth of employment subcenters in the Chicago metropolitan area from 1970 to 2000. I also use employment forecasts generated by the Northeastern Illinois Planning Commission to identify subcenters for 2020. Chicago had nine subcenters in 1970. The number of subcenters rose to 13 in 1980, 15 in 1990, and 32 in 2000, and is projected to drop to 24 in 2020. Existing subcenters are becoming larger and are particularly likely to expand along major expressways. I use a formal cluster analysis to categorize the subcenters by employment mix in 1980, 1990, and 2000. Although Chicago's subcenters had high concentrations of manufacturing jobs in the past, the industry mix now closely resembles that of the overall metropolitan area.

I use distance from the nearest subcenter as an explanatory variable in employment and population density regressions (density is the number of workers or residents per acre). The results imply that the traditional city center still has a significant and widespread influence on densities in the Chicago metropolitan area. Firms tend to locate near important parts of the transportation system--near highway interchanges and rail stations and along freight rail lines. Subcenters also have pronounced effects on the distribution of jobs: Employment density rises significantly near subcenters. However, apart from O'Hare Airport, Chicago's subcenters are still not large enough to increase population density in neighboring areas. Construction of high-density housing near subcenters could potentially reduce aggregate commuting costs.

Subcenters are not unique to the Chicago metropolitan area. In related work, McMillen and Smith (2004) have identified subcenters in 62 large American urban areas in 1990. All but 14 of these cities have employment centers. The Los Angeles and New York metropolitan areas have the most subcenters, with 46 in Los Angeles and 38 in New York. In all 62 of these urban areas, employment density continues to decline significantly with distance from the traditional city center. Employment density also declines significantly with distance from the nearest subcenter in those cities following a polycentric form. Using the subcenter count as the dependent variable for a Poisson regression, I find that the number of subcenters rises with the urban area's population, and cities with higher commuting costs tend to have more subcenters.

Subcenters in the Chicago metro area

Subcenters are areas outside the traditional central business district with employment levels large enough to have significant effects on the overall spatial distribution of jobs and population. Subcenter locations are not always obvious or easy to identify beforehand. Areas near the city center with high employment density may not differ significantly from surrounding sites. Remote sites with relatively high employment densities may not have significant effects on the spatial distribution of jobs and population. Researchers such as McDonald (1987), Giuliano and Small (1991), Craig and Ng (2001), and McMillen (2001) have proposed procedures that objectively identify subcenter sites using standard data sources.

In this article, I use Giuliano and Small's (1991) approach to identify subcenters in the Chicago metropolitan area between 1970 and 2000 and to predict subcenter sites in 2020. Analyzing the Los Angeles metropolitan area, Giuliano and Small define a subcenter as a set of contiguous tracts that each have at least ten employees per acre and together have at least 10,000 employees. (1) The number of subcenters is sensitive to these two cutoffs. Higher minimum density levels or higher values for total employment produce fewer subcenters. To ensure reasonable results, one needs local knowledge to guide the choice of cutoffs. After some experimentation, I chose cutoff points of 15 employees per acre and 10,000 total workers. These values produce a reasonable number of subcenters in each period. McMillen and Smith (2004) provide a detailed explanation of the subcenter identification procedure.

Data on employment and population were provided by the Northeastern Illinois Planning Commission (NIPC). NIPC conducts decennial land use surveys for the six-county Chicago primary metropolitan statistical area. The six counties are Cook, DuPage, Kane, Lake, McHenry, and Will. The unit of observation is the quarter section, which is 160 acres or one-quarter of a square mile. There are slightly more than 15,000 quarter sections in these six counties. NIPC provided employment data for 1970, 1980, 1990, and 2000, and forecasts for 2020. Population data are not yet available for quarter sections in 2000, although forecasts are available for 2020. Comparisons over time for individual quarter sections are not completely reliable because NIPC has changed its methodology. In 1970 and 2020, NIPC reports employment data for any quarter section with jobs. In 1980 and 1990, only quarter sections with ten or more employees are included in the dataset, whereas the minimum employment level is eight in 2000. Due to this limi tation, the dataset has more tracts with positive values for employment in 1970 than in 1980-2000, despite the general decentralization of the Chicago metropolitan area over this time.

Figures 1 and 2 show the subcenter sites. The number of subcenters rises from nine in 1970 to 13 in 1980, 15 in 1990, and 32 in 2000. The NIPC employment forecasts lead to a prediction of 24 subcenters in 2020. Figure 1, panel A shows that in 1970 there was a subcenter in Hyde Park on the south side of Chicago, along with a ring of subcenters that nearly encircles the city. The number and geographic scope of the subcenters expand over time. O'Hare Airport is the center of a large conglomeration of subcenter employment. Another group of subcenters spreads along the I-88 toll way running west out of the city. In 2000 (panel D), small subcenters appear at the fringes of the metropolitan area in Kane County and Will County. These sites are in the old satellite cities of Elgin, St. Charles, Aurora, and Joliet. The NIPC forecasts suggest that the satellite Cities will not continue to qualify for subcenter status in 2020, although the accuracy of this forecast appears questionable in light of the ongoing decentraliz ation of employment in the Chicago metropolitan area. In 2020, also, several formerly separate subcenters along I-88 and near O'Hare are predicted to merge (figure 2). The general pattern of figure 1 is one of rapidly expanding subcenters, with most of the growth occurring near O'Hare Airport and along the major highways serving the city.

Subcenter clusters

Employment data are available by sector for 1980, 1990, and 2000. Table 1 presents data on the total number of jobs and the distribution of employment across five sectors in the subcenters identified for these years. The sectors are manufacturing; retail; services; transportation, communication, and utilities (TCU); finance, insurance, and real estate (FIRE); and government (federal, state, and local). I also use these sectors as headings for groups of similar subcenters that I identify using a formal cluster analysis. The cluster analysis (2) categorizes subcenters by looking for groups with similar employment compositions. The cluster analysis is performed for a given number of clusters, leaving it to the analyst to specify the appropriate number. Experimentation suggested that specifying five groups produces reasonable results, with clusters that are dominated by jobs in one of the five primary employment categories. Table 1 groups the subcenters by cluster in each year, with the subcenter sites identified by the municipalities (or neighborhoods within Chicago) in which they are located.

In 1980, eight of the 13 subcenters were dominated by manufacturing jobs. Traditional manufacturing sites such as Cicero, the Clearing District of Chicago, and Franklin Park appear as subcenters, along with newer suburban sites such as Elk Grove Village, Niles-Skokie, and Schaumburg. The manufacturing subcenters are generally larger than the service, TCU, and government subcenters, with total employment ranging from 13,430 in Rosemont to 46,740 in Franklin Park-Melrose Park. Although these subcenters are dominated by manufacturing, they also can include significant numbers of other types of jobs. For example, 28.89 percent of the Albany Park subcenter's jobs are in the FIRE sector, compared with 55.05 percent in manufacturing. The Clearing-West Lawn and Schaumburg subcenters have many retail jobs, representing 27.73 percent and 32.04 percent of the jobs in those subcenters, respectively. Rosemont is a diversified subcenter, having a similar number of jobs in manufacturing, service, TCU, and FIRE. Of the remai ning subcenters in 1980, three specialize in the service sector (Evanston, Oak Brook, and the Hyde Park area of Chicago, which includes the University of Chicago), one specializes in TCU (O'Hare), and one specializes in government (Broadview-Maywood-Qak Park). Maywood has a significant county governmental facility, Broadview has several township offices, and Oak Park, which is fairly large in population, has several village and township offices. Oak Brook, which is the site of a regional shopping mall and is near the intersection of the Tri-State and East-West tollways, also includes many retail and TCU jobs: These two sectors account for 22.95 percent and 21.78 percent of the jobs in the subcenter, respectively.

Table 1 shows that the subcenters continue to be dominated by manufacturing jobs in 1990, although the locations have changed somewhat. Whereas the manufacturing subcenters were formerly concentrated in Chicago and in the near western suburbs, by 1990 they are more apt to be in the northwestern suburbs and near O'Hare Airport. New manufacturing sites in this area include Addison, Arlington Heights, and Palatine. Another new manufacturing subcenter appears in the rapidly growing western suburb of Naperville. These manufacturing subcenters range in size from 10,120 in Naperville to 95,420 in Elk Grove Village-Schaumburg. Several of the subcenters also include many TCU jobs, although they are placed in another category: TCU accounts for 38.51 percent of the jobs in the Addison subcenter, 20.21 percent in Bedford Park-Chicago Lawn-West Lawn, 22.97 percent in Des Plaines-Rosemont, 27.10 percent in Elk Grove Village-Schaumburg, 24.90 percent in Naperville, 21.65 percent in Niles-Skokie, and 40.82 percent in Palatin e. Five subcenters specialize in service employment in 1990: The sector accounts for 48.78 percent of the employment in Bellwood-Broadview-Maywood, 32.47 percent in Deerfield-Northbrook, 57.74 percent in Evanston, 39.40 percent in Oak Brook, and 98.80 percent at the University of Chicago. The O'Hare subcenter continues to be dominated by TCU employment in 1990. None of the subcenters is placed in the government category in 1990.

The list grows to 32 subcenters in 2000 from 15 in 1990. The number of manufacturing subcenters falls to six-Addison, Glenview, North Chicago, Schaumburg, St. Charles, and Wheeling. All the manufacturing subcenters are now in more distant suburbs. Retail appears as a subcenter category in 2000, with sites in Deerfield-Northbrook (classified as service in 1990), Franklin Park, Hoffman Estates, and Melrose Park. The Hoffman Estates subcenter is a result of the movement of the Sears corporate headquarters out of Chicago. The number of service sector subcenters also increases significantly, with sites in Aurora, Broadview-Forest Park, Cicero-Oak Park, Elk Grove Village, Evanston, Glenbard, Joliet, Lincolnshire, Lisle--Naperville, and Oak Brook. In addition, TCU accounts for five subcenters in 2000, one subcenter specializes in FIRE, and six have large concentrations of government employment. The largest subcenters are in Schaumburg (82,092 employees) and Elk Grove Village (101,012 employees). In 2000, the subcent er job mix closely resembles the employment composition of the full metropolitan area. (3)

Employment and population density in Chicago

The spatial distribution of jobs and residences can be summarized by regressing measures of employment and population density on a set of explanatory variables, including distance from Chicago's traditional city center and measures of proximity to subcenter sites. Population density functions have a long history in urban economics, dating back to Clark (1951). Issues involved in estimation and a review of studies up to the late 1980s are reviewed in McDonald (1989). Employment density functions are estimated less frequently. Prominent examples include Booth (1999), Combes (2000), McDonald (1985), McDonald and Prather (1994), McMillen and McDonald (1997), and Small and Song (1994). With the natural logarithm of density as the dependent variable, the coefficient for distance from the central business district (CBD) or city center is referred to as the "CBD gradient." The gradient measures the percentage change in density associated with a one-mile increase in distance from the city center. It is a simple measur e of centralization: Density declines rapidly with distance in a highly centralized city, leading to large negative values for the estimated CBD gradient. Empirical studies suggest that most cities in the world have become increasingly decentralized over the last century, although employment generally remains more centralized than population.

Explanatory variables for the estimated density functions include distance from the traditional city center at the intersection of State and Madison streets, distance from O'Hare Airport, and distance from the nearest quarter section that is part of a subcenter. Distance from the nearest subcenter enters the estimating equations in inverse form, because I expect the effect of proximity to a subcenter to decline rapidly with distance. Proximity to subcenters increases densities if the coefficient for this variable is positive, and the effect rises over time if the coefficient becomes larger over time.

Other explanatory variables have localized effects on densities that can be accounted for using simple dummy variables. I include dummy variables that equal one when a quarter section is within one-third of a mile and between one-third and one mile of the following sites: a highway interchange, a commuter rail station, an elevated train line (the "el"), a station on an electric line serving the South Side, and Lake Michigan. I distinguish between commuter rail, el, and electric train lines because they have different areas and clienteles. The commuter rail lines primarily serve the suburbs, and have long intervals between stops. El lines are nearly entirely within the City of Chicago, and have frequent stops. The electric train line is something of a hybrid. It runs from downtown Chicago to the distant southern suburbs, along with a separate spur to Northwest Indiana. Although it primarily serves suburbanites, it resembles the el in making frequent stops within the city.

Table 2 presents detailed employment density estimates. The results indicate that employment fell by 5.6 percent with each mile from the Chicago city center in 1970. The rate of decline falls to 2.2 percent in 1980 as Chicago becomes more decentralized, and remains at about that level for 1990 (2.3 percent) and 2000 (2.2 percent again). The rate of decline is expected to be 1.9 percent per mile in 2020, based on NIPC employment forecasts. With the exception of 2000, proximity to O'Hare also increases employment density. Employment density is estimated to decline by 1.0 percent per mile in 1980, 0.9 percent in 1990, and a forecasted 3.4 percent in 2020.

Other results in table 2 are much as expected. Employment density is higher near highway interchanges. Densities are estimated to be 30.6 percent higher within one-third of a mile of a highway interchange in 1970, compared with 37.9 percent in 2000, and a forecasted 40.5 percent in 2020. Densities decline somewhat in the next two-thirds of a mile from a highway interchange. In 1970, densities are 18.1 percent higher in the ring from one-third to one mile of a highway interchange than in more distant sites, compared with 21.6 percent in 2000 and a forecasted 13.6 percent in 2020. Similarly, densities are higher near commuter rail stations. For example, in 1970 employment density is estimated to be 85.2 percent higher within one-third of a mile of a commuter station and 50.6 percent higher in the one-third to one-mile ring, compared with more distant locations. Commuter train stations decline in importance in subsequent years. In 2020, employment density is expected to be 54.5 percent higher within one-third of a mile of a commuter station and 9.4 percent higher in the one-third to one-mile ring. Proximity to stations on the electric line has similar effects on employment, except the effect is confined to the initial zero to one-third of a mile ring.

Lake Michigan has little or no effect on employment density. Quarter sections through which the Chicago River or the Sanitary and Ship Canal runs tend to have high employment density. In 2000, densities are estimated to be 58.3 percent higher in quarter sections with the river or canal. Although sites within Chicago had higher densities from 1970 to 1990, the effect declines from a 103.5 percent increase in 1970 to 39.6 percent in 1980 to 13.5 in 1990. After controlling for other explanatory variables, city locations do not have higher employment density in 2000 or 2020.

The final set of results in table 2 includes the effects of proximity to subcenters on employment density. The 1970 and 2020 regressions include a single variable representing the inverse of distance from the nearest subcenter. The regressions confirm the importance of subcenters in accounting for the spatial distribution of employment density. Letting d represent the distance from the nearest subcenter, the marginal effect of distance is -0.774/[d.sup.2] in 1970 and a forecasted -0.568/[d.sup.2] in 2020. The minimum value for d is 0.25. Thus, the estimated marginal effect of distance from the nearest subcenter in 1970 is -12.38 at subcenter sites, with the effect falling to -0.77 after one mile, and -0.19 after two miles. Comparable values for 2020 are -9.09, -0.57, and -0.142, respectively. Although subcenters do not affect employment over as wide an area as the traditional CBD, the high t-values of 19.209 in 1970 and 27.253 in 2020 indicate that they are critically important determinants of the spatial dis tribution ofjobs in the Chicago area.

For the years with data on employment sectors (1980, 1990, and 2000), I include separate explanatory variables for each cluster type. For these years, the regressions include dummy variables indicating the sector for the closest subcenter and interactions between these dummy variables and the inverse of distance from the subcenter. The dummy variables are generally not statistically significant. The coefficients for the inverse of distance from the nearest subcenter again indicate that employment densities rise significantly near subcenters. In 1980, the marginal effect of distance from the nearest subcenter is -0.471 at a distance of one mile when the nearest subcenter is in the government cluster, compared with -0.564 for service subcenters, -0.754 for TCU, and -0.665 for manufacturing. In 1980, these marginal effects are -0.600 for service, -2.5 10 for TCU, and -0.670 for manufacturing. In 2000, the marginal effect at one mile from a subcenter is -0.758 for retail, -0.645 for government, -0.784 for service , -0.756 for TCU, and -0.768 for manufacturing. The results are all highly significant. What is more surprising is that, with the exception of the TCU cluster in 1990, the estimated marginal effects do not vary much across sectors.

Table 3 presents abbreviated results for comparable population density function estimates. Population density is estimated to decline by 7.3 percent with each mile from the Chicago city center in 1970, compared with 7.8 percent in 1980, 7.2 percent in 1990, and a forecasted 6.6 percent in 2020 (recall that population data are not yet available for 2000 at the quarter section level). These results are somewhat surprising in their implication that the CBD gradient is now larger for population than for jobs after controlling for the effects of other variables. O'Hare Airport also has a significant effect on population density. Controlling for other variables, each additional mile from O'Hare reduces population density by 4.9 percent in 1970, 4.6 percent in 1980, 6.0 percent in 1990, and a forecasted 5.9 percent in 2020.

In keeping with the results of McMillen and McDonald (2000), proximity to employment subcenters is estimated to reduce population density. Each additional mile from the nearest employment subcenter increases density by 16.4 percent in 1970, 44.9 percent in 1980, 36.2 percent in 1990, and a forecasted 47.3 percent in 2002. This result has two explanations. First, our density measures are gross rather than net, meaning that density is measured per acre of total land area rather than per acre of residential land area. Densities are low near subcenters because by definition much of the land area in subcenters is in nonresidential use.

Second, although subcenters are getting bigger, they are not yet large enough in the Chicago area to lead to large increases in population density in neighboring sites. Subcenter employment has increased primarily through an increase in the number of subcenters rather than by the creation of a few larger subcenters that rival the traditional CBD in their effects on density patterns.

Subcenters in other metro areas

Subcenters are not only a Chicago phenomenon. Studies by Anderson and Bogart (2001), Bogart and Ferry (1999), Cervero and Wu (1997, 1998), Craig and Ng (2001), Giuliano and Small (1991), McMillen (2001), and Small and Song (1994) have identified subcenters in Cleveland, Dallas, Houston, Indianapolis, Los Angeles, New Orleans, St. Louis, and the San Francisco Bay Area. Recently, Baumont, Ertur, and LeGallo (2002) and Muniz, Galindo, and Garcia (2003) have extended the analysis to the European cities of Dijon, France and Barcelona, Spain.

The remainder of this section summarizes the results of a recent study by McMillen and Smith (2004), which is the first to apply a single subcenter identification procedure to a large number of metropolitan areas. They use a variant of the Giuliano and Small (1991) procedure to identify subcenters in 62 large U.S. metropolitan areas. The data come from the urban element of the Census Transportation Planning Package, which is produced by the Department of Transportation's Bureau of Transportation Statistics (BTS). The BTS obtained special tabulations of 1990 U.S. Census data to match Census data with the BTS geographic unit, called the Transportation Analysis Zone. These zones, which vary across metropolitan areas, are typically smaller than Census tracts or zip codes and often are the same as Census blocks.

Figure 3 shows the distribution of the number of subcenters across the 62 metropolitan areas. Fourteen of the metropolitan areas have no subcenters. Eight metropolitan areas--Boston, Chicago, Dallas--Fort Worth, Los Angeles, New York, the San Francisco Bay Area, Seattle, and Washington, DC--have at least ten subcenters. The two largest cities, New York and Los Angeles, have the most subcenters with 38 and 46, respectively. Chicago is next with 15.

For a subset of the 62 metropolitan areas, table 4 presents the results of simple regressions of the natural logarithm of employment density on distance from the traditional central business district and the inverse of distance from the nearest zone that is part of a subcenter. The [R.sup.2]s for the regression indicate that these two variables alone account for no less than 21.7 percent of the variation in employment density (in San Francisco), with an average of 38.3 percent and a maximum of 57.0 percent (in Washington, DC). The traditional CBD still has a tremendous impact on employment densities. For example, employment densities in Atlanta are estimated to decline by 21.3 percent with each additional mile from the CBD after controlling for proximity to subcenters. In table 4, the average CBD gradient is -12.8 percent, with a range of -4.2 percent in Chicago to--22.7 percent in Kansas City.

The coefficients for the inverse of distance from the nearest subcenter zone indicate that employment densities are higher near subcenters. For example, in Atlanta the estimated marginal effect of distance from the nearest subcenter is estimated to be -0.482/[d.sub.2], where d is distance. The marginal effect of distance is -7.71, -.48, and -.12 for sites that are one-quarter mile, one mile, and two miles, respectively, from the nearest subcenter in Atlanta. The average coefficient for distance from the nearest subcenter is 0.417 in table 4, with a range of 0.172 (New York) to 0.649 (Chicago). These results imply that the rate of decline in employment densities with distance from the nearest subcenter is highest in Chicago and lowest in New York.

Theoretical and empirical models of subcenter formation have thus far developed in relative isolation. Theoretical models have focused on examining the equilibrium spatial configuration of polycentric cities rather than on producing empirically testable, comparative static results. Models such as those developed by An as and Kim (1996), Berliant and Konishi (2000), Fujita, Krugman, and Mon (1999), Fujita and Ogawa (1982), Fujita, Thisse, and Zenou (1997), Helsley and Sullivan (1991), Henderson and Mitra (1996), Konishi (2000), Wieand (1987), and Yinger (1992) emphasize the role that population and commuting cost play in altering the equilibrium spatial configuration of a city. The primary prediction is that the equilibrium number of subcenters tends to rise with population and commuting costs.

This prediction can be tested for our sample of 62 metropolitan areas using the number of subcenters as the dependent variable. Poisson regression is the appropriate estimation procedure for this type of count data (Cameron and Trivedi, 2001). The key explanatory variables are population and commuting costs. Population, which is measured over the full metropolitan area, ranges from 127,855 in Laredo, Texas, to 16,885,598 in New York. I use two measures of commuting cost. The first is a travel time index developed by the Texas Transportation Institute for its Mobility Monitoring Program. It is designed as a measure of peak-period congestion. The Travel Rate Index exceeds 1.0 if it takes longer on average to make a trip in congested periods than at other times of the day. As an alternative, I also use a measure of highway capacity-thousands of miles traveled on average daily by all vehicles per mile of freeway lanes (DVMTLANE). This index focuses on average travel time across the day, whereas the travel time in dex focuses on travel at peak commuting times. Its advantage is that it has a greater claim to being exogenous or predetermined: The highway capacity in most American cities is a direct result of federal highway programs from the 1950s and 1960s. Strict exogeneity is not essential because I am estimating an equilibrium relationship. The correlation is high among all of the indexes available from the Texas Transportation Institute's Urban Mobility Study, and the results are not sensitive to the choice.

Other explanatory variables control for differences among cities. I include the central city's proportion of the urban area's population, because subcenters may be more likely to form when there are more suburbs. Competition among suburbs for firms may produce subcenters, whereas a large central city may adopt policies to encourage the continued dominance of the traditional CBD. The median income of the central city has ambiguous effects on subcenter formation. On the one hand, high income suggests a vibrant central city, which may discourage subcenter formation. But incomes in the central city and suburbs are highly correlated, and subcenters may be more likely to form if higher income increases the aversion to long commutes.

I include the average tract size in the regressions, because McMillen and Smith (2004) find that the subcenter identification procedure tends to find more subcenters when tract sizes are small. I include the last two variables, median house age and the age of the central city, because analysts such as Garreau (1991) have suggested that subcenters will come to dominate American cities in the future. Thus, newer cities may be more likely to have already developed subcenters. Median house age, as reported by the 1990 U.S. Census for 1990, is one measure of a city's age. I also use a variable suggested by Brueckner (1986) to measure city age: the number of years since the central city first reached 25 percent of its 1990 population level.

Table 5 displays the Poisson regression results. The estimated coefficients are interpreted as semi-elasticities. For example, the estimated coefficient for population in model 1 indicates that an additional million in population raises the expected number of subcenters by 14.8 percent. This estimate is stable across the three alternative model specifications, rising to 15.1 percent when I use DVMTLANE in place of the travel rate index to measure commuting cost and to 17.3 percent when I use only population and DVMTLANE as explanatory variables. The travel rate index and DVMTLANE have the expected positive signs, indicating that higher commuting cost leads to more subcenters. The coefficients for DVMTLANE indicate that an additional thousand miles traveled on average per mile of freeway lane raises the expected number of subcenters by 9.4 percent in model 2 and 9.3 percent in model 3.

The remaining explanatory variables are not important determinants of the number of subcenters in this sample. Metropolitan areas with large central cities tend to have fewer subcenters, but estimated coefficients for other explanatory variables--median income, average tract size, median house age, and age of the central city--are statistically insignificant. The pseudo-[R.sup.2]s for the regressions (Cameron and Windmeijer, 1996) imply that the explanatory variables account for approximately 80 percent of the variation in the natural logarithm of the number of subcenters. Table 5 suggests a strong, simple empirical regularity in the number of subcenters in large metropolitan areas: The number of subcenters rises with population and commuting costs.

Conclusion

The traditional central business district is still the largest single employment site in most metropolitan areas. However, urban areas have become increasingly decentralized over time, and many cities now have more jobs in the suburbs than in the central city. Jobs are not spread randomly about the suburban landscape. Firms tend to locate at sites with ready access to the transportation system. Large employment subcenters have developed in many metropolitan areas that offer agglomeration economies to firms, while potentially reducing commuting times for suburban workers.

This article has documented the growth of employment subcenters in the Chicago metropolitan area between 1970 and 2000 and used forecasts of future employment to predict subcenter sites in 2020. A cluster analysis suggests that the employment mix in the subcenters has changed from predominantly manufacturing in 1970 to a mix that now closely resembles that of the overall metropolitan area. A regression analysis of employment density in the Chicago metropolitan area suggests that density rises near highway interchanges, rail stations, and along freight rail lines. Employment density also rises significantly in the area around employment subcenters.

Subcenters are found throughout the United States. Chicago had only 15 subcenters in 1990, New York had 38, and Los Angeles had 46. Of 62 large metropolitan areas analyzed in this article, 48 had at least one subcenter. The number of subcenters has a remarkably predictable pattern across the 62 urban areas. Poisson regression results imply that the number of subcenters rises with population and commuting costs. Thus, as cities grow, one can expect that subcenters will develop as firms congregate near intersections of major highways and in formerly satellite cities. Although new subcenters do not offer the same level of agglomeration economies as the traditional central city, they do offer lower land costs, easy access to highways, and the possibility of reduced wages for suburban workers whose commuting costs are reduced.

[FIGURE 3 OMITTED]

TABLE 1

Subcenter characteristics

 Subcenter employment composition (%)
 Total
Subcenter Cluster employment Mfg. Retail

1980

Albany Park-Jefferson mfg 14,640 55.05 0.41
 Park-North Park
Cicero-Austin mfg 28,210 62.96 17.23
Clearing-West Lawn mfg 10,890 45.36 27.73
Elk Grove Village mfg 37,030 39.08 7.05
Franklin Park-MeLrose Park mfg 46,740 68.66 11.49
Niles-Skokie mfg 40,800 65.66 5.96
Rosemont mfg 13,430 18.63 8.59
Schaumburg mfg 23,000 46.22 32.04
Evanston serv 22,430 3.79 12.26
Oak Brook serv 27,500 12.36 22.95
University of Chicago serv 15,300 0.07 0.26
 (Hyde Park)
O'Hare tcu 11,970 0.00 20.55
Broadview-Maywood-Oak Park govt 22,260 7.23 10.42

1990

Addison mfg 11,790 42.32 0.85
Arlington Heights mfg 15,270 55.73 4.98
Bedford Park-Chicago mfg 16,230 49.23 5.67
 Lawn-West Lawn
Des Plaines-Rosemont mfg 44,070 24.95 8.46
Elk Grove Village-Schaumburg mfg 95,420 33.03 11.04
Elmhurst-Franklin
 Park-Melrose
 Park-Northlake mfg 50,250 46.61 14.31
Naperville mfg 10,120 17.00 13.54
Niles-Skokie mfg 27,620 45.08 5.25
Palatine mfg 10,290 49.17 2.82
Bellwood-Broadview-Maywood serv 21,730 17.67 1.47
Deerfield-Northbrook serv 26,730 17.92 22.15
Evanston serv 25,580 7.00 12.51
Oak Brook serv 76,760 7.43 18.88
University of serv 16,670 0.00 0.54
 Chicago (Hyde Park)
O'Hare tcu 40,340 0.00 9.22

2000

Addison mfg 29,593 33.12 8.01
Glenview mfg 15,215 40.47 5.49
North Chicago mfg 19,432 88.30 0.00
Schaumburg mfg 82,092 40.01 13.11
St. Charles mfg 10,815 51.20 16.98
Wheeling mfg 10,595 24.68 1.52
Deerfield-Northbrook retl 51,253 4.06 49.45
Franklin Park retl 25,064 30.93 47.12
Hoffman Estates retl 17,355 0.00 100.00
Melrose Park retl 54,550 6.19 71.37
Aurora-South serv 10,570 0.52 1.96
Broadview-Forest Park serv 28,119 8.70 0.79
Cicero-Oak Park serv 15,609 3.57 5.45
Elk Grove Village serv 101,012 20.92 4.19
Evanston serv 46,957 1.08 5.40
Glenbard serv 28,242 3.83 16.07
Joliet serv 10,917 0.35 5.06
Lincolnshire serv 33,121 5.64 3.34
Lisle-Naperville serv 34,197 8.76 14.42
Oak Brook serv 78,810 3.56 19.17
Bedford Park tcu 18,790 4.28 0.00
Bensenville-Elmhurst tcu 29,253 17.71 9.65
Midway Airport tcu 20,183 12.35 15.40
O'Hare tcu 61,527 0.00 9.57
Vernon Hills tcu 13,599 11.42 9.25
Prospect Heights fire 20,913 4.19 6.16
Arlington Heights govt 14,270 5.88 1.97
Aurora-North govt 14,268 0.00 0.00
Des Plaines-Rosemont govt 67,565 19.27 2.82
Elgin govt 26,119 11.58 0.89
Niles-Skokie-Northern Chicago govt 59,806 30.68 4.03
Norridge-Norwood Park govt 16,662 16.20 1.75

 Subcenter employment composition (%)

Subcenter Services TCU FIRE Government

1980

Albany Park-Jefferson 2.53 3.48 28.89 6.63
 Park-North Park
Cicero-Austin 4.86 8.29 0.00 2.30
Clearing-West Lawn 1.65 10.65 3.86 10.74
Elk Grove Village 3.83 44.56 0.16 0.00
Franklin Park-MeLrose Park 2.55 9.52 0.06 5.88
Niles-Skokie 5.27 17.28 1.81 0.20
Rosemont 24.11 25.89 17.96 1.55
Schaumburg 4.13 6.61 6.00 4.91
Evanston 51.67 2.41 25.28 4.15
Oak Brook 30.25 21.78 10.40 0.55
University of Chicago 96.08 0.33 0.07 2.75
 (Hyde Park)
O'Hare 10.69 51.04 0.00 17.71
Broadview-Maywood-Oak Park 36.21 12.62 5.75 27.18

1990

Addison 7.46 38.51 0.34 1.10
Arlington Heights 7.60 7.99 4.58 0.00
Bedford Park-Chicago 12.14 20.21 0.25 12.26
 Lawn-West Lawn
Des Plaines-Rosemont 25.86 22.97 10.29 3.53
Elk Grove Village-Schaumburg 15.05 27.10 7.68 1.07
Elmhurst-Franklin
 Park-Melrose
 Park-Northlake 13.47 14.83 1.59 6.79
Naperville 22.73 24.90 2.87 12.45
Niles-Skokie 14.45 21.65 5.54 0.91
Palatine 4.76 40.82 2.04 0.00
Bellwood-Broadview-Maywood 48.78 12.43 0.18 17.35
Deerfield-Northbrook 32.47 16.46 7.00 0.19
Evanston 57.74 2.15 13.88 3.91
Oak Brook 39.40 20.64 8.91 0.81
University of 98.80 0.12 0.00 0.00
 Chicago (Hyde Park)
O'Hare 7.68 76.03 0.05 6.79

2000

Addison 10.57 38.03 0.56 0.00
Glenview 24.96 23.35 0.20 0.00
North Chicago 0.00 11.70 0.00 0.00
Schaumburg 19.39 6.80 3.55 0.00
St. Charles 16.38 6.03 2.64 4.28
Wheeling 25.16 16.64 0.28 0.00
Deerfield-Northbrook 23.79 14.07 3.80 1.50
Franklin Park 2.29 16.66 0.00 0.37
Hoffman Estates 0.00 0.00 0.00 0.00
Melrose Park 16.18 4.41 0.92 0.00
Aurora-South 50.23 1.42 13.52 19.01
Broadview-Forest Park 88.33 1.64 0.00 0.00
Cicero-Oak Park 63.58 3.13 8.50 2.94
Elk Grove Village 36.98 22.73 9.88 1.11
Evanston 72.00 0.37 1.09 14.01
Glenbard 57.40 13.26 8.31 0.00
Joliet 43.67 2.83 4.09 21.89
Lincolnshire 78.27 11.85 0.00 0.00
Lisle-Naperville 40.79 17.02 16.16 0.66
Oak Brook 49.74 12.15 13.24 0.05
Bedford Park 1.37 94.10 0.00 0.00
Bensenville-Elmhurst 28.82 37.45 2.22 0.00
Midway Airport 14.16 35.18 22.22 0.00
O'Hare 3.02 87.37 0.00 0.00
Vernon Hills 8.93 45.97 24.42 0.00
Prospect Heights 1.32 2.55 85.77 0.00
Arlington Heights 23.75 5.05 1.85 30.07
Aurora-North 0.00 0.00 0.00 99.41
Des Plaines-Rosemont 28.33 16.19 4.72 27.69
Elgin 12.68 0.50 0.00 61.50
Niles-Skokie-Northern Chicago 23.16 8.18 1.83 27.63
Norridge-Norwood Park 31.53 2.04 0.18 46.61

Notes: Mfg. Is manufacturing; TCU is transportation, communications, and
utilities; and FIRE is finance, Insurance, and real estate.

Source: Northeastern Illinois Planning Commission, 1970-2000, decennial
land use surveys.

TABLE 2

Total employment density

 1970 1980

Miles from city center -0.056 -0.022 *
 (15.742) * (7.685)
Miles from O'Hare Airport -0.005 -0.010 *
 (1.599) (3.521)
0 - 1/3 mile from highway interchange 0.306 * 0.266 *
 (3.132) (3.759)
1/3 - 1 mile from highway interchange 0.181 * 0.259 *
 (2.996) (5.682)
0 - 1/3 mile from commuter rail station 0.852 * 0.541 *
 (5.858) (5.102)
1/3 - 1 mile from commuter rail station 0.506 * 0.182 *
 (8.123) (3.839)
0 - 1/3 mile from el station 0.937 * 0.770 *
 (5.752) (6.401)
1/3 - 1 mile from el station 0.557 * 0.291 *
 (4.877) (3.400)
0 - 1/3 mile from station on electric line 0.805 * 0.553 *
 (2.887) (2.711)
1/3 - 1 mile from station on electric line 0.173 0.173 **
 (1.244) (1.671)
0 - 1/3 mile from Lake Michigan -0.207 0.015
 (1.054) (0.090)
1/3 - 1 mile from Lake Michigan 0.276 * 0.223 *
 (2.019) (2.084)
Chicago River or canal runs through tract 0.386 0.433 *
 (1.552) (2.532)
Freight rail line within tract 0.723 * 0.398 *
 (12.893) (9.305)
Within City of Chicago 1.035 * 0.396 *
 (11.917) (5.909)
inverse of distance from the nearest subcenter 0.774 *
 (19.209)
Nearest subcenter is in retail cluster

Nearest subcenter is in government cluster 0.164
 (1.123)
Nearest subcenter is in service cluster -0.037
 (0.823)
Nearest subcenter is in TCU cluster 0.431
 (1.200)
Nearest subcenter is in FIRE cluster

Inverse of distance from nearest
 subcenter x retail cluster
Inverse of distance from nearest 0.471 *
 subcenter x government cluster (5.161)
Inverse of distance from nearest 0.564 *
 subcenter x service cluster (11.493)
Inverse of distance from nearest 0.754 *
 subcenter x TCU cluster (4.016)
Inverse of distance from nearest 0.665 *
 subcenter x manufacturing cluster (23.708)
Inverse of distance from nearest
 subcenter x FIRE cluster
Constant -0.325 * 0.253 *
 (3.553) (3.345)
[R.sup.2] 0.425 0.364
Number of observations 6,081 5,220

 1990 2000

Miles from city center -0.023 * -0.022 *
 (9.311) (6.334)
Miles from O'Hare Airport -0.009 * 0.005
 (3.461) (1.423)
0 - 1/3 mile from highway interchange 0.287 * 0.379 *
 (4.478) (5.048)
1/3 - 1 mile from highway interchange 0.180 * 0.216 *
 (4.309) (4.514)
0 - 1/3 mile from commuter rail station 0.576 * 0.608 *
 (5.632) (5.141)
1/3 - 1 mile from commuter rail station 0.106 * 0.127 *
 (2.414) (2.520)
0 - 1/3 mile from el station 1.038 * 0.551 *
 (9.147) (4.232)
1/3 - 1 mile from el station 0.592 * 0.146
 (7.339) (1.558)
0 - 1/3 mile from station on electric line 0.572 * 0.564 *
 (2.952) (2.560)
1/3 - 1 mile from station on electric line -0.036 0.027
 (0.376) (0.233)
0 - 1/3 mile from Lake Michigan -0.228 0.197
 (1.478) (1.065)
1/3 - 1 mile from Lake Michigan 0.005 0.100
 (0.049) (0.863)
Chicago River or canal runs through tract 0.284 ** 0.583 *
 (1.719) (2.906)
Freight rail line within tract 0.356 * 0.250 *
 (9.122) (5.546)
Within City of Chicago 0.135 * -0.044
 (2.174) (0.601)
inverse of distance from the nearest subcenter

Nearest subcenter is in retail cluster 0.044
 (0.448)
Nearest subcenter is in government cluster 0.208 *
 (2.195)
Nearest subcenter is in service cluster 0.047 -0.209 *
 (1.001) (2.746)
Nearest subcenter is in TCU cluster -4.978 * -0.141
 (1.965) (1.642)
Nearest subcenter is in FIRE cluster -1.218 *
 (2.390)
Inverse of distance from nearest 0.758 *
 subcenter x retail cluster (13.840)
Inverse of distance from nearest 0.645 *
 subcenter x government cluster (13.685)
Inverse of distance from nearest 0.600 * 0.784 *
 subcenter x service cluster (19.102) (24.082)
Inverse of distance from nearest 2.510 * 0.756 *
 subcenter x TCU cluster (3.707) (11.237)
Inverse of distance from nearest 0.670 * 0.768 *
 subcenter x manufacturing cluster (28.658) (17.003)
Inverse of distance from nearest 0.992 *
 subcenter x FIRE cluster (3.707)
Constant 0.403 * 0.078 *
 (5.677) (0.799)
[R.sup.2] 0.390 0.314
Number of observations 5,817 5,649

 2020

Miles from city center -0.019 *
 (7.448)
Miles from O'Hare Airport -0.034 *
 13.523)
0 - 1/3 mile from highway interchange 0.405 *
 (5.678)
1/3 - 1 mile from highway interchange 0.136 *
 (2.974)
0 - 1/3 mile from commuter rail station 0.545 *
 (4.619)
1/3 - 1 mile from commuter rail station 0.094 **
 (1.891)
0 - 1/3 mile from el station 1.152 *
 (8.634)
1/3 - 1 mile from el station 0.500 *
 (5.349)
0 - 1/3 mile from station on electric line 0.770 *
 (3.431)
1/3 - 1 mile from station on electric line 0.329 *
 (3.012)
0 - 1/3 mile from Lake Michigan 0.173
 (0.991)
1/3 - 1 mile from Lake Michigan 0.267 *
 (2.307)
Chicago River or canal runs through tract -0.020
 (0.108)
Freight rail line within tract 0.430 *
 (10.105)
Within City of Chicago 0.065
 (0.936)
inverse of distance from the nearest subcenter 0.568 *
 (27.253)
Nearest subcenter is in retail cluster

Nearest subcenter is in government cluster

Nearest subcenter is in service cluster

Nearest subcenter is in TCU cluster

Nearest subcenter is in FIRE cluster

Inverse of distance from nearest
 subcenter x retail cluster
Inverse of distance from nearest
 subcenter x government cluster
Inverse of distance from nearest
 subcenter x service cluster
Inverse of distance from nearest
 subcenter x TCU cluster
Inverse of distance from nearest
 subcenter x manufacturing cluster
Inverse of distance from nearest
 subcenter x FIRE cluster
Constant 0.840 *
 (11.308)
[R.sup.2] 0.376
Number of observations 7,522

Notes: Absolute t-values are in parentheses. The dependent variable is
the natural logarithm of employment density per acre.

"*" indicates significance at the 5 percent level.

"**" indicates significance at the 10 percent level.

Source: Author's calculations based on data from the Northeastern
Illinois Planning Commission, 1970-2000, decennial land use surveys.

TABLE 3

Population density

 1970 1980 1990 2020

Miles from city center -0.073 -0.078 -0.072 -0.066
 (30.963) (31.107) (29.670) (28.345)

Miles from O'Hare Airport -0.049 -0.046 -0.060 -0.059
 (22.795) (20.052) (25.880) (25.942)

Inverse of distance from
 nearest subcenter -0.164 -0.449 -0.362 -0.473
 (4.305) (13.263) (12.589) (18.731)

[R.sub.2] 0.512 0.423 0.429 0.361

Number of observations 10369 10942 11129 11687

Notes: Absolute t-values are in parentheses. The dependent variable is
the natural logarithm of population density. Other explanatory variables
include dummy variables representing locations within the City Chicago
and proximity to highways, commuter rail lines, el lines, electric train
lines, Lake Michigan, the Chicago River and the Sanitary and Ship Canal,
and freight rail lines.

Source: Author's calculations based on data from the Northeastern
illinois Planning Commisison, 1970-2000, decennial land use surveys.

TABLE 4

Employment density functions of selected metro areas, 1990

 No.of CBD Subcenter
 subcenters coefficient t-value coefficient

Atlanta 4 -0.213 -34.307 0.482
Boston 11 -0.097 -31.519 0.353
Chicago 15 -0.042 -28.760 0.649
Cincinnati 3 -0.170 -20.370 0.297
Cieveiand 3 -0.138 -24.385 0.199
Dailas 12 -0.089 -34.565 0.532
Denver 5 -0.095 -19.545 0.419
Detroit 8 -0.106 -35.274 0.484
Houston 8 -0.118 -33.938 0.583
Kansas City 2 -0.227 -26.767 0.487
Los Angeies 46 -0.048 -16.779 0.449
Minneapoiis-St. Paui 7 -0.201 -34.608 0.373
New York 38 -0.097 -77.606 0.172
Philacleiphia 4 -0.109 -25.083 0.527
Phoenix 5 -0.206 -31.049 0.308
San Diego 6 -0.090 -15.121 0.335
San Francisco 12 -0.056 -24.195 0.378
Seattie 14 -0.133 -21.009 0.438
St. Louis 5 -0.165 -24.630 0.453
Washington, DC 10 -0.153 -55.863 0.416


 t-value [R.sup.2] n

Atlanta 9.266 0.569 943
Boston 19.013 0.267 3744
Chicago 36.607 0.340 5935
Cincinnati 3.656 0.313 958
Cieveiand 3.153 0.399 991
Dailas 26.049 0.297 4379
Denver 8.864 0.277 1336
Detroit 19.342 0.388 2688
Houston 14.747 0.399 2128
Kansas City 5.896 0.529 732
Los Angeies 20.846 0.201 3051
Minneapoiis-St. Paui 11.020 0.559 1187
New York 17.631 0.306 14831
Philacleiphia 9.072 0.363 1350
Phoenix 7.923 0.545 996
San Diego 6.369 0.299 632
San Francisco 16.080 0.217 2913
Seattie 11.255 0.404 828
St. Louis 7.378 0.420 995
Washington, DC 22.782 0.570 3090

Note: The explanatory variables include an Intercept, distance from the
city center, and the Inverse of distance from the nearest subcenter.

Source: Author's calculations based upon data from the U.S. Department
of Transportation, Bureau of Transportation Statistics, Census
Transportaion Planning Package.

TABLE 5

Poisson regressions: Number of Subcenters

 Number of subcenters
 Model 1 Model 2 Model 3

Metro population (millions) 0.148 * 0.151 * 0.173 *
 (7.015) (7.670) (12.846)
Travel Rate Index 1.223 *
 (2.878)
DVMTLANE 0.094 0.093
 (3.190) (5.441)
Proportion of metro -1.479 * -1.490 * -1.710 *
 population in central city (2.494) (2.522) (3.694)
Median income in central city 0.029 0.027
 ($1,000) (1.512) (1.409)
Average tract size (sq. miles) -0.053 -0.046
 (1.293) (1.125)
Median house age (10 yrs.) 0.058 0.026
 (0.403) (0.175)
Central city age (10 yrs.) -0.006 0.012
 (0.135) (0.256)
Constant -1.045 -0.627 0.124
 (1.390) (0.900) (0.419)
Log-likelihood value -199.217 -118.177 -120.395
[R.sup.2] 0.811 0.816 0.806

Notes: Each regression has 62 observations, Absolute z-values are in
parentheses below the estimated coefficients. An asterisk indicates
significance at the 5 percent level.

Source: Author's calculations based upon data from the U.S. Department
of Transportation, Bureau of Transportation Statistics, Census
Transportation Planning Package, and from the U.S. Department of
Commerce, Bureau of the Census.

NOTES

(1.) The tracts analyzed by Giuliano and Small (1991) are transportation analysis zones, as defined by the Southern California Association of Governments. The average area of the tracts is about 1.75 square miles.

(2.) Performed using the program STATA.

(3.) In 2000, 33.7 percent of the jobs in the Chicago metropolitan area were in the service sector, 19.3 percent were in retail, 11.3 percent were in manufacturing, 11.1 percent were in TCU, and 4.3 percent were in the government sector.

REFERENCES

Anas, A., and I. Kim, 1996, "General equilibrium models of polycentric urban land use with endogenous congestion and job agglomeration," Journal of Urban Economics, Vol. 40, No. 2, pp. 232-256.

Anderson, N., and W. Bogart, 2001, "The structure of sprawl: Identifying and characterizing employment centers in polycentric metropolitan areas," American Journal of Economics and Sociology, Vol. 60, No. 1, pp. 147-169.

Baumont, C., C. Ertur, and J. LeGallo, 2002, "Intra-urban spatial distributions of population and employment: The case of the agglomeration of Dijon, 1999," Universite de Bourgone, manuscript.

Berliant, M., and H. Konishi, 2000, "The endogenous foundations of a city: Population agglomeration and marketplaces in a location-specific production economy," Regional Science and Urban Economics, Vol. 30, No. 3, pp. 289-324.

Bogart, W., and W. Ferry, 1999, "Employment centres in Greater Cleveland: Evidence of evolution of a formerly monocentric city," Urban Studies, Vol. 36, No. 12, pp. 2099-2110.

Booth, D., 1999, "Spatial patterns in the economic development of the Mountain West," Growth and Change, Vol. 30, No. 3, pp. 384-405.

Brueckner, J. K., 1986, "A switching regression analysis of urban population densities," Journal of Urban Economics, Vol. 19, No. 2, pp. 174-189.

Cameron, A. C., and P. K. Trivedi, 2001, "Essentials of count data regression," in A Companion Guide to Theoretical Econometrics, B. H. Baltagi (ed.), Malden, MA: Blackwell Publishers.

Cameron, A. C., and F. A. G. Windmeijer, 1996, "R-squared measures for count data regression models with applications to health-care utilization," Journal of Business and Economic Statistics, Vol. 14, No. 2, pp. 209-220.

Cervero, R. and K. Wu, 1998, "Subcentering and commuting: Evidence from the San Francisco Bay area, 1980-90," Urban Studies, Vol. 35, No. 7, pp. 1059-1076.

_____, 1997, "Polycentrism, commuting, and residential location in the San Francisco Bay area," Environment and Planning A, Vol. 29, No. 5, pp. 865-886.

Clark, C., 1951, "Urban population densities," Journal of the Royal Statistical Society: Series A, Vol. 114, pp. 490-496.

Combes, P., 2000, "Economic structure and local growth: France, 1984-1993," Journal of Urban Economics, Vol. 47, No. 3, pp. 329-355.

Craig, S. G, and P. Ng, 2001, "Using quantile smoothing splines to identify employment subcenters in a multicentric urban area," Journal of Urban Economics, Vol. 49, No. 1, pp. 100-120.

Fujita, M., P. Krugman, and T. Mori, 1999, "On the evolution of hierarchical urban systems," European Economic Review, Vol. 43, No. 2, pp. 209-251.

Fujita, M., and H. Ogawa, 1982, "Multiple equilibria and structural transition of non-monocentric urban configurations," Regional Science and Urban Economics, Vol. 12, No. 2, pp. 161-196.

Fujita, M., J. F. Thisse, and Y. Zenou, 1997, "On the endogenous formation of secondary employment centers in a city," Journal of Urban Economics, Vol. 41, No. 3, pp. 337-358.

Garreau, J., 1998, Edge City: Life on the New Frontier, New York: Doubleday.

Giuliano, G, and K. A. Small, 1991, "Subcenters in the Los Angeles region," Regional Science and Urban Economics, Vol. 21, No. 2, pp. 163-182.

Helsley, R., and A. Sullivan, 1991, "Urban subcenter formation," Regional Science and Urban Economics, Vol. 21, No. 2, pp. 255-275.

Henderson, J. V., and A. Mitra, 1996, "The new urban landscape: Developers and edge cities," Regional Science and Urban Economics, Vol. 26, No. 6, pp. 613-643.

Konishi, H., 2000, "Formation of hub cities: Transportation cost advantage and population agglomeration," Journal of Urban Economics, Vol. 48, No. 1, pp. 1-28.

McDonald, J. F., 1989, "Econometric studies of urban population density: A survey," Journal of Urban Economics, Vol. 26, No. 3, pp. 361-385.

_____, 1987, "The identification of urban employment subcenters," Journal of Urban Economics, Vol. 21, No. 2, pp. 242-258.

_____, 1985, "The intensity of land use in urban employment sectors: Chicago 1956-1970," Journal of Urban Economics Vol. 18, No. 3, pp. 261-277.

McDonald, J. F., and P. J. Prather, 1994, "Suburban employment centers: The case of Chicago," Urban Studies, Vol. 31, No. 2, pp. 201-218.

McMillen, D. P., 2001, "Nonparametric employment subcenter identification," Journal of Urban Economics, Vol. 50, No. 3, pp. 448-473.

McMillen, D. P., and J. F. McDonald, 2000, "Employment subcenters and subsequent real estate development in suburban Chicago," Journal of Urban Economics, Vol. 48, No. 1, pp. 135-157.

_____, 1997, "A nonparametric analysis of employment density in a polycentric city," Journal of Regional Science, Vol. 37, No. 4, pp. 591-612.

McMillen, D. P., and S. Smith, 2004, "The number of subcenters in large urban areas," Journal of Urban Economics, forthcoming.

Muniz, I., A. Galindo, and M. Garcia, 2003, "Cubic spline population density functions and subcenter delimitation: The case of Barcelona," Urban Studies, forthcoming.

Small, K. A., and S. Song, 1994, "Population and employment densities: Structure and change," Journal of Urban Economics, Vol. 36, No. 3, pp. 292-313.

Wieand, K., 1987, "An extension of the monocentric urban spatial equilibrium model to a multicenter setting: The case of the two-center city," Journal of Urban Economics, Vol. 21, No. 3, pp. 259-271.

Yinger, J., 1992, "City and suburb: Urban models with more than one employment center," Journal of Urban Economics, Vol. 31, No. 2, pp. 181-205.

Daniel P. McMillen is a professor of economics at the University of Illinois at Chicago and a consultant to the Federal Reserve Bank of Chicago.