Legacy regions, not legacy cities: Growth and decline in city-centered regional economies

ABSTRACT

Advocates adopted the term legacy city for older former industrial central cities to shift policy discussions from their social liabilities to their economic and physical assets. In examining this concept, we place U.S. central cities within a regional context, using cluster-discriminant analysis to distinguish between groups of metropolitan statistical areas, which proxy regional economies. Cluster analysis subsets metropolitan areas according to their position on several theory-driven dimensions. Discriminant analysis then identifies the variables most closely associated with each subset’s statistical grouping. The results respond to three questions: (1) Are regions with legacy characteristics homogeneous in population size? (2) Do socioeconomically troubled regions share legacy characteristics? (3) Does the typology of metropolitan regions provide insights into the performance of regional clusters? The findings suggest that private investment and public policy must change regional economic development paths while intentionally including distressed jurisdictions and populations before the futures of central cities that experienced severe population losses can shift to a positive trajectory.

Acknowledgments

It is customary to thank the reviewers for their contributions. In our case, we mean it. One reviewer offered suggestions on the variable specification that improved the performance of the model. Another pushed us hard to understand the difference between legacy regions and regions with legacy characteristics. The reviewer was correct. We also benefited from a very patient, thoughtful, and encouraging editor.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The American Assembly was founded by Dwight Eisenhower in 1950 when he was the president of Columbia University as a forum for nonpartisan public-policy discussions. Eisenhower held “the vision of a great cultural center where business, professional, and governmental leaders could meet from time to time to discuss and reach conclusions concerning problems of a social and political nature” (T. Jacobs, 1992, p. 457). Eisenhower wrote in a letter that explained his proposed assembly’s mission on September 14, 1950, that he had come “out of World War II” with “a profound conviction that America was in danger for two reasons.” The first was “The Communist threat from without,” and the second was “The failure of most of us to remember that the basic values of democracy were won only through sacrifice and to recognize the dangers of indifference and of ignorance” (T. Jacobs, 1992, p. 459).

2. Mallach developed the metric for the assembly’s 2011 convening resulting in a list of 48 cities (Mallach, 2012). Griffin (2015) added another 17 cities by dropping the requirement that the central city have at a population of at least 50,000 in 2000.

3. Figure 1.1 in Tighe and Ryberg-Webster (2019) lists the liabilities, burdens and problems of decline in legacy cities along with the assets, benefits and opportunities inherent in their built environments.

4. Schilling and Velasco (2020) listed the threshold sizes of central cities used in the work that followed the American Assembly’s original report (2011). The threshold population used by Mallach and Brachman (2013) was 50,000, as did the original research (American Assembly, 2011; Mallach, 2012). The Funders Network (2017) examined mid-sized cities with populations between 50,000 and 250,000; while Hollingsworth and Goebel (2017) used a population range from 30,000 to 200,000 in work for the Lincoln Land Institute. Alan Berube (2019) chose a population range from 20,000 to 200,000.

5. The study universe consists of all MSAs that remained unchanged between 2003 and 2015, excepting Fairbanks, Anchorage, and Honolulu. They were dropped because of their idiosyncratic economic bases and geographic separation from the system of regions in the U.S. mainland.

6. Variable selection is extremely import when cluster-discriminant analysis is used because the cluster analysis is a non-statistical technique. Cluster analysis minimizes within-group distances based on all variables included and cannot distinguish between variables statistically. Therefore, variables that do not have theoretical reasons for inclusion can distort the results of the clustering process, as can variables with extremely skewed distributions. The mathematical basis of cluster analysis makes it difficult to spot specification error, thereby increasing the importance of having a theoretical justification for entering variables into the analysis (Hill et al., 1998; Hill & Brennan, 2000). Our analysis is an explicit test of the legacy region hypothesis based on the relevant literature.

7. A complete description of each variable—year, data source, and geographic scale—is available in Table A1 of Appendix A.

8. The U.S. Bureau of the Census divisions and regions are Northeast (New England and Middle Atlantic), Midwest (East North Central and West North Central). South (South Atlantic, East South Central, and Wet South Central), and West (Mountain and Pacific).

9. The “access index” was constructed from the USDOT’s Transportation and Health Tool and is an average of each MSA’s commute mode share measures for commutes by foot or via public transportation.

10. This variable is measured as the absolute value of the percent change. It is only calculated for those MSAs with central city populations that are currently below their peak. For MSAs where the primary central city population is currently at its peak the variable is measured as 0% decline. Sixty-two percent (219) of MSAs in the study universe have a central city whose 2010 population was its peak population. For these MSAs the value for the variable was recorded as −0.0001. This avoids division by zero.

11. Ward’s minimum variance method was used to merge clusters as the clustering procedure moves from one step to the next. Ward’s method merges the two clusters that will result in the smallest post-merger, within-cluster, variance (Aldenderfer & Blashfield, 1984).

12. The m-score was first used in Furdell et al. (2005). We standardized the data in this research using z-scores and m-scores and compared the results. The effect of the skewed distributions drastically changed the candidate and preferred cluster solutions and their cluster groups with the m-score giving superior results.

13. The PSD is the interquartile range divided by 1.349. In the case of a normal distribution the ratio of the IQR to 1.349 equals the standard deviation.

14. We use the slope and acceleration statistics derived from the agglomeration coefficients to identify the optimal solution to the cluster analysis to avoid researcher judgment or bias (Everitt, 1993; Hill & Brennan, 2000; Ketchen & Shook, 1996). The decision rules are explained in more detail by Hill et al. (1998).

15. The three consolidations that differentiate these two solutions took place among the research university MSAs, two subsets of rapidly growing metros in the large median set of regions, and New Orleans combining with a subset of slow-growing metropolitan areas with weak assets.

16. The discriminant score is calculated by multiplying each function’s standardized canonical coefficients (which are generated during the analysis) by the standardized value for each variable, for a given MSA. This generates a single predicted value for each observation (MSA) and discriminant function, which in our analysis generates a matrix of discriminant scores with N rows and K-1 columns (in our model, 351 × 9). The centroid is the within-cluster mean of each discriminant score for a given function. Calculating centroids generates a matrix with K rows and K-1 columns (in our model, 10 × 9; see Table 2).

17. The 8-cluster solution was the runner-up candidate in this analysis. Tables B1 and B2 in Appendix B correspond with Table 2 and Table 3 in the main body, and delineate the clusters that would have been formed had we used K = 8 to guide our analysis.

18. Many MSAs have a primary central city identified as legacy city in the literature, and a number have other core municipalities with legacy assets and liabilities. Detroit is just one example. The Lincoln Land Institute identifies Detroit as a legacy city along with Warren and seven much smaller municipalities—Dearborn Heights, Hamtramck, Highland Park, Lincoln Park, Pontiac, Roseville, and Taylor. This list of municipalities makes it clear that the challenges of Southeast Michigan are regional, not purely municipal. See: https://www.lincolninst.edu/research-data/data-toolkits/legacy-cities/comparative-cities-map

19. We selected 2005 as the base year for this comparison because it was the year before the foreshocks to the Great Recession occurred.

Additional information

Funding

The Ohio State University’s discovery theme on sustainable and resilient economies provided research support.

Notes on contributors

Andrew J. Van Leuven

Andrew J. Van Leuven is a postdoctoral fellow in the Division of Applied Social Sciences, University of Missouri–Columbia. He works with Dr. Sarah Low on regional economic development and entrepreneurship research. He earned his PhD in public policy and management from the John Glenn College of Public Affairs at The Ohio State University. His research centers on downtown revitalization and other questions surrounding local economic development in non-metropolitan contexts.

Edward W. Hill

Edward W. Hill is Professor of Economic Development at The Ohio State University’s John Glenn College of Public Affairs and Senior Research Associate in the College of Engineering’s Ohio Manufacturing Institute. Ned edited Economic Development Quarterly for 11 years and was the Chair of the Advisory Board of the National Institute of Standards and Technology’s Manufacturing Extension Partnership. In addition, he was the Dean of the Levin College of Urban Affairs at Cleveland State University.