Spatial structure of city population growth

Overview of U.S. domestic migration flows

The most recent ACS county-to-county flow dataset²⁶ reports that about 45.6 million people migrated to the U.S. per year during the period 2015–2019, which corresponds to 14.2% of the U.S. population²⁷. Approximately 43.5 million annual moves corresponded to domestic migration (moves within the U.S.²⁸), while 2.1 million accounted for inflows of individuals from other countries (viz. international immigration).

With respect to domestic migration, 25.7 million people per year migrated within the same county, thus showing that the highest share of domestic flows (59%) is intra-county. Annually, about 10.4 people moved between different counties within the same state, thus intra-state flows account for 24% of the domestic migration (Supplementary Fig. 1), mainly driven by the search for more affordable housing, better jobs, and for family reasons such as change in marital status²⁹. Long distance moves, captured by inter-state flows, represent the remaining 17% of domestic flows, which comprises about 7.5 million moves per year. Here, we will refer to these domestic migration flows as inflows or outflows, and netflows (inflows-outflows).

The United States Office of Management and Budget (OMB) classifies counties as metropolitan, micropolitan, or neither³⁰. A metropolitan statistical area contains a core urban area of at least 50,000 population. A metro area represents a functional delineation of an urban area with a network of strong socioeconomic ties, and provision of infrastructure services^31,32,33. A micropolitan statistical area contains an urban core of at lest 10,000 but less than 50,000 inhabitants. There are over 380 metropolitan statistical areas in the U.S., each composed of one or more counties, accounting for about 86% of the total U.S. population and comprising approximately 28% of the land area of the country. For this reason, our analysis focuses on the growth dynamics of MSA counties. Supplementary Fig. 2 shows the 3141 counties (administrative subdivisions of the states) in the U.S., comprising about 321 million inhabitants in the starting year of the ACS 5-Year survey period (2015–2019) of our analysis²⁶.

Population growth has two components, namely natural growth and migration. Natural growth accounts for births minus deaths, and migration comprises domestic and international migration. With recent trends showing that births and natural increase have declined in the U.S. and in recent years contribute less to overall city population growth^34,35, migration patterns become more relevant to the study of city population growth. Because the ACS flow files contain international inflows only, the relative importance of migrations on population growth is here addressed by x = ∣Inflows−Outflows∣/∣Births−Deaths∣ (Supplementary Figs. 3, 4), which is the ratio between domestic netflows and natural growth. The statistical distribution of this quantity computed for all U.S. counties is well fitted by a lognormal distribution, and shows that x≥1 for 76.5% of counties. For most counties, domestic migration dominates population growth, and understanding the spatial structure of domestic netflows (and their distribution within a city) is crucial to the comprehension of the mechanisms behind the heterogeneity of city population growth.

At this spatial granularity, we observe a strong heterogeneity among the U.S. counties (Supplementary Fig. 2) for the period 2015 − 2019, along with examples of specific MSAs. In particular, the relative dispersion of counties relative growth due to netflows is higher than one for about 85% of the metro areas, indicating a large heterogeneity within the same city and pointing towards the spatial structure of domestic migration. The observed difference in the netflows stresses the relevance of our approach: counties belonging to the same city may have specific growth rates due to population flow patterns, thus indicating preferential flow destinations and pinpointing the direction in which the city has expanded.

Heterogeneity of inter- and intra-city flows

Inter-city flows represent the major component of the total flows (~55%), while intra-city flows represent ~25%. Flows between metro and micro areas, and between metro and non-statistical areas are the smallest components, with ~13% and ~7%, respectively. Given that about 80% of the domestic migration are composed of intra- and inter-city flows, we will focus our attention on describing the structure of intra- and inter-city flows, but in the Supplementary Information we offer a brief analysis of flows between metro and micro areas, and between metro and non-statistical areas.

Inter-city flows are not uniform across the U.S. cities. The most intense annual netflows (>2000 people per year), accounting for approximately 17% of the entire inter-city U.S. netflows, are mainly from New York and Chicago to California and Florida (Fig. 2), and from Los Angeles to neighboring cities. Notably, netflows among the Midwestern cities are mostly negative and below the threshold we set. These flows are mainly responsible for increasing or decreasing the population of a given city. Intra-city flow patterns, illustrated with the 7 most populous U.S. cities with more than 5 counties, are also non-uniform.

Our analysis reveals that city centers (defined as the core county with the highest population density) are more likely to have negative netflows, indicating that people are leaving the central regions of cities. The arrows in Fig. 2 indicate the direction of the most intense netflows, supporting this finding and highlighting that there is a trend of people moving from internal to external regions, contributing to population growth and spatial expansion of U.S. cities. In fact, we found no correlation between relative population growth (viz. population growth by county size) and distance from the core county (Supplementary Fig. 5A) for the 46 cities with more than 5 counties, with relative growth about 0.03 ± 0.05. On the other hand, we found that relative natural growth (Supplementary Fig. 5B) is negatively correlated with the distance to core county, thus natural growth is less relevant as a component of growth in the outer regions of cities. Consequently, our results show that not only the contribution of each component of growth changes with distance to core county, but also that the internal redistribution of people is an important mechanism of growth, mainly in the external counties.

We also examined variability in inter- and intra-city flows within the 50 states (Supplementary Fig. 6). Total flows within a state increase, as expected, with the state population. Two special cases are, however, of interest: (1) two states (Vermont and Rhode Island) with small populations have only one MSA, in which case within-state inter-city flows are zero; and (2) nearly 40%, or 149, of MSAs have only one county, in which case intra-city flows could not be estimated. For all other cases, we observe on average an equal split between inter- and intra-city flows, but with considerable variability among the states, with a mean about 0.5 and standard deviation about 0.2. A generalization of the intra- and inter-city migratory patterns for all 46 cities with more than 5 counties shows that the percentage of migrants from intra- and inter-city flows are of the same order of magnitude (Fig. 3).

Apart from the core county, flows from the same city correspond to about 50% of the inflow of people in the counties, presenting a slightly positive correlation with their distance from the city center (Fig. 3A). The low percentage for the core county indicates that it is not the major destination of flows from the same city. The percentage of inflows from other cities is higher in the core county and decays as we move towards the suburbs (Fig. 3B). The moderate negative correlation of this percentage with the distance reveals that inflows from other cities are more likely to concentrate in the core regions of a city.

The percentage of outflows directed from the core county to other counties within the same city has a slightly negative correlation with the distance of the origin county to the city center, so it is more likely to find intra-city flows with outflows from internal regions (Fig. 3C). The core county is an exception again, suggesting that it is less likely that someone leaving the core county will move to another county within the same city. The slightly negative correlation of the percentage of outflows directed to other cities suggests that there is a trend of people leaving the core county and the central regions to move to other cities (Fig. 3D). The high percentage of inflows (Fig. 3B) and outflows (Fig. 3D) in the central region due to inter-city flows implies that the central regions of cities are more dynamic and diverse and that people tend to move to counties with similar levels of urbanization. The same pattern is observed for flows between metro and micro areas, and for metro and non-statistical areas, allowing us to conclude that people moving from rural areas are more likely to move to the external regions of a city (Supplementary Fig. 7).

The positive correlation of the relative growth with the distance due to intra-city flows (Fig. 3E) shows that the resulting intra-city redistribution of people, given by the difference between inflows and outflows, is such that there is a trend from core county to the external counties (viz. suburbs). When compared to the relative growth due to inter-city flows (Fig. 3F), which do not show any trend and that have negative values for the most distant counties, it becomes clear that intra-city flows play a major role in the population increase observed in outer regions of cities. Interestingly, large circle and square dots in Fig. 3E and F suggest that the loss of people due to inter-city netflows is more intense than the gain of people due to intra-city netflows in some external counties of the largest metro areas, thus explaining the population decline in some outer regions of New York and Chicago (as shown in Fig. 2B and C).

The population growth due to intra-city flows is also depicted in Fig. 4. The concentration of flows below the diagonal captures the heterogeneity and the preferential destination of intra-city netflows. We observe that people are more likely to move to lower population density counties when moving from one place to another within the same city, as exemplified by 7 cities in panel A. Panel B summarizes this analysis for the 46 cities with more than 5 counties by showing the fraction ({{{{{{{mathcal{F}}}}}}}}) of intra-city netflows to lower density counties. We note that more than 93% of the cities have ({{{{{{{mathcal{F}}}}}}}} > 0.5) and that there is a positive correlation of ({{{{{{{mathcal{F}}}}}}}}) with the city population, and C shows the rank of cities according to the fraction of intra-city netflows to lower density counties.

Population density does not seem to play a major role in driving flows between counties of different cities. The fraction of inter-city netflows to lower density counties is about 57% when we consider all the 384 MSAs. The heterogeneity in the inter-city netflow pattern can be assessed by analyzing ({{{{{{{mathcal{F}}}}}}}}) versus the population of the destination city (Fig. 5A, B) and ({{{{{{{mathcal{F}}}}}}}}) versus the population of the origin city (Fig. 5C, D). The negative correlation of ({{{{{{{mathcal{F}}}}}}}}) with the population of the destination city in panel A indicates that inflows are more likely to come from lower density counties as the destination city size increases. The positive correlation of ({{{{{{{mathcal{F}}}}}}}}) with the population of the origin city in panel C reveals that outflows tend to be directed to lower density counties as the origin city size increases. The trends observed in panels A and C reveal that inter-city flows are more likely between counties with different population densities rather than between counties with similar population densities. Panels B and D show the rank order of cities according to a function of the destination city size and the origin city size, respectively.

We would expect that there might be preferential locations within a given city to which people move due to various factors such as lower costs of housing and employment opportunities. However, it seems that house prices have little to no effect on intra-city netflows (Supplementary Fig. 8). While the fraction of intra-city netflows to counties with less expensive houses is about 0.8 for cities like New York, Chicago and Washington, this fraction is about 0.2 for cities like Dallas, Houston and Philadelphia. The lack of a clear national pattern highlights the specificity of each city and the heterogeneity of the regional housing market in the U.S.^36,37. On the other hand, the fraction of intra-city netflows to counties with lower unemployment rates is higher than 0.5 for about 2/3 of the cities (Supplementary Fig. 9), thus showing that people are more likely to move to counties with lower unemployment rates.

Statistical structure of inter-city flows

Intra-city flows capture the internal redistribution of population, without altering the total city population. In this context, we focus on inter-city flows to investigate whether or not extreme flows play an important role in shaping the growth of counties as observed at the city level⁵. For cities, Verbavatz and Barthelemy⁵ introduce a stochastic equation to describe population growth, composed of two terms. The first term accounts for out-of-system growth, which includes natural growth and international migration, and the second term accounts for the growth due to domestic netflows. They find that total netflows adjusted by population size can be well approximated by a Lévy distribution, and this heavy-tailed distribution indicates that rare and extreme inter-city flows (viz. migratory shocks) dominate city population growth.

Here, we find that, for counties, the distribution of total netflows adjusted by population size, which is represented by ζ_i and captures the intensity of inter-city migratory flows (see the section “Methods” for details), can be approximated by a Gaussian distribution (Fig. 6). The lack of a heavy tail in the empirical distribution of ζ_i suggests the absence of extreme flows at the county level, thus indicating that the growth of counties can be described by smoother migratory process than cities. Given that cities do experience migratory shocks⁵, our findings indicate that cities redistribute inflows among its different counties, leading to a spill-over effect that dampens flow shocks at the county level.

Heterogeneity of international inflows

The highest share of international inflows is concentrated in large cities. About 40% of the international inflows are destined to the top 10 (~2.6%) largest metro areas of the U.S. New York is the first with 8.5% of international inflows, followed by Los Angeles and Miami with 5.4% and 5.0%, respectively. Indeed, international inflows Y_k scale superlinearly with the population S_k of the metro area k (Fig. 7A), thus larger cities have more immigrants per capita than smaller cities.

Interestingly, this gain with scale is also observed in socioeconomic city metrics as crime, GDP, innovation and wealth creation due to the manifestation of nonlinear agglomeration phenomena^38,39,40. Using Y = Y₀S^θ as a null model, we can compute deviations from the average behavior by means of residuals given by (log ({Y}_{k}/{Y}_{0}{S}_{k}^{theta }))³⁸. The rank of the residues (Fig. 7B) shows that college towns are among the top metro areas receiving more international inflows than expected, while large cities as Los Angeles, New York, Atlanta, and Chicago are among the metro areas receiving less international inflows than expected.

The spatial distribution of international inflows within cities is shown in Supplementary Fig. 10. The highest share of inflows is concentrated at core counties, and the percentage of inflows decreases dramatically with the distance from the core county. This result suggests that inflow of international migrants is an important component of population growth, particularly at the core regions of large cities.

Robustness of our findings

Patterns of population redistribution change from time to time in the U.S., and are affected by several factors. For instance, in the 1960s non-metropolitan counties lost about 3 million people due to outflows to metropolitan counties, while the reverse trend was observed in the 1970s when non-metropolitan counties experienced net inflows of about 2.6 million people⁴¹. Wardwell and Brown in⁴¹ indicate that three factors might be among the main reasons of such change, namely economic decentralization, preference for rural living, and modernization of rural life. The temporal influence of factors as socioeconomic conditions, transportation infrastructure, natural amenities, and land use and development on population growth in rural and suburban areas is explored in⁴². Changes in rural migration patterns are also studied in⁴³, where age-specific rural migration patterns from 1950 to 1995 are analyzed. In⁴⁴, the authors explore redistribution trends across U.S. counties from 1980 to 1995 split into three five year periods (1980–1985, 1985–1990, 1990–1995), and⁴⁵ analyzes changes in age-specific nationwide migration patterns from 1950 to 2010.

The spatial structure of migration patterns may indeed change from time to time; our results correspond to the current intra- and inter-city redistribution trends, based on the most recent ACS migration flow files. We present a thorough empirical and statistical analysis of domestic migration flows among U.S. cities ans counties. Our study also introduces a framework that can be used for analyzing and comparing internal redistribution of people across different time periods. Indeed, we extended our analysis for two other time periods, 2005–2009 and 2010–2014. With respect to the spatial distribution of intra- and inter-city flows, similar trends are observed in both periods (Supplementary Figs. 12, 13), namely inter-city flows are responsible for the highest share of inflows to core counties, and intra-city flows are responsible for the highest share of inflows to external counties. We also explored the role of population density in driving netflows between counties within the same metro area in 2005–2009 and 2010–2014. The results (Supplementary Figs. 14 and 15) indicate that 95.7% of cities were dominated by intra-city moves to lower density counties in 2005–2009, and this percentage dropped to 76.1% in 2010–2014. Our findings indicate that the trends we report here are taking place since 2005 but with different intensities.

The robustness of our findings is checked with additional migration data from the Internal Revenue Service (IRS), which reports the year-to-year address changes on individual tax returns filled with the IRS⁴⁶. The results obtained with the analysis of IRS datasets from periods 2015–2016, 2016–2017, 2017–2018, 2018–2019 (Supplementary Figs. 16, 17, 18, 19), reveal similar trends to those we found using ACS data. Particularly, we observe that, for all periods considered, the correlation between intra-city netflow/S and distance to core county is stronger than we found with ACS data, thus highlighting the role of intra-city flows in driving population to external regions of cities. The main difference between both datasets is in the percentage of intra- and inter-city inflows and outflows: while ACS data indicates that both flows have approximately the same contribution to the total flows, the IRS data indicates that, besides the core county, intra-city flows are responsible for about 80% of inflows and outflows of metro areas.

Source: Ecology - nature.com