Economic complexity and divergent population growth by race and rurality

Abstract

Economic complexity (EC) measures the diversification and domestic comparative advantage of industry in an economy. Originally applied to studies of global economic development, dominant frameworks suggest the extent of a place’s economic capabilities underlies its economic and population growth potential and ascribe to the universalistic notion that economic growth generates population growth. Limited research has extended the observed linkages between EC and economic and population growth in a subnational context, focusing solely on metropolitan/micropolitan contexts and neglecting potential spatialized forces that might promote variation in the relationship, especially in rural areas. After computing EC estimates for Commuting Zones (CZs), a typology inclusive of metropolitan, micropolitan, and non-metropolitan areas, we use spatial modeling techniques to investigate whether economic complexity relates to population growth and net in-migration uniformly by rurality and for ethnoracial groups. In contrast to universalistic assumptions and our expectations that complexity would be less predictive of growth in non-metro contexts, we find that higher EC more strongly associates with population growth and net in-migration in non-metro contexts. Our findings also suggest that there is a racial dimension to EC, with higher EC CZs experiencing net-inflows of White populations and net-outflows of Asian, Black, Hispanic, and Indigenous populations. Study results suggest spatially and racially disparate implications of regional economic growth and development, challenging conventional assumptions of the widely distributed benefits of economic development on less resourced ethnoracial groups, especially in non-metro contexts.

1.0. Introduction

Classical theory in economic development posits that growth in private enterprise results in aggregate economic well-being for communities and their residents through economic multiplier effects (North 1955; Tiebout 1956; Conway 2021). This convention perpetuates universalistic assumptions about economic growth translating into beneficial outcomes for communities such as population growth. Growth in private investment, economic output, tax base, jobs, and average income is often treated as an unambiguous victory for the public and its residents. However, how these core tenants of development translate across distinct contexts of place (i.e., rurality) and between different social groups (i.e., race, class, nativity) remain underexplored.

Previous research asserts industry diversification and domestic comparative advantage underlie healthy regional economies and predict future regional economic and population growth (Hidalgo & Hausmann 2009; Hausmann et al. 2014). This logic suggests a place characterized by an economy with a wide array of industries (diversification) may be better prepared to weather economic shocks to certain industries, while a place characterized by an economy with a domestic comparative advantage in a given industry may be able to capitalize on its relative competency by growing an industry cluster such as in semiconductor manufacturing or biotechnology. Economic complexity interacts these two concepts into a single measure by considering the diversity of industries for which a region holds a comparative advantage. Economic complexity has been used to empirically characterize the extent of economic capabilities within a place’s economy and to assess an area’s economic and population growth potential (Hidalgo and Hausmann 2009).

Research on economic complexity has largely focused on the relationship between combined considerations of diversification and comparative advantage on economic growth at an international scale and with urban leanings. Scholars have documented a positive relationship between economic complexity and growth in a series of economic outcomes including output, incomes, and labor productivity in addition to population growth (Hidalgo & Hausmann 2009; Ourens 2012; Hausmann et al. 2014; Stojkoski et al. 2016; Sbardella et al. 2018; Tacchella et al. 2018; Escobari et al. 2019; Daboin et al. 2019; Koch 2021; Hidalgo 2023). Limited research applying economic complexity regionally within the United States largely confirms the relationship between complexity and various forms of growth, yet is conducted at the metropolitan and micropolitan level, effectively excluding non-metropolitan economies from consideration (Escobari et al. 2019; Daboin et al. 2019; Fritz & Manduca 2021). Consequently, it remains unclear whether variation in economic complexity underlies disparate patterns of economic and population growth across the rural-urban gradient. Further, previous research has not explored heterogeneity in complexity-growth associations for specific subgroups of residents. Given racialized patterns of population settlement and economic development, established complexity-growth relationships may differ for White and historically marginalized ethnoracial groups.

The exclusion of non-metro areas in the subnational regional growth and development literature is due in part to suppression of economic output and workforce data estimates for small population areas, and in part to conceptual indifference since non-metro economic activity is often conceptualized as peripheral to metropolitan economic engines (Stauber 2001; Lobao 2014; Goetz, Partridge, & Stephens 2018). As a result, the extent to which economic processes along the rural-urban continuum shape population growth and stagnation is under-theorized and under-explored. This limits our understanding of the forces generating and perpetuating disparate development trajectories for metro and non-metro places and for historically marginalized ethnoracial groups. We address this conceptual and empirical limitation by using imputed workforce data to address issues of data suppression, and by building on research examining how the rural-urban continuum challenges the conceptual indifference to non-metro economic activity (e.g., Lichter & Ziliak 2017; Lichter et al. 2021) and investigating racialized spatial patterns of population distribution, growth, and economies (e.g., Foulkes & Schafft 2010; Lichter et al. 2022).

Considering the troubling decline of many rural economies over the preceding decades and the subsequent increasing divide in economic opportunities afforded to residents of rural communities and especially among historically marginalized ethnoracial groups (Kusmin 2016; Cromartie 2018; Johnson & Lichter 2019), we assert the importance of economic complexity in understanding mechanisms leading to rural-urban and racial divergence as a consequence of development. In this study, we ask: does economic complexity predict growth uniformly across places and for differently advantaged ethnoracial groups? Using contiguous labor market areas (commuting zones) to allow for simultaneous consideration of places along the rural-urban continuum, we first assess whether the prevailing relationship between economic complexity and population growth holds true in non-metro economies or whether differences in the economic, demographic, and spatial contexts of non-metro economies result in disparate patterns of growth. We then assess whether the prevailing relationship between economic complexity and population growth applies equally to all ethnoracial groups or whether the relationship varies systematically and in ways that advantage some ethnoracial groups over others.

We find economic complexity to be a stronger predictor of future growth in non-metro contexts, and Asian, Black, Hispanic, and Indigenous populations are more likely to remain in and move to lower complexity places while Non-Hispanic White populations are more likely to move to and remain in higher complexity places. These robust results emphasize the unequal nature of where and who benefits from aggregate economic growth and development.

2.1. Divergent economic development and economic complexity

Local industrial structure contributes to the general well-being of places and their populations (Lobao & Shulman 1991; Barnes & Blevins 1992; Lobao et al. 1993; Tolbert et al. 1998; McLaughlin & Stokes 2002; Albrecht & Albrecht 2010; Hooks et al. 2016). In the long-run, regional economic stagnation and decline has an array of community, institutional, and individual-level impacts. Examples include reducing the revenue capacity of local governments, along with education and local service quality, and worsening the quantity and quality of available employment, the prevalence and persistence of poverty, social mobility opportunities, mental and physical health outcomes, and family stability (Friedman & Lichter 1998; Peters & Fisher 2002; Green & Sanchez 2007; Albrecht & Albrecht 2010; Curtis et al. 2013; Curtis et al. 2019; Connor et al. 2024).

Conventional development theory posits that regions with a relatively diverse set of specialized, exporting industries (i.e., higher industrial diversity and tradable sector comparative advantage) are better-suited for economic stability and future economic growth (North 1955; Tiebout 1956; Hidalgo & Hausmann 2009; Conway 2021). These two components – economic comparative advantage and industrial diversification – are jointly considered through a single measure of an economy’s “economic complexity.”

Tradable sector (often referred to as “export sector”) industries generate products and services generally intended for use outside a given locality (e.g., manufacturing, agriculture, mining, software) and contrast with non-tradable sector (often referred to as “non-export sector” or “local sector”) industries primarily servicing businesses and individuals inside a given locality (e.g., primary and secondary education, healthcare, utilities). Production in the tradable sector is theorized to have a widely distributed benefit on a place and its residents, as externally traded goods and services bring in economic resources to a regional economy, creating economic multipliers throughout the non-tradable sector (North 1955; Tiebout 1956; Conway 2021). While classical economic development theory and analysis almost solely focus on the tradable sector, in recent decades scholars have recognized the importance of the non-export economy in driving regional economic growth (Markusen & Schrock 2006; Kay et al. 2007). Further, reliance on only a few industries for a substantial portion of a locality’s tradable sector activity and employment can result in economic instability and distress during shocks to industry or business cycles (Malizia & Ke 1993; Wagner & Deller 1998).

Originally developed to explain the productive capabilities of national economies, Hidalgo & Hausmann (2009) are widely credited with popularizing economic complexity as a measure of the relative international economic development of countries. In addition to the usefulness of economic complexity as a descriptive measure of the economic diversity and domestic comparative advantage of a place’s industrial structure, Hidalgo and Hausmann established the relationship between higher economic complexity levels and higher economic growth and average incomes at the country level (Hidalgo & Hausmann 2009; Hausmann et al. 2014). This relationship has largely been corroborated at the international scale (Ourens 2012; Stojkoski et al. 2016; Sbardella et al. 2018; Tacchella et al. 2018; Koch 2021; Hidalgo 2023) and extended to the subnational region scale in a variety of contexts (Poncet & Waldemar 2013; Chavez et al. 2017; Davies & Mare 2020; Mealy & Coyle 2021; Mewes & Broekel 2022).

Beyond more traditional applications of economic complexity to growth and incomes, economic complexity has been instructive in understanding differences in income inequality (Hartmann et al. 2017; Lee & Vu 2019; Zhu et al. 2020), greenhouse gas emissions (Neagu & Teodoru 2019; Romero & Gramkow 2019; Lapatinas et al. 2021), and even health outcomes (Vu 2020) among countries. Several studies have demonstrated linkages between higher economic complexity and subsequent growth in economic output, average incomes, and technological capabilities, lower income inequality and carbon emissions, along with positive health outcomes at the regional level within the US (Balland & Rigby 2017; Mealy et al. 2018; Mealy et al. 2019; Fritz & Manduca 2021; Gomez-Lievano & Patterson-Lomba 2021; Lo Turco & Maggioni 2021). In general, previous research on economic complexity suggests higher complexity aligns with greater economic prosperity.

A separate literature demonstrates the economic prosperity of places and populations has clear spatial and temporal patterning within the US, as evidenced by strong spatial clustering in a variety of markers of community and population well-being. For instance, high poverty communities tend to exist near other high poverty places and remain impoverished for lasting stretches of time (Cotter et al. 2002; Glasmeier 2006; Beale & Gibbs 2006; Curtis et al. 2013; Call & Voss 2016).

Most critical for the current study, economic development and the impacts of shifting economic structures as a result of increasing urbanization, globalization, and technological advancement create ripple effects in local and regional economic realities and industrial structures that are distinct between urban and rural places (Albrecht 2014; Curtis et al. 2019). While metropolitan areas in the US have mostly experienced aggregate (albeit unequal) economic growth in the second half of the twentieth century, the rural US has experienced substantial industrial restructuring in agriculture, mining, forestry and fishing, and manufacturing which has resulted in aggregate economic decline (Albrecht 2020).

In addition to these diverging trends in economic compositions, growth, and development across spatial contexts, previous research demonstrates different ethnoracial groups experience uneven benefits from economic growth and development. Overall, areas with larger concentrations of historically marginalized ethnoracial groups have higher levels of poverty, higher racial inequality, and less specialized and diversified economic structures, such as economies primarily focused on agriculture, food processing, and lower-skilled manufacturing (Tomaskovic-Devey and Roscigno 1997; O’Connell 2012; Davis et al. 2016; Curtis and O’Connell 2017). Uneven prosperity between and within areas is also inseparable from issues of racial equity. For instance, greater residential segregation and long-distance migration patterns in the late 20th and early 21st century – namely among Black and White Americans – mirrored spatial shifts in industrial compositions, with higher-wage professional sector growth coinciding with White residential location and, in contrast, growth in lower-wage, less specialized industries coinciding with the location of historically-marginalized residents (Wilson 1996; Collins & Mayer 2010; Curtis et al. 2013). Moreover, while Black, Hispanic, Indigenous, and Asian Americans make up only around one-fifth of the population in prosperous non-metro areas, these historically marginalized ethnoracial groups comprise nearly one-half of residents in non-metro areas marked by persistent poverty (Dobis et al. 2021; see also Jensen 1994).

Historical contexts of racialized institutional structures and processes in the US continue to influence the spatial variation of economic, social, and community outcomes for every marginalized racial group in the US (Albrecht 2014; Gabriel and Tolnay 2017; O’Connell et al. 2020). Institutional forces have generated persistent patterns of settlement among racially marginalized groups within economically marginalized places. Tomes of research on poverty, economic development, and various scales of human mobility have documented the relative inability of marginalized populations to gain access to more prosperous places characterized by higher development, higher pay, and more material resources. For instance, historical geographies of forced migration, chattel slavery, residential segregation, and economic exclusion, especially in the Southern and Northern US, have shaped the opportunity structures for Black Americans and resulted in widening Black-White income gaps (Tribble 1996; White 2018; O’Connell et al. 2020). Populations indigenous to the US continue to experience the influence of state control and economic exclusion. Forced relocation to reservations primarily in the Western US combined with government control of land ownership has produced Indigenous communities with little economic infrastructure and therefore few economic opportunities (Snipp 1986; Miller 2012; Albrecht 2014). The combination of historically shifting borders and immigration policies have influenced spatially patterned economic opportunities for many Hispanic/Latine citizens and migrants in almost every region of the US (Lichter et al. 2010; Lichter et al. 2012; Albrecht 2014).

The extant scholarship on economic complexity suggests higher levels of complexity as antidotes for economic stagnation (Hidalgo & Hausmann 2009; Hausmann et al. 2014). This research has largely focused on examining these outcomes and associations in metropolitan areas (Daboin et. al 2019; Escobai et al. 2019; Fritz & Manduca 2021). However, differences in historical and contemporary trends in economic and industrial development between places along the rural-urban continuum and among different racialized populations within the US suggest that the assumed benefits of economic complexity might not be experienced universally.

2.2. Economic development and population growth

While industries and skills have continued to cluster in a handful of mostly urban economic centers, rural places have experienced substantial heterogeneity in development opportunities for economic and industrial growth with implications for population growth. Some rural places experience declines in tax base and the “brain drain” of its most educated populations to other regions with greater economic opportunities (Berry & Glaeser 2005; Glaeser & Gottlieb 2009). Yet other rural places experience economic and industrial stability creating opportunities to attract new populations (McGranahan & Wojan 2007). Research suggests whether a non-metro place grows or declines partially depends on its access to centers of transportation and natural amenities that might attract remote workers and other populations seeking recreational opportunities (McGranahan & Wojan 2007; Deller et al. 2008; Partridge 2010; Chi & Marcouiller 2011; Chi & Marcouiller 2012; Rickman & Wang 2017).

Escobari et al. (2019, see also Daboin et al. 2019) offer a new use case for economic complexity: foreshadowing future population growth directly. As argued by Escobari et al. (2019), population growth is a crucial indicator of regional economic development. The argument builds on a core assumption of “spatial equilibrium” in labor markets and economic development borrowed from urban economists and asserts that workers will migrate to cities where economic opportunities are greater, i.e., real wages are higher (Glaeser & Gottlieb 2009).

The lens of population change through net migration elucidates diverging trends in the prosperity of place, particularly between non-metro and metropolitan contexts. While economic development literature frequently conceptualizes population growth as an indicator of the prosperity of a place, population growth is not always an indicator of prosperity. The growth of populations can be complicated by non-economic factors that attract populations to a place and the resources retaining populations once they arrive (Isserman et al. 2009). Examining net migration (the balance of people moving into or out of a place) adds additional nuance to understanding the relationship between economic complexity and population growth and provides an opportunity to identify additional economic and non-economic push and pull factors that influence population movement. The relationship between place and equitable economic opportunity by race and ethnicity is deeply rooted in the United States, underscoring the importance of understanding the forces driving prosperity, stagnation, and decline for different populations and their communities.

In aggregate, populations in rural places in the United States have been declining. Approximately half of all non-metro counties in the United States experienced population peaks in the first half of the 20th Century (Johnson & Lichter 2019). Most recently, rural populations declined by 0.6 percent between 2010 and 2020, compared with 8.8 percent growth in populations in urban areas (Dobis et al. 2021). This trend primarily results from larger net loss through people migrating away from non-metro communities than the smaller positive natural increase (i.e., births exceeding deaths) among people who remain in the same communities. In contrast, metropolitan centers have exhibited consistent population growth over the last century as a result of urbanization, natural increase, and international in-migration (Johnson & Lichter 2019; Dobis et al. 2021).

Subsequent stagnation and decline for many non-metro counties is largely driven by changes to economic development through the consolidation of agricultural firms and the movement of manufacturing operations into and then subsequently out of metro adjacent rural areas (Fuguitt 1985; Frey 1987; Albrecht 2014). In contrast, the development of natural amenities has led to an increase in retirement migration to rural destinations and a resulting influx of younger-aged workers in service and construction industries (McGranahan & Wojan 2007; Deller et al. 2008; Partridge 2010; Chi & Marcouiller 2011; Chi & Marcouiller 2012; Rickman & Wang 2017). Population decline has been more pronounced in more remote non-metro contexts without access to natural amenities or transportation centers that can generate economic opportunities that frequently precede in-migration to communities (McGranahan et al. 2010; Albrecht 2014; Johnson & Lichter 2019).

Further, place-based variation in economic development patterns and subsequent population shifts have widened existing gaps between more and less prosperous places in metro and non-metro contexts. Non-metro economies characterized by persistent poverty saw population reductions of 5.7 percent, compared with metro counties characterized by persistent poverty, which grew at 5.8 percent (Dobis et al. 2021). This non-metro/metro divergence is suggestive of disparate processes underlying economic prosperity and population growth trajectories for places along the rural-urban continuum.

Concentration of poverty in rural contexts partly results from the unique and temporally consistent migration patterns of under-resourced communities in the United States that are dissociated from economic development (Foulkes & Schafft 2010; O’Connell & Shoff 2014; Lichter et al. 2022). Counter to economic conceptualizations of migration that theorize mobility is driven by a rational choice among actors to seek better economic opportunity, scholars have found that populations living with poverty in rural areas also move to other rural places characterized by high proportions of the population living in poverty (Foulkes & Schafft 2010; Lichter et al. 2022).

Additionally, given the majority of the population living in or around the federal poverty line are members of historically-marginalized ethnoracial groups, and these groups are disproportionately likely to live in rural, urban, and suburban communities marked by persistent poverty, it is plausible that forces driving migration differ among White populations relative to Black, Hispanic, Indigenous, and Asian American populations (Lichter et al. 2012; O’Connell & Shoff 2014). These operative racialized forces may reflect the greater capacity of the White population to locate in more economically prosperous contexts (Mayblin & Turner 2020; Robertson & Roberts 2022).

In recent decades, Hispanic populations have increased substantially in non-traditional rural and suburban destinations often characterized by less economically complex industrial structures (e.g., agriculture, food processing, lower-skilled manufacturing), slowing aggregate declines in rural populations. Coinciding “white flight” from these new destinations has resulted in greater instances of racial and ethnic segregation in these communities (Lichter et al. 2010; Crowder et al. 2011; Lichter et al. 2012; Lichter & Johnson 2020). Meanwhile, trends of net in-migration of Black Americans to the southeastern US may be characterized by a set of contradictory economic and non-economic push and pull factors resulting in a higher influx of African Americans born in the southeast to more economically prosperous communities in the southeast and a higher influx of African Americans born outside the southeast to less economically prosperous communities in the southeast (Brown & Cromartie 2006; Hunt et al. 2008; Ambinakudige et al. 2012).

2.2.1. Spatial differentiation in economic and population dynamics

Despite the well-established nature of subnational heterogeneity in economic, socioeconomic, and demographic trends and rural-urban divergence, non-metro areas are often treated with conceptual indifference. Commonly, non-metro areas are excluded from region-level analyses through the use of metropolitan statistical areas as the geographic unit of analysis, evident in the subnational US literature on connection between economic complexity and economic and population growth (Daboin et al. 2019; Escobari et al. 2019; Fritz & Manduca 2021). Perhaps more problematic, non-metro areas are unintentionally subjected to the periphery through the lumping of non-metro areas into geographic units containing metropolitan areas (Stauber 2001; Lobao 2004; Hooks et al. 2016; Goetz, Partridge, & Stephens 2018). Diverse scholarship has challenged this conceptual indifference, acknowledging that interactions between economic, demographic, social, and spatial processes may vary across contexts, particularly between non-metro and metropolitan areas. Examples applying the concept of spatial heterogeneity include the study of subnational income inequality, migration, intergenerational mobility, and poverty (Curtis et al. 2012; Chetty et al. 2014; Lichter and Ziliak 2017; Thiede et al. 2018; VanHeuvelen 2018; Curtis et al. 2019; Lichter et al. 2020). In this study, we set out to address two central research questions: 1) are associations between economic complexity and local population growth found in global contexts substantiated across metro and non-metro contexts, and 2) does higher economic complexity predict subsequent net migration patterns of Asian, Black, Hispanic, Indigenous, and White residents uniformly?

Building on insights from existing literature on the variation in economic development between and within places and populations along the rural-urban continuum, we first anticipate a positive relationship between higher economic complexity and future population growth will hold stronger in more urban commuting zones, as growth and net in-migration patterns will be more closely intertwined with traditional explanations of economic migration than in more rural commuting zones where other forces driving attraction and retention of residents might have greater explanatory power, such as amenity migration or non-economic factors such as family ties and place connectedness.

Second, given that prior studies have established populations of color are more likely to reside in places characterized by persistent poverty and economic deprivation underwritten by spatio-economic and socio-spatial histories of exploitation and marginalization, we anticipate this racialized spatial patterning to extend to economic complexity. We expect Asian, Hispanic, and especially Black and Indigenous populations to migrate to and live in lower complexity regions at higher rates than non-Hispanic White populations. Due to their comparatively privileged position in the US economic system, we expect non-Hispanic White residents to migrate to and reside in higher complexity regions characterized by higher economic capabilities and probable future economic growth and prosperity at higher rates than other ethnoracial groups.

3.0. Methodology

3.1. Measuring economic complexity (EC)

Most studies incorporating measures of economic complexity either replicate Hidalgo & Hausmann’s (2009) calculation methodology, or for those conducting cross-national analyses, use EC figures directly provided by Hausmann et. al (2014). Some exceptions apply, particularly in the extension of complexity to subnational regional context (Davies & Mare 2020; Fritz & Manduca 2021; Mewes & Brokel 2022). While some of these works use substantially different formulas for calculating economic complexity (Davies & Mare 2020; Mewes & Brokel 2022), others differ in their consideration of revealed comparative advantage (RCA) (Ourens 2012; Stojkoski et al. 2016; Chavez et al. 2017; Mealy et al. 2018; Mealy et al. 2019; Fritz & Manduca 2021). Most commonly, a binary specification of RCA is used, as regions with an above average share of a given industry activity receive a 1, and those with a below average share are given a 0.¹

For each region-industry combination (RCA _{r, i}), we assess whether a region (r) holds a revealed comparative advantage in a particular industry (i) by computing whether the share of the measured industry activity $\left(X_{r,i}/X_r\right)$ in a region (e.g., employment, output, payroll, etc.) exceeds that of the region overall $\left(X_i/X\right)$.

$$RC{A}_{r,i}=1\left[\frac{X_{r,i}/X_r}{X_i/X}\ge 1\right]$$ (1.1)

The final form of the EC calculation is summarized in Equation 1.3, with k_r denoting the diversification of region’s industrial structure (i.e., number of industries with an RCA of 1) and k_i denoting the ubiquity of a given industry (i.e., number of regions with an RCA of 1 for given industry).

$$E{C}_r=-\frac{{\sum }_i{k}_{i}RC{A}_{r,i}}{k_r}$$ (1.3)

The inputs used in calculating EC vary based on several key considerations such as the type of data used (e.g., exports, output, employment, patents, etc.), the preferred industry classification aggregation (i.e., 6-digit NAICS level, 4-digit ISIC, or 4-digit SITC), whether tradable sector industries should be the sole focus of complexity analysis, and the preferred geographic aggregation.² We opt to use industry employment counts in computing the EC, permitting inclusion of the non-tradable sector in this work.³ As presented in Appendix A, our results are robust to the decision to use employment counts rather than alternative underlying data. Finally, we log transform then subsequently standardize raw EC estimates into z-scores for ease of interpretation, where a positive z-score denotes higher levels of EC.⁴

With the exception of Mealy et al. 2018, which computes complexity at the state level, U.S.-focused complexity estimates have been conducted at the metropolitan statistical area (MSA) level, sometimes including micropolitan statistical areas (Daboin et. al 2019; Escobai et al. 2019; Fritz & Manduca 2021). This aggregation is selected due to its social meaningfulness as politically-defined units that reflect the commuting patterns of residents into an urban core, and its widespread acceptance in measuring urban areas across the fields of demography and economics. However, a major limitation of metropolitan and micropolitan area definitions is their exclusion of non-metro counties. Due to its coverage of all US counties, metro and non-metro, we have opted to use Commuting Zones as our geographic unit. Originally designed by the USDA’s Economic Research Service (ERS), Commuting Zones (CZs) seek to aggregate counties by their commuting patterns in order to “contain” a given labor market shed, incorporating the labor market patterns of each rural, suburban, and urban counties (Fowler & Jensen 2020).⁵

In order to maximize local industry employment coverage, especially our inclusion of sometimes small-population, non-CBSA counties for which data are more commonly imputed, we opt to use employment figures to the 4-digit NAICS level. Nonetheless, we compute and present the sensitivity of our EC estimates to NAICS level in Appendix A, finding no differences in significance or direction of results between EC computations at the 3, 4, 5, and 6-digit NAICS level.

3.2. Data sources

3.2.1. Outcome variables: Population Growth and Ethnoracial Net Migration

Population growth rates are computed from Census Population Estimates Program (PEP) data at the CZ level. Estimates from the University of Wisconsin–Madison Applied Population Laboratory’s (APL) analysis of Decennial Census and PEP Estimates Base data is used to compute overall and “prime-aged” (25-54-year-olds) net migration rates for each CZ, in addition to disaggregated estimates for migration rate variables for Hispanic and non-Hispanic Asian, Black, Indigenous, and White residents (Egan-Robertson et al. 2023). These net migration estimates are highly advantageous to alternative disaggregated net migration estimates garnered from sources such as the Census Bureau’s American Community Survey due to use of the high-resolution Decennial Census and PEP Estimates Base data, which is particularly important considering data accuracy concerns in small population geographies.

3.2.2. Independent variable

We use an imputed version of the U.S. Census County Business Patterns (CBP) employment count data by industry for input into economic complexity calculations. These data are aggregated from the county to the CZ-level. CBP notably omits estimates for county-industry observations that risk disclosure of individual organization data; as such, data for certain county-industry groupings are unavailable in CBP, particularly for less populated areas and more narrow industry definitions. Imputation by Eckert et al. (2020) offers mostly complete coverage of county-industry groupings, with the exception of most government, crop and livestock production, and rail transportation industries.

We supplement these employment counts with data from the Census of Agriculture to integrate employment levels for crop and livestock production, but we are unable to include employment estimates of government and rail transportation in our calculations of EC. Since employment data is unavailable at the subnational level for detailed industry definitions, we multiply our calculation of the national share of producers and workers per farm in a given industry by the number of farms in a given industry at the county level to estimate worker count for each CZ-industry at the 4 and 5-digit NAICS level.

3.2.3. Moderator

We conceptualize the rurality of a region as a potential modifier of the economic-population growth relationship. The extent of a CZs rurality is expressed both as a dummy and continuous variable calculated as a population-weighted average of county-level rural-urban continuum codes (RUCC) from the Economic Research Service’s (ERS).⁶ The 617 CZs are comprised of 144 metropolitan and 473 non-metropolitan CZs. The rescaled distribution of CZs across continuous RUCC is depicted in Figure 1.

Figure 1 –

Distribution of Commuting Zones (CZs) Across Binary and Continuous Categorization

Source: USDA Economic Research Service RUC codes, 2013.

3.2.4. Covariates

Our multivariate modeling strategy includes several covariates. The log of a CZs population is obtained from PEP data. The median age, employment rate, log of a CZs median individual income, the proportion of a CZs population that identify as non-Hispanic White, and the proportion that are foreign-born are each received from the Census American Community Survey (ACS) 5-year estimates.

Notably, the industrial structure is not parameterized directly in our models. However, as employment concentration underlies the computation of EC, industrial structure is reflected indirectly in EC. A regions’ amenities have been established as an important predictor of migration, particularly in non-metropolitan areas (McGranahan & Wojan 2007; Deller et al. 2008; Partridge 2010; Chi & Marcouiller 2011; Chi & Marcouiller 2012; Rickman & Wang 2017). Due to our inclusion of non-tradable sectors in the calculation of EC, regional amenities are indirectly captured in EC, mostly through hospitality-related and other services. Due to this grouping alongside other hospitality and service industries, regional amenities are unlikely to be core drivers of model results through either domestic comparative advantage or industrial diversification components of EC.

3.3. Modeling the role of complexity on population growth

We employ a series of spatial models to address two core research questions: 1) are global associations between economic complexity and local population growth substantiated across metro and non-metro contexts, and 2) does higher economic complexity predict subsequent net migration patterns of Asian, Black, Hispanic, Indigenous, and White residents uniformly?

To address these questions, we first test the bivariate relationship between complexity and population growth, along with each of our aggregated and disaggregated net migration and prime-aged net migration variables. This approach allows us to disentangle the extent to which migration or natural population change drive population growth, stagnation, and decline. As net migration estimates are constrained to years in which the decennial census is conducted, we estimate economic complexity in 2010 and measure the net migration rate from 2010 to 2020. Population growth rate estimates are available at annual increments; as such, we measure both five-year and ten-year population growth from 2010.⁷

Shifting to multivariate modeling, we find strong evidence of spatial autocorrelation in OLS model residuals, which violates the independent observations assumption and could result in downward-biased standard errors and inaccurate parameter estimates (Cliff and Ord 1973; Cliff and Ord 1981). Consequently, we use spatial error models to account for spatial patterning of omitted or immeasurable explanatory predictors of population growth and net migration (i.e., shared climate conditions or proximity to amenities of neighboring CZs) (Ward & Gleditsch 2018).⁸

Finally, while global models are instructive of nation-wide associations in local labor market dynamics, associations likely differ for regional and spatial contexts with varying underlying characteristics, thereby questioning the validity of the “constancy assumption” in global models that asserts relationships between tested variables apply to all observations (Freedman et al. 1991). Spatial regime models allow us to explore heterogeneity through both contiguous and non-contiguous partition strategies while maintaining the underlying spatial structure of our data operationalized in global spatial error models (O’Loughlin et al., 1994; LeSage 1999; Curtis and O’Connell 2017). We group regimes by each CZs degree of rurality, using both binary and ordinal categorization of non-metro categories.

4.0. Results

4.1. Descriptive analysis of complexity and population growth

The spatial distribution of economic complexity and population growth are mapped in Figures 2 and 3. Descriptive statistics are presented in Appendix B and C. In the maps, we observe clear spatial patterning in both EC and population growth, characterized by positive and statistically significant spatial autocorrelation (Moran’s I values). Clusters of low EC emerge in the Great Plains as well as parts of the West and central Appalachia. Low population growth emerges in all areas of the US with the exception of the West and South. The spatial distribution of EC and population growth overlap most strongly in the central part of the US and most weakly in the areas within the West, suggestive of spatial differentiation in the relationship between economic growth and population growth.

Figure 2 –

Map of Commuting Zone (CZ) Economic Complexity for US Counties

Source: EC data comes from authors’ calculations using 2010 CBP via Eckert et al. (2020). *** p < 0.001; ** p < 0.01; * p < 0.05.

Figure 3 –

Map of 5-year and 10-year Population Growth for US Commuting Zones

Source: 10-year population growth data comes from Census PEP 2010–2020. *** p < 0.001; ** p < 0.01; * p < 0.05.

Source: 5-year population growth data comes from Census PEP 2010–2015. *** p < 0.001; ** p < 0.01; * p < 0.05.

4.2. Complexity, population change, and heterogeneity by race

Bivariate analysis indicates a moderate, statistically significant association between higher economic complexity and higher population growth, along with similar complexity–migration associations.⁹ This association persists in our multivariate spatial error analysis presented in Table 1. Higher EC holds a significant, positive association with higher population growth over the short- and long-term (5- and 10-years, respectively) net of covariates. A one standard deviation increase in a CZ’s EC is associated with a 0.04 and 0.02 standard deviation increase in population growth over 10-year and 5-year periods.

Table 1 –

Spatial Error Analysis of Economic Complexity on Population Growth and Net In-Migration for US Commuting Zones

	Population Growtha		Net In-Migration Rateb
	10-yr	5-yr	Aggregate	NH Asian	NH Black	Hispanic	NH Indigenous	NH White
(Intercept)	−0.15	−0.24	5.40	−211.25	58.89	−119.31	−61.73	9.93
	(0.28)	(0.15)	(23.29)	(363.88)	(321.17)	(216.07)	(90.21)	(25.24)
ECc	0.04***	0.02***	3.07***	−15.17**	−17.76***	−11.12***	−2.78*	3.24***
	(0.00)	(0.00)	(0.31)	(4.67)	(4.14)	(2.83)	(1.17)	(0.33)
Lambda	0.61***	0.67***	0.62***	−0.02	0.36***	0.60***	0.46***	0.57***
	(0.04)	(0.04)	(0.04)	(0.07)	(0.05)	(0.04)	(0.05)	(0.04)
Non-metro (= 1)d	−0.02**	−0.01	−1.48**	−3.59	1.20	−10.75*	−0.22	−0.83
	(0.01)	(0.00)	(0.57)	(10.38)	(8.33)	(5.31)	(2.28)	(0.62)
Log incomee	0.01	0.03*	−3.05	14.20	−33.36	20.63	3.55	−2.27
	(0.03)	(0.02)	(2.42)	(37.69)	(33.31)	(22.41)	(9.36)	(2.62)
% NH Whitef	0.08*	0.03	12.13***	−45.53	−83.58**	−68.29**	8.98	8.00**
	(0.03)	(0.02)	(2.47)	(28.66)	(29.76)	(22.69)	(8.81)	(2.60)
% Foreign borng	0.13	0.10*	13.15	−164.99	−230.65**	−244.23***	12.19	−21.20**
	(0.08)	(0.05)	(6.98)	(86.46)	(86.45)	(64.30)	(25.30)	(7.40)
Median ageh	−0.00**	−0.00***	0.33***	2.25*	3.97***	−0.33	0.67**	0.28***
	(0.00)	(0.00)	(0.07)	(0.98)	(0.89)	(0.62)	(0.25)	(0.07)
Employment ratei	0.17*	0.07*	6.68	115.35	439.09***	40.24	10.19	−7.01
	(0.07)	(0.04)	(5.66)	(70.61)	(69.39)	(52.08)	(20.33)	(5.98)
nobs	617	617	615	614	615	615	615	615
R-squared	0.52	0.56	0.54	0.05	0.30	0.36	0.21	0.47
AIC	−1,717	−2,481	3,733	7,202	6,968	6,471	5,396	3,828
BIC	−1,673	−2,437	3,777	7,246	7,013	6,516	5,441	3,873
deviance	1.98	0.56	13,791	4,319,294	2,830,151	1,190,838	215,456	16,378
logLik	868	1,251	−1,857	−3,591	−3,474	−3,226	−2,688	−1,904
nobs.1	617	617	615	614	615	615	615	615

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^***p < 0.001

^**p < 0.01

^*p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.USDA ERS RUCC with author rescaling

^e.-i.Census ACS 2006–2010.

Our analysis of net migration and prime-aged net migration yields consistent results. A one standard deviation increase in a CZ’s EC is associated with a 3.07 net migrant increase per 100 residents (approximately equal to 0.44 standard deviations). In addition to demonstrating an aggregate association between economic complexity levels and subsequent population growth and net migration, these initial results suggest a relatively strong connection between EC and net migration compared with combined elements of migration, births, and deaths captured in population growth measures.

As hypothesized, analysis of the complexity–migration relationship for disaggregated ethnoracial groups reveals divergences in the positive association characterizing the total population and all migrant groups combined. While similar moderate, significant positive associations between EC and net migration persist for White migration, statistically significant negative associations between EC and net migration emerge for Asian, Black, Hispanic, and Indigenous migration. A one standard deviation increase in EC corresponds with a 3.24 increase in White migrants per 100 residents (0.46 of a standard deviation), compared with a 15.17 (0.18 standard deviations), 17.76 (0.22 standard deviations), 11.12 (0.21 standard deviations), and 2.78 (0.13 standard deviations) decrease in Asian, Black, Hispanic, and Indigenous migrants per 100 residents, respectively. This suggests that while White populations tend to locate and maintain residence in places with more complex economies, historically marginalized populations are more likely to leave and not move to places with more complex economies, instead moving to and maintaining residence in less economically complex places.

4.3. Complexity, population change, and heterogeneity by rurality

Results from the global model indicate the rurality of a CZ is negatively associated with higher population growth and net in-migration. Further, testing interaction terms between complexity and rurality provides evidence that the complexity–growth association is strongest among non-metro economies. We more systematically explore the heterogeneity of the complexity–growth relationship by rurality through a spatial regime approach. Results are reported in Table 2. Contrary to our a priori expectations, we find that the relationship between higher EC and higher population growth and net migration persists in non-metro labor markets. Coefficients in non-metro models featuring population growth outcome variables are unchanged, while coefficients increased to 3.38 net migrants per 100 residents (0.48 standard deviations) in non-metro models with net migration rates as outcome variables. In contrast, the association is more tenuous for metropolitan economies, as the level of magnitude and statistical significance are materially lower: around half the size as in non-metro population growth models, and 0.19 standard deviations lower in non-metro net migration models. We discuss these unexpected results in the following section.

Table 2 –

Spatial Regime Analysis of Economic Complexity on Population Growth and Net Migration for US Commuting Zones

	Population Growtha		Net In-Migration Rateb
	10-yr	5-yr	Aggregate	NH Asian	NH Black	Hispanic	NH Indig.	NH White
Intercept (Non-metro)	−0.36	−0.33*	−13.08	−159.09	201.18	−29.76	−47.09	−10.15
	(0.30)	(0.16)	(25.35)	(399.69)	(350.49)	(236.20)	(98.86)	(27.42)
Intercept (Metro)	−0.07	−0.06	23.31	75.10	−195.49	−16.52	7.12	76.30
	(0.68)	(0.36)	(56.47)	(941.28)	(803.40)	(527.52)	(223.78)	(61.39)
EC (Non-metro)c	0.04***	0.02***	3.38***	−18.58***	−21.05***	−13.66***	−3.20*	3.45***
	(0.00)	(0.00)	(0.33)	(4.97)	(4.43)	(3.03)	(1.26)	(0.35)
EC (Metro)	0.02*	0.01**	2.05**	4.33	−0.67	−1.32	−0.98	2.58**
	(0.01)	(0.00)	(0.75)	(14.16)	(11.06)	(7.02)	(3.02)	(0.82)
lambda	0.61***	0.67***	0.61***	−0.06	0.35***	0.59***	0.47***	0.57***
	(0.04)	(0.04)	(0.04)	(0.07)	(0.05)	(0.04)	(0.05)	(0.04)
Log income (Non-metro)d	0.03	0.04*	−1.31	8.89	−49.12	10.34	1.96	−0.28
	(0.03)	(0.02)	(2.62)	(41.23)	(36.23)	(24.44)	(10.23)	(2.84)
Log income (Metro)	0.00	0.00	−5.38	−6.93	14.33	7.12	−5.15	−11.19
	(0.07)	(0.04)	(6.10)	(103.31)	(87.04)	(56.99)	(24.21)	(6.63)
% NH White (Non-metro)e	0.07*	0.03	12.09***	−61.04*	−90.42**	−67.85**	5.91	7.01**
	(0.03)	(0.02)	(2.51)	(30.96)	(30.89)	(23.13)	(9.17)	(2.66)
% NH White (Metro)	0.03	−0.01	8.18	15.91	−27.98	−61.94	24.88	9.47
	(0.06)	(0.03)	(4.77)	(76.86)	(65.57)	(44.36)	(18.50)	(5.15)
% foreign born (Non-metro)f	0.20*	0.14**	20.09*	−312.94**	−345.47**	−315.17***	−19.06	−22.45**
	(0.10)	(0.05)	(8.11)	(113.71)	(106.05)	(75.15)	(30.64)	(8.68)
% foreign born (Metro)	0.09	0.06	7.09	−27.90	−126.69	−183.31	65.30	−9.22
	(0.13)	(0.07)	(10.87)	(148.65)	(139.35)	(100.54)	(40.60)	(11.60)
Median age (Non-metro)g	−0.00**	−0.00***	0.33***	2.18*	4.08***	−0.46	0.68*	0.27***
	(0.00)	(0.00)	(0.07)	(1.05)	(0.94)	(0.65)	(0.27)	(0.07)
Median age (Metro)	−0.00	−0.00	0.57**	0.57	0.69	0.20	0.70	0.70***
	(0.00)	(0.00)	(0.18)	(2.92)	(2.50)	(1.67)	(0.70)	(0.19)
Employment rate (Non-metro)h	0.16*	0.07	5.16	150.89*	483.40***	62.25	18.62	−6.40
	(0.07)	(0.04)	(5.92)	(74.64)	(73.27)	(54.56)	(21.67)	(6.28)
Employment rate (Metro)	0.17	0.12	8.26	−29.92	126.92	31.41	13.13	6.40
	(0.17)	(0.09)	(14.37)	(246.91)	(203.81)	(134.10)	(56.70)	(15.59)
nobs	617	617	615	614	615	615	615	615
r.squared	0.53	0.57	0.54	0.06	0.31	0.36	0.22	0.48
AIC	−1,717	−2,482	3,733	7,207	6,970	6,477	5,404	3,827
BIC	−1,646	−2,411	3,804	7,278	7,040	6,547	5,475	3,898
deviance	1.95	0.55	13,570	4,271,621	2,785,463	1,183,435	213,492	16,023
logLik	874	1257	−1,851	−3,588	−3,469	−3,222	−2,686	−1,897
nobs.1	617	617	615	614	615	615	615	615

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^***p < 0.001

^**p < 0.01

^*p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.-h.Census ACS 2006–2010.

Partitioning the rural-urban comparison by varying levels of RUCC aggregation, we find that medium- and small-population CZs exhibit the strongest associations between EC and both population growth and net migration. Urban areas of 20,000 residents or larger not included in metropolitan statistical areas demonstrate the strongest associations between complexity and growth (RUCC 4-5).¹⁰ Together, these findings suggest stronger associations between economic complexity and population growth among more rural places as compared to more urban places, the latter of which have been the exclusive focus of prior studies. In non-metro contexts, lagging economic complexity is more acutely tied to declining population, indicative of the importance of economic narratives of stagnation and decline.

Findings are consistent for models disaggregating migration by rurality and race. Net migration for White residents has the strongest association with higher EC levels in non-metro areas, and more modest associations with higher EC in metro areas. Meanwhile, the same directional associations are maintained in these rurality-partitioned models, and non-metro economies are more strongly consistent with general trends in comparison to non-significant or weak associations in metropolitan economies. These findings suggest racialized patterns in the economic-population growth relationship are more pronounced among more rural areas and in ways that align higher economic growth with higher growth in the White population.¹¹

5.0. Discussion

Previous research using economic complexity has primarily focused on examining outcomes of economic development using international and subnational metropolitan contexts, effectively excluding consideration of how EC associates with outcomes in non-metropolitan contexts. Use of high-resolution, race-specific net migration estimates in the US allows for first-of-its-kind analysis of the connection between EC as an operationalization of relative economic development and disparate population trends between White and historically marginalized ethnoracial groups. Additionally, aggregation of county-level workforce and demographic data to commuting zones permitted analysis of contiguous US regions, allowing for the study of potential heterogeneity along the rural-urban continuum.

Given long-term trends in economic divergence and compositional differences in the size and scope of metropolitan and non-metropolitan economies, larger and more urbanized regional economies hold higher levels of economic complexity, and experience higher population growth and positive net migration. We set out to answer the question: do more economically complex places sustain and grow their populations to a greater extent? We find in general, yes. Our aggregate findings underscore the material role of greater industrial diversification and domestic comparative advantage in encouraging population growth through resident attraction and retention. Results from our metro-non-metro regime model suggest this pattern is particularly pronounced for relatively under-resourced non-metropolitan economies that have suffered from economic divergence, population stagnation, and decline in preceding decades. However, we caution against treating economic complexity as a “silver bullet,” as our findings illuminate unequal access to these dynamic economies along lines of rurality and race.

Contrary to our expectations that the associations between complexity and population growth via the net inflow of new residents would be stronger in metropolitan areas characterized by higher EC and higher population growth, the complexity-migration associations are strongest outside of metropolitan economies. This suggests that other explanations for resident attraction and retention accounted for in our spatial error term are more salient in metropolitan compared with non-metropolitan contexts, perhaps including amenity and non-economic factors (e.g., family ties and place connectedness). In more-rural commuting zones, a lack of economic diversity and domestic comparative advantage are prohibitive to population growth, suggesting that economic stagnation or decline has more severe consequences for the maintenance of population in more-rural areas than in more-urban areas with alternative paths to growing and sustaining their populations. Observed heterogeneity in demographic and economic processes across the rural-urban continuum, especially at the cusp of the traditional rural-urban cutoff, underscores the crucial importance of temporally consistent and flexible definitions of rurality in the context of reclassification of many non-metro areas as metropolitan during recent decadal reclassification.

Further, we find disparate directional associations between economic complexity, population growth, and migration by race and ethnicity. In line with our presuppositions, non-Hispanic White populations appear to locate in greener economic pastures at higher rates. Meanwhile, historically marginalized ethnoracial groups are more likely to move to and reside in places with less dynamic economies. These results highlight racial inequalities in access to greater economic opportunity and raise the broader question of who benefits from aggregate “economic development”?

From the historical Great Migration to suburbanization in the late 20th Century to contemporary sponsored labor migrations, migration has been historically rationalized as a possible pathway by which individuals and households can escape the confines of place and improve their economic conditions, especially when afflicted with lower labor market opportunities and worse public services often found in more distressed economic contexts (Tickamyer & Duncan 1990; Spring et al. 2015). However, for historically-marginalized populations – particularly those in non-metropolitan contexts – our contemporary findings suggest departure from a previous geographic context to a more prosperous one is comparatively fraught, as pathways to opportunity are constrained by historical and contemporary structural forces (Brooks et al. 2010; McGranahan et al. 2010; Chetty et al. 2014; Curtis & O’Connell 2017; Krause & Reeves 2017). At the place level, migration is celebrated as a pathway toward community economic development, including among places with historically contracting population size and economic infrastructure. Our findings suggest this pathway toward greater economic development might apply exclusively to places with larger White populations and higher White migration, thereby challenging the broad assumption that economic development benefits all places and populations similarly.

The relationship between economic complexity and population growth is but one branch of the extensive lattice of literature on economic growth and population well-being. Our findings enrich broader discussions of the tenuous ties between economic growth, development, and distributed prosperity. As economic complexity emerges as a macroeconomic indicator of development that can be tied to aggregate levels of socioeconomic and demographic change, and national, regional, and local economies alike seek to increase their industrial diversity and domestic comparative advantage, our analysis serves as an important reminder that development and growth are not uniform across different place contexts and demographic groups. While analyses in this study focus on regional economic growth and population change, economic complexity could be a useful tool for understanding racial disparities in prosperity at various spatial scales, including micro-level displacement and widening racial inequalities driven by broader economic growth (i.e., the process of economic growth and development, gentrification, and displacement).

Our findings demonstrate the benefits of a more dynamic local economy are not evenly distributed. Rising levels of economic output, population growth, and average incomes may be indicative of “development” in the eyes of elected officials, economic development administrators, and private stakeholders, yet one is left to ask whether “a rising tide raises all boats” (Kennedy 1963)? The complexity–growth association demonstrated throughout various global and aggregated models presented in this work and others masks an unequal share of the spoils by long over-resourced subpopulations and types of places, suggestive of concerning consequences of traditional economic development in exacerbating racial and spatial inequalities.

10.0. Appendices

10.1. Appendix A - Robustness Tests

10.1. Specification of EC

We conduct a series of tests to determine robustness of estimates to alternative specifications of economic complexity. EC estimates computed using regional industry sales and employment (as used in our main analyses) hold a pairwise correlation of 0.75. There are minor, non-contradictory differences in model results when using sales and employment data.

When using varying aggregations of industry employment classifications, including the NAICS 3-digit, 5-digit, and 6-digit levels, we find no differences in significance or direction compared with results obtained with EC estimates computed using the 4-digit NAICS level (as used in our main analyses).

Additionally, we examined the robustness of results across time. Our results are mostly robust when adjusting our temporal range from base period 2010 and outcome periods 2015 and 2020 to base period 2000 and outcome periods 2005 and 2010.

In order to demonstrate comparability to previous subnational EC research, we compare our results garnered at the commuting zone regional aggregation to alternative results at the metropolitan and micropolitan area level and find minor, non-contradictory differences between CZ and CBSA models.

Finally, EC estimates computed using Fritz and Manduca’s (2021) alternative specification of revealed comparative advantage are substantially the same as estimates yielded from traditional binary specification of RCA. The two different EC estimates hold a pairwise correlation of 0.91 and virtually no differences in model results when using the different RCA specification methods presented in Equations 1.1 and 1.2.

$$RC{A}_{r,i}=1if\left[\frac{X_{r,i}/X_r}{X_i/X}\ge 1\right]or{X}_{r,i}>50$$ (1.2)

10.1.2. Selection of rurality covariate

We tested multivariate models using three different definitions of CZ scale in addition to our standard covariates – a metro–non-metro dummy, a continuous weighted mean RUCC code, and log population. While metro–non-metro differences are among the core interests of our analysis, population size is highly correlated with our measure of economic complexity. Compositionally, more populous CZs have a strong tendency to hold higher EC levels, as larger economies typically have the capacity to support a wider variety of industries than smaller economies. Due to this compositional reality, we examine diagnostics for multicollinearity and seek to ascertain whether a mediation relationship properly explains the interrelation between complexity and population size in their association to population growth.

As discussed, EC is highly correlated with log population.¹ With a VIF of 5.00, log population is problematically multicollinear with other independent variables in our model, while our other specifications of rurality – a non-metro dummy and continuous RUCC code – hold a VIF of 1.67 and 3.37. Using mediation analysis between our main predictor variable (EC), log population, and our outcome variable (10-year population growth), we find that while EC and log population together hold significant total effect with population growth, EC holds a significant average direct effect (ADE), yet the average causal mediation effect (ACME) is not significant. Results are consistent for our net migration outcome variable. Meanwhile, there is a very small, statistically significant mediation effect of log population on our 5-year population growth outcome variable; however, this mediation effect is miniscule, accounting for less than 0.1% of the total effect. Jointly considering high levels of multicollinearity and virtually no mediation effect on our outcome variables, we conclude that log population is less useful of an explanatory variable than our non-metro dummy and continuous RUCC specifications, and opt to use these latter two definitions of rurality instead of log population in our core analyses.

10.1.3. Sensitivity of model results from Table 1 and 2

10.1.3.1. Sensitivity to geographic aggregation – commuting zones and counties

We run alternative specifications of our spatial error and spatial regime models disaggregating our geographic unit to the county-level in lieu of the commuting zone for the measurement of population growth, net migration rates, and covariates. We continue to use CZ-level EC. Spatial error and regime results are robust to main models with the exception of models for 5-year population growth and net-migration of Asian residents, which do not maintain statistical significance at a 0.05 level.

It is important to note that ethnoracial group specific net migration rates are far more unstable than CZ aggregated rates, particularly in counties with small populations of historically marginalized ethnoracial groups. For example, fewer than 5% of US counties had a non-Hispanic Asian population share of at least 5%. The standard deviation of net migration rate per 100 residents for Asian residents rises from 86.3 to 430.0 when data is shifted to county-level aggregation. Further, model fit statistics substantially weaken for ethnoracial group specific models at the county-level, likely due to the strong increase in variability of net migration rate estimates for these small groups at the county level. Due to these reasons, we do not present ethnoracial group specific county-level models.

Table A1 –

Spatial Error Analysis of Economic Complexity on Population Growth and Net In-Migration for US Counties

	Population Growtha		Net In-Migration Rateb
	10-yr	5-yr	Aggregate
(Intercept)	−0.66***	−0.39***	−40.40***
	(0.100)	(0.050)	(9.370)
ECc	0.01***	0.00	1.31***
	(0.000)	(0.000)	(0.200)
Lambda	0.64***	0.66***	0.63***
	(0.020)	(0.020)	(0.020)
Non-metro (= 1)d	−0.03***	−0.01***	−2.61***
	(0.000)	(0.000)	(0.320)
Log incomee	0.06***	0.04***	1.81
	(0.010)	(0.010)	(0.990)
% NH Whitef	0.13***	0.05***	16.08***
	(0.010)	(0.010)	(1.260)
% Foreign borng	0.31***	0.19***	20.93***
	(0.040)	(0.020)	(3.510)
Median ageh	−0.00***	−0.00***	0.06*
	(0.000)	(0.000)	(0.030)
Employment ratei	0.21***	0.09***	13.03***
	(0.030)	(0.010)	(2.490)
nobs	3,079	3,079	3074
R-squared	0.540	0.560	0.470
AIC	−7,553	−11,484	20,289
BIC	−7,493	−11,424	20,350
deviance	14.07	3.89	120,546
logLik	3,786	5,752	−10,135
nobs.1	3,079	3,079	3074

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^***p < 0.001

^**p < 0.01

^*p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.USDA ERS RUCC with author rescaling

^e.-i.Census ACS 2006–2010.

Table A2 –

Spatial Regime Analysis of Economic Complexity on Population Growth and Net Migration for US Commuting Zones, Metropolitan and Non-metropolitan areas

	Population Growtha		Net In-Migration Rateb
	10-yr	5-yr	Aggregate
Intercept (Non-metro)	−0.25*	−0.23***	−6.2
	(0.120)	(0.060)	(11.210)
Intercept (Metro)	−1.44***	−0.70***	−102.51***
	(0.170)	(0.090)	(15.380)
EC (Non-metro)c	0.01***	0.00	1.19***
	(0.000)	(0.000)	(0.230)
EC (Metro)	0.01*	0.00	1.07**
	(0.000)	(0.000)	(0.370)
lambda	0.65***	0.68***	0.64***
	(0.020)	(0.020)	(0.020)
Log income (Non-metro)d	0.02	0.03***	−1.84
	(0.010)	(0.010)	(1.180)
Log income (Metro)	0.15***	0.07***	8.06***
	(0.020)	(0.010)	(1.690)
% NH White (Non-metro)e	0.14***	0.05***	16.82***
	(0.020)	(0.010)	(1.390)
% NH White (Metro)	0.13***	0.06***	16.11***
	(0.020)	(0.010)	(1.860)
% foreign born (Non-metro)f	0.20***	0.13***	8.04
	(0.050)	(0.020)	(4.230)
% foreign born (Metro)	0.38***	0.25***	31.47***
	(0.050)	(0.030)	(5.040)
Median age (Non-metro)g	−0.00***	−0.00***	0.10**
	(0.000)	(0.000)	(0.040)
Median age (Metro)	−0.01***	−0.00***	−0.05
	(0.000)	(0.000)	(0.060)
Employment rate (Non-metro)h	0.18***	0.08***	10.31***
	(0.030)	(0.020)	(2.730)
Employment rate (Metro)	0.23***	0.10***	16.72***
	(0.050)	(0.030)	(4.390)
nobs	3,079	3,079	3,074
r.squared	0.56	0.57	0.49
AIC	(7,653)	(11,543)	20,206
BIC	(7,557)	(11,446)	20,303
deviance	13.51	3.78	116,425
logLik	3,843	5,787	(10,087)
nobs.1	3,079	3,079	3,074

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^***p < 0.001

^**p < 0.01

^*p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.-h.Census ACS 2006–2010.

10.1.3.2. Sensitivity along RUCC continuum

In addition to binary specification of metro/non-metro regimes, we conduct spatial regime analysis by a variety of groupings of RUCC, including nine individual regimes for each RUCC.

Table A3 –

Spatial Regime Analysis of Economic Complexity on Population Growth and Net Migration for US Commuting Zones, RUCC codes

	Population Growtha		Net In-Migration Rateb
	10-yr	5-yr	Aggregate
Intercept (RUCC 1 = most urban)	−0.51	−0.24	−13.44
	(1.750)	(0.930)	(145.390)
Intercept (RUCC 2)	−2.23	−1.07	−89.81
	(1.330)	(0.710)	(110.650)
Intercept (RUCC 3)	−1.00	−0.34	−54.37
	(1.060)	(0.560)	(88.310)
Intercept (RUCC 4)	−0.66	−0.46	16.58
	(0.750)	(0.400)	(62.600)
Intercept (RUCC 5)	−0.08	−0.12	10.77
	(0.710)	(0.380)	(59.490)
Intercept (RUCC 6)	−0.71	−0.28	−86.42
	(0.720)	(0.380)	(60.070)
Intercept (RUCC 7)	−0.76	−0.47	−40.57
	(0.500)	(0.260)	(41.410)
Intercept (RUCC 8)	−0.78	−0.94*	−51.71
	(0.590)	(0.310)	(49.910)
Intercept (RUCC 9 = most rural)	0.26	−0.02	5.48
	(0.890)	(0.470)	(74.160)
ECc (RUCC 1 = most urban)	0.00	0.00	−0.12
	(0.020)	(0.010)	(2.000)
ECc (RUCC 2)	0.01	0.01	0.61
	(0.020)	(0.010)	(1.540)
ECc (RUCC 3)	0.03	0.02	3.82*
	(0.020)	(0.010)	(1.320)
ECc (RUCC 4)	0.05**	0.02	4.10*
	(0.020)	(0.010)	(1.270)
ECc (RUCC 5)	0.07**	0.03**	4.91**
	(0.010)	(0.010)	(1.170)
ECc (RUCC 6)	0.05**	0.02**	3.43*
	(0.010)	(0.010)	(1.150)
ECc (RUCC 7)	0.03*	0.02**	2.42*
	(0.010)	(0.000)	(0.770)
ECc (RUCC 8)	0.04**	0.01*	3.65**
	(0.010)	(0.000)	(0.700)
ECc (RUCC 9 = most rural)	0.02	0.01	1.96
	(0.010)	(0.010)	(1.110)
lambda	0.61**	0.68**	0.62**
	(0.040)	(0.040)	(0.040)
Log incomee (RUCC 1 = most urban)	0.00	0.00	−5.99
	(0.210)	(0.110)	(17.380)
Log incomee (RUCC 2)	0.22	0.1	6.92
	(0.140)	(0.070)	(11.670)
Log incomee (RUCC 3)	0.11	0.04	3.18
	(0.110)	(0.060)	(9.580)
Log incomee (RUCC 4)	0.05	0.04	−5.47
	(0.080)	(0.040)	(6.580)
Log incomee (RUCC 5)	0.01	0.01	−3.81
	(0.080)	(0.040)	(6.500)
Log incomee (RUCC 6)	0.07	0.04	7.73
	(0.070)	(0.040)	(6.210)
Log incomee (RUCC 7)	0.06	0.06	0.51
	(0.050)	(0.030)	(4.240)
Log incomee (RUCC 8)	0.08	0.11**	3.29
	(0.060)	(0.030)	(5.190)
Log incomee (RUCC 9 = most rural)	−0.05	0.01	−5.46
	(0.090)	(0.050)	(7.380)
% NH Whitef (RUCC 1 = most urban)	−0.06	−0.03	0.58
	(0.190)	(0.100)	(15.430)
% NH Whitef (RUCC 2)	−0.15	−0.09	−6.89
	(0.120)	(0.060)	(9.790)
% NH Whitef (RUCC 3)	0.13	0.05	15.94*
	(0.070)	(0.040)	(5.980)
% NH Whitef (RUCC 4)	−0.03	−0.03	3.98
	(0.070)	(0.040)	(5.980)
% NH Whitef (RUCC 5)	0.09	0.03	10.36
	(0.050)	(0.030)	(4.180)
% NH Whitef (RUCC 6)	0.13	0.07	17.65**
	(0.060)	(0.030)	(5.140)
% NH Whitef (RUCC 7)	0.05	0.06	10.13*
	(0.040)	(0.020)	(3.640)
% NH Whitef (RUCC 8)	0.00	−0.01	9.51
	(0.050)	(0.030)	(4.470)
% NH Whitef (RUCC 9 = most rural)	0.00	0.05	8.32
	(0.160)	(0.090)	(13.310)
% Foreign borng (RUCC 1 = most urban)	−0.07	0.04	1.07
	(0.260)	(0.140)	(22.090)
% Foreign borng (RUCC 2)	−0.07	0.02	−10.62
	(0.230)	(0.120)	(18.920)
% Foreign borng (RUCC 3)	0.43	0.2	29.43
	(0.190)	(0.110)	(16.250)
% Foreign borng (RUCC 4)	0.69	0.27	52.40
	(0.270)	(0.150)	(22.570)
% Foreign borng (RUCC 5)	0.49	0.25	35.60
	(0.200)	(0.110)	(16.870)
% Foreign borng (RUCC 6)	0.39	0.16	18.23
	(0.220)	(0.120)	(18.650)
% Foreign borng (RUCC 7)	0.18	0.23	7.6
	(0.200)	(0.110)	(16.310)
% Foreign borng (RUCC 8)	−0.27	−0.23	−25.5
	(0.230)	(0.120)	(19.050)
% Foreign borng (RUCC 9 = most rural)	0.13	0.22	25.39
	(0.350)	(0.190)	(28.970)
Median ageh (RUCC 1 = most urban)	0.01	0.00	1.1
	(0.010)	(0.000)	(0.770)
Median ageh (RUCC 2)	0.00	0.00	0.83*
	(0.000)	(0.000)	(0.310)
Median ageh (RUCC 3)	−0.01	0.00	0.26
	(0.000)	(0.000)	(0.220)
Median ageh (RUCC 4)	0.00	0.00	0.50
	(0.000)	(0.000)	(0.200)
Median ageh (RUCC 5)	−0.01	−0.00	0.22
	(0.000)	(0.000)	(0.180)
Median ageh (RUCC 6)	0.00	−0.00*	0.00
	(0.000)	(0.000)	(0.200)
Median ageh (RUCC 7)	0.00	−0.00**	0.50**
	(0.000)	(0.000)	(0.130)
Median ageh (RUCC 8)	0.00	−0.00**	0.28
	(0.000)	(0.000)	(0.120)
Median ageh (RUCC 9 = most rural)	0.00	0.00	0.80*
	(0.000)	(0.000)	(0.250)
Employment ratei (RUCC 1 = most urban)	0.7	0.43	62.93
	(0.680)	(0.360)	(57.030)
Employment ratei (RUCC 2)	−0.06	0.07	−6.89
	(0.310)	(0.170)	(26.070)
Employment ratei (RUCC 3)	0.00	0.02	−3.16
	(0.240)	(0.130)	(20.100)
Employment ratei (RUCC 4)	0.34	0.21	25.98
	(0.170)	(0.090)	(14.450)
Employment ratei (RUCC 5)	0.2	0.11	17.91
	(0.160)	(0.090)	(13.300)
Employment ratei (RUCC 6)	0.01	0.01	−12.34
	(0.140)	(0.080)	(11.940)
Employment ratei (RUCC 7)	0.21	0.03	10.19
	(0.110)	(0.060)	(9.210)
Employment ratei (RUCC 8)	0.18	0.03	0.45
	(0.110)	(0.060)	(9.740)
Employment ratei (RUCC 9 = most rural)	0.2	0.06	9.98
	(0.210)	(0.110)	(17.520)
nobs	617	617	615
r.squared	0.58	0.61	0.59
AIC	−1,687	−2,447	3,761
BIC	−1,400	−2,159	4,049
deviance	1.74	0.5	12,062
logLik	909	1,288	−1,816
nobs.1	617	617	615

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^**p < 0.001

^*p < 0.01

p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.-h.Census ACS 2006–2010.

10.1.3.3. Sensitivity to binary rurality threshold choice

In addition to our core binary specification of metro/non-metro regimes using a threshold of 4.0 to categorize continuous CZ codes, we replicate binary spatial regime results for aggregate outcome variables from Table 2 for alternative thresholds ranging from 2.5 to 5.5. Results demonstrate relatively stronger associations between complexity and population growth and in-migration in non-metropolitan areas are robust to shifting the population-weighted CZ RUC threshold from 4.0 to 2.5, 3.0, 3.5, 4.5, 5.0, and 5.5. In line with previous findings depicted in Table A3 defining spatial regimes by each RUC code in lieu of binary metropolitan/non-metropolitan classifications, we find binary results with alternative thresholds closer to 2.5 yield greater contrasts between metropolitan and non-metropolitan regimes (in terms of significance and magnitude). Setting thresholds closer to 5.5 yields greater similarity between metropolitan and non-metropolitan regimes, albeit differences in magnitude are observed (particularly minor in 5.5, but more pronounced in 4.5 and 5.0).

Table A4 –

Spatial Regime Analysis of Economic Complexity on Population Growth and Net Migration for US Commuting Zones, Metropolitan and Non-metropolitan areas with Alternative Rurality Thresholds

	5-year Population Growtha
	2.5	3.0	3.5	4.0 (original)	4.5	5.0	5.5
Intercept (Non-metro)	−0.41 (−0.28)	−0.41 (−0.29)	−0.37 (−0.3)	−0.36 (−0.3)	−0.3 (−0.32)	−0.51 (−0.33)	−0.5 (−0.34)
Intercept (Metro)	−0.85 (−1.06)	−0.4 (−0.94)	−0.41 (−0.75)	−0.07 (−0.68)	−0.27 (−0.55)	0.11 (−0.48)	−0.12 (−0.44)
EC (Non-metro)c	0.04*** (0.00)	0.04*** (0.00)	0.04*** (0.00)	0.04*** (0.00)	0.04*** (0.00)	0.04*** (0.00)	0.04*** (0.00)
EC (Metro)	0.00 (−0.01)	0.00 (−0.01)	0.01 (−0.01)	0.02* (−0.01)	0.02** (−0.01)	0.03*** (−0.01)	0.03*** (−0.01)
lambda	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)
Log income (Non-metro)d	0.04 (−0.03)	0.04 (−0.03)	0.03 (−0.03)	0.03 (−0.03)	0.03 (−0.03)	0.05 (−0.03)	0.04 (−0.03)
Log income (Metro)	0.07 (−0.11)	0.03 (−0.1)	0.04 (−0.08)	0.00 (−0.07)	0.02 (−0.06)	−0.01 (−0.05)	0.01 (−0.05)
% NH White (Non-metro)e	0.07* (−0.03)	0.07* (−0.03)	0.07* (−0.03)	0.07* (−0.03)	0.08** (−0.03)	0.08* (−0.03)	0.08* (−0.03)
% NH White (Metro)	−0.11 (−0.1)	0.00 (−0.08)	0.02 (−0.06)	0.03 (−0.06)	0.02 (−0.05)	0.03 (−0.04)	0.06 (−0.04)
% foreign born (Non-metro)f	0.24** (−0.09)	0.22* (−0.09)	0.24* (−0.09)	0.20* (−0.1)	0.20* (−0.1)	0.20* (−0.1)	0.18 (−0.1)
% foreign born (Metro)	−0.14 (−0.17)	0.05 (−0.15)	0.00 (−0.13)	0.09 (−0.13)	0.07 (−0.12)	0.11 (−0.12)	0.12 (−0.12)
Median age (Non-metro)g	−0.00** (0.00)	−0.00** (0.00)	−0.00** (0.00)	−0.00** (0.00)	−0.00** (0.00)	−0.00* (0.00)	0.00 (0.00)
Median age (Metro)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	−0.00* (0.00)
Employment rate (Non-metro)h	0.15* (−0.07)	0.15* (−0.07)	0.15* (−0.07)	0.16* (−0.07)	0.16* (−0.07)	0.13 (−0.07)	0.13 (−0.08)
Employment rate (Metro)	0.16 (−0.3)	0.25 (−0.25)	0.12 (−0.19)	0.17 (−0.17)	0.2 (−0.15)	0.23 (−0.12)	0.2 (−0.11)
nobs	617	617	617	617	617	617	617
r.squared	0.54	0.53	0.54	0.53	0.53	0.53	0.52
AIC	−1,727	−1,723	−1,724	−1,717	−1,717	−1,714	−1,710
BIC	−1,656	−1,652	−1,653	−1,646	−1,647	−1,643	−1,639
deviance	1.91	1.92	1.92	1.95	1.94	1.96	1.97
logLik	879	877	878	875	875	873	871
nobs.1	617	617	617	617	617	617	617
	10-year Population Growtha
	2.5	3.0	3.5	4.0 (original)	4.5	5.0	5.5
Intercept (Non-metro)	−0.33* (−0.15)	−0.34* (−0.15)	−0.33* (−0.16)	−0.33* (−0.16)	−0.27 (−0.17)	−0.3 (−0.17)	−0.33 (−0.18)
Intercept (Metro)	−0.47 (−0.57)	−0.22 (−0.5)	−0.14 (−0.4)	−0.06 (−0.36)	−0.22 (−0.29)	−0.16 (−0.25)	−0.18 (−0.23)
EC (Non-metro)c	0.02*** (0.00)	0.02*** (0.00)	0.02*** (0.00)	0.02*** (0.00)	0.02*** (0.00)	0.02*** (0.00)	0.02*** (0.00)
EC (Metro)	0.00 (−0.01)	0.00 (−0.01)	0.01 (−0.01)	0.01** (0.00)	0.01** (0.00)	0.01*** (0.00)	0.02*** (0.00)
lambda	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)
Log income (Non-metro)d	0.04* (−0.02)	0.04* (−0.02)	0.04* (−0.02)	0.04* (−0.02)	0.03 (−0.02)	0.04* (−0.02)	0.04* (−0.02)
Log income (Metro)	0.04 (−0.06)	0.02 (−0.05)	0.01 (−0.04)	0.00 (−0.04)	0.02 (−0.03)	0.02 (−0.03)	0.02 (−0.03)
% NH White (Non-metro)e	0.03 (−0.02)	0.03 (−0.02)	0.03 (−0.02)	0.03 (−0.02)	0.03* (−0.02)	0.04* (−0.02)	0.04* (−0.02)
% NH White (Metro)	−0.08 (−0.05)	−0.02 (−0.04)	−0.02 (−0.03)	−0.01 (−0.03)	−0.01 (−0.03)	−0.01 (−0.02)	0.00 (−0.02)
% foreign born (Non-metro)f	0.15** (−0.05)	0.14** (−0.05)	0.15** (−0.05)	0.14** (−0.05)	0.14** (−0.05)	0.14** (−0.05)	0.14** (−0.05)
% foreign born (Metro)	−0.03 (−0.09)	0.06 (−0.08)	0.02 (−0.07)	0.06 (−0.07)	0.06 (−0.07)	0.07 (−0.07)	0.08 (−0.06)
Median age (Non-metro)g	−0.00*** (0.00)	−0.00*** (0.00)	−0.00*** (0.00)	−0.00*** (0.00)	−0.00*** (0.00)	−0.00*** (0.00)	−0.00*** (0.00)
Median age (Metro)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)	−0.00* (0.00)
Employment rate (Non-metro)h	0.07 (−0.04)	0.07 (−0.04)	0.06 (−0.04)	0.07 (−0.04)	0.06 (−0.04)	0.06 (−0.04)	0.05 (−0.04)
Employment rate (Metro)	0.18 (−0.16)	0.19 (−0.13)	0.12 (−0.1)	0.12 (−0.09)	0.13 (−0.08)	0.14* (−0.07)	0.14* (−0.06)
nobs	617	617	617	617	617	617	617
r.squared	0.57	0.57	0.57	0.57	0.57	0.57	0.57
AIC	−2,488	−2,484	−2,485	−2,482	−2,480	−2,480	−2,476
BIC	−2,417	−2,413	−2,414	−2,411	−2,409	−2,409	−2,405
deviance	0.55	0.55	0.55	0.55	0.55	0.55	0.56
logLik	1,260	1,258	1,259	1,257	1,256	1,256	1,254
nobs.1	617	617	617	617	617	617	617
	Aggregate Net In-Migration Rateb
	2.5	3.0	3.5	4.0 (original)	4.5	5.0	5.5
Intercept (Non-metro)	−18.27 (−23.7)	−16.3 (−24.05)	−12.63 (−24.87)	−13.08 (−25.35)	−15.25 (−26.73)	−33.89 (−27.4)	−29.21 (−28.32)
Intercept (Metro)	−39.37 (−88.5)	−2.91 (−78.71)	−5.49 (−62.89)	23.31 (−56.47)	18.46 (−46.31)	49.63 (−40.21)	25.17 (−36.74)
EC (Non-metro)c	3.64*** (−0.3)	3.60*** (−0.31)	3.49*** (−0.32)	3.38*** (−0.33)	3.32*** (−0.34)	3.07*** (−0.36)	2.92*** (−0.38)
EC (Metro)	0.17 (−1.18)	0.03 (−1.08)	1.33 (−0.8)	2.05** (−0.75)	2.10** (−0.68)	2.74*** (−0.61)	2.90*** (−0.56)
lambda	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)	0.61*** (−0.04)
Log income (Non-metro)d	−0.78 (−2.46)	−1.01 (−2.5)	−1.38 (−2.58)	−1.31 (−2.62)	−1.13 (−2.75)	0.71 (−2.82)	0.14 (−2.9)
Log income (Metro)	1.25 (−9.56)	−2.22 (−8.5)	−1.62 (−6.79)	−5.38 (−6.1)	−4.92 (−5.05)	−7.81 (−4.33)	−5.53 (−3.98)
% NH White (Non-metro)e	11.87*** (−2.44)	11.93*** (−2.46)	12.09*** (−2.49)	12.09*** (−2.51)	12.42*** (−2.55)	12.63*** (−2.61)	13.70*** (−2.8)
% NH White (Metro)	−1.83 (−8.51)	3.26 (−6.47)	5.88 (−5.11)	8.18 (−4.77)	7.42 (−4.29)	7.61* (−3.74)	8.52* (−3.34)
% foreign born (Non-metro)f	22.94** (−7.48)	21.16** (−7.58)	22.30** (−7.93)	20.09* (−8.11)	20.33* (−8.24)	20.12* (−8.34)	19.29* (−8.49)
% foreign born (Metro)	−7.94 (−14.65)	3.36 (−12.41)	−0.56 (−11.2)	7.09 (−10.87)	7.24 (−10.36)	9.29 (−10.04)	10.31 (−9.78)
Median age (Non-metro)g	0.33*** (−0.07)	0.34*** (−0.07)	0.34*** (−0.07)	0.33*** (−0.07)	0.34*** (−0.07)	0.36*** (−0.07)	0.37*** (−0.08)
Median age (Metro)	0.76* (−0.3)	0.55* (−0.24)	0.49** (−0.19)	0.57** (−0.18)	0.53*** (−0.16)	0.40** (−0.13)	0.39** (−0.12)
Employment rate (Non-metro)h	4.96 (−5.7)	5.13 (−5.74)	4.85 (−5.84)	5.16 (−5.92)	4.39 (−6.05)	1.92 (−6.24)	1.76 (−6.35)
Employment rate (Metro)	5.04 (−24.89)	10.59 (−20.75)	2.64 (−15.87)	8.26 (−14.37)	12.38 (−12.4)	16.95 (−10.15)	17.65 (−9.42)
nobs	615	615	615	615	615	615	615
r.squared	0.55	0.55	0.55	0.54	0.54	0.54	0.54
AIC	3,724	3,723	3,727	3,733	3,734	3,736	3,739
BIC	3,794	3,794	3,798	3,804	3,804	3,807	3,810
deviance	13,356	13,323	13,412	13,570	13,569	13,636	13,673
logLik	−1,846	−1,846	−1,847	−1,851	−1,851	−1,852	−1,853
nobs.1	615	615	615	615	615	615	615

Note: All continuous variables are mean-centered and scaled by 1 standard deviation, except Net In-Migration Rates which are reported as net migration rates per 100 residents.

^***p < 0.001

^**p < 0.01

^*p < 0.05.

Source:

^a.Census PEP 2010–2020

^b.Decennial Census 2010–2020 via APL

^c.Authors’ calculations using 2010 CBP via Eckert et al. (2020)

^d.-h.Census ACS 2006–2010.

10.2. Appendix B - Descriptive Analysis of Independent Variables

10.2.1. EC by rurality

EC (CZs)	Obs.	Mean	Median	Std. Dev.	Min.	Max.
Metropolitan	144	1.087	1.005	0.682	−0.528	3.024
Non-metropolitan	473	−0.324	−0.195	0.831	−3.374	1.192
Overall	617	0.005	0.045	0.997	−3.374	3.024

Source: EC data comes from authors’ calculations using 2010 CBP via Eckert et al. (2020)

10.2.2. Covariates

(CZs)	Obs.	Mean	Median	Std. Dev.	Min.	Max.
RUCC (rescaled, most rural = 1)a	617	0.57	0.63	0.27	0	1
Rurality ( non-metro = 1)a	617	0.77	1	0.42	0	1
Log income (median)b	617	10.2	10.2	0.13	9.75	10.7
% NH Whiteb	617	0.77	0.84	0.18	0.09	0.97
% Foreign bornb	617	0.053	0.037	0.054	0.004	0.479
Median ageb	617	41.5	41.2	4.13	26.1	59
Employment rateb	617	0.54	0.54	0.07	0.33	0.86

Source:

^a.USDA ERS RUCC with author rescaling

^b.Census ACS 2006–2010

10.3. Appendix C - Descriptive Analysis of 10-year Population Growth by Rurality

EC (CZs)	Obs.	Mean	Median	Std. Dev.	Min.	Max.
Metropolitan	473	1.087	1.005	0.682	−0.528	3.024
Non-metropolitan	144	−0.324	−0.195	0.831	−3.374	1.192
Overall	617	0.005	0.045	0.997	−3.374	3.024

Source: 10-year population growth data comes from Census PEP 2010–2020