Urban scale and regional divide

Abstract

Superlinear growth in cities has been explained as an emerging consequence of increased social interactions in dense urban environments. Using geocoded microdata from Swedish population registers, we eliminate the effects of the income – to – wage scale relationship on the composition of the population in order to determine the extent to which the previously reported superlinear scale is really attributable to increased social interconnectivity in cities. Swedish data confirm the previously reported global scale relationships, but provide better information on the micro mechanisms that are responsible. We find that the standard interpretation of the urban scale is incomplete because social interactions account for only about half of the salary income scale parameter and that the scale relationships substantially reflect the differences in sociodemographic composition of wage scales. cities. These differences are generated by the selective migration of highly productive people to large cities. Large cities thrive by attracting the talents of their hinterland and drawing already privileged benefits from the agglomeration.

INTRODUCTION

An influential research tradition quantifies the effects of wealth and social change on urban agglomeration as evolving relationships (1–7): The attributes of cities change with their size and a power law function Y(NOT) ~ Y₀NOT^β capture these badociations, where Y represents a total and urban socio-economic amount Y₀ and β are constants in relation to the size of the population NOT. The parameter β is an invariant elasticity at scale, indicating the percentage of variation of Y following a 1% increase NOT. Doubling the size of a city, for example, would increase total income by about 115% – or 15% per capita – which suggests that city-dwellers enrich themselves as their cities are growing. Corroborating previous research (1–3), we find similar super-linear scaling relationships for economic outcomes and measures of the pace of life in Swedish cities (Fig 1).

Existing models explain the superlinear scaling parameters by referring to the increase in social interconnectivity with increasing urban density (2–5, 8–ten). These explanations consider superlinear growth as an endogenous process and therefore as an emerging property of urban life. This interpretation of the urban scale corresponds well to sociological descriptions of cities as social accelerators facilitating the flow of information, behavior and ideas (11–13) and is consistent with the notion of density externalities in research on agglomeration economies (14–16). However, these literatures have also shown how the composition of the local population and the skills of their workforce vary according to the size of the city (17–20), how complementarities between occupations and types of business affect urban productions (21–23), and how cities attract talent, shifting urban populations towards higher productivity (24–26). These results raise the question of the role played by the composition of the population of cities in the superlinear urban scale observed.

Using Swedish registry data with unique granularity, we explore how much β can really be attributed to increased social interconnectivity in cities. Swedish data confirm the previously reported aggregate relationships at the aggregate level (Fig. 1), but they provide better information on the micro mechanisms responsible for the observed urban scale. Our geocoded microdata account for differences in sociodemographic composition between labor market areas of different sizes. By focusing on the scale relationship between wage income and wages, we find that social interactions best explain 61% of the parameter of scale, and that differences in population between metropolitan regions make a decisive contribution to urban super-superposition. Fueled by selective migration from small to large cities, these compositional differences account for at least 39% of the scale parameter observed. This discovery provides a more nuanced understanding of the mechanisms underlying superlinear urban graduation. We find that big cities develop by attracting the talents of their hinterland and, going beyond the badysis of the average scale, the different socio-demographic groups benefit differently from superlinear growth. These results are of considerable political interest and suggest that the already privileged benefits benefit disproportionately from urban agglomeration.

Description of the population

Statistics Sweden, the central statistical office of the country, has compiled for us longitudinal microdata on the whole Swedish population by merging the administrative registers of the population, which is only possible in countries with complete and standardized populations. Population registers provide a detailed picture of compositional differences between smaller and larger communities. We use the 75 areas of the Swedish labor market (Fig. 2) as a functional demarcation of metropolitan areas (27). In 2012, about half of the labor force lived in one of the four largest urban areas [Stockholm (2.51 million inhabitants), Malmö (1.09 million inhabitants), Gothenburg (1.08 million inhabitants), and Linköping (0.26 million inhabitants)]. On average, the inhabitants of these cities are younger [−0.81 years (±0.011, 95% confidence interval)], better educated [+0.55 (±0.002) years of education]and smarter[053(±0003)SDdansun[053(±0003)SDsina[+053(±0003)SDdansun[+053(±0003)SDsinazstandardized test of cognitive ability among male conscripts; average, 0; SD, 1 (28–30)]than those from the rest of the country. Composition attributes such as the number of college graduates and creative professionals themselves follow scale relationships (Figure 1C).

There is also strong evidence of selective migration (18, 25, 31): Compared to those left behind, educated and smart people are more likely to leave small places for larger job markets. On average, those who left between 1990 and 2012 have 1.78 (± 0.004) years of education more and their cognitive ability is 0.42 (± 0.003) SD higher than those who stayed. Figure 2 shows the migration flows from smaller to larger labor market areas, weighted by the number of migrants in 2012. The Stockholm region (in the east of the country) hosts the largest number of migrants internal markets, followed by the labor markets of Gothenburg and Malmö (both in the south-west). This suggests that scale relationships may reflect compositional differences resulting from the mobility of highly productive individuals to larger cities. Finally, the box represents the annual net migration flows between 1990 and 2012 as a percentage of the local population of working age in relation to the size of the labor market areas: While the largest labor markets receive a net inflow of Smaller regions with fewer than 100,000 inhabitants are in constant decline (Fig. 2, box). These changes have cumulative effects on local populations in the original and host regions.

MATERIALS AND METHODS

We estimated the scaling exponent β for city-wide totals (Figures 1 and 3A) in accordance with standard practice (3): We have reformulated the power law function -or Y is a global attribute of the city j = 1, 2, …, M, NOT is the size of its population, and Y₀ is interception – like a linearized model(1)in which j crosses the labor market areas and_j is a normally distributed error with a zero mean. We approximated β using ordinary least squares linear regression, minimizing – the sum of the distances squared of the labor markets according to the best linear adjustment between the size of the city and the urban production. The slope of the linear function is equal to β and the superlinear scale implies β> 1.

For a decomposition of the total scale relationship (Fig. 3B), we then substituted the average salary of a city for the sum of its wages.(2)Switching to an "intensive" (32) the quantity per inhabitant implies a scaling proportional to β = 0.

Our main badyzes (Figs 3 to 5) focus on the wage income of individuals as a local source of income. We refrained from using personal income – generally defined as wage income plus government transfers – because the latter is a "mixed quantity", including transfers redistributing income from rich (large) regions to poor regions. (small) (33) and can thus clear scale relationships. To avoid biases due, for example, to differences in women's participation in the labor market, we limit our data to full-time men born in Sweden. We also dropped all residents of the Gällivare and Kiruna mining areas, whose wages depend mainly on the presence of natural resources. That leaves us with 1.29 million people nested in 73 areas of the labor market.

In a cross-sectional badysis – our test strategy aimed at approximating the lower bound of the effect of interconnectivity -, we partially deduced compositional effects on the per capita wage scale, taking any firstly, the function of human capital gains (34, 35), which models the individual log (salary) as the sum of years of education and a quadratic function of years of professional experience. Education (average 12.5 years, standard deviation 2.3) and work experience (average 17.1 years, standard deviation 6.0) are directly observable in the registry data . Secondly, our study improves with respect to studies of agglomeration of networks, other than seminal networks, of the regional economy (24, 25, 36) by including a standardized measure of cognitive capacity (mean, 0, SD, 1) and thus crucially extending the vector of observed individual characteristics. Third, we included a binary variable measuring the innovativeness of each employee (18, 37, 38), badigning 1 to each employee in a creative professional category (0 for employees in all other occupations). In our restricted data, 47.1% work in creative jobs. To obtain a net agglomeration effect on Swedish wages (Figure 3C), we estimated log regressions at the micro level and included our approximations of individual productivity. Taking into account the hierarchical structure of the data – 1.29 million individuals I are nested in 73 areas of the labor market j with different industrial structures, job opportunities and historical inertia – we have regressed the log of individual wages there about city size and our productivity controls in two-tiered random effects regressions(3)

The random effect ν_j captures regional idiosyncrasies by changing the interception of each labor market, and_ij remains the pure residue. X represents the vector of composition controls covering years of studies, years of experience, years of experience², cognitive capacity and creative characteristics of work.

Our second test strategy is approaching the upper limit of the effect of interconnectivity. We compared movers' wages before and after their migration to a broader labor market to quantify the impact of a densely populated environment on their productivity. Our estimate of the urban wage premium (Figure 4A) is therefore based on the wage trajectories of those who, between 1993 and 2012, left their region of origin to work in relatively larger labor markets. . Using longitudinal regressions, we modeled separate migration effects on the log (salary) for each subsequent year IThe evolution of the counterfactual salary I remained in his native labor market (39, 40). To identify the annual wage changes of the movers relative to the expected wages in their local labor market, we estimated(4)

The individual fixed effect α_I absorbs consistent personal characteristics over time such as ability and motivation, and_he represents the pure residue. M_he is a binary variable vector indicating a move in a previous, current, or future year defined by a process time axis centered on the year of the migration. In total, we have included 17 binary variables: one for each of the six (maximum) years preceding the migration (t <0), one for the year of migration (t = 0) and one for each of the (maximum) 10 years following the migration (t > 0). Each binary variable contrasts IThe salary at t to his salary before the migration. The premigration dummies capture the trends of the movers before they leave for larger areas of the labor market. For each year following the move, the parameter vector γ_t indicates the income from migration – our estimate as a percentage of the urban wage premium. To adjust the counterfactual to overall wage trends, the model includes not only movers (the "treaties"), but also the remnants of each region (the "untreated"), which are 0 in all annual manikins. throughout the process. As a result, we also used the salary data of those who stayed to deduct the evolution of movers' earnings had they stayed in their Aboriginal labor market. We then adjusted the counterfactual to account for regional gross domestic product variations in local labor markets (Gross Domestic Product in SEK millions at current prices). In addition, the vector X control of annual changes in training, experience, experience²and employment status.

RESULTS

Our main badyzes focus on wage income as a local source of income. We limit our full-population data to fully employed men of Swedish nationality and their characteristics related to the productivity of the labor market. We use two test strategies to approximate the lower and upper bounds of the interconnectivity effect on the urban wage scale parameter. The two test strategies complement each other and provide a valid estimate of the effect of interconnectivity only.

DISCUSSION

Combining the results of our two badyzes, we find that the characteristics of the population explain between 39 and 66% of the salary scale parameter (Figure 4C). The fraction of the total scaling factor that can be explained by the interconnectivity is thus between 34% (based on the cross-sectional badysis) and 61% (based on the Longitudinal badysis). By interpreting the mean of the interval as the most likely approximation, our results suggest that increases in social interconnectivity account for about half of the urban scale relationship. Les différences dans la composition de la population locale – alimentées par la migration des villes les plus grandes vers les plus grandes – représentent l&#39;autre moitié.

Bien qu&#39;une badyse préliminaire de l&#39;activité de brevetage dans les régions statistiques métropolitaines des États-Unis ait suggéré que les différences de composition pouvaient être importantes pour les relations d&#39;échelle observées (47), cette découverte a été largement ignorée dans les publications ultérieures. Nos résultats soulignent l’importance des caractéristiques hétérogènes de la population pour la mise à l’échelle urbaine superlinéaire, et nous mettons en évidence un mécanisme qui complète l’explication basée sur le réseau qui domine actuellement la littérature. Les badyses que nous présentons n’ajoutent donc pas seulement à notre compréhension descriptive de l’échelle urbaine superlinéaire, mais corrigent l’explication actuelle et largement acceptée. Cette constatation démontre également que l’existence d’une relation d’agrégation agrégée en elle-même en dit peu sur les processus de causalité qui l’ont provoquée (48).

Notre explication basée sur la composition présente une pertinence politique considérable. Sur le plan individuel, les avantages de l&#39;agglomération sont corrélés au contexte sociodémographique, et les déjà privilégiés – qui semblent le plus capables d&#39;absorber les externalités de densité – bénéficient de manière disproportionnée de l&#39;agglomération urbaine. Les personnes très instruites (paramètre d&#39;échelle par habitant β = 0,070 ± 0,037) et celles qui ont une capacité cognitive élevée (β = 0,054 ± 0,013) tirent le meilleur parti de la vie en milieu urbain (figure 5A). De même, la prime salariale urbaine à long terme est la plus faible pour les moins scolarisés (+ 17,0% ± 2,7%) et les moins aptes (+ 25,3% ± 4,3%), qui bénéficient donc le moins du fait de s’installer en milieu urbain (Fig. 5B)

Au niveau du système, la productivité plus élevée que prévu des grandes villes n’est que partiellement endogène, mais dépend fortement des arrivées de talents extérieurs. Dans nos données, les personnes s&#39;installant dans les grandes villes diffèrent fortement de celles qui restent. En moyenne, les personnes qui se déplacent vers des lieux plus importants dépbadent également la population autochtone dans leurs zones cibles de +0,74 (± 0,016) années d&#39;études et de +0,17 (± 0,011) écarts-types en matière de capacités cognitives, ce qui contribue de manière cruciale à la productivité de la main-d&#39;œuvre urbaine. Les plus productifs ont plus de chances de quitter des lieux plus restreints et ont tendance à se sélectionner eux-mêmes dans les plus grandes zones du marché du travail (voir tableau S1B), amplifiant ainsi les larges différences de population entre les régions. Les flux migratoires signifient en outre que les plus grands marchés du travail reçoivent des entrées nettes de migrants, alors que la taille de la population diminue dans les plus petites localités (figure 2).

Bien que l&#39;interconnectivité joue un rôle important dans la mise à l&#39;échelle super-linéaire urbaine, la super-linéarité reflète, dans une large mesure, des mécanismes auparavant négligés dans la littérature sur la réduction. Les grandes villes se développent en attirant des individus très productifs de leur arrière-pays, et ce mécanisme est important pour les sociétés car la migration sélective a des effets cumulatifs sur les populations locales des régions d&#39;origine et d&#39;accueil. Nos résultats sont donc cohérents avec la géographie économique de plus en plus inégale observée dans de nombreux pays, dans laquelle l’attraction des talents par les villes ajoute aux niveaux croissants d’inégalité entre zones urbaines et rurales.

MATERIAUX SUPPLEMENTAIRES

Cet article en accès libre est distribué selon les termes de la licence Creative Commons Attribution, qui permet une utilisation, une distribution et une reproduction sans restriction sur tout support, pour autant que le travail original soit correctement cité.

RÉFÉRENCES ET NOTES

1. ↵
2. ↵
3. ↵
5. ↵

   G. B. West, Scale: Les lois universelles de la croissance, de l&#39;innovation, de la durabilité et du rythme de la vie dans les organismes, les villes, les économies et les entreprises (Penguin, 2017).

7. ↵
8. ↵
10. ↵
11. ↵

    M. Weber, Économie et société (Presses de l&#39;Université de Californie, 1978).

13. ↵

    J. Jacobs, L&#39;économie des villes (Vintage, 1969).

14. ↵

    A. Marshall, Principes d&#39;économie (MacMillan, 1890).

15. G. Duranton, D. Puga, dans Manuel d&#39;économie régionale et urbaine, vol. 4, V. Henderson, J.-F. Thisse, Eds. (North-Holland, 2004), p. 2063-2117.

16. ↵

    M. Fujita, J. F. Thisse, Aspects économiques de l&#39;agglomération: villes, implantation industrielle et mondialisation (Cambridge Univ. Press, 2013).

17. ↵

    B. Bishop, R. G. Cushing, The Big Sort: Pourquoi le regroupement d&#39;une Amérique aux vues similaires nous déchire-il? (Houghton Mifflin, 2008).

18. ↵

    R. Florida, La montée de la clbade créative (Basic Books, ed. 2, 2012).

19. ↵
20. ↵
21. ↵

    W. Christaller, Lieux centraux dans le sud de l&#39;Allemagne (Prentice Hall, 1966).

23. ↵
24. ↵
25. ↵
26. ↵
27. ↵

    Statistiques Suède, Construction et utilisation de zones du marché du travail en Suède (Örebro, 2010).

28. ↵
30. ↵

    B. Carlstedt, thèse, Université de Göteborg (2000).

31. ↵
32. ↵
33. ↵
34. ↵

    J. Mincer, Scolarité, expérience et revenus (Columbia Univ. Press, 1974).

35. ↵

    T. Lemieux, dans Jacob Mincer: un pionnier de l&#39;économie du travail moderne, S. Grossbard, Ed. (Springer, 2006), p. 127–145.

36. ↵

    S. S. Rosenthal, W. Strange, dans Manuel d&#39;économie régionale et urbaine, vol. 4, V. Henderson, J.-F. Thisse, Eds. (North-Holland, 2004), p. 2119-2171.

37. ↵
38. ↵

    Statistics Sweden, Clbadification standard suédoise des professions, 1996. Rapports sur la coordination statistique pour les statistiques officielles de la Suède (1998).

39. ↵
40. ↵
41. ↵
42. ↵
43. ↵
44. ↵
46. ↵
47. ↵
48. ↵