Ageing places: convergence and the role of the foreign population

ABSTRACT

Using data for over four decades, we examine the ageing of Swedish municipalities and if there has been convergence or divergence across time and space, where we differentiate across the urban–rural hierarchy. As migration is claimed to be a moderating factor in the ageing of places, we assess how the share of immigrants relates to ageing patterns. Our findings show that the share of older individuals increases in more peripheral localities and that there has been convergence across municipalities. However, the share of foreign population is negatively related to the ageing profile of a place.

KEYWORDS:

• ageing
• location
• international migration
• convergence

JEL:

• J10
• J14
• R11
• R12

1. INTRODUCTION

By 2030, 25% of the European Union’s (EU) population will be over the age of 65. National, regional and local governments in the EU are increasingly concerned about this transformation (Bloom et al., 2003; Tynkkynen et al., 2022), as it creates social and economic challenges for current and future working-age generations (Azzali et al., 2022). Changes in regional population-age structures that diminish the relative or absolute size of the working-age population will have negative economic consequences in the affected regions and localities since they tend to increase the dependency ration (Thalassinos et al., 2019; Van der Gaag & de Beer, 2015). An increase in the 65+ population also leads to increased social welfare and healthcare costs for the elderly (Kluge et al., 2019).

The responsibility for financing these costs is often divided among the central, regional and local governments, so different regions and localities are affected differently by their ageing population. Overall, however, the ageing process leads to pressures especially on the budgets of regional and municipal councils that are responsible for healthcare and care of the elderly, respectively. To fund the expected increased expenditures, regions and municipalities will need to increase their taxes, cut expenditures in other policy areas, and/or get larger subsidies from the national budget. These expenses in public consumption and investment may slow down economic growth (Teixeira et al., 2017), especially if labour is sucked from the private sector into public sector activities at a time when the active labour force is growing very slowly and will eventually decline.

It is thus not surprising that dealing with population ageing has become a main topic of debate in demographics as well as economics (Pani-Harreman et al., 2021; Seco Matos et al., 2018). An increasing number of studies document how ageing influences population structures in regions and localities (Faggian et al., 2017; Gutiérrez Posada et al., 2018). Population ageing in every country is, from a spatial perspective, a very heterogenous process with substantial differences in timing and speed among regions and cities (Gutiérrez Posada et al., 2018; Kashnitsky et al., 2017). Kashnitsky et al. (2021) find age convergence, indicating that regions become more demographically similar over time. Other authors claim the existence of a spatial divergence process (Fratesi & Riggi, 2007; Gregory & Patuelli, 2015; Kanbur & Rapoport, 2005). However, if the ageing process within a country shows increasingly large divergent trends over time, indicating locations that have a growing younger population, regional disparities will arise, leading to ageing divergence. Thus, the study of spatial convergence/divergence in population age structures is of great interest to policymakers at the regional and local levels as it provides them with a better foundation for medium- and long-term planning and decision-making regarding population growth, changes in age structure, and the potential need for future investments in housing, healthcare facilities, homes for the elderly and various other infrastructure. In addition, in countries with a rapidly ageing population, migration – especially international migration – can become a component of such population changes (Czaika & de Haas, 2014; Findlay & Wahba, 2013). However, this raises a fundamental question: Is the immigration of younger generations the right solution to manage an ageing and declining population of native workers (Bouvier, 2001; Cuaresma et al., 2015)?

To assess whether convergence or divergence is occurring, this paper analyses the ageing process in Sweden during the period 1970–2020 and the role that international migration played in the ageing profile of places. The results are of substantial interest from both a scientific and a political point of view, since there exist political goals in Sweden concerning overall spatial equality, though at the same time, there is a clear risk that spatial demographic trends often tend to increase spatial inequality. We use Sweden as the focus of this study because it is a country that experienced the most rapid ageing during the twentieth century when individuals above the age of 65 doubled (Bengtsson & Scott, 2011). At the same time, Sweden’s share of individuals above the age of 80 did not change between 2001 and 2021, whereas it increased in all other EU states during the same period.¹ Thus, Sweden has experienced a complex ageing process, making it a good example for research.

This paper contributes to the literature on the demographic structure of locations, specifically on the spatial dynamics of population structures. The population structure in Swedish regions and localities is, besides nativity and mortality, determined by both internal migration (Rees et al., 2013) and international immigration (cf. Czaika & de Haas, 2014). It is important to emphasise that substantial differences exist in how these processes affect regions and localities across the spectrum from central urban areas to peripheral rural areas (Gutiérrez Posada et al., 2018). Thus, this paper contributes to knowledge about the two major demographic transitions in developed countries today: ageing and urbanisation (Kashnitsky et al., 2021).

Our findings show that, overall, Swedish municipalities have experienced convergence whereby they have become more similar in their ageing structure to those with an initial position of having older populations and thus less growth in ageing over time. There are, however, differences across the urban–rural continuum where the share of older individuals increased in smaller and more peripheral localities. Our results further show that the share of foreign populations within a municipality is negatively related to the ageing profile of a place and that a large inflow of immigrants leads to ageing divergence across municipalities.

2. THE CONVERGENCE AND DIVERGENCE DYNAMICS OF POPULATION STRUCTURES

In closed regions of a country, the dynamics of population structures are governed by the factors determining nativity and mortality. Due to regional variations in these factors, we get different patterns of convergence and divergence over time. But when we enlarge the scope to include interregional interaction in terms of migration between regions, we introduce a new element that can influence the patterns of convergence or divergence. In general, migration processes can be expected to generate diverging patterns of ageing (Smailes et al., 2014) since the propensity to migrate between regions varies among age cohorts according to their life stage (Bloom et al., 2003). While urban (growth) regions tend to attract the young, particularly highly educated people of working age, which rejuvenates their populations, more rural regions and localities usually observe that their population ages over time (Fratesi & Riggi, 2007). However, a factor that can mitigate this spatial age divergence effect is that people living in more urban regions have on average a longer life expectancy than people in more rural regions (Kibele, 2014), which would tend to lead slowly to convergence. Nevertheless, a basic assumption is that population migration leads to a divergence of population age structures in regions.

Furthermore, the factors that motivate migration to distinct types of locations at various stages of the household life cycle are nuanced (Whilser et al., 2008). Early life-cycle migration among adults is determined by education and jobs, and the willingness to migrate between regions is relatively high. Later, when couples start a family, the willingness to migrate to another region declines. After retirement, migration restrictions are reduced for many people as equity in homes, particularly in high-cost agglomerations, can be cashed in to finance migration to high-amenity localities (Mueser & Graves, 1995).

Age- and education-selective migration is associated with spatial spillovers which, together with the migration effects on nativity and mortality, tend to increase the dispersion between regions and localities in terms of population age structure, education level, and spatial economic growth. In general, higher growth and the related higher income are attractors for migrants. This dynamic migration process tends to generate a ‘lock-in effect’ for ageing regions and localities, making it exceedingly difficult for them to reverse the processes of migration and ageing. As a result, there is a trend towards a pattern of specialisation among localities divided between either ‘working’ and ‘rentier’ localities (Yong Chen & Rosenthal, 2008), that is, localities with a younger versus those with an older population, indicating that age convergence is not a universally general pattern.

Gaigné and Thisse (2009) formally modelled the potential role of an ageing population in altering the size distribution and specialisation patterns of settlements. The results of the model are conditioned by the level of mobility among the elderly. In general, they find that while there will be some segregation of older individuals and workers, the system of locations will continue as it is. This finding was later confirmed by, among others, Brown et al. (2011) and Philip and MacLeod (2018). The ‘working localities’ maintain primacy in the sense that they will be larger than the ‘rentier localities’ dominated by the non-working older population. In spatial terms of ageing, this model points in the direction of no large-scale ageing convergence but instead a specialisation among localities for the young and localities for the old, that is, a population divergence. Thus, there is a sort of ‘path dependence’ where growing and declining regions are locked into different paths that are hard to break away from, as found by previous studies (MacKinnon et al., 2019; Martin & Sunley, 2006). These demographic processes are exceedingly difficult to be influenced by public policy measures.

2.1. International migration and changes in spatial population structures

A basic assumption in the discussion above is that the migration of younger adults might contribute to the lock-in effect of population structures in more and less urbanised localities, that is, leading to a divergence of population age structures in the system of regions. However, demographic shocks coming from large-scale migration and asylum-seeking immigrants may shake up and possibly even reverse this path dependence, since immigrants tend, on average, to be younger than the population at large.² But overall, the precise effects of immigration on the population age structures in regions depend on how they are distributed over localities in a country.

A critical question in this discussion is how the internal migration of international immigrants in Sweden affects the system of localities, specifically how its influence on age convergence/divergence in the regions and localities. In principle, there exist two models governing the spatial location of asylum immigrants: either the immigrants themselves determine where to locate, or the authorities determine where they can live. In some time periods, a mixed model has been used in Sweden, whereby immigrants with family or friends already in Sweden were allowed to live with them, and the rest were allocated to various locations in the country having vacant housing. In many of the more rural and often ‘greying’ municipalities, the politicians have been quite positive about receiving asylum immigrants, since they see them as helping to solve problems they had with their ageing population, a shrinking labour force, depopulation, and an outmigration of younger people. As a rational solution, they perceive it easier to influence local demographic development through in-migration of asylum immigrants than, obviously, to influence the rates of nativity and mortality. However, we must also acknowledge that over time, there is often a substantial secondary migration where immigrants move again because they are unsatisfied with their initial location or because they get a job at another location. Thus, for regions with an ageing and declining population, it is not enough to attract immigrants; they must also take measures to retain them (Derwing & Krahn, 2008). This secondary migration involves to a large extent moving to the largest cities. This is a general European trend and points to the importance of agglomeration economies and associated amenities, as well as to the importance of higher wages and lower unemployment rates for attracting immigrants to larger cities (Viñuela, 2022). If immigrants concentrate in larger cities, this may actually sharpen prevailing international differences (Golini & Di Bartolomeo, 2009).³

Voluntary initial settlement patterns of international immigrants have been analysed in many studies (Åslund, 2005) based upon the traditional household location models (Greenwood, 1997), where labour market conditions and the existence of ethnic networks are the most crucial factors. From an historical perspective, we know that immigrants often end up in large urban places where there is a large supply of vacant jobs and various urban amenities (Buch et al., 2014). However, in recent decades, we find new residential patterns among immigrants, including a higher share of especially non-European migrants living in non-metropolitan cities and towns (Malmberg et al., 2018). Over time though, these numbers tend to shrink since there is an out-migration trend towards the largest cities where most immigrants mainly target low-skilled jobs in the service sector (Rauhut, 2007).

The share of immigrants in some places can, in principle, reach a sufficiently high enough value to alter the existing systems of localities (Davezies, 2008) and influence the patterns of spatial age convergence/divergence. However, in cases where asylum migrants are free to settle where they like, we expect the lion’s share to choose the larger regions, thereby contributing to increased population structure divergence between regions. On the other hand, in cases where the central government either totally or partly decides where asylum migrants will settle, there is the potential for population structure convergence between regions. Due to recent large-scale asylum immigration to Sweden, we investigate in our analysis the possibility of demographic structures in localities being unstable enough that random fluctuations due to immigration play a significant role.

3. AN AGEING CONVERGENCE/DIVERGENCE MODEL

One standard approach on the convergence/divergence question goes back to Baumol (1986) and focuses on establishing a relationship between relative changes in a specific variable at different localities over time from its initial value (Barro & Sala-i-Martin, 2003). In this approach, the conclusion is inferred from the value of the β-coefficient. A negative β-coefficient means that localities have been experiencing a convergence process. Locations with initially lower levels grow faster than localities with initially higher levels, that is, localities with lower values are catching up with the leaders. Such a development is known as β-convergence. In our empirical exercise, we relate the change in the demographic profile in a region to its initial position in each location, as illustrated in equation (1): (1) $\log \left(\frac{y_i^{\tau }}{y_i^0}\right)={\alpha }_0+{\beta }_0\log (y_i^0)+{\epsilon }_i$(1)

where $y_i^0$ is the share of people over 65 or average age in location i in the initial year and $y_i^{\tau }$ is the share of people over 65 or average age in location i in the final year. $\log \left(\frac{y_i^{\tau }}{y_i^0}\right)$ is the logarithm of the growth of the age variable in location i from year 0 to year τ. The estimated parameter in equation 1 is known as the unconditional β-convergence parameter. Usually, equation (1) is complemented with a number of control variables representing various characteristics $x_i^n$ of each location. These characteristics are expected to influence the spatial concentration/deconcentration of older people by functioning as attractors or repellers, following Gutiérrez Posada et al. (2018). When such control variables are included, the conditional β-convergence is estimated as presented in equation (2): (2) $\log \left(\frac{y_i^{\tau }}{y_i^0}\right)={\alpha }_0+{\beta }_0\log (y_i^0)+\sum _{n=1}^N{\gamma }^n{x}_i^n+{\epsilon }_i$(2)

In a second step, we apply a σ-convergence approach and interpret convergence between localities as a reduction in the dispersion (Quah, 1993a). Sigma-convergence is a stricter convergence test, though closely related to β-convergence. The connotation of σ-convergence implies that the variance of a group of indicators gradually decreases over time, which represents a weakening of the fluctuation of indicators for this group of economic samples. Thus, β-convergence is a necessary but not sufficient condition for σ-convergence (Sala-i-Martin, 1996b). The occurrence of σ-convergence also depends on the influence of the variance of the disturbance term (Young, Higgins, & Levy, 2008). This implies that β-convergence may take place due to random fluctuations even if there is no reduction in the variance according to the σ-convergence value. As a measure of σ-convergence, we apply the coefficient of variation, as this is the most common approach to measure it (Monfort, 2008).⁴ Both measures are needed to fully confirm the existence of convergence.⁵

4. AGEING AND THE CARE OF THE ELDERLY IN SWEDEN

Sweden, like other countries in Europe, is an ageing nation, with the number of very old people especially increasing. Between 1990 and 2014, the Swedish population increased by 10.7%, but during the same period the number of women 75+ increased by 14.7% and the number of men 75+ increased by 31.5%.⁶ Interestingly, during the period 2009–20 the share of people 65+ increased from 17.8% to 20.0%.⁷ This ageing process has taken place even though Sweden has experienced substantial population growth over the last 30 years; between 1990 and 2020, the Swedish population increased by 20.2%. In Sweden, the average person retires at the age of 63.5 years. At this age, the average man is expected to live another 19 years and the average women another 22 years.

The responsibility for services aimed at the elderly is divided between Sweden’s regions and municipalities. Healthcare for the elderly is the responsibility of the 21 regions. The 290 municipalities organise and finance their social care, which consists of homecare services and care homes. Both the regions and the municipalities are financed by a flat income tax that varies between regions and particularly between municipalities. In addition, both regions and municipalities receive subsidies from the national state.

When people retire and go from income-based work to a pension income, their income typically drops around 30%. The consequences of this are far reaching, particularly in a country like Sweden where both regions and municipalities finance their activities mainly using a flat income tax on personal income. An ageing population thus not only implies a cost increase for regions and municipalities, but also a severe decrease in their tax base as the working age population declines with a concomitant loss of working income compared with pension income.

4.1. Immigration and immigration policy in Sweden, 1970–2000

During the 1950s and 1960s, immigration to Sweden was for the most part highly regulated labour immigration for people living in countries other than the Nordic nations. In the late 1960s, however, regulated labour immigration was halted. Beginning in the 1970s, Sweden started to accept United Nations (UN) quota-based refugees. From the end of 1970s, immigration became dominated by refugees. Between 1970 and 2002, Sweden has employed different settlement policies for all refugees it accepted. Between the mid-1980s and mid-1990s, the spatial immigration policy in Sweden had as its main goal reducing ethnic segregation among refugees and asylum seekers. A new distributional policy – ‘the whole Sweden policy’ – was introduced, which directed that refugees should be distributed randomly over all Swedish municipalities. Refugees were assigned to an initial location to alleviate administrative burdens in the larger cities and to improve integration into Swedish society (Andersson & Solid, 2003). This geographical allocation policy was administered by The Swedish Migration Agency and was mandatory for all refuges except those who came to reunite with their families already in Sweden. However, localised housing shortages allowed certain municipalities to refuse receiving immigrants according to plans. After being assigned to an initial location, refugees could move freely within Sweden, and many immigrants later moved to the country’s main metropolitan areas.

Starting in 1994, new legislation was implemented where refugees were free to choose where to reside right from their entry. They could choose to stay at an asylum centre run by The Swedish Migration Agency or find a place themselves. Hence, depending on the locational choices immigrants made in different years, Sweden had varying settlement patterns (Andersson et al., 2010). Later, the housing allowance was taken away from those refugees who chose to find their own housing. This change did not reduce the share of refugees who chose to find their own housing in one of the metropolitan areas (Esaiasson & Sohlberg, 2018). The migration patterns of refugees varied a lot depending upon, among other things, home country and even home region in the home country. This was due to a significant heterogeneity among refugees coming from the same country in terms of their individual networks (Aradhya et al., 2017).

5. SPATIAL AGEING DYNAMICS IN SWEDEN

Our empirical analysis is conducted at the municipal level using annual data. The data have restricted public access and originates from Statistics Sweden covering the period 1968–2020. Starting at the general level and examining the ageing development in Sweden overall, Figure 1 illustrates the average age across municipalities and the share of inhabitants over the age of 65. The figure presents a striking development where we see a large demographic change towards older inhabitants, a development common among the majority of developed countries in the EU (Grundy & Murphy, 2017; Rechel et al., 2013).

Figure 1. Average age in Sweden (a) and share of the population over the age of 65 years (b), 1968–2020.

the figures show a line with an increasing trend but has a S-curved shape. The line representing the share of 65+ has a more pronounced S-shape where the line between 1990 and 2005 is more like a horizontal line.

Interesting to note are ‘S’-like curves, indicating a brief slowing down of the ageing process in the 1990s. This change is more pronounced using the share of inhabitants over the age of 65. Two possible reasons might explain this. The first relates to the large inflow of younger immigrants and especially refugees from the Balkans and the Middle East that started in the mid-1990s, skewing the ageing ratios a small amount. However, this must be only part of the story, given that during the 2010s Sweden also experienced large inflows of immigrants, but we do not see the same change take shape for those years. Another explanation might lie in the high birth rates that happened from the end of the 1980s through to the mid-1990s⁸ compared with the low birth rates during the 1920s and beginning of 1930s (Myrdal & Myrdal, 1935) which created a shortfall in 80 and 90 year olds.

To understand the spatial ageing patterns, Figure 2 presents the share of the population above the age of 65 in 1970 and 2020, and the growth in the share of the population above 65 across the years for all Swedish municipalities. The maps show a significant difference between the municipality with the highest share of 65+ and the municipality with the lowest share. The share of older individuals in different municipalities tends to be related to the degree of urbanisation, where older individuals usually concentrate in more rural locations, as found in previous studies (Burholt & Dobbs, 2012; Skinner & Winterton, 2017).⁹ The spatial differences in terms of the share of older people are so large that we can talk about a spatial disequilibrium. This contrast emphasises the importance of analysing processes of the convergence/divergence of population cohorts at a fine spatial scale.

Figure 2. Share of people aged 65+ years in Swedish municipalities in 1970 (a) and 2020 (b), and percentage change from 1970 to 2020 (c).

the figures show Sweden where all municipalities are identified and shaded in different colors depending on the ageing profile. Darker color indicates a higher age.

A striking observation from Figure 2 is the strong spatial distribution regarding the demographic structures. In 1970, the municipalities with the higher share of older individuals were in the most rural parts of Sweden, both in different parts of the country. Typical for the municipalities with the highest share in 2020 is that they are very sparsely populated and located in the northern part of Sweden, where the growth of older individuals has been the highest. There are also pockets of rural municipalities in southern Sweden with a very high share of older people in their population. The municipalities with the lowest share of elderly are, on the other hand, in the large city regions around Stockholm, Gothenburg and Malmö. The change in the ageing structure is also visible through the change in the level of the share of 65+ between 1970 and 2020, where the maximum increases substantial in the levels for the four quantiles in 2020.

To further describe, compare and visually present changes in the demographic profile of all localities over time, we calculate for various years the cumulative densities and plot the distributions of localities arranged in ascending orders. Figure 3 shows the cumulative distribution function using the average age in the Swedish municipalities at 10-year intervals for the period 1970–2020. The age profile is clearly moving towards a higher average age and where the spread across municipalities diminishes over time, similar to the finding by Yuan Chen et al. (2019).

Figure 3. Cumulative distribution functions of the average age in Swedish municipalities, 1970–2020.

the figure constitutes of six parallel lines where all lines have a S-shape. When moving from left to right the lines becomes more straight with shorter tails.

6. EMPIRICAL FINDINGS

We perform several different estimations to capture the ageing structure in Sweden. The variables used in the analysis are presented in Table 1 along with summary statistics. To capture the age profile, we use the average age in the municipality as well as the share of inhabitants over the age of 65 and their growth across different years. We also include several control variables. First, we include a variable that captures the geographical position of the municipality, size of the own municipality and size of the surrounding locations. We use an accessibility measure following Johansson et al. (2002, 2003) where a municipality’s accessibility is based on the own size, size of locations within the same region, and the size of all other locations in Sweden, weighted by the time distance by car (Accessibility, population). The use of these accessibility measures has several advantages such as enabling us to take into account the infrastructure system, mitigate issues with spatial autocorrelation (Andersson & Gråsjö, 2009), allow us to link to the functional and the spatial components of an urban system (Bertuglia & Occelli, 2000), and accounts for the cost of overcoming space in order to seize opportunities in other localities (Bertuglia & Occelli, 2000).

Table 1. Variables used in the empirical estimations and summary statistics.

Variable	Mean	SD
Growth of the share of the population 65+ between 1970 and 2020	0.514	0.275
Growth of the average age between 1970 and 2020	0.157	0.077
Share of the population 65+, 1970 (t = 0)	0.149	0.041
Share of the population 65+, 2020 (t = τ)	0.241	0.046
Average age, 1970 (t = 0)	37.35	3.34
Average age, 2020 (t = τ)	43.62	2.71
Accessibility, population, 1970	56,033.76	74,964.26
Share of foreign born, 1973	0.038	0.034
Net migration of individuals 65+, 1970	1.56	48.77

A positive net migration of individuals 65+ in a municipality is assumed to relate to a convergence process. The immigrant population on average usually has a lower average age than the ethnic population, but here it suggests that a higher share of foreign-born population also relates to the convergence/divergence process. The first data for this variable are from 1973 and thus used as the initial period. The size of the municipality in terms of number of inhabitants relates to the convergence and is controlled for (total population). From the descriptive statistics in Table 1, we observe that the average age has increased from 37 to 43 years from 1970 to 2020, an increase of approximately 17%. The average growth of the average age is 16%, but the average growth in share of inhabitants over 65 is as high as 51%.

6.1. Convergence/divergence in ageing

To determine if there has been convergence or divergence in Sweden since 1970, we estimate equation (1) (overall convergence) and equation (2) (conditional convergence) using ordinary least squares. Table 2 shows the finding using the share of inhabitants 65+ and the average age in the municipality as proxies for ageing. The results show that there has been convergence in the ageing of the municipalities, confirming the findings of Gutiérrez Posada et al. (2018), Kashnitsky et al. (2021) and Van der Gaag and de Beer (2015). Furthermore, the size of the accessibility in a municipality in the initial period is negatively related to the growth of an ageing population, while the share of foreign born is negatively related to the share of population above the age of 65.¹⁰ The share of foreign-born shows a negative relationship with the share of inhabitants over the age of 65.

Table 2. Convergence and conditional convergence using the share of 65+ and the average age as proxies for the demographic profile of the municipalities.

	Beta	Conditional	Beta	Conditional
	Share 65+	Share 65+	Average age	Average age
Share of the population 65+, 1970 (ln)	−0.640***	−0.844***
	(0.031)	(0.027)
Average age, 1970 (ln)			−0.613***	−0.847***
			(0.028)	(0.028)
Accessibility population, 1970 (ln)		−0.123***		−0.039***
		(0.0091)		(0.003)
Share of foreign born, 1973 (ln)		−0.004***		0.001
		(0.009)		(0.003)
Net migration of individuals 65+, 1970		3.81e–05		−2.75e–06
		(0.0001)		(4.07e–05)
Constant	−0.738***	0.151	2.375***	3.626***
	(0.059)	(0.117)	(0.100)	(0.113)
Observations	283	283	278	274
R^2	0.676	0.841	0.582	0.786

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are shown in parentheses.

To obtain a more nuanced and comprehensive picture of the relationship between the initial value and the growth, that is, the convergence pattern, we use quantile regressions. The findings presented in Table 3 show that there are differences in convergence across the distribution of the outcome variable, that is, the growth of the average age or growth in the share of inhabitants over the age of 65.

Table 3. Convergence using the share of 65+ and average age as proxies for the demographic profile of the municipalities: quantile regressions.

	Share 65+	Share 65+	Share 65+	Average age	Average age	Average age
	0.25	0.50	0.75	0.25	0.50	0.75
Share of the population 65+, 1970 (ln)	−0.554***	−0.615***	−0.702***
	(0.0456)	(0.0568)	(0.0261)
Average age, 1970 (ln)				−0.577***	−0.606***	−0.595***
				(0.0439)	(0.0301)	(0.0418)
Constant	−0.667***	−0.676***	−0.754***	2.212***	2.351***	2.338***
	(0.0935)	(0.102)	(0.0513)	(0.158)	(0.110)	(0.152)
Observations	283	283	283	278	278	278
Pseudo-R^2	0.296	0.375	0.471	0.258	0.345	0.396

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are shown in parentheses. The findings are robust to including control variables.

Figure 4 shows the coefficient of variation as a measure of the σ-convergence for the full period. Again, we use the average age and the share of inhabitants over the age of 65 as proxies for the demographic profile of the municipalities over time. Sigma-convergence relates to a reduction in the disparities across municipalities over time, while β-convergence estimates if there is catching-up across municipalities. Important to note is that the coefficient of variation is more sensitive in the upper end of the distribution (Monfort, 2008). As visualised, there has been convergence across the Swedish municipalities from the 1970s up to the mid-1990s (using average age as demographic proxy) while for the share of inhabitants over the age of 65, the convergence continues until the 2010s. Thus, the disparities across municipalities decreased during these periods.

Figure 4. Sigma convergence across Swedish municipalities for the period 1970–2020 using average age and the share 65+.

the figure shows two lines that has the shape of a hook (pointing to the right). The darker line has its turning point to the left of the light grey line.

6.2. Regional differences

Table 4 uses different types of municipalities to assess the change in convergence. The urban–rural hierarchy is captured by dividing the Swedish municipalities into three categories (Tillväxtanalys, 2014): Metropolitan (municipalities with more than 80% of the population in urban areas and a population size of at least 500,000 (including adjacent municipalities)), Cities (municipalities with more than 50% in urban areas and with varying accessibility to a larger city (defined as having at least 50,000 inhabitants)) and Rural locations (municipalities with more than 50% in rural areas and with varying accessibility to a larger city (defined as having at least 50,000 inhabitants)). The results show that convergence has a non-linear relationship to the level of urbanisation in the municipalities, that is, the coefficients are significantly different from each other for the metropolitan, city, and rural locations. The larger convergence takes place in the most urban locations and the most rural areas.¹¹ These results confirm those by Gutiérrez Posada et al. (2018) and Kashnitsky et al. (2017). Again, we observe a negative relationship between the share of foreign-born and the ageing in a municipality for metropolitan and rural municipalities.

Table 4. Convergence and conditional convergence using the share of 65+ and average age as proxies for the demographic profile of the municipalities: estimations performed for different regional categories.

	Beta	Conditional	Beta	Conditional
	Share 65+	Share 65+	Average age	Average age
Metropolitan
Share of the population 65+, 1970 (ln)	−0.722***	−0.972***
	(0.041)	(0.034)
Average age, 1970 (ln)			−0.711***	−0.934***
			(0.041)	(0.033)
Accessibility population, 1970 (ln)		−0.152***		−0.041***
		(0.022)		(0.006)
Share of foreign born, 1973 (ln)		−0.101***		−0.026**
		(0.024)		(0.007)
Net migration of individuals 65+, 1970		4.51e–05		−2.86e–05
		(0.0001)		(2.28e–05)
Constant	−1.012***	−0.129	2.699***	3.881***
	(0.095)	(0.263)	(0.144)	(0.148)
Observations	79	79	82	78
R^2	0.828	0.928	0.788	0.909
Cities
Share of the population 65+, 1970 (ln)	−0.728***	−0.775***
	(0.059)	(0.055)
Average age, 1970 (ln)			−0.682***	−0.756***
			(0.065)	(0.058)
Accessibility population, 1970 (ln)		−0.150***		−0.050***
		(0.012)		(0.005)
Share of foreign born, 1973 (ln)		0.035***		0.011***
		(0.010)		(0.004)
Net migration of individuals 65+, 1970		−0.001		−0.0003
		(0.001)		(0.0002)
Constant	−0.874***	0.680	2.631***	3.444***
	(0.109)	(0.177)	(0.238)	(0.214)
Observations	155	155	151	151
R^2	0.580	0.807	0.434	0.718
Rural locations
Share of the population 65+, 1970 (ln)	−0.848***	−0.911***
	(0.118)	(0.078)
Average age, 1970 (ln)			−0.737***	−0.902***
			(0.111)	(0.080)
Accessibility population, 1970 (ln)		−0.066***		−0.024**
		(0.018)		(0.008)
Share of foreign born, 1973 (ln)		−0.037***		−0.010
		(0.019)		(0.005)
Net migration of individuals 65+, 1970		−0.001		−0.001
		(0.002)		(0.0004)
Constant	−1.056***	−0.722**	2.850***	3.628***
	(0.209)	(0.328)	(0.408)	(0.280)
Observations	49	49	45	45
R^2	0.596	0.759	0.513	0.714

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are shown in parentheses.

Figure 5 displays the σ-convergence using the three distinct types of regions, which follow different patterns and face different initial positions. Starting with the most urbanised municipalities, captured by the category Metropolitan, the municipalities show sharp convergence until the 2000s using average age and 2010s using the share of inhabitants over the age of 65. After that, the convergence process levels off and even shows signs of divergence. Municipalities classified as Cities share similar patterns as Metropolitan but the convergence is less pronounced. Locations that are the least urbanised, Rural locations, diverge from this pattern where for the full period they show signs of convergence using the average age, albeit at a smaller scale. For rural locations using the share of individuals over 65, rural locations show divergence from 1990 to the mid-2000s.

Figure 5. Sigma convergence across Swedish municipalities for the period 1970–2020 using average age (a) and the share aged 65+ (b) across the regional hierarchy.

the two figures have both three lines each. The black line in each figure has the shape of a hook, the light grey resembles a stretched U and the dark grey lines are almost straight lines with a small negative slope.

6.3. The foreign-born population and demographic profile of municipalities

To further explore how the share of foreign-born in a location relates to the overall ageing and to the convergence process, we conduct several estimations. The group of foreign-born is heterogeneous, but we are unfortunately not able to separate asylum seekers from labour-driven immigration, so they are all treated as one group. This aspect of how immigration relates to the ageing of a country is not new and has been covered for a long time within demography (for longer discussions, see, e.g., Gesano & Strozza, 2011; and Bengtsson & Scott, 2011). Fewer studies, however, focus on the regional level. We start by simply estimating the relationship between the foreign-born population and the demographic profile of the municipality over time where we look at the ageing pattern relationship between 1973 and 2020. The outcome variables capturing the age profile are the same as previously used: the average age in a municipality and the share of inhabitants over the age of 65. We explore the relationship using a fixed-effects model where the same set of control variables previously described are added, along with time-fixed effects. The main difference using a fixed-effects model is that we now want to use the difference that occurred annually within each municipality, holding unobservable factors fixed. Instead of looking at an initial year and the growth across years, we utilise all available information and the incremental differences that constantly occur. The fixed-effects model is less sensitive to discrepancies in the initial year. In addition, by isolating the time-invariant factors at the municipal level, more factors are controlled for. The findings are indicative of a negative correlation between the share of foreign born and the ageing of a location, similar to the findings in McDonald and Kippen (2001), Collantes et al. (2014) and Green et al. (2009).

Table 6 shows how the share of foreign born relates to the demographic profile across the different regional categories, again using a fixed-effects model as in Table 5. We observe a negative relationship between the share of foreign-born and the ageing profile across the urban–rural hierarchy, albeit stronger in rural locations where the coefficients when using the share of individuals above 65 for the metropolitan and rural locations are significantly different from each other. Using the average age as the outcome shows significant differences for the coefficients across all regional categories.

Table 5. Relationship between foreign born and demographic profile of the municipalities proxied by the share of 65+ and the average age, estimated using a fixed effect model over the period 1973–2020.

	Share 65+	Average age
Share of foreign born	−0.255***	−14.271***
	(0.008)	(0.596)
Total population (ln)	−0.045***	−1.792***
	(0.001)	(0.078)
Net migration of individuals 65+	−7.49e–06***	−0.007***
	(4.60e–06)	(0.0003)
Constant	0.583***	54.114***
	(0.010)	(0.763)
Year fixed effect	Yes	Yes
Observations	15,151	15,151
Number of units	289	289
R^2 within	0.764	0.794

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors are shown in parentheses.

Table 6. Relationship between foreign born and demographic profile of the municipalities proxied by the share of 65+ and average age, along the urban–rural hierarchy, estimated using a fixed effect model over the period 1973–2020.

	Share 65+			Average age
	Metropolitan	Cities	Rural locations	Metropolitan	Cities	Rural locations
Share of foreign born	−0.201***	−0.223***	−0.298***	−10.601***	−14.593***	−18.819***
	(0.019)	(0.010)	(0.016)	(1.446)	(0.686)	(1.030)
Constant	0.246***	1.151***	1.081***	22.511***	101.099***	78.893***
	(0.027)	(0.017)	(0.028)	(2.052)	(2.052)	(1.799)
Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Observations	4312	8242	2597	4279	8192	2556
Number of units	84	156	53	84	156	49
R^2 within	0.661	0.845	0.827	0.724	0.869	0.874

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Standard errors are shown in parentheses.

After establishing that there is a link between the share of foreign-born and the age profile in a location, we now turn to assess if the inflow of immigrants is related to age convergence or divergence. To estimate the convergence rate that took place in a municipality due to a large inflow of immigrants, we make use of the different unexpected inflows of immigrants that Sweden experienced during this period. To detect unusual inflows, we make use of data from the Sweden’s migration agency over the period 1980–2014 (see Figure A1 in Appendix A in the supplemental data online). Visible from the figure is that during 1993 and 1994, there was a sharp increase of immigrant inflow during these years due to the war in the former Yugoslavia. The increase and magnitude of the inflow was large and in many cases unexpected (Backman et al., 2021; Bråmå, 2006). Table 7 shows the convergence in age when dividing municipalities into diverse groups based on the inflow of immigrants during 1993–94. To define municipalities that received a large inflow of immigrants, we use the total number of immigrants and the share of immigrants. We define the municipalities in the top quartile in 1993 and/or 1994 as municipalities having a large inflow. We then evaluate the age convergence using the two groups of municipalities starting from the year of 1993 and assessing the growth until 2020, like the previous convergence tests. The results in Table 7 indicates that the municipalities that received a large inflow of individuals with a foreign background display a slower ageing of their municipalities compared with municipalities that did not receive the same number of foreign individuals. Thus, the inflow of immigrants is associated with an ageing divergence process where locations that initially were characterised by older inhabitants experience a slower ageing and de-accelerating differences across municipalities.

Table 7. Convergence and conditional convergence using the share of 65+ and average age as proxies for the demographic profile of the municipalities: estimations performed for diverse groups of municipalities based on the inflow of immigrants in 1993 and 1994.

	Beta Share 65+	Beta Share 65+	Conditional Share 65+	Conditional Share 65+	Beta Average age	Beta Average age	Conditional Average age	Conditional Average age
	Absolute inflow	Share of the population	Absolute inflow	Share of the population	Absolute inflow	Share of the population	Absolute inflow	Share of the population
Large inflow
Share of the population 65+, 1993 (ln)	−0.448***	−0.557***	−0.715***	−0.842***
	(0.079)	(0.065)	(0.078)	(0.045)
Average age, 1993 (ln)					−0.210***	−0.372***	−0.531***	−0.747***
					(0.098)	(0.088)	(0.092)	(0.052)
Accessibility population, 1993 (ln)			−0.162***	−0.165***			−0.047***	−0.046***
			(0.023)	(0.016)			(0.007)	(0.004)
Share of foreign born, 1993 (ln)			−0.009	−0.128***			0.010	−0.022*
			(0.038)	(0.044)			(0.074)	(0.012)
Net migration of individuals 65+, 1993			0.0002**	2.09e–5			3.25e–5	−1.16–5
			(0.0001)	(5.26e–5)			(2.85e–5)	(1.36e–5)
Constant	−0.574***	−0.763***	0.826**	0.305	0.841***	1.436***	2.588	3.293
	(0.147)	(0.132)	(0.339)	(0.219)	(0.356)	(0.319)	(0.355)	(0.210)
Observations	72	71	72	71	72	71	71	71
R^2	0.302	0.487	0.754	0.845	0.049	0.178	0.771	0.738
Others
Share of the population 65+, 1993 (ln)	−0.443***	−0.402***	−0.649***	−0.605***
	(0.035)	(0.034)	(0.046)	(0.047)
Average age, 1993 (ln)					−0.206***	−0.172***	−0.488***	−0.436***
					(0.041)	(0.048)	(0.063)	(0.068)
Accessibility population, 1993 (ln)			−0.079***	−0.068***			−0.024***	−0.023***
			(0.015)	(0.013)			(0.004)	(0.004)
Share of foreign born, 1993 (ln)			−0.019	−0.034			−0.003	−0.007
			(0.017)	(0.019)			(0.005)	(0.007)
Net migration of individuals 65+, 1993			0.001**	0.001**			0.0004**	0.0003***
			(0.001)	(0.0003)			(0.0001)	(0.0001)
Constant	−0.459***	−0.397***	−0.019	−0.111	0.861***	0.736***	2.153***	1.926***
	(0.058)	(0.057)	(0.143)	(0.153)	(0.151)	(0.176)	(0.263)	(0.288)
Observations	213	214	213	214	214	215	213	214
R^2	0.472	0.337	0.609	0.472	0.090	0.049	0.298	0.225

Note: ***p < 0.01, **p < 0.05, *p < 0.1. Robust standard errors are shown in parentheses.

7. CONCLUSIONS

This paper assesses the demographic change in Sweden from 1970 to 2020 to capture what the interregional ageing process in a developed country looks like. Sweden is used due to its fascinating ageing development where individuals above the age of 65 doubled in the twentieth century (Bengtsson & Scott, 2011) but the share of individuals above the age of 80 did not change between 2001 and 2021, an exception among EU member states. We find an overall level of convergence in the ageing across municipalities in Sweden. The results imply that locations that initially have a younger population and thus a lower level of ageing experience a higher increase in their aged population. The regional disparities in ageing are therefore decreasing over time. One potential reason for this is an overall ageing process that has taken place, making it evident that most locations in Sweden are facing a demographic transition that will have social and economic consequences.

Our findings further show that even though there is an overall ageing convergence, there are differences across the urban–rural hierarchy. The demographic transition has been more pronounced in rural areas than in urban localities. The big risk in the future for these regions will be that their economic growth might decline and even turn negative as their labour force grows more slowly, while more labour is demanded for healthcare and care of the elderly (Feldstein, 2006).

A special interest in this paper has been to assess if and how the inflow of foreign-born immigrants relates to the ageing outcome. We find that a higher share of foreign-born in a municipality is negatively related to a location becoming older, and that rural municipalities tend to benefit more in terms of rejuvenation from an increased share of foreign-born. A larger inflow of foreign-born further leads to a divergence of age across municipalities where these locations display slower ageing population growth.

Certainly, the sustainability of regional and municipal budgets will be a hotly debated issue in the coming years as the number and share of older people increases, especially since the regional and municipal taxes in Sweden are a percentage of the public’s taxable income. This development might create conflict between different spending priorities in regional and local budgets.

This study is not without caveats that might guide future research in this area. As with most studies that focus on one context, external validity is warranted. Although Sweden shares commonalities with many other developed countries, it also has some divergent age patterns, thus making it important to explore ageing patterns in other countries. In addition, this paper explores the role of immigration in intra-country ageing as a proxy for an external shock to ageing, but other external shocks can be further explored, such as changes in policies related to migration and emigration of individuals, and even a crisis like the COVID-19. By no means is the immigrant population a homogenous group, where the reasons to relocate diverge, including both labour market reasons and being forced to leave due to natural disaster, war, or political/religious persecution. Thus, the location pattern between these groups is most likely different and thereby relates differently to the ageing process within a country, making it also worthy of further study.

DATA AVAILABILITY

The research data are confidential.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Additional information

Funding

The authors are grateful for the financial support from the Marianne and Marcus Wallenbergs Foundation, Sweden [grant number MMW 2018.0049], for the project grant ‘Ageing and Entrepreneurship’.

Notes

1. See https://ec.europa.eu/eurostat/cache/digpub/demography/bloc-1c.html?lang=en

2. Of course, large-scale immigration has numerous effects on the receiving country. It has, for example, been discussed whether large-scale emigration even alleviates the economic burden of countries associated with ageing populations (Bonin et al., 2000).

3. We can observe that generally in Europe, immigrants are more concentrated to densely populated areas than the native-born population (Organisation for Economic Co-operation and Development (OECD), 2016).

4. Defined as the standard deviation divided by the mean.

5. What convergence measure to use has been hotly debated after it was pointed out that estimations of β-convergence due to the regression to the mean may be statistically flawed (Quah, 1993b). Sala-i-Martin (1996a) conjectured that analyses of β-convergence are valuable to evaluate the outcomes of analyses of σ-convergence and that it is not motivation enough to disregard β-convergence measures just because of the risk of some econometric problems.

6. See https://www.scb.se/hitta-statistik/statistik-efter-amne/befolkning/befolkningens-sammansattning/befolkningsstatistik/

7. See https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Population_structure_and_ageing

8. See https://www.macrotrends.net/countries/SWE/sweden/birth-rate

9. A lower average age of the population in urbanized municipalities results in higher birth rates and lower mortality rates than in more rural municipalities, which functions as a centripetal force that drives the concentration dynamics (Grafeneder-Weissteiner & Prettner, 2013). The overall patterns illustrated are equal to what we should expect according to what the literature on urban economics indicates about demographic processes (Rozenfeld et al., 2011).

10. We further accounted for non-linear effects by adding the squared value of the share of the population over 65 and the average age. The findings show signs of increasing marginal effects.

11. Once again, we tested for non-linearities, and the results indicated marginal increasing effects only when using average age for the regional categories Metropolitan and Cities.