1. Data Sources and Variables

The dataset of this research consists of the records of the total population, age and gender characteristics of the population, birth and migration at the district level obtained from the Address Based Population Registration System (ABPRS) of the Turkish Statistical Institute (TurkStat) for 2020. These data provided by TurkStat were used by the authors to calculate the structural ageing level (structural ageing-SA) and the proportion of very old population (aged 80 and over) in the old population (aged 65 and over) to represent longevity (very olderly population ratio-VOEPR) , variables consisting of the general fertility rate (GFR) , in-migration ratio (IMR) and out-migration ratio (OMR) were obtained. Demographic factors affecting ageing were analyzed to establish ageing zones with the help of these variables.

Fig. 1 Statistical regions (NUTS I), provinces and districts of Turkey. 

A high level of structural ageing in a given area can be considered as a sign of ageing in that area (^{Atkins & Tonts, 2016}; ^{Chen & Hao, 2014}; ^{Lutz et al., 2008}; ^{Zhang & Chen, 2014}). Structural ageing is widely used in the identification, monitoring and comparison of spatial patterns, especially at varying scales (^{Kiniorska & Pytel, 2022}; ^{Kurek 2003}; ^{Walford & Kurek 2010}; ^{Wilson, 2015}; ^{Wu et al., 2019}, ²⁰²¹). The combination of declining fertility and increasing longevity (life expectancy at birth) has led to an increasing number and proportion of older people in many parts of the world (^{Kinsella & Velkoff, 2001}). Thus, total fertility rate and life expectancy at birth are considered to be important indicators of a society's ageing level (^{Kinsella & Phillips, 2005}; ^{Kurek, 2007}). Therefore, it is taken as a variable when distinguishing ageing regions. However, the lack of district level statistics on these two indicators in Turkey has made it necessary to produce a new variable referring to longevity (VEPR) and to use the general fertility rate (GFR).

Internal migration is one of the major factors affecting the age structure of the population in a locality. Internal migration has been found to affect ageing by offsetting or even reversing regional differences in birth and death rates (^{Chen et al., 2020}; ^{Moore & Pacey, 2004}). As a result of age-selectivity, out-migration usually leads to the departure of young people from an area and leaves older people behind, while in-migration leads to an increase in the young population and a decrease in the proportion of older people in the target areas (see ^{Ananta & Arifin, 2009}; ^{Arifin & Ananta, 2016}; ^{Wu et al., 2021}; ^{Zheng & Yang, 2016}). Therefore, in-migration and out-migration variables affect the age structure and ageing process of the population in two ways. Since the effect of internal migration on ageing cannot be overlooked in a country like Turkey where a powerful migration from rural to urban areas has taken place in the last 50-60 years, the variables of internal migration rate (IMR) and external migration rate (OMR) have been included in the grouping analysis.

2. Analysis Technique: Grouping Analysis

Spatial grouping analysis was used in this study to establish ageing regions. According to this analysis, groups emerge when nearby or neighboring measures tend to be similar. The identification of spatial groups facilitates hypothesizing about the relationship between the patterns of a geographic phenomenon and the processes underlying the phenomenon. Group detection fulfills the dual task of examining whether the distribution is clustered, and revealing the location and extent of groups. Group detection is normally accomplished by assessing how the investigated patterns differ from a random pattern. If nearby values are more similar to each other than a random pattern, or if there are more observations within a neighborhood than for a random distribution, the distribution under study is claimed to cluster. A neighborhood with more similar values or more observations defines a group (^{Lu, 2009}, p. 317).

Spatial grouping analysis has recently been added to traditional cluster analysis. Spatial grouping analysis differs from traditional clustering analysis to a certain extent. One of the most important differences is that spatial grouping analysis is mainly based on the spatial location of geographical phenomena and the attribute information of the relevant geographical phenomenon. Secondly, the spatial aggregation of objects in space according to their different subcomponents into groups and various subgroups occupying different spatial units are analyzed. This is because the formation of spatial subgroups is a product of the geographical environment (^{Jiang, 2021}). The identification of groups from the spatial patterns of a multi-component phenomenon such as ageing is possible through spatial grouping analysis rather than traditional cluster analysis.

Objects with more than one attribute can be classified using spatial statistical analysis sets in the Geographical Information System - GIS environment (^{Kastreva & Patarchanova, 2021}). The most important advantages of grouping analysis using GIS programs are the mapping of groups and the visualization of group locations and distribution areas. Thus, groups can be identified and mapped by analyzing different layers with location information and multiple attribute data together (^{Kastreva & Kastreva, 2020}). Indeed, spatial grouping analysis is used in demographic research to serve such a purpose (^{Kastreva & Patarchanova, 2021}).

Grouping analysis provides useful results in the establishment of ageing regions. Therefore, it is necessary to distinguish between areas that are similar and dissimilar in terms of the level of ageing and its determinants at a given spatial scale. Accordingly, it is expected that the similarities between all units within a group are maximized, which is the basic principle of region definition and distinction. On the other hand, similarities must be maximum within a region (group) and minimum between regions. Therefore, areas that are similar to each other remain within the region, while those that are dissimilar are excluded. This leads us to distinguish and define the region, in a way, by “drawing a border” (fig. 2). This process, which must be done separately for each variable to be included in the analysis, results in the separation of more than one region. It should be reduced to a single layer, since a separate layer of region boundaries established for all variables will not be useful. This can be possible by reanalyzing the region separations generated for each variable. However, more than one variable can be processed with grouping analysis, and a result can be reached directly.

Fig. 2 Schematic representation of grouping analysis. Source: ^{Environmental Systems Research Institute} [ESRI] 

This study professes that the data of five different layers (variables) related to ageing and its determinants according to the spatial units of districts can be analyzed in a statistically grouped way to determine the tendency to form regions. In this framework, the distribution of structural ageing and fertility, longevity and in-migration and out-migration, which are effective in this distribution, were considered together and Grouping Analysis, one of the Spatial Statistics techniques of ArcGIS 10.5 software, was used to determine which districts are spatially similar (or in the same group) and which districts are different (in different groups). This analysis makes a grouping based on the distribution of the data according to the predetermined number of groups from the processed different data, as the number of groups is left to the user's preference in version 10.5 of the ArcGIS. The aim was to ensure that the differences within the group are minimized and the differences between the groups are maximized.

Many algorithms are used in statistics to determine a specific clustering and the results are tested with various performance indicators (^{Karaoğlan, 2018}). However, K-means clustering, one of the non-hierarchical multivariate statistical analysis techniques which is often preferred in social science research was also preferred in this study. In fact, K-means clustering is used in the background of the spatial grouping analysis of ArcGIS 10.5 software (ESRI, 2022). The number of groups in the analysis was determined using the Calinski-Harabasz pseudo F-statistic, which is a ratio reflecting the similarity within groups and the difference between groups, and this number was determined as four. Thus, 973 districts in Turkey were divided into four different groups in terms of ageing and its determinants.

If the spatial position of the groups obtained in the grouping analysis is to be taken into account (within the Spatial Constraints tab), the spatial weight matrix (neighborhood, distance, etc.) can be selected and included in the analysis as in other spatial statistical analyses. However, in this study, no spatial weight matrix was used in order to identify districts that exhibit ageing characteristics that are completely different from their surroundings. Another reason for not using a spatial weight matrix was the desire to maximize the similarity within the group. Thus, restrictions such as spatial proximity or neighborhood were overcome.

The output of the grouping analysis is a report containing the descriptive statistics of each group and variable and a map showing which group the spatial units are in (^{Kastreva & Paratachanova, 2021}). The final report of the analysis in this study includes descriptive statistics of all the variables in the classification of four regions, descriptive statistics of each region separately, and a parallel boxplot graph generated from the statistical indicators of the groups.

1. The Spatial Patterns of Demographic Variables Used to Distinguish Ageing Regions in Turkey

This section summarizes the spatial patterns of variables used to distinguish the ageing regions in Turkey to demonstrate the background of the grouping analysis better.

Structural ageing (SA): As can be seen in figure 3, the country is roughly divided into two main areas in terms of structural ageing. The first area includes districts in the Northeast, Central East, and Southeast Anatolia regions of Turkey, the coastal areas of the East Marmara, Aegean and Mediterranean regions, West Anatolia, Istanbul and other metropolitan urban areas. These districts are often areas with structural ageing levels (i.e. young and mature age structure) of less than 10% and sometimes even less than 5%. On the other hand, the second area includes mostly districts located in the West Black Sea and East Black Sea, Central Anatolia, West Marmara and Aegean regions. These districts are rural areas with structural ageing levels (hence old or very old age structure), above 10%, sometimes 15% or even 20% (^{Yakar & Özgür, 2022}).

Fig. 3 Structural ageing (SA)(Ratio of olderly population in total population) by districts in Turkey (2020). 

General fertility rate (GFR): The spatial pattern of the overall fertility rate in Turkey is shown in figure 4. According to this, the country is again divided into two main areas. The first area includes the districts in the eastern regions of the country. Accordingly, the GFR exceeds 52 per thousand (average for districts) and often exceeds 75 per thousand in the districts of the Southeast Anatolia Region and the neighboring provinces of Central East Anatolia, such as Van, Bitlis and Muş, and those in Northeast Anatolia, such as Ağrı and Erzurum (fig. 4). These districts have historically been high fertility areas in Turkey, and as such have been always being distinctly different from the rest of the country (^{Yüceşahin & Özgür, 2008}). Secondly, there are districts where the GFR is below 52 per thousand and partly below 39 per thousand. The areas in this latter group, which have the lowest GFR values in the country, are mostly concentrated in the West Black Sea, Aegean, West Marmara and East Marmara regions (fig. 4).

Fig. 4 General fertility rate (GFR) by districts in Turkey. 

Out-migration ratio (OMR): the spatial differences in outward migration are clearly evident in figure 5. Accordingly, the cumulatively high out-migration ratios of northern and northeastern Turkey are immediately marked: In most districts of the West Black Sea, East Black Sea, Northeast Anatolia, Central Anatolia, the OMR exceeds 56% or even 69%. Since the mid-20^th century, these have been the most out-migrating areas due to difficult natural conditions, insufficient economic resources and social opportunities (^{Filiz, 2008}; ^{Tümertekin, 1968}). In contrast, districts in western and southern Turkey have relatively lower levels of OMR: In most districts in the Aegean, West Marmara, East Marmara, West Anatolia and Mediterranean regions, and partly in Southeast Anatolia, the OMR falls below 43%, and often below 30% (fig. 5), as they are the developed districts of Western Turkey with the lowest OMRs, reflecting development differences (^{Ministry of Industry and Technology, 2022}).

Fig. 5 Out-migration ratio (OMR) by districts in Turkey. 

In-migration ratio (IMR): The cumulatively highest in-migration ratios in the country are observed in western and southern Turkey: Istanbul, Aegean, West Marmara, East Marmara, East Marmara, West Anatolia and Mediterranean regions (fig. 6). The metropolitan districts in these regions have long been the destination of migration in Turkey, thanks to urban job opportunities in industrial, touristic and various other sectors (^{Yakar, 2015}). The IMR in these districts exceeds 40%, and even 60% in the developed metropolitan districts of Istanbul, Ankara, Izmir and Kocaeli. On the other hand, districts in the rest of the country have low levels of IMR, mostly below 10% (fig. 6 ). These areas correspond to districts in Turkey with low levels of socio-economic development and population loss through negative net migration rates (^{Yakar & Saraçlı, 2010}).

Fig. 6 In-migration ratio (IMR) by districts in Turkey. 

Very olderly population ratio (VEPR): The pattern of very old (olderly) population ratios in Türkiye does not display inter-regional differences that are as marked as the other variables. In each region of the country, there are districts with high VEPRs. In particular, some districts in the East Black Sea, East Marmara, Central East Anatolia, Aegean and Mediterranean regions have VEPRs above 24% and in some cases over 27%. The VEPR is below 18% in districts mostly located in the eastern and central parts of Türkiye (fig. 7).

Fig. 7 Very olderly population ratio (VEPR) by districts in Turkey. 

2. Ageing Regions of Turkey

Descriptive statistics and R² values of all variables included in the analysis are the most important statistical indicators resulting from the grouping analysis (table I). The calculated R² values approaching 1 indicate that the distinguishing capacity of that variable in grouping analysis is high compared to other variables. When this value is close to zero, it indicates that the distinguishing capacity of that variable in the formation of groups has weakened (^{Kastreva & Patarchanova, 2021}; ^{Ruda, 2016}). Accordingly, among the R² values of ageing patterns and variables, the highest value corresponds to the in-migration ratio (R² = 0.76). This is followed by the structural level of ageing, the general fertility rate, the out-migration ratio and the R² value of the proportion of the very olderly population (table I). These R² values of the variables also indicate that they can be used in grouping analysis to establish ageing regions.

The main output of the grouping analysis is the identification of groups with the variables included in the analysis. Two separate outputs consisting of a statistical table and a map are generated as a result of the analysis. The table contains the descriptive statistics of each variable in the groups resulting from the analysis according to the number of determined groups (table II). These statistics show how the variables are distributed within and between the groups.

The shared value of each variable in the groups is obtained by dividing the range value (a measure of central dispersion, calculated by subtracting the minimum from the maximum in the data set) of the relevant variable in the group in which it falls by the global (country-wide in table I) range value of the relevant variable. This tells us how much of the global range value is covered by the range value of any variable in the group. As a result, a high share value indicates high coverage and low distinctiveness in the formation of that group, since it will cover a large part of the global range value¹. Therefore, when the share value of the variables within each group is ranked from low to high, a sequence emerges from the most important variable to the least important variable in the formation of that group. On the other hand, when the variables are ranked from the highest share value to the lowest one, a sequence casting a reflection/similarity for the overall country from high to low is obtained.

Four different regions were established as a result of the grouping analysis. The districts that each of these regions covers are presented in figure 8, their descriptive statistics are presented in table II and the visualization of these statistics is presented in the boxplot in figure 9. All these outputs of the analysis give us the opportunity to define the characteristics of the ageing regions.

The First Region includes 155 districts in the east and southeast of Turkey (fig. 8 ). The average SA and IMR values for the region are well below the national average, while the VEPR is slightly below average. Furthermore, the OMR is slightly above average, while the GFR is well above average (tables I and II; fig. 9). The two variables that most distinguish the first region from the others are structural ageing and the in-migration ratio. The analysis shows that the other variables have low distinguishing capacity for the region. However, at this point, it should be noted that the region's high GFR rejuvenates the population, leading to a very low structural ageing in the total population (there is a negative correlation between the fertility rate and the proportion of olderly population). The first region has an OMR that is slightly above Turkey's average. However, although out-migration is known to increase ageing, the high fertility level reduces the negative impact of out-migration on ageing in this region, as the migrants are mostly young people. In conclusion, it can be established that Region 1 is characterized by high GFR and OMR values and low SA and IMR values (table II, fig. 9).

Fig. 8 Ageing regions of Turkey. 

Fig. 9 Boxplot of variables by region generated in grouping analysis. 

The Second Region mostly consists of the metropolitan and urban districts in Türkiye. There are 173 districts in this region, which consists of the central districts of metropolitan cities such as Istanbul, Ankara, Izmir, Bursa, Konya, Konya, Antalya, Adana, Gaziantep, Kayseri, Eskisehir, Denizli and Mersin, and districts that have an urban population density in some provinces such as Çanakkale, Edirne, Zonguldak, Osmaniye (table 2, fig. 8). Almost all districts in this region are concentrated in the western half of the country. This pattern is associated with urbanization dynamics due to internal migration. The IMR of the region is the highest among all regions. According to the analysis, the average in-migration ratio of the districts in the region is 70%. Therefore, the most important characteristic of the second region is the high in-migration ratio (fig. 9). The SA, VEPR, GFR and OMR values of the region are below the national averages. The share value of the second region, which largely constitutes the centers of attraction of urban migration in Türkiye, is as high as 0.78. Therefore, IMR is a variable with high coverage but a low distinguishing capacity for this region. On the other hand, considering that those migrating to urban areas in Türkiye are mostly young people (especially between the ages of 20-40), the proportion of the young population in the total population has naturally increased, which has curbed the increase in the proportion of the olderly population to some extent. Therefore, while urban areas were expected to have a higher proportion of olderly due to their fertility transition taking place earlier, the high levels of in-migration in the districts of the region have slowed down the ageing transition. In the second region, the district average of GFR with 44.6‰ is below the national level. The fact that GFR has the lowest share value of 0.29 among the other variables indicates the importance of fertility in distinguishing this region. The fact that fertility is lower than the national threshold in this predominantly urban region of Türkiye suggests that while it should accelerate ageing, the full impact has not occurred yet.

This is a sign of slow ageing from the bottom in the region and is related to in-migration slowing the pace of ageing. The average OMR in the second region is 19.6%, the lowest among all regions. The share value of OMR is 0.55 and this variable has a moderate effect in differentiating the region. The average structural ageing in the districts of the region corresponds to the national average of 9.7%. In addition, the proportion of very olderly population in this region is the lowest among all regions (17.5%) (table II).

The Third Region has the most coverage with 356 districts. This region, which is widely distributed in Turkey, includes districts mainly in the central, southern and western parts of the country, largely outside the urban areas of the second region (fig. 8). This region has the closest indicators to the national averages for all variables (fig. 9). The GFR, IMR and OMR values of this region are slightly below the national average, while the SA and VEPR values are almost commensurate with the average values (table II). The average GFR of the third region was 47.8‰. GFR is the most important distinguishing indicator of the region with a share value of 0.29. The low fertility rate has an increasing effect on ageing in the region. IMR (with a value of 14.19%) is the second most important determinant in the third region after GFR (with a share value of 0.47). These two variables characterize the region. In this region, the SA with a value of 15.04% and the VEPR with a value of 20.6% almost correspond to the national average. The fact that the third region is located close to the median value for each variable in the boxplot in terms of both the structural ageing and migration, moreover, fertility indicators shows that this region has the closest representation to national values (fig. 9).

The Fourth Region includes 289 districts, mostly located in the West and East Black Sea and Northeast Anatolia (in the north and northeast of the country), in the eastern parts of Central Anatolia and in mountainous areas. The first characteristic feature of the districts of this region is that they are represented by a low IMR. As shown in table II, the average in-migration ratio of the districts in this region is 9.8%. This value is even less than the average of half of districts in Turkey (table I and II). However, observations regarding retirees among those who migrated to the region suggest that in-migration (return migration of the olderly) may have had an increasing effect on ageing. Despite the low IMR, the region's out-migration ratio reaches the highest level among the regions with a value of 66.65%. The high OMR (especially the labor migration of youth) has accelerated ageing in this region. The GFR of the fourth region was below the national rate at 44.7%. High out-migration has contributed to a decline in fertility, especially as it leads to a loss of young population. The average SA of this region is 23.07% (table II), making the fourth region the oldest in Turkey. The average VEPR among the olderly population in the region is 23.59%, which is above the national threshold. As a result, the fourth region corresponds to the area with the highest SA and VEPR values in Turkey as a result of the ageing-enhancing effect of low fertility and high out-migration.

The ageing transition is relatively slow in this region where high fertility rates have an impact on the age structure, and the districts of the region correspond to areas that will experience population ageing last as a result of declining fertility levels. The second region which is comprised of districts with metropolitan areas and urban center characteristics (urban Turkey) represent a mature-older position in terms of the ageing level. This region, where the substantial effects of high in-migration have delayed ageing for the time being, can be labeled as having a "mature" population. The second region is the most populous region in Turkey in terms of total population and olderly population, but is not yet in the olderly population category. Therefore, this region is of critical importance for the future of national ageing and is expected to be the main determinant of changes in the level of national ageing in the future. The third region, which is notable by low fertility and in-migration ratios and contains the largest number of districts among the regions, can be characterized as "aged". Considering all the districts in Turkey, this region is composed of spatial units that represent the country as a whole in terms of ageing that are getting older as the proportion of the olderly population in the early stages of ageing increases rapidly. The fourth region is characterized by low fertility levels, marked by a loss of young population as a result of a long period of high out-migration. At the same time, the olderly population tends to live longer in the region. As an area undergoing the advanced stages of the ageing transition in Turkey, the fourth region can be labeled as "very aged".