Keywords

1 Introduction

In 2005, Japan experienced its lowest Total Fertility Rate (TFR) in modern history, with a rate of 1.26. Japan could experience a decrease of 11 million people within a decade, which represents approximately 10% of the population. Despite variations in TFR by region, TFR is commonly used as a benchmark in studies related to declining birthrates. Sasai (2005) argued that regional fertility rates exhibit diverse levels and patterns of change, and that two major demographic factors, namely marriage trends and couple’s fertility, can explain the TFR to some extent [1]. However, a geographically weighted regression (GWR) analysis conducted by Kamata and Iwasawa (2009) demonstrated that, despite regional variations, most studies tend to use least squares (OLS) regression, and that the factors contributing to TFR are largely the same across regions [2].

Nevertheless, a high TFR in a region does not necessarily guarantee population sustainability in the future. Maeda (2005) suggested that TFR is a ratio of the number of children to the number of women, and that even if the number of children decreases, the TFR remains unchanged as long as the number of women in the denominator also decreases [3]. Therefore, the number of women aged 15–49 is an important factor to consider when examining TFR. The scatter plot below shows TFR on the X-axis and FMI (an indicator of women moving in and out of the region) on the Y-axis, and will be discussed later (Fig. 1).

Fig. 1.
figure 1

Scatter plot of TFR and FMI, and pie plot of clusters

Full size image

We conducted a cluster analysis based on TFR and FMI, and the resulting clusters are named as shown in the legend on the upper right of Fig. 2. The histogram in Fig. 2 displays the future population for each cluster, with the year 2015 as the base (set to 100). Regions with high TFR and FMI, which are considered ideal for maintaining population, are able to sustain their population. However, regions with high TFR and low FMI, which we named the “High TFR type,” are unable to maintain their population. This highlights the importance of not only considering TFR, but also FMI when addressing the issue of declining birthrate and population.

Fig. 2.
figure 2

Histogram of future population by clusters

Full size image

This paper examines the issue of declining birthrate from two perspectives: TFR and the movement of women in and out of regions, taking into account regional differences. We provide a brief overview of each region in Japan: Hokkaido, Tohoku, Kanto, Chubu, Kinki, Chugoku, and Kyushu. Hokkaido, located at the northernmost tip of Japan, is the largest prefecture and boasts flourishing dairy and fishery industries. Tohoku is the northernmost region of Honshu, the main island of Japan, and is known for its thriving agriculture industry. The Kanto region is situated south of Tohoku and is home to Tokyo and other major urban centers. Chubu is a prosperous region located between Tokyo and Osaka and is home to many large manufacturing companies. Kinki is situated west of Chubu and includes cities such as Osaka, Kyoto, and Kobe. Chugoku, also known for its manufacturing industry, is located west of Kinki. Finally, Kyushu, located at the western end of Japan, has one of the highest birth rates in the country and is one of the fastest growing regions.

2 Methods

2.1 Data Source and Data Processing

Table 1. All data we used
Full size table

We obtained statistical data for all cities, towns, and villages in Japan from the Japanese Statistics Bureau, excluding the evacuation zone near the Fukushima nuclear power plant. Geographic data was acquired from the Ministry of Land, Infrastructure, Transport and Tourism, as well as previous research, and university campus location data was obtained from the university’s website. To calculate the distance between regions A and B, assuming the Earth to be a sphere, we used the following formula based on latitude (lat) and longitude (lon) (Table 1).

$$ Distance = 6371{*}\arccos \left( {\begin{array}{*{20}c} {\cos \left( {latA} \right) + \cos \left( {latB} \right)*\cos \left( {lonB - lonA} \right)} \\ { + \sin \left( {latA} \right)*\sin \left( {latB} \right)} \\ \end{array} } \right) $$

The variables that contain the word “expense” in their item names are related to municipal expenditures. For instance, the housing expense ratio measures the proportion of expenditure allocated to the development of residential areas in relation to the total expenditure of the municipality.

2.2 Create an Indicator of Women Moving in and Moving Out

We developed a novel indicator, the Female Migration Indicator (FMI), by dividing the number of women aged 15–49 moving into a region by the number of women aged 15–49 moving out. The purpose of FMI was to capture population flows in and out of a region, and we employed FMI as the response variable in our regression analysis.

2.3 Analysis Procedure

This study aims to explore the relationship between Female Migration Indicator (FMI) and Total Fertility Rate (TFR) in Japan using three different regression models: OLS regression with geographic variables for TFR (TOwG), OLS regression without geographic variables for TFR (TOwoG), and GWR regression for TFR (TG). Similarly, we employed three regression models to examine the relationship between FMI and TFR, which are OLS regression with geographic variables for FMI (FOwG), OLS regression without geographic variables for FMI (FOwoG), and GWR for FMI (FG). The best model for each indicator was selected by adding explanatory variables based on increasing values of partial correlation coefficient and considering R-squared, Adjusted R-squared, and AIC. To avoid multicollinearity, explanatory variables with VIF scores over 4 were excluded. The bandwidth for the kernel function used in GWR was set to a fixed type, which is a constant bandwidth from the regression point. The results of this study provide insights into the relationship between FMI and TFR, as well as the impact of geographic variables on the relationship.

3 Result

3.1 TOwG and TOwoG

Table 2 presents a summary of the statistical analysis for both TOwG and TOwoG models, which were found to be statistically significant based on the probability of F-statistic. The R-squared and Adjusted R-squared values indicate that geographic variables play a crucial role in predicting TFR, as evidenced by their substantially different scores (Fig. 3).

Table 2. Summary of result of TOwG and TowoG
Full size table
Fig. 3.
figure 3

Coefficients of TOwG and TOwoG

Full size image

Focusing on the coefficients of the OLS regression analysis with geographic variables (TOwG), indicated by the blue bars, we observe that longitude has a large negative score, suggesting that TFR tends to be lower in the eastern area and higher in the western area. The percentage of university graduates and the number of university campuses within a 20 km radius also have negative scores, while FMI, child welfare expense ratio, and the number of retail stores have positive values.

Similarly, the coefficients of the OLS regression analysis without geographic variables (TOwoG), represented by the orange bars, exhibit similar scores to those of TOwG. Additionally, the coefficients of agriculture, forestry, and fisheries expense ratio, and nuclear family household ratio have positive values, while the percentage of female legislators and the average annual salary have negative values. Interestingly, our findings differ from those of previous studies. For example, the coefficient of annual salary being negative implies that as the annual income decreases, the TFR increases, and vice versa, while previous studies suggest that declining incomes contribute to declining birthrates [4], and 40% of unmarried people under the age of 49 do not desire to have children due to economic issues [5].

3.2 FOwG and FOwoG

Table 3 presents the results of FOwG and FOwoG analyses, both of which are statistically significant at a significance level of 5%. In contrast to the TFR regression analysis, there is little difference in R-squared values between the two models.

Table 3. Summary of result of FOwG and FOwoG
Full size table

The coefficients bar graph below illustrates that FOwG and FOwoG have similar coefficients values. Percentage of university graduates has the largest positive value. The number of university campuses within 20 km also has a positive score, suggesting a positive effect of universities on FMI. Conversely, the number of persons per household has the largest negative value, indicating that regions with many single-person households tend to have high FMI, whereas those with many large families tend to indicate low FMI. Regarding the blue bar, latitude has a positive coefficient, suggesting high FMI in the north and low in the south, although the coefficient is relatively small, indicating a weak relationship. As for the orange bar, the female employment rate has a positive value (Fig. 4).

Fig. 4.
figure 4

Coefficients of FOwG and FOwoG

Full size image

3.3 TG and FG Result

Table 4 is summary of TG and FG. From R-squared and AICc, these GWR models are improved from OLS regressions analyses. The right two columns of Table 4 show results of Leung’s F test [6]. Leung’s F test has three types. F1 test is the test that confirms whether GWR is better fitted than OLS which is TOwoG or FOwoG in this paper. F2 test verifies whether there is statistically significant difference between GWR and OLS. F3 test decides if the regional differences in coefficients are statistically significant. Table 4 shows that both GWR models, TG and FG, are better fitted model and the difference with OLS are statistically significant. The result of F3 test is shown in Table 5. Variables marked with * indicate regional differences, and variables marked in yellow do not indicate regional differences.

To add a note about TG and FG, Ogasawara village was excluded because optimize bandwidth was too small to regress TFR and FMI of there. The following section, statistically significant coefficients of TG and FG are visualized, provided that Tokyo’s island areas and part of Kagoshima’s island areas are excluded from the visualization because isolated islands tend to be outliers.

Table 4. Summary of Result of TG and FG
Full size table
Table 5. The result of F3 test
Full size table

3.4 TG Coefficients

Figure 5 illustrates the regional variation of coefficients. Most regions exhibit positive values of child welfare expense ratio and FMI, and negative values of the number of university campuses within 20 km. Other variables have both positive and negative values in certain regions. Moving on to Fig. 6, the coefficient of child welfare expense ratio has the highest value in the Kyushu region and the lowest in the southern Hokkaido region, where it is negative. The Northern Tohoku and Kanto regions have small but positive values. In other words, most regions show a positive correlation between TFR and child welfare expense ratio, indicating that the government should consider providing better financial support for children. In Hokkaido and Kanto regions, where the coefficient value is positive but close to zero, raising child welfare expenses should be approached with caution. The coefficient of FMI is also mostly positive across regions, with particularly high values in the Kanto, Chubu, and Kinki regions.

Fig. 5.
figure 5

Coefficients of TG

Full size image

It was found that the coefficient of the number of university campuses within 20 km has negative values in most regions, consistent with Tsutsumi’s (2020) previous research indicating that an increase in university students, who do not give birth, can lead to a decrease in TFR. On the other hand, the Agriculture, Forestry and Fisheries expense ratio is high in Kinki region, with most regions having a positive value, except for Chubu region.

While areas with high agricultural expenses tend to score high TFR, this trend does not exist in Chubu and Kyushu regions. Additionally, the percentage of female legislators, often used as a measure of women’s social advancement, has a negative value in most areas and is believed to decrease TFR. Even in the area with the highest coefficient, the score is almost 0, indicating that the percentage of female legislators does not significantly affect TFR.

Fig. 6.
figure 6

Map of Japan with TG coefficients displayed

Full size image

The largest female employment rate is found in Tottori Prefecture, while the lowest is in the Kyushu region. Other regions have small positive values, except for Kyushu. Contrary to common belief, the advancement of women in society does not necessarily accelerate the decline in birthrate, as shown by the results of the percentage of female legislators. The nuclear family household ratio has a slightly positive value in the Kanto, Chubu, and Kansai regions, and values closer to 0 in other regions. In terms of industry factors, the ratio of workers in the secondary industry has a positive coefficient in 80% of the regions, but is negative in parts of the Hokkaido, Miyagi, and Kyushu regions. The tertiary industry has a negative coefficient in 80% of the areas, with positive values only in the Kinki and Chugoku regions. Lastly, focusing on the intercept, it represents the TFR without the influence of explanatory variables. Regional differences in the intercept imply disparities that are not explained by these variables. The left figure shows that intercept values tend to be higher in the west and lower in the east, which cannot be fully explained by the variables used. The right figure is a scatter plot of intercept and TFR, with a simple regression analysis showing that 19% of the variance of TFR is explained by the intercept and 50% by explanatory variables in the TG analysis (Fig. 7).

Fig. 7.
figure 7

Map of Japan with TG intercept displayed and scatter plot of TFR and intercept

Full size image

3.5 FG Coefficients

Figure 8 is violin plot of FG coefficients. As can be seen from the figure, percentage of university graduates has largest value, and TFR is the second largest. Number of persons per household and ratio of workers in secondary industry are negative value in more than 85% region.

Focusing on TFR, most regions have positive coefficients, especially in the Kanto region. Although not as high as Kanto, Chubu and Kinki regions also have relatively high coefficients. Comparing the previous analysis, there seems to be a correlation between TFR and FMI in the Kanto region. This is because the coefficients of FMI in TG were high in Kanto, and the coefficients of TFR in FG were also high.

In fact, there is a correlation in the Tokyo metropolitan area, excluding Tokyo’s 23 wards and remote islands, with a correlation coefficient of 0.47. In Kanagawa, Chiba, and Saitama, the correlation coefficient is 0.57. If we expand the scope slightly, there is also a correlation in areas such as Tokyo’s metropolitan area, excluding Tokyo’s 23 wards, Aichi, Kyoto, Osaka, and Hyogo, with a correlation coefficient of 0.32. In all regions in Japan, the correlation coefficient is 0.11. Based on these results, it can be said that there is a tendency for women to gather in areas with high TFR or for areas where women gather to have a high TFR in urban areas.

The percentage of university graduates shows the highest coefficient in Kyushu region, particularly in Fukuoka, and is moderately high in Hokkaido region and Tohoku region, while it has the lowest coefficient in Honshu, except for Tohoku region. The strong tendency in Fukuoka is due to the high concentration of people from Kyushu region and numerous universities. The housing expense ratio is relatively high in Kanto and Chubu regions. The number of persons per household has the only positive value in Tohoku region, while it has a negative value in other areas, especially low in Chubu and Kinki regions. Although the OLS regression analysis showed a negative coefficient for the number of persons per household, there are positive coefficients that vary by region.

Fig. 8.
figure 8

FG Coefficients

Full size image

The Kyushu region has the lowest child welfare expense ratio, whereas it had the highest ratio in the previous TG analysis. This is due to the lack of correlation between TFR and FMI in Kyushu, where TFR remains around 1.5 to 2 regardless of the value of FMI. The relationship between FMI and female employment rate seems to be weak as the value is close to zero in most areas. The ratio of workers in the secondary industry is high in Kyushu, Chubu, and eastern Tohoku regions, possibly due to the presence of semiconductor factories. The rise of the secondary industry is unlikely to be a factor in attracting women to the area, but rather it may attract men who, in turn, create new economic zones that increase the population of both men and women (Fig. 9).

Fig. 9.
figure 9

Map of Japan with FG coefficients displayed

Full size image

3.6 Successful Policies for Region Maintaining Population

In this section, we examine policies adopted by regions with high TFR and FMI. The first region is Nagaizumi Town, Shizuoka Prefecture. Located about 100 km from Tokyo and 300 km from Osaka, the town has a well-developed transportation network and high TFR and FMI. Instead of implementing measures to combat the declining birthrate, the town focused on attracting companies to establish a stable financial base and utilized its transportation convenience to develop the town, leading to an increase in population and birth rates. By keeping companies local, the population has remained steady.

The second region is Tsukuba City, Ibaraki Prefecture. Located about 40 km from Tokyo, the city has a day/night population ratio of 86% and serves as a bedroom community. With 20 university campuses within a 20 km radius, the city is expected to attract many students. The city also focuses on attracting companies and promoting the relocation of factories and research institutes. Childcare support is also a priority.

After examining successful municipalities and their policies to prevent declining birthrates, it was found that the common approach was to reduce outflows or increase inflows of population by attracting companies. This approach was often paired with childcare support policies that raise the TFR. By increasing the number of children in the numerator while maintaining the population in the denominator of the TFR, the government has succeeded in maintaining the population.

On the other hand, according to Maeda (2005), Nagaoka City in Niigata Prefecture and Tohno City in Iwate Prefecture focused on child-rearing support in order to counter the declining birthrate, and although they were able to improve TFR as a result, their future populations declined, indicating that they could prevent the declining birthrate in the long run [3]. The reason for this is thought to be that the government skipped the first step and started with the third step.

4 Discussion

The result of TG provides a good explanation of TFR, but the factors contributing to TFR vary by region, requiring tailored policies rather than a uniform approach. The west high/east low trend observed in the intercept cannot be fully explained by the explanatory variables. For instance, in the Kanto region, TFR does not rise as high as in west Japan, and policies to promote women’s in-migration could improve TFR, given the highest coefficient was FMI. Economic factors such as child welfare expense and average annual salary had little impact on TFR, indicating other factors such as working hours, opportunities for social interaction, and psychology could be contributing to the decline. The coefficient of the percentage of female legislators was negative, contrary to expectations, and the ratio of workers in the tertiary industry negatively impacted FMI due to longer working hours and less personal time. The coefficient of the percentage of university graduates was high in most regions, indicating increasing employment opportunities for university graduates could increase FMI. However, in the Kanto region, the intercept was extremely high, and the coefficient of TFR was the largest, suggesting that improving TFR could lead to even higher FMI. Ultimately, focusing on the intercept and coefficient can provide insight into raising TFR and FMI and addressing declining birth rates in each region.

5 Conclusions

This study aimed to examine the decline in fertility by focusing on TFR and FMI. To achieve this goal, OLS regression analysis was conducted to identify the essential variables that explain TFR and FMI. We used GWR analysis to vary the intercept and coefficients, leading to a high explanatory power model that could be tailored to different regions. By analyzing successful municipalities, we observed a common trend of enhancing FMI and TFR to improve the birthrate. Future studies should explore variables that can explain the west high/east low trend in TFR and establish a causal relationship instead of relying on correlation. In addition, an analysis accounting for the impact of COVID-19 will be necessary.