Spatial spillover effects of fiscal expenditure and economic opportunity on internal migration in Taiwan

Abstract

The purpose of this study is to understand the determining factors of in-migrant population registration in various counties or cities in Taiwan. Panel data from counties and cities in Taiwan from 2000 to 2021 were adopted, and spatial econometric analysis was conducted using the spatial Durbin model. In terms of spatial spillover effects, the in-migrant population of each county or city is influenced by the fiscal expenditure and economic opportunity factors of neighboring counties or cities, such as (1) expenditure on education, science and culture, (2) number of employed population, (3) average annual regular income per household and (4) average disposable income per household.

Keywords

fiscal expenditure economic opportunity in-migration spatial autocorrelation spatial Durbin model

Introduction

Government fiscal policy is one of the primary means through which governments achieve economic policy goals and provide public resources and services. The impact of public resources and services on population mobility, especially on the family-oriented mobility of the population, is not a simplistic social process. Studies on the influence of public resources and services on population mobility can be traced back to the “vote with their feet” theory proposed by Tiebout. Tiebout (1956) introduced the theory of “vote with their feet,” suggesting that consumers (voters) “vote” with their mobility by choosing localities or communities, implying that each locality or community has a mix of taxes and public goods. This system of “voting with feet” can mimic the operation of market mechanisms. Tiebout (1956) posited that generally speaking, wherever a locality or community can effectively provide public products and services that align with the consumer preferences of certain individuals, it will experience an influx of population; if it cannot do so effectively, it will face the phenomenon of population outflow. Tiebout’s proposition initiated the prologue of the study in local public economics and has become an important reference in urban economics research. Subsequently, Oates (1969) conducted empirical research and development on the “vote with their feet” theory. The core of his research can be summarized as the capitalization of public services. The basic logic of his study is that the level of local public services will affect the residential choices of inhabitants.

The issue of regional disparity has always been one of the significant concerns for geographers, economists and government administrators (Gu et al., 2021). Regional disparity is a universal phenomenon in economic development, and the internal migration of residents directly reflects the economic and social developmental disparities (Gu et al., 2021). Cross-regional internal migration has a tremendous impact on regional economic and social development (Gu et al., 2021). Internal migration—the mobility of residents between different locations and regions within a country—results in a major redistribution of the national population (Charles-Edwards et al., 2019). However, in terms of the spatial dependency of domestic population migration, the migration behavior of the resident population in a geographical region is not independent; it has some spatial correlation with adjacent geographical areas. In other words, the distribution of domestic population migration behavior has certain spatial patterns. The population migration in different geographical regions within a country will be affected by some local factors and the neighborhood effect, meaning the influences of adjacent areas.

There is a wide range of studies about internal migration exploring different influential factors. Some studies done in Asia include the following: Zhao’s (2023) study, which posits that young people migrating to major cities like Beijing, Shanghai, Guangzhou and Shenzhen in China show a positive correlation with income realization; the economic benefits obtained by young people are greater compared to migrating to other places within the country. This study also validated the “escalator region hypothesis”¹ first proposed by Fielding (1992). Lu (2023) explored the causal impact of the equalization of basic public health and medical services in China on the long-term urban settlement intentions of the floating population. Zhou and Guo (2023) argue that internal population mobility in China has led to a large influx of rural populations into cities, and compared to urban residents, they face more structural and policy-based disadvantages in the cities. In Indonesia, Malamassam et al. (2021) suggest that highly educated people are mainly concentrated in large cities or metropolitan areas. The migration decision-making process of educated migrants to smaller cities is driven by promising career developments. Yu et al. (2019) believe that employment opportunities and wages have led to an imbalance in China’s population migration patterns, although amenities have also become an important pull factor for migration. Kone et al. (2018) think that policies between different states within India create obstacles to domestic resident migration, such as those related to healthcare and education. Park and Kim (2015) explored the factors affecting the migration decisions of elderly people in South Korea. Additionally, some studies explore the impact caused by domestic internal migration, including Long et al. (2022) who believe that during the process of internal migration in China, the modes of production and consumption usually change, ultimately leading to alterations in ecological footprints. Ma and Tang (2020) researched the impact of population mobility between cities in China on local welfare, suggesting that further relaxing the Hukou restrictions in major Chinese cities could improve the welfare of local residents.

However, upon reviewing past literature, there is relatively limited research discussing the influencing factors of internal migration behaviors within Taiwan. Taiwan currently has six special municipalities, 13 counties and three cities, making up a total of 22 administrative regions as the highest level of local autonomous entities. Additionally, among the studies with Taiwan as the background, fewer studies utilize the perspective of spatial effects to explore relevant issues. Hence, this paper primarily analyzes two aspects: Local fiscal expenditure and economic opportunities.

Firstly, regarding fiscal expenditure, past research has already discussed that there is a significant relationship between government spending and population migration (Bonin et al., 2000; Day, 1992; Fox et al., 1989; Tiebout, 1956). Based on Tiebout’s (1956) “vote with their feet” theory, this thesis specifically explores the impact of expenditures on economic development, education, science and culture, social welfare, community development and environmental protection on in-migrant populations.

Secondly, in terms of economic opportunities, previous studies have elucidated that economic opportunities have a significant impact on population migration (Lee, 1966; Ravenstein, 1885; Todaro, 1969). The studies on the impact of economic factors on population migration are mostly focused on the influence of economic development opportunities. This paper, based on Ravenstein’s (1885) “The Laws of Migration” and Todaro’s (1969) “Model of Labor Migration,” specifically investigates the impact of the number of registered companies, employed population, average annual regular income per household and average disposable income per household of in-migrant populations. In addition, the empirical model of this study also includes medical and educational-related control variables.

Therefore, the research objectives of this study are as follows: (1) to explore whether there is an overall spatial autocorrelation clustering phenomenon of the in-migrant populations in various counties and cities in Taiwan, and the clustering conditions of specific counties and cities; (2) to investigate the direct effects of local fiscal expenditure and economic opportunities on the in-migrant populations in local counties and cities; and (3) to study the spatial spillover effects of neighboring counties’ and cities' fiscal expenditure and economic opportunities on the in-migrant populations in the local counties and cities.

The next section will elucidate the development of the hypotheses, the data and research sample, the research variables and empirical models. The results of the empirical analysis include descriptive statistics and the results of correlation analysis, spatial autocorrelation tests, LISA cluster diagrams, and the analysis results of the spatial Durbin model, direct effects and spatial spillover effects. Finally, the research findings and policy implications concerning internal migration in Taiwan are presented.

Literature review

Fiscal expenditure and in-migration population

Local finance is an essential component of national finance, and the fiscal efficiency of local governments directly determines the quality of local public services. It reflects the level of administrative efficiency of local governments and is related to the vital interests of the masses. A region with more comprehensive environmental regulations on infrastructure not only influences its development but also has greater attractiveness to the capital, labor and other production elements of neighboring areas (Boarnet, 1998). Government expenditure is one of the primary tools of macroeconomic stabilization policies. Regional differences in living function or quality are also one of the significant factors affecting population migration. Through fiscal expenditure, various constructions can be carried out locally to create a suitable living environment and promote various developments to attract more in-migrants. Past studies have shown that regional living function or quality does have a substantial impact on population migration (Cushing, 1987; Porell, 1982). For example, Porell (1982) believed that people tend to migrate to regions with higher levels of public services, that is, areas with better living function or quality. Cushing (1987) found that people value regional comfort and would rather forsake wages to choose to live in more comfortable areas. Regarding the air quality of the living environment, Zhang et al. (2022) believe that, compared to cities with dirty air, migrant workers are more likely to settle in cities with clean air. Air pollution only affects the permanent settlement intentions of the floating population and does not affect their short-term stay intentions.

Moreover, Fox et al. (1989) and Day (1992) posit that the differences in the scale and structure of fiscal expenditure have diverse impacts on the migration behavior of residents. Bonin et al. (2000) further point out that fiscal expenditure elevates the motivation for population migration by directly affecting the level of public services. However, excessive in-migration can have effects on the economy and society of the destination area. Therefore, the fiscal policies and other administrative measures employed by local governments not only have direct or indirect impacts on the endowment and mobility of production elements, such as labor, capital, technology and information, but also bring significant external effects to the living environment. Naturally, this would attract the population from other counties or cities to register as residents in this county or city. Gu et al. (2021) believe that per capita fiscal expenditure is positively correlated with the willingness to migrate in. Hence, this study establishes the following hypothesis (direct effect):

Hypothesis 1a

An increase in the fiscal expenditure of this county or city has a significant impact on the in-migration population of this county or city.

Case et al. (1993) explored the interrelated effects of fiscal policy and budget spillovers in the United States (USA). They suggested that neighbor effects could be considered as the degree of similarity between local governments, implying that public expenditure can influence each other, primarily due to certain similar characteristics existing among local governments. The higher the similarity between local governments, the stronger the spillover or demonstrative effects likely generated by each other’s public expenditure. Through empirical analysis, Pereira and Roca-Sagalés (2003) and Cohen and Morrison Paul (2004) found that the public expenditure of a region, especially infrastructure construction, has a positive spatial spillover effect on the economic growth of neighboring areas. López et al. (2017) believe that the key function of local governments is to provide a wide range of public services, and these services create spatial spillovers between neighboring cities. Kameda et al. (2021) also argue that due to the strong interdependence of local economies in the absence of border effects, government expenditure in local economies easily spills over to other local economies. Banzhaf and Walsh (2008) validated Tiebout’s theory, suggesting that improvements in public goods lead to an increase in population density in neighboring areas. Therefore, this study establishes the following hypothesis (spatial spillover effect):

Hypothesis 1b

An increase in the fiscal expenditure of neighboring counties or cities has a significant impact on the in-migration population of this county or city.

Economic opportunity and in-migration population

Migration refers to the movement of migrants between two places over a period of time and can be viewed as a behavior of people changing their residence (Clark et al., 1996). Additionally, it is a reaction to the economic, social and demographic forces in the environment (Bogue, 1969). In earlier literature, after observing the data on the place of birth and residence of the population, Ravenstein (1885) proposed the “Laws of Migration,” believing that the factors attracting migration are mostly economic incentives (dominance of economic motive). Under this theory, the number of migrants increases with the development of industry, commerce and transportation, and long-distance migration generally tends to move to industrially and commercially developed urban centers.

On one hand, the viewpoint of neoclassical economics explains the reasons for migration from economic factors, believing that differences in the labor market, such as job opportunities and wage structure differences, are the main factors causing population movement, reflecting a kind of “labor migration.” For instance, Hicks (1932) regarded economic benefits as the main influencing factor for population migration, and Lee (1966) noticed that human migration behavior is affected by push and pull factors and established a push-pull system. The push-pull system explains that in a market economy, with free movement of population, people migrate and emigrate to improve living conditions—that is, to have better economic opportunities (Lee, 1966). Todaro (1969) proposed a model of labor migration, suggesting that labor in rural areas would be influenced by job development opportunities in urban areas, generating migratory behavior. Davies et al. (2001) explored inter-state population migration in the USA from 1986 to 1997, considering that per capita income has a positive impact on in-migration, indicating that people tend to migrate to regions with relatively high average income or expected higher economic opportunities.

Meanwhile, the “new economics of labor migration” theory believes that the purpose of migration is to maximize household income (Ryndyk, 2020). Conway and Houtenville (1998) investigated the influence of living costs, amenities and the public sector on immigration decisions. Chen and Coulson (2002) explored the reasons for the change in the number of urban population migrations in China, thinking that the development environment and development potential created by enterprises have a significant impact on attracting immigrants. Yu et al. (2019) investigated the impact of amenities and economic opportunities on migration patterns in China, finding that most migrants tend to migrate to major cities in the eastern coastal areas and a few major cities inland, to seek more development opportunities. Li and Pan (2023) believed that Foreign Direct Investment (FDI) encourages internal migration by investing in labor-intensive export sectors, and this impact varies over time and space. Huo et al. (2016) proposed a dynamic model based on migration rules, which considered migration cost as a function of migration distance, and higher economic benefits would promote migration. Chen and Coulson (2002) believed that per capita urban total income has a significant impact on migration. Therefore, this study establishes the following hypothesis (direct effect):

Hypothesis 2a

The creation of economic opportunities in this county or city has a significant impact on the in-migrant population of this county or city.

As previously discussed, the spillover effect of environmental regulations has always been the focus of many scholars (Boarnet, 1998; Cohen and Morrison Paul, 2004; Frère et al., 2014; López et al., 2017; Pereira and Roca-Sagalés, 2003). When the living environment or economic development opportunities in neighboring counties or cities are worse than those in the local county or city, residents of neighboring counties or cities will migrate to the local county or city to seek better economic development opportunities or a more comfortable living environment. Therefore, this study establishes the following hypothesis (spatial spillover effect):

Hypothesis 2b

The creation of economic opportunities in neighboring counties or cities has a significant impact on the in-migrant population in this county or city.

Methodology

Data and sample

This study utilizes panel data from the years 2000 to 2021 with all 22 administrative regions in Taiwan as samples. The data are mainly derived from the County/City Major Statistical Indicators Query System found in the Statistical Information Network, which is published annually by the Directorate General of Budget, Accounting and Statistics (DGBAS) of the Executive Yuan (2023). Table 1 shows the statistical data of in-migration and out-migration populations in Taiwan from 2000 to 2021. In 2000, the in-migrant population in each county and city of Taiwan was 1,381,843, which has been decreasing ever since. After the outbreak of COVID-19 in 2020, the in-migrant population in each county and city dropped to 946,251 and further decreased to 920,703 in 2021. On the other hand, the out-migrant population in each county and city of Taiwan was 1,376,912 in 2000 and has also been decreasing ever since. It reached a historical low of 939,240 in 2017. After the outbreak of COVID-19 in 2020, the out-migrant population in each county and city decreased to 980,229 and then increased again to 1,076,713 in 2021.

Table 1.

Numbers of in-migrants and out-migrants in Taiwan from 2000 to 2021.

Year	In-migrants	Out-migrants
2000	1,381,843	1,376,912
2001	1,373,363	1,377,174
2002	1,452,575	1,456,261
2003	1,273,526	1,286,021
2004	1,264,937	1,261,692
2005	1,442,018	1,427,213
2006	1,414,340	1,376,816
2007	1,173,040	1,154,510
2008	1,196,407	1,172,845
2009	1,198,561	1,163,548
2010	1,213,899	1,192,662
2011	1,128,449	1,109,372
2012	1,118,006	1,102,326
2013	1,079,702	1,065,212
2014	1,073,783	1,060,001
2015	1,004,108	995,527
2016	964,536	952,829
2017	948,049	939,240
2018	984,815	975,927
2019	972,043	959,325
2020	946,251	980,229
2021	920,703	1,076,713

Source: County/City Major Statistical Indicators Query System found in the Statistical Information Network (DGBAS of the Executive Yuan, 2023).

Table 2 displays the comparison of in-migration and out-migration populations in the six municipalities of Taiwan during the presidential election years, namely, 2000, 2004, 2008, 2012, 2016 and 2020 also lifted from the County/City Major Statistical Indicators Query System (DGBAS of the Executive Yuan, 2023). The highest in-migration and out-migration populations in Taiwan are mostly concentrated in the six municipalities of New Taipei City, Taipei City, Kaohsiung City, Taichung City, Taoyuan City and Tainan City. Regardless of in-migration or out-migration, New Taipei City has the highest population. As of August 2023, the populations of the cities are New Taipei City, 4,034,102; Taipei City, 2,506,788; Taoyuan City, 2,309,770; Taichung City, 2,839,768; Tainan City, 1,859,728; and Kaohsiung City, 2,737,582. Among them, New Taipei City has the highest population and is closely bordered by Taipei City and Taoyuan City, forming the largest living circle in Taiwan. The total population of the three cities is 8,850,660, and this living circle, whether in terms of local government budgets or economic development, is the most developed region in Taiwan.

Table 2.

Statistics of in-migration and out-migration population registrations in Taiwan’s six municipalities across six stages.

Year	City	In-migrants	Out-migrants	Year	City	In-migrants	Out-migrants
2000	New Taipei City	[1] 251,936	[1] 224,816	2012	New Taipei City	[1] 198,847	[1] 196,560
	Taipei City	[2] 192,667	[2] 208,194		Taipei City	[2] 165,994	[2] 156,655
	Kaohsiung City	[3] 189,254	[3] 190,312		Kaohsiung City	[4] 138,552	[3] 140,381
	Taichung City	[4] 171,953	[4] 157,364		Taichung City	[3] 138,734	[4] 131,542
	Taoyuan City	[5] 115,888	[6] 93,076		Taoyuan City	[5] 102,780	[5] 94,813
	Tainan City	[6] 112,821	[5] 114,295		Tainan City	[6] 80,439	[6] 79,635
2004	New Taipei City	[1] 226,483	[1] 214,888	2016	New Taipei City	[1] 161,943	[1] 164,454
	Taipei City	[3] 174,294	[2] 201,261		Taipei City	[3] 127,081	[2] 146,197
	Kaohsiung City	[2] 175,655	[3] 183,079		Kaohsiung City	[4] 117,099	[3] 117,038
	Taichung City	[4] 152,251	[4] 146,285		Taichung City	[2] 128,155	[4] 114,095
	Taoyuan City	[5] 115,448	[6] 91,879		Taoyuan City	[5] 108,929	[5] 78,095
	Tainan City	[6] 98,538	[5] 98,677		Tainan City	[6] 68,842	[6] 67,814
2008	New Taipei City	[1] 217,631	[1] 196,289	2020	New Taipei City	[1] 173,605	[1] 163,028
	Taipei City	[2] 163,599	[2] 175,030		Taipei City	[3] 117,562	[2] 162,002
	Kaohsiung City	[3] 153,778	[3] 154,257		Kaohsiung City	[4] 113,084	[4] 117,567
	Taichung City	[4] 148,328	[4] 140,871		Taichung City	[2] 123,707	[3] 120,288
	Taoyuan City	[5] 106,317	[5] 92,102		Taoyuan City	[5] 105,073	[5] 93,567
	Tainan City	[6] 87,515	[6] 86,304		Tainan City	[6] 68,800	[6] 70,245

Source: County/City Major Statistical Indicators Query System found in the Statistical Information Network (DGBAS of the Executive Yuan, 2023).

Research variables

This research utilizes empirical models, and the definitions and explanations of dependent variables and independent variables are as follows:

Dependent variable

1. In-Migration Population (persons) ( ${I M}_{i t}$ ): Refers to the number of registered in-migrants in the t year in the county or city of Taiwan (not including the number of in-migrants due to address changes).

Independent variables for fiscal expenditure

1. Economic Development Expenditure (in millions of New Taiwan dollars (TWD)) ( ${E D E}_{i t}$ ): Refers to the expenditure related to agriculture, industry, transportation and other economic services undertaken by the county or city in the i county or city of Taiwan in the t year.

2. Education, Science and Culture Expenditure (in TWD millions) ( ${E S C E}_{i t}$ ): Refers to the expenditure related to education, science and culture services and subsidies undertaken by the county or city in the i county or city of Taiwan in the t year.

3. Social Welfare Expenditure (in TWD millions) ( ${S W E}_{i t}$ ): Refers to the expenditure related to social insurance, social assistance, welfare services, national employment, and health care services and subsidies undertaken by the county or city in the i county or city of Taiwan in the t year.

4. Community Development and Environmental Protection Expenditure (in TWD millions) ( ${C E E}_{i t}$ ): Refers to the expenditure related to community development, and environmental protection services and subsidies undertaken by the county or city in the i county or city of Taiwan in the t year.

Independent variables for economic opportunity

1. Number of Company Registrations ( ${N C R}_{i t}$ ): Refers to the number of newly registered companies in the i county or city of Taiwan in the t year, as defined by company law.

2. Employed Population (in thousand persons) ( ${E P}_{i t}$ ): Refers to the number of people aged 15 and over who are engaged in paid work or are unpaid family workers working for 15 hours or more in the i county or city of Taiwan in the t year.

3. Average Annual Regular Income Per Household (TWD) ( ${R I}_{i t}$ ): Refers to the total regular income of an average household in the i county or city of Taiwan in the t year. The calculation formula is (Total Regular Income ÷ Total Number of Households).

4. Average Disposable Income Per Household (TWD) ( ${D I}_{i t}$ ): Refers to the average disposable income per household in the i county or city of Taiwan in the t year. The calculation formula is (Disposable Income ÷ Total Number of Households).

Control variables

In-migrants driven by economic opportunities also consider the quality of life, such as education and health conditions, in cities. Firstly, regarding education conditions, mobile populations with higher education levels are generally more willing to settle in destination cities than their peers (Lao et al., 2022). Those migrating with children have the strongest intention to move internally, considering the education of their children in the destination cities (Lao et al., 2022). Secondly, regarding health conditions, the supply of health services is a significant determinant of elderly migration (Norman and Boyle, 2014). Gu et al. (2022) believe that the provision of health services has a push-pull effect on the mobility of elderly people in China, especially for those whose health conditions are deteriorating with age. Therefore, this research introduces two control variables related to education and health to verify the robustness of the empirical analysis results.

1. Percentage of Civilian Population Aged 15 and Above with College Education or Above (percent) ( ${C P C}_{i t})$ : Refers to the percentage of the civilian population aged 15 and above with college education or above in the i county or city of Taiwan in the t year. The calculation formula is (Number of Civilian Population Aged 15 and Above with College Education or Above ÷ Civilian Population Aged 15 and Above) × 100.

2. Average Number of People Served Per Medical Institution (persons/institution) ( ${A P M}_{i t})$ : Refers to the average number of people served per medical institution in the i county or city of Taiwan in the t year. The calculation formula is (Registered Population ÷ Number of Medical Institutions).

Empirical model

LeSage and Pace (2009) introduced the Spatial Durbin Model (SDM), which includes lagged terms of the dependent variable as well as lagged terms of the independent variables. The SDM constructed in this study is as follows:

{I M}_{i t} = ρ \sum_{j = 1}^{N} W_{i j} {I M}_{j t} + α + β_{1} {E D E}_{i t} + β_{2} {E S C E}_{i t} + β_{3} {S W E}_{i t} + β_{4} {C E E}_{i t} + β_{5} {N C R}_{i t} + β_{6} {E P}_{i t} + β_{7} {R I}_{i t} + β_{8} {D I}_{i t} + β_{9} {C P C}_{i t} + β_{10} {A P M}_{i t} + θ_{1} \sum_{j = 1}^{N} W_{i j} {E D E}_{j t} + θ_{2} \sum_{j = 1}^{N} W_{i j} {E S C E}_{j t} + θ_{3} \sum_{j = 1}^{N} W_{i j} {S W E}_{j t} + θ_{4} \sum_{j = 1}^{N} W_{i j} {C E E}_{j t} + θ_{5} \sum_{j = 1}^{N} W_{i j} {N C R}_{j t} + θ_{6} \sum_{j = 1}^{N} W_{i j} {E P}_{j t} + θ_{7} \sum_{j = 1}^{N} W_{i j} {R I}_{j t} + θ_{8} \sum_{j = 1}^{N} W_{i j} {D I}_{j t} + θ_{9} \sum_{j = 1}^{N} W_{i j} {C P C}_{j t} + θ_{10} \sum_{j = 1}^{N} W_{i j} {A P M}_{j t} + μ_{i} + ε_{i t}

i \neq j

Herein,

{I M}_{i t}

is the dependent variable.

i, j represent the various counties and cities in Taiwan.

t represents the years (t = 2000∼2021).

W is the spatial weight matrix, a square matrix symmetric from top-right to bottom-left, with the number of rows and columns equal to the number of counties and cities (this study includes 22 counties and cities). This study adopts queen contiguity to define spatial adjacency. Queen contiguity implies that counties and cities are adjacent if their edges or corners touch (Sawada, 2004). The “contiguity matrix” is used to define neighboring relationships: If two counties or cities are defined as “neighboring,” its value is 1; otherwise, it is 0, and the diagonal is also 0 (a place does not neighbor itself). Taiwan has three island counties, and these do not have neighboring relationships with other counties, their value is 0. The spatial weight matrix is defined as

W_{i j} = {\begin{cases} 1, \\ 0, \end{cases} \begin{array}{l} r e g i o n s i a n d j a r e n e i g h b o r i n g r e g i o n s \\ r e g i o n s i a n d j a r e n o t n e i g h b o r i n g r e g i o n s \end{array}

$ρ$ is the spatial lag coefficient, the coefficient of the spatially autocorrelated term of the dependent variable, reflecting the direction and degree of the impact of the dependent variable ${I M}_{j t}$ in neighboring counties or cities on the dependent variable ${I M}_{i t}$ of the county or city in question. By testing the spatial lag coefficient $ρ$ , we can further explore the spatial spillover effects of neighboring counties or cities on the county or city in question. A being statistically significant suggests a clear spatial dependence between dependent variables. If $ρ$ ≠ 0, it indeed has a spatial relationship with neighboring counties or cities. $ρ$ > 0 indicates the presence of positive spatial spillover effects, creating positive spatial spillover. The magnitude of $ρ$ reflects the degree of spatial diffusion or spatial spillover interactions among the various counties or cities.

$W_{i j} {I M}_{j t}$ is the spatially autocorrelated term matrix of the dependent variable and is an endogenous variable, showing the impact of the dependent variable ${I M}_{j t}$ in neighboring counties or cities on the dependent variable ${I M}_{i t}$ of the county or city in question.

$α$ is the constant term.

$β$ is the coefficient to be estimated, representing the original effect coefficient.

${E D E}_{i t}$ , ${E S C E}_{i t}$ , ${S W E}_{i t}$ , ${C E E}_{i t}$ , ${N C R}_{i t}$ , ${E P}_{i t}$ , ${R I}_{i t}$ ${D I}_{i t}$ , ${C P C}_{i t}$ and ${A P M}_{i t}$ are independent variables.

$β_{1} {E D E}_{i t}$ , $β_{2} {E S C E}_{i t}$ $β_{3} {S W E}_{i t}$ $β_{4} {C E E}_{i t}$ , $β_{5} {N C R}_{i t}$ , $β_{6} {E P}_{i t}$ , $β_{7} {R I}_{i t}$ , $β_{8} {D I}_{i t}$ , ${β_{9} C P C}_{i t}$ ,, and $β_{10} {A P M}_{i t}$ reflect the impact of each independent variable on ${I M}_{i t}$ .

$θ$ is the coefficient to be estimated, the coefficient of the spatially autocorrelated term of the independent variable, reflecting the impact of all independent variables ${E D E}_{j t}$ , ${E S C E}_{j t}$ , ${S W E}_{j t}$ , ${C E E}_{j t}$ , ${N C R}_{j t}$ , ${E P}_{j t}$ , ${R I}_{j t}$ , ${D I}_{j t}$ , ${C P C}_{j t}$ and ${A P M}_{j t}$ in neighboring counties or cities on the dependent variable ${I M}_{i t}$ of the county or city in question. If the symbol of $θ$ is positive, it indicates that the neighboring county or city effect has a positive impact on the dependent variable ${I M}_{i t}$ of the county or city in question, demonstrating the existence of spatial spillover effects between neighboring counties or cities. If the symbol of $θ$ is negative, it means there are competitive effects between neighboring counties or cities.

${W_{i j} E D E}_{j t}$ , $W_{i j} {E S C E}_{j t}$ , ${W_{i j} S W E}_{j t}$ , $W_{i j} {C E E}_{j t}$ , $W_{i j} {N C R}_{j t}$ , $W_{i j} {E P}_{j t}$ , $W_{i j} {R I}_{j t}$ , $W_{i j} {D I}_{j t}$ , $W_{i j} {C P C}_{j t}$ and $W_{i j} {A P M}_{j t}$ are the spatially autocorrelated term matrices of the independent variables, representing all independent variables’ spatial effects, showing the impact of all independent variables in neighboring counties or cities on the dependent variable ${I M}_{i t}$ of the county or city in question.

$μ_{i}$ is the spatial (individual) effect, the individual effect of county or city i.

$ε_{i t}$ is the independently and identically distributed random error term, the spatially autocorrelated error term.

Results

Descriptive statistics

Table 3 lists the variables in the model of this study. Table 4 lists the analysis result. The variables used in the table were based on raw data.

Table 3.

Summary of descriptive statistics.

Variables	Obs.	Mean	Std. Dev.	Min.	25th Percentile	Median	75th Percentile	Max.
IM	484	52,737.51	60,475.29	684.00	15,498.50	23,442.50	83,028.50	266,683.00
EDE	484	7,027.69	7,420.06	541.91	2,484.00	4,135.07	8,063.99	46,357.08
ESCE	484	14,875.77	16,344.70	400.57	4,604.01	7,359.75	18,488.28	73,161.88
SWE	484	5,947.65	7,899.58	180.82	1,698.41	2,727.11	5,628.98	44,720.84
CEE	484	2,340.23	3,722.81	63.44	307.43	674.69	1,881.97	20,164.01
NCR	484	28,687.92	44,824.14	37.00	3,884.00	6,070.50	31,705.50	184,128.00
EP	484	481.46	494.68	2.00	148.00	251.00	842.50	2,002.00
RI	484	1,107,636.44	238,909.22	715,008.00	932,404.00	932,404.00	1,221,556.00	1,861,512.00
DI	484	881,097.71	176,661.52	568,409.00	748,706.50	844,582.00	974,217.00	1,430,572.00
CPC	484	32.11	12.13	8.56	23.72	30.22	37.95	80.22
APM	484	1,345.83	442.37	647.21	1,303.63	1,303.63	1,449.54	3,364.50

Notes: TWD: New Taiwan dollar (equal to USD 0.031).

Obs.: Observations. IM: In-Migration Population (in persons); EDE: Economic Development Expenditure (in million TWD); ESCE: Education, Science, and Culture Expenditure (in million TWD); SWE: Social Welfare Expenditure (in million TWD); CEE: Community Development and Environmental Protection Expenditure (in million TWD); NCR: Number of Company Registrations; EP: Employed Population (in thousand persons); RI: Average Annual Regular Income Per Household (TWD); DI: Average Disposable Income Per Household (TWD); CPC: Percentage of Civilian Population Aged 15 and Above with College Education or Above (%); APM: Average Number of People Served Per Medical Institution (persons/institution).

Table 4.

Pearson correlation analysis.

	IM	EDE	ESCE	SWE	CEE	NCR	EP	RI	DI	CPC	APM
IM	1
EDE	0.750^**	1
ESCE	0.849^**	0.900^**	1
SWE	0.711^**	0.852^**	0.932^**	1
CEE	0.803^**	0.871^**	0.932^**	0.916^**	1
NCR	0.887^**	0.842^**	0.947^**	0.873^**	0.921^**	1
EP	0.938^**	0.812^**	0.913^**	0.793^**	0.849^**	0.885^**	1
RI	0.355^**	0.416^**	0.497^**	0.492^**	0.527^**	0.548^**	0.343^**	1
DI	0.343^**	0.414^**	0.492^**	0.485^**	0.517^**	0.538^**	0.338^**	0.993^**	1
CPC	0.338^**	0.488^**	0.568^**	0.594^**	0.581^**	0.575^**	0.403^**	0.734^**	0.730^**	1
APM	−0.353^**	−0.331^**	−0.414^**	−0.378^**	−0.364^**	−0.390^**	−0.387^**	−0.246^**	−0.221^**	−0.546^**	1

Notes: ^**p < 0.01.

TWD: New Taiwan dollar (equal to USD 0.031).

IM: In-Migration Population (in persons); EDE: Economic Development Expenditure (in million TWD); ESCE: Education, Science, and Culture Expenditure (in million TWD); SWE: Social Welfare Expenditure (in million TWD); CEE: Community Development and Environmental Protection Expenditure (in million TWD); NCR: Number of Company Registrations; EP: Employed Population (in thousand persons); RI: Average Annual Regular Income Per Household (TWD); DI: Average Disposable Income Per Household (TWD); CPC: Percentage of Civilian Population Aged 15 and Above with College Education or Above (%); APM: Average Number of People Served Per Medical Institution (persons/institution).

Spatial autocorrelation test results

Moran’s I values range between [−1, 1]. A value closer to 1 indicates a stronger positive spatial autocorrelation, and a value closer to −1 indicates a stronger negative spatial autocorrelation (Moran, 1950). Observing from Table 5, more than half of the Moran’s I values have reached a significance level of 0.05. Except for 2005, the number of in-migrant households in various counties and cities in Taiwan from 2000 to 2021 all reached a significance level of 0.1. Therefore, this study is still suitable for analysis using spatial econometric models. The trend of spatial autocorrelation in the number of in-migrant households in Taiwan over the past 22 years shows a gradual decline from 2010, revealing that the degree of spatial autocorrelation among various counties and cities in Taiwan is weakening over time.

Table 5.

Spatial autocorrelation indicators from 2000 to 2021.

Year	Moran’s I		Year	Moran’s I
Year	I	p-value	Year	I	p-value
2000	0.279	0.054	2011	0.341	0.021
2001	0.273	0.053	2012	0.328	0.026
2002	0.261	0.069	2013	0.312	0.034
2003	0.286	0.048	2014	0.311	0.035
2004	0.286	0.049	2015	0.275	0.061
2005	0.143	0.262	2016	0.272	0.062
2006	0.281	0.055	2017	0.271	0.063
2007	0.298	0.042	2018	0.289	0.048
2008	0.290	0.046	2019	0.295	0.043
2009	0.324	0.026	2020	0.281	0.052
2010	0.359	0.016	2021	0.259	0.070

Moran’s I does not display the spatial clustering between different counties and cities and cannot illustrate the spatial autocorrelation features of the cities, but Local Indicators of Spatial Association (LISA) can supplement this gap. Anselin (1995) divided LISA into four quadrants according to the degree of spatial clustering, representing the spatial relationship between the local county and neighboring counties. The results of LISA statistical testing are presented through maps, distinguishing regions reaching significance with different colors, allowing observation of variations in LISA clustering at different times and understanding the changing spatial structure over time.

In the LISA clustering map, the first quadrant represents hot spots where both the local and neighboring counties have high in-migration, denoted as High-High (HH). The third quadrant represents cold spots where both have low in-migration, denoted as Low-Low (LL). The second quadrant represents a low value surrounded by high values, with the local county having low in-migration and neighboring counties high, denoted as Low-High (LH). The fourth quadrant represents a high value surrounded by low values, with the local county having high in-migration and neighboring counties low, denoted as High-Low (HL).

Figure 1 displays a LISA cluster map in six phases to illustrate the changes in the in-migration population across various counties and cities in Taiwan. The years represented by these six stages all coincide with the years when presidential elections were held. From 2000 to 2020, Taiwan experienced two political party turnovers (Central Election Commission, 2023). Some counties and cities underwent changes through central political party alternations. For instance, before 2016, whether the Kuomintang (KMT) or the Democratic Progressive Party (DPP) governed at the central and local levels, the clustering condition in Yilan County was not significant. However, after 2016, the clustering status of Yilan County belongs to Low-High (LH), indicating that the in-migrant population of Yilan County was lower while the in-migrant population of neighboring counties and cities was higher after 2016. Meanwhile, from 2000 to 2020, regardless of which political party governed at the central and local levels, Keelung City always exhibited a Low-High (LH) clustering status, indicating a lower in-migrant population in Keelung City and a higher in-migrant population in neighboring counties and cities during these years.

Figure 1.

LISA cluster diagrams of in-migrant household changes in various counties and cities of Taiwan across six stages.

SDM analysis results

LeSage and Pace (2009) and Elhorst (2010) contend that the SDM can be simplified to either the spatial lag model (SLM) or the spatial error model (SEM). Model selection is conducted through testing the following hypotheses. If H₀: $θ$ = 0 is rejected, SDM cannot be simplified to SLM, making the selection of SDM more reasonable. If H₀: $θ$ = 0 is rejected, SDM cannot be simplified to SEM; hence, choosing SDM is more reasonable. Between SLM and SDM with spatial fixed-effects, the Wald test shows $x^{2}$ = 47.38, p < 0.001, and the Likelihood-ratio test shows $x^{2}$ = 45.10, p < 0.001, suggesting that choosing SDM with spatial fixed-effects is more reasonable. Similarly, between SEM and SDM with spatial fixed-effects, the Wald test indicates $x^{2}$ = 40.32, p < 0.001 and the Likelihood-ratio test reveals $x^{2}$ = 38.77, p < 0.001, leading to the selection of SDM with spatial fixed-effects as more suitable. Likewise, comparisons between SDM with time fixed-effects, SLM and SEM reveal that SDM with time fixed-effects is the more reasonable selection. Comparing SDM with spatial and time fixed-effects, SLM and SEM, analysis indicates that choosing SDM with spatial and time fixed-effects is more suitable. Between SDM with random-effects, SLM and SEM, the analysis results suggest choosing SDM with random-effects as more reasonable.

Table 6 presents the estimation results of four SDMs. This study must choose among these four models. Regarding the overall fit, spatial econometric models are estimated using the maximum likelihood estimation (MLE), thereby deriving the maximum likeness and non-linear test statistics in the models and maximizing their coefficients. The fit tests based on non-linear principles such as Log Likelihood (LIK), Akaike’s (1973) Akaike Information Criterion (AIC) and Schwarz’s (1978) Schwartz Bayesian Information Criterion (SBC or BIC) can be used as indicators to check the fitting degree. The model with the highest Log Likelihood value or the smallest AIC or BIC value is deemed the best model.

Table 6.

Spatial Durbin model analysis results.

Variables	Model 1 SDM with spatial fixed-effects		Model 2 SDM with time fixed-effects		Model 3 SDM with spatial and time fixed-effects		Model 4 SDM with random-effects
Variables	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
EDE	0.989^***	0.134	−0.033	0.216	0.874^***	0.136	0.946^***	0.138
ESCE	−0.140	0.181	−1.254^***	0.286	−0.246	0.191	−0.201	0.186
SWE	−0.016	0.202	−1.559^***	0.315	0.002	0.199	−0.067	0.206
CEE	−1.150^*	0.482	1.851^**	0.683	−0.737	0.484	−1.183^*	0.493
NCR	0.108	0.179	0.651^***	0.087	0.141	0.183	0.178	0.181
EP	−178.473^***	16.736	114.004^***	6.781	−174.254^***	16.333	−165.675^***	17.234
RI	0.030	0.020	0.045	0.028	0.038	0.020	0.028	0.021
DI	−0.027	0.025	−0.061	0.037	−0.036	0.026	−0.026	0.026
CPC	−184.932	153.517	−134.055	195.961	−841.396^**	268.960	−170.564	156.878
APM	2.060	2.938	0.019	2.247	1.789	2.935	1.680	3.005
W × EDE	0.081	0.200	−0.621	0.334	0.055	0.209	0.089	0.203
W × ESCE	−1.122^***	0.256	−1.780^***	0.460	−1.260^***	0.304	−1.126^***	0.262
W × SWE	0.312	0.296	1.138^*	0.537	0.213	0.329	0.425	0.304
W × CEE	0.540	0.612	1.495	1.041	1.246	0.649	0.411	0.628
W × NCR	−0.048	0.281	−0.567^*	0.253	−0.167	0.319	−0.029	0.286
W × EP	87.480^**	25.511	54.681^**	17.730	86.014^**	28.809	82.615^**	25.254
W × RI	0.080^**	0.026	0.062	0.043	0.115^***	0.028	0.081^**	0.027
W × DI	−0.073^*	0.035	−0.042	0.061	−0.112^**	0.038	−0.073^*	0.036
W × CPC	−203.442	190.533	−913.031^**	293.331	−486.157^*	208.803	−220.222	194.804
W × APM	8.540	9.623	1.182	4.675	12.293	9.763	11.919	9.682
Constant							51179.79	31242.00
n		484		484		484		484
Spatial $ρ$	0.245^***	0.056	0.180	0.132	0.128^*	0.064	0.247^***	0.056
Log likelihood	−5021.24		−5291.47		−5000.94		−5127.25
AIC	10064.49		10598.93		10023.88		10276.49
BIC	10110.49		10632.39		10069.89		10322.50

Notes: ^*p < 0.05; ^**p < 0.01; ^***p < 0.001.

SDM: Spatial Durbin model.

Coef.: Coefficient; Std. Err.: Standard Error.

W is a spatial weight matrix.

TWD: New Taiwan dollar (equal to USD 0.031).

Model 3, with its Log likelihood, AIC and BIC values, outperforms Model 1, Model 2 and Model 4. Notably, the spatial lag coefficient $ρ$ of Model 3 reaches a significant level ( $ρ$ = 0.128, p < 0.05), indicating significant spatial autocorrelation in the distribution of in-migrants across the various counties and cities, thereby reaffirming the rationality of incorporating spatial effects into econometric models. Therefore, this study ultimately decides to adopt Model 3 (SDM with spatial and time fixed-effects) as the basis for analysis.

Decomposition results of the SDM with spatial and time fixed-effects

LeSage and Pace (2009) pointed out that in SDMs considering spatial interaction effects, utilizing the regression results of spatial estimation parameters directly to judge the existence of spatial spillover effects might lead to incorrect conclusions due to the presence of feedback effects (FE). Since SDMs incorporate spatially lagged independent and dependent variables, coefficient estimation results cannot directly reflect their marginal effects and are difficult to accurately measure the direct impact of independent variables on dependent variables (Elhorst, 2010). Therefore, it is essential to decompose SDMs into direct effects, indirect effects (spatial spillover effects) and total effects. Table 7 displays the results of decomposition of the SDM with spatial and temporal fixed-effects.

Table 7.

Direct, indirect, and total effects of SDM with spatial and time fixed-effects.

Variables	Direct Effect		Indirect Effect		Total Effect
Variables	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Independent variables of fiscal expenditure
EDE	0.883^***	0.141	0.151	0.194	1.034^***	0.272
ESCE	−0.290	0.186	−1.226^***	0.320	−1.516^***	0.401
SWE	0.030	0.188	0.180	0.307	0.211	0.305
CEE	−0.728	0.469	1.136	0.656	0.407	0.873
Independent variables of economic opportunity
NCR	0.143	0.173	−0.156	0.297	−0.013	0.348
EP	−172.720^***	16.295	62.296^*	25.266	−110.424^***	26.400
RI	0.041	0.021	0.117^***	0.029	0.158^***	0.040
DI	−0.040	0.026	−0.115^**	0.039	−0.155^**	0.051
Control variables
CPC	−836.219^**	266.924	−572.428^*	246.131	−1408.647^***	402.595
APM	2.137	2.958	12.000	9.409	14.137	9.985

Note: ^*p < 0.05; ^**p < 0.01; ^***p < 0.001.

Coef.: Coefficient; Std. Err.: Standard Error.

TWD: New Taiwan dollar (equal to USD 0.031).

Regarding the determinant factors of in-migration population related to fiscal expenditures, firstly, an increase in expenditures for economic development (in TWD millions) will drive the growth of the in-migrant population in the local counties or cities. Every additional unit increase in economic development expenditure will lead to a 0.883 unit increase in the in-migrant population in local counties or cities. However, the growth in economic development expenditure does not have a significant impact on neighboring counties or cities. Secondly, increases or decreases in expenditure on education, science and culture (in NTD millions) will not affect the growth of the in-migrant population in the local counties or cities. However, it brings about a negative spatial spillover effect on neighboring counties or cities. For every additional unit increase in education, science and culture expenditure in neighboring counties or cities, the in-migrant population in the local counties or cities decreases by 1.226 units.

In terms of economic opportunities affecting in-migration decisions, firstly, the growth in employment population (in thousand persons) will suppress the growth of the in-migrant population in local counties or cities. For every additional unit increase in the employment population in local counties or cities, the in-migrant population decreases by 172.72 units. However, the growth in employment population brings a positive spillover effect on neighboring counties or cities, with every additional unit increase leading to a 62.296 unit increase in the in-migrant population in local counties or cities. Secondly, increases or decreases in average annual regular income per household (TWD) do not affect the growth of the in-migrant population in the local counties or cities. However, it brings a positive spillover effect on neighboring counties or cities. For every additional unit increase in average annual regular income per household in neighboring counties or cities, the in-migrant population in the local counties or cities increases by 0.117 units. Thirdly, increases or decreases in average disposable income per household (TWD) do not impact the growth of the in-migrant population in the local counties or cities. However, it brings a negative spillover effect on neighboring counties or cities. For every additional unit increase in average disposable income per household in neighboring counties or cities, the in-migrant population in the local counties or cities decreases by 0.115 units.

Concerning control variables, the growth in the proportion (percent) of the civilian population aged 15 and above with an education level of college or above will suppress the growth of the in-migrant population in the local counties or cities. For every additional unit increase in this proportion in local counties or cities, the in-migrant population decreases by 836.219 units. Additionally, the growth in this proportion brings a negative spillover effect on neighboring counties or cities. For every additional unit increase in neighboring counties or cities, the in-migrant population in the local counties or cities decreases by 572.428 units.

Conclusion

Research findings on internal migration in Taiwan

From 2000 to 2020, Taiwan underwent various political party turnovers, reflecting the deep-seated impact of different policies and ideologies of the parties on population migration and clustering conditions across different counties and cities. For instance, the clustering condition in Yilan County after 2016 underwent a significant change, exhibiting a Low-High characteristic of a relatively sparse population, while neighboring counties showed a more concentrated population. Simultaneously, Keelung City consistently displayed this characteristic throughout the observation period. This study found that regardless of which political party is governing or the implementation of proposed policies, the current Low-High condition of in-migrant population in these two counties and cities cannot be altered. That is to say, irrespective of which political party is in power at both central and local levels, they are impotent in effecting change in the clustering condition of the in-migrant population.

Concerning fiscal expenditures, studies reveal that local counties’ economic development expenditure has a notable positive effect on in-migrant populations. This could be due to economic development investments stimulating local employment and economic activities, attracting more in-migration. However, it doesn’t have a significant impact on the in-migration of neighboring counties and cities. Conversely, the expenditure on education, science and culture does not have a significant effect on the local counties' in-migrant populations, but it generates negative spatial spillover effects on neighboring counties, likely due to the concentration of educational resources attracting populations from surrounding areas.

In terms of economic opportunities, an increase in the number of employed individuals tends to suppress the growth of in-migration population in the local counties and cities. This implies that more job opportunities do not necessarily attract more in-migrant households, possibly because local residents can find employment nearby without needing to migrate. However, for neighboring counties, an increase in the number of employed individuals will bring about a positive spillover effect, possibly due to residents from surrounding areas migrating in search of better employment opportunities. Meanwhile, changes in the average annual regular income per household and average disposable income per household do not directly impact the in-migrant population in the local counties and cities, but rather exert both positive and negative spillover effects on the in-migrant population in neighboring counties. This could be because the local living costs (for instance, housing prices and rents) might grow proportionally with income, resulting in residents not experiencing an increase in disposable income or an enhancement in quality of life even with higher incomes.

In terms of educational levels, an increase in the proportion of populations with a college degree or above suppresses the growth of in-migrant populations in the local counties and generates negative spillover effects on neighboring counties. This may be because individuals with higher education tend to reside and work in metropolitan areas with abundant educational resources and numerous employment opportunities.

In conclusion, the migration and agglomeration conditions in Taiwan are shaped by multifarious factors, including the policy changes brought by party alternations, fiscal expenditures, economic opportunities and educational levels. The interaction of these factors determines the divergences in the internal migration and agglomeration patterns across various counties and cities. To promote regional balanced development and optimize population structures, it is imperative to delve into the influential mechanisms of these factors to formulate more scientific and precise development strategies and policies.

Policies to manage internal migration in Taiwan

In Taiwan, managing internal migration is crucial for achieving balanced regional development, maintaining social harmony and addressing disparities between different regions. Below are some policies for managing internal migration that Taiwan can consider implementing:

Balanced regional development and area investment

Implement comprehensive planning nationwide, strengthen inter-regional coordination and connection and establish integrated objectives for balanced regional development. By mutually reinforcing policies, mitigating and alleviating developmental disparities, and curbing excessive population concentration. Increase investments in infrastructure like transportation, communication and public services, such as education and health, especially in economically lagging regions, to enhance their attractiveness.

Expansion of the labor market and employment

Formulate corresponding policies to promote industrial transformation and upgrading, guiding enterprises and investors to relocate to regions with high out-migration rates, and offering diverse employment opportunities. Implement flexible and diverse labor policies, strengthening adjustments to the labor market to ensure a balance in labor supply and demand in all regions.

Enhancement of educational resources and training

Promote balanced national education development by providing more educational resources, especially to remote and underdeveloped areas. Actively develop diversified vocational education and training opportunities, enhancing the professional skills of the workforce to meet the industrial development needs of different regions.

Optimization of living environment

Implement and refine various housing support policies, provide affordable housing, ensuring the residential needs of all social strata are met. Strengthen urban and rural planning, and improve public facilities and the environment, making people willing to reside in these areas.

Enhancement of medical resource allocation and promotion of health

Strive to improve and increase medical resources in remote and underdeveloped areas, enhancing the level of medical services for local residents. By optimizing health resources and strengthening public health education, raising the health level and quality of life of the population.

Through these comprehensive and profound policies, Taiwan can not only manage and guide internal migration more effectively but also promote comprehensive economic and social progress through coordinating the development of various regions and constructing a more harmonious and stable society.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research authorship and/or publication of this article.

Funding

The author(s) received no financial support for the research authorship and/or publication of this article.

ORCID iD

Hao-Chen Huang

Note

References

Akaike

(1973) Information theory and an extension of the maximum likelihood principle. In: Petrov

Csaki

(eds) 2nd International Symposium on Information Theory. Budapest: Akademiai Kiado, pp. 267-281.

Anselin

(1995) Local indicators of spatial association—LISA. Geographical Analysis 27(2): 93-115. DOI: 10.1111/j.1538-4632.1995.tb00338.x.

Banzhaf

Walsh

(2008) Do people vote with their feet? An empirical test of Tiebout’s mechanism. American Economic Review 98(3): 843-863. DOI: 10.1257/aer.98.3.843.

Boarnet

(1998) Spillovers and the locational effects of public infrastructure. Journal of Regional Science 38(3): 381-400. DOI: 10.1111/0022-4146.00099.

Bogue

(1969) Principles of Demography. New York: John Wiley and Sons Inc.

Bonin

Raffelhüschen

Walliser

(2000) Can immigration alleviate the demographic burden?. FinanzArchiv 57(1): 1-21. Available at: https://www.jstor.org/stable/40912912.

Case

Rosen

Hines

JR (1993) Budget spillovers and fiscal policy interdependence: Evidence from the states. Journal of Public Economics 52(3): 285-307. DOI: 10.1016/0047-2727(93)90036-S.

Central Election Commission (2023) Election and Referendum Database . Dataset, Central Election Commission. Available at: https://db.cec.gov.tw/ElecTable/Election?type=President (accessed 25 December 2023).

Charles-Edwards

Bell

Bernard

et al. (2019) Internal migration in the countries of Asia: Levels, ages and spatial impacts. Asian Population Studies 15(2): 150-171. DOI: 10.1080/17441730.2019.1619256.

10.

Chen

Coulson

(2002) Determinants of urban migration: Evidence from Chinese cities. Urban Studies 39(12): 2189-2197. DOI: 10.1080/0042098022000033818.

11.

Clark

Knapp

White

(1996) Personal and location-specific characteristics and elderly interstate migration. Growth and Change 27(3): 327-351. DOI: 10.1111/j.1468-2257.1996.tb00909.x.

12.

Cohen

Paul

CJM

(2004) Public infrastructure investment, interstate spatial spillovers, and manufacturing costs. The Review of Economics and Statistics 86(2): 551-560. Available at: https://www.jstor.org/stable/3211646.

13.

Conway

Houtenville

(1998) Do the elderly “vote with their feet?”. Public Choice 97(4): 663-685. Available at: https://www.jstor.org/stable/30024453.

14.

Cushing

(1987) Location-specific amenities, topography, and population migration. The Annals of Regional Science 21(2): 74-85. Available at: https://pubmed.ncbi.nlm.nih.gov/12281051/.

15.

Davies

Greenwood

(2001) A conditional logit approach to U.S. state-to-state migration. Journal of Regional Science 41(2): 337-360. DOI: 10.1111/0022-4146.00220.

16.

Day

(1992) Interprovincial migration and local public goods. The Canadian Journal of Economics 25(1): 123-144. DOI: 10.2307/135714.

17.

Department of Land Administration, Ministry of the Interior (2023) Square domain business. Available at: https://www.land.moi.gov.tw/chhtml/content/68?mcid=3224 (accessed 25 December 2023).

18.

Directorate General of Budget, Accounting and Statistics (DGBAS) of the Executive Yuan (2023) County/City Major Statistical Indicators Query System Found in the Statistical Information Network. Dataset, DGBAS. Available at: https://winstacity.dgbas.gov.tw/DgbasWeb/ZWeb/StateFile_ZWeb.aspx (accessed 25 December 2023).

19.

Elhorst

(2010) Applied spatial econometrics: Raising the bar. Spatial Economic Analysis 5(1): 9-28. DOI: 10.1080/17421770903541772.

20.

Fielding

(1992) Migration and social mobility: South East England as an escalator region. Regional Studies 26(1): 1-15. DOI: 10.1080/00343409212331346741.

21.

Fox

Herzog

HW Schlottman

(1989) Metropolitan fiscal structure and migration. Journal of Regional Science 29(4): 523-536. DOI: 10.1111/j.1467-9787.1989.tb01242.x.

22.

Frère

Leprince

Paty

(2014) The impact of intermunicipal cooperation on local public spending. Urban Studies 51(8): 1741-1760. DOI: 10.1177/0042098013499080.

23.

Jie

Lao

(2022) Health service disparity, push-pull effect, and elderly migration in ageing China. Habitat International 125: 102581. DOI: 10.1016/j.habitatint.2022.102581.

24.

Jie

et al. (2021) What drives migrants to settle in Chinese cities: A panel data analysis. Applied Spatial Analysis and Policy 14: 297-314. DOI: 10.1007/s12061-020-09358-z.

25.

Hicks

(1932) The Theory of Wages. London: Macmillan.

26.

Huo

Wang

Zhao

et al. (2016) Statistical characteristics of dynamics for population migration driven by the economic interests. Physica A: Statistical Mechanics and its Applications 451: 123-134. DOI: 10.1016/j.physa.2016.01.054.

27.

Kameda

Namba

Tsuruga

(2021) Decomposing local fiscal multipliers: Evidence from Japan. Japan and the World Economy 57: 101053. DOI: 10.1016/j.japwor.2021.101053.

28.

Kone

Liu

Mattoo

et al. (2018) Internal borders and migration in India. Journal of Economic Geography 18(4): 729-759. Available at: https://www.jstor.org/stable/26501174.

29.

Lao

Zhao

(2022) Revisiting Hukou transfer intentions among floating population in Chinese cities: Spatial differences and multi-level determinants. Sage Open 12(2): 215824402210979. DOI: 10.1177/21582440221097926.

30.

Lee

(1966) A theory of migration. Demography 3(1): 47-57. DOI: 10.2307/2060063.

31.

LeSage

Pace

(2009) Introduction to Spatial Econometrics. New York: Chapman and Hall/CRC Press.

32.

Pan

(2023) Globalization-driven internal migration in China: The impact of foreign direct investment and exports since 2000. Population, Space and Place 29(2): e2612. DOI: 10.1002/psp.2612.

33.

Long

Liu

(2022) Internal migration and associated carbon emission changes: Evidence from cities in China. Energy Economics 110: 106010. DOI: 10.1016/j.eneco.2022.106010.

34.

López

Martínez-Ortiz

Cegarra-Navarro

(2017) Spatial spillovers in public expenditure on a municipal level in Spain. The Annals of Regional Science 58: 39-65. DOI: 10.1007/s00168-016-0780-7.

35.

(2023) The impact of equalisation of basic public health and medical services on the long-term urban settlement intentions of internal migrants in China. Asia Pacific Viewpoint 64(1): 2-16. DOI: 10.1111/apv.12349.

36.

Tang

(2020) Geography, trade, and internal migration in China. Journal of Urban Economics 115: 103181. DOI: 10.1016/j.jue.2019.06.004.

37.

Malamassam

Hidayati

Setiawan

et al. (2021) “Move backward to make a step forward”: Understanding the migration of the highly educated to Sorong City, West Papua, Indonesia. Asian and Pacific Migration Journal 30(4): 485-499. DOI: 10.1177/01171968211069722.

38.

Moran

PAP

(1950) Notes on continuous stochastic phenomena. Biometrika 37(1/2): 17-23. DOI: 10.2307/2332142.

39.

Norman

Boyle

(2014) Are health inequalities between differently deprived areas evident at different ages? A longitudinal study of census records in England and Wales, 1991-2001. Health & Place 26: 88-93. DOI: 10.1016/j.healthplace.2013.12.010.

40.

Oates

(1969) The effects of property taxes and local public spending on property values: An empirical study of tax capitalization and the Tiebout hypothesis. Journal of Political Economy 77(6): 957-971. Available at: https://www.jstor.org/stable/1837209.

41.

Park

Kim

(2015) Internal migration of the elderly in Korea: A multilevel logit analysis of their migration decision. Asian and Pacific Migration Journal 24(2): 187-212. DOI: 10.1177/0117196815583759.

42.

Pereira

Roca-Sagalés

(2003) Spillover effects of public capital formation: Evidence from the Spanish regions. Journal of Urban Economics 53(2): 238-256. DOI: 10.1016/S0094-1190(02)00517-X.

43.

Porell

(1982) Intermetropolitan migration and quality of life. Journal of Regional Science 22(2): 137-158. DOI: 10.1111/j.1467-9787.1982.tb00741.x.

44.

Ravenstein

(1885) The laws of migration. Journal of the Statistical Society of London 48(2): 167-235. DOI: 10.2307/2979181.

45.

Ryndyk

(2020) The role of labour market integration in migrants’ decisions about family reunification: A comparative study of Polish migrants in Norway, Sweden, and the UK. Comparative Migration Studies 8(17). DOI: 10.1186/s40878-020-00177-2.

46.

Sawada

(2004) Global Spatial Autocorrelation Indices- Moran’s I, Geary’s C and the General Cross-Product Statistic. Research paper. Ottawa: University of Ottawa.

47.

Schwarz

(1978) Estimating the dimension of a model. The Annals of Statistics 6(2): 461-464. Available at: https://www.jstor.org/stable/2958889.

48.

Tiebout

(1956) A pure theory of local expenditures. Journal of Political Economy 64(5): 416-424. Available at: https://www.jstor.org/stable/1826343.

49.

Todaro

(1969) A model of labor migration and urban unemployment in less developed countries. The American Economic Review 59(1): 138-148. Available at: https://www.jstor.org/stable/1811100.

50.

Zhang

Tao

et al. (2019) Amenities, economic opportunities and patterns of migration at the city level in China. Asian and Pacific Migration Journal 28(1): 3-27. DOI: 10.1177/0117196819832054.

51.

Zhang

Liao

et al. (2022) The effect of air pollution on migrants’ permanent settlement intention: Evidence from China. Journal of Cleaner Production 373: 133832. DOI: 10.1016/j.jclepro.2022.133832.

52.

Zhao

(2023) Moving to ‘Bei-Shang-Guang’? Internal migration, local Hukou barriers and labour market outcomes among college graduates in China. Population, Space and Place 29(5): e2651. DOI: 10.1002/psp.2651.

53.

Zhou

Guo

(2023) Housing characteristics and health in urban China: A comparative study of rural migrants and urban locals. Population, Space and Place 29(1): e2600. DOI: 10.1002/psp.2600.