County Town Comprehensive Service Functions in China: Measurement, Spatio-Temporal Divergence Evolution, and Heterogeneity of Influencing Factors

Abstract

Strengthening the service function of small towns, using its fundamental role in the urban system to drive rural development, is the main issue that needs to be addressed urgently in numerous developing countries. County towns are unique types of small towns in China. Analyzing the spatial-temporal patterns and differentiation mechanisms of comprehensive service functions of county towns in China from a geographic point of view can not only provide a basis for the macro-control of county towns but also provide typical regional research results for the study of urban systems and urban–rural coordination in developing countries. Based on Point of Interest (POI) data of 1788 county towns in China, this study analyzes the evolution of spatial and temporal differentiation of comprehensive service functions and influencing factors by using modeling methods such as Getis-Ord Gi* analysis, the random forest model, and Multiscale Geographically Weighted Regression (MGWR). The obtained results show that (1) from 2012 to 2021, the average value of the comprehensive service function index (CSFI) of county towns in China shows a significant increase, and the proportion of county towns with medium–high service levels and above increases from 3.41% to 54.50%; (2) spatially, the comprehensive service function of county towns is characterized by the basic pattern of “high east, low west; high south, low north”, which keeps getting stronger. During the study period, eastern China has always been a high-level region, northwestern and southwestern China have always been low-level regions, and northeastern China has been a stagnant region, while central, northern, and southern China have been fast-growing regions; (3) county general public budget revenues, value added of secondary industry, GDP per capita, county town resident population, altitude, and GDP per capita of affiliated prefecture-level cities to which it belongs are the key factors influencing the comprehensive service function of county towns in China. The county general public budget revenue indicator, which represents the governmental capacity, has the strongest influence; and (4) the results of the MGWR analysis indicate that there is spatial and temporal heterogeneity in the intensity of the above-mentioned key influencing factors on the development of comprehensive service functions of county towns in China. Based on this finding, differentiated strategies should be proposed to policy makers and urban planners in different regions in order to effectively enhance the level of comprehensive service functions of county towns in China.

1. Introduction

The comprehensive service function of cities and towns refers to the tasks and roles of cities and small towns in providing production and lifestyle services within a certain area in order to meet the needs of human beings for their own survival and development [1]. Although the service function of small towns has a small radius, it can directly radiate to the adjacent rural areas; its well-developed service function is a prerequisite for controlling the excessive influx of a rural population into large cities, and is also the basis for the sustainable development of regional urbanization [2,3]. From the viewpoint of the experience of town construction in developed countries, whether it is a development model of European countries [4], which is based on small towns, or the models of Japan and New Zealand [5,6], which are based on large cities and promotes the development of small towns in a balanced way [5], it has been thoroughly proven that giving importance to the construction of service functions in small towns is an inevitable choice for promoting urban–rural integration and the healthy and sustainable development of urbanization [7,8,9]. However, after World War II, developing countries developed rapidly, influenced by unbalanced development theories such as the “urban–rural dual structure”, which has overly emphasized the polarizing role of large cities and has deprived small towns of opportunities for development, resulting in the shrinking of the comprehensive service functions of small towns [10]. Therefore, the enhancement of the role of small towns as service centers in rural areas has become a policy objective in many developing countries [5,11,12,13]. Research on the comprehensive service functions of small towns is of theoretical and practicable significance for the future development of small towns and rural areas in developing countries [14].

This paper summarizes the research on the service functions of towns [2,5,6,8,12,13,15,16,17,18,19,20] and has found that the existing studies on the comprehensive service functions of towns and cities mainly focus on large cities and rapidly developing typical cities [2,5,6,8,9,12,15], with few studies on the service functions of small towns focusing on developed countries [21]. Additionally, the content of these studies often focuses on the single service functions of small towns, such as shopping [19,22,23], health care [24,25], education [26,27,28], etc., or the provision of services to specific groups of people, such as the elderly [25,29], women [26,30], etc., in small towns. In recent years, although a few scholars have carried out preliminary discussions on the comprehensive service functions of small towns by drawing on the research paradigm of the comprehensive service functions of large cities [31], the importance of research on the comprehensive service functions of small towns in developing countries has not received the attention it deserves.

China is a developing country with a large population, small land area, and relatively low per capita income [32]. Since 1994, its rapidly advancing urbanization has shown a bias towards the development of large cities and the development of polar nucleation, which has led to the weakening of various types of basic support roles for small towns in the national urban system, thereby leading to the prominent hollowing out and decay of the countryside [32,33,34]. To eradicate this problem, since 2012, the Chinese government has been promoting “New Urbanization” that focuses on the overall sustainable development of urban and rural areas, emphasizing the coordinated development of large, medium, and small cities and towns to promote the development of the countryside and narrow the development gap between urban and rural areas, thereby making it clear that small towns have a fundamental role in China’s new urbanization pattern [35,36]. Among them, county towns, as fourth-level settlements in China’s five-tier urban system of “central government, province (autonomous regions and municipalities directly under the central government), prefecture-level cities, county towns, and townships”, are a unique type of small town that combine the characteristics of big cities and villages [37]. They are the “top priority” in China’s New Urbanization and Rural Revitalization Projects [31,37]. However, county towns in China are generally characterized by a small construction scale, weak service functions, and a narrow service scope, which make them difficult to attract, satisfy, and retain people. Furthermore, the radiation-driven effect is difficult to extend to cover the vast countryside, resulting in their unsatisfactory ability to serve as carriers of New Urbanization [38].

In the current context, grasping the spatial-temporal characteristics of the development of comprehensive service functions in county towns in China and the factors influencing them is a prerequisite for carrying out the governance and planning of county towns’ comprehensive service function upgrading in accordance with local conditions, which is of great significance for the advancement of the New Urbanization strategy in China. This paper, on the basis of a general theoretical analysis of the spatio-temporal differentiation of county towns’ comprehensive service functions and their causes, utilizes the POI (Point of Interest) Big Data set from 2012 and 2021, as well as China’s officially published survey data, to measure the level of China’s county towns’ comprehensive service capabilities, and further explores the spatio-temporal differences in the comprehensive service functions of China’s county towns, as well as the spatio-temporal heterogeneity of its influencing factors. The results of this study can not only provide a basis for the macro-control of the service function of county towns and the New Urbanization of county regions in China but also make up for the lack of research on small towns in developing countries in the theoretical system of international comprehensive urban service functions. This research idea could also provide a basis for research on the comprehensive service function of small towns in other developing countries.

2. Theoretical Analysis of Spatio-Temporal Differentiation of Comprehensive Service Functions of County Towns

Christaller presented a center ground theory that is a classic for understanding the comprehensive service function of towns, with a theory that different towns have different service functions, types, levels, and scopes [39,40]. In other words, the larger the town, the greater its influence over distance, and the wider the scope of its services, the stronger the agglomeration of the comprehensive service function [41,42]. The service functions of small towns in the region can also exhibit considerable variability and complexity due to the heterogeneity of their agglomeration sizes. As the scale of small towns (population size, economic size, land size, etc.) is significantly affected by natural endowments and transportation conditions of the region in which they are located, the initial scale of their own development, the government’s role in promoting them, and the external role from the surrounding large cities [35,39,43,44,45], the spatial-temporal characteristics of these influencing factors also determine the spatial and temporal pattern of comprehensive service functions of small towns. In terms of county towns, as a unique type of small town in China, the spatio-temporal characteristics and influencing factors of their comprehensive service functions are also compatible with the aforementioned law (Figure 1).

Natural endowment and transportation conditions are the fundamental guarantee for the formation of county towns and the concentration of comprehensive service functions. County towns located in plains, resource-rich, and conveniently located areas can support larger populations and industrial agglomeration, and therefore have stronger comprehensive service functions [43,44]. County towns located in resource-poor mountainous and hilly areas face practical constraints in terms of industrial and agricultural production, construction of transportation infrastructure, and space for the development of county towns. This limits the scale of agglomeration and the development of comprehensive service functions of county towns [46].

The initial size of a county town is an important basis for its development and functional agglomeration. According to the “principle of the ratchet effect of initial benefits” and the “principle of circular cumulative causality”, each stage of county town development depends on the previous stage of development, and decisions on further agglomeration of factors into a county town are made based on what the town has to offer [47]. In other words, a county town with a strong industry and good infrastructure will always represent a better breeding ground for population and industry than a county town with weak economy, poor infrastructure, and serious population outflow, which encourages the self-growth of a functional agglomeration in the town [48].

The role of the government is to encourage the expansion of county towns and the improvement of comprehensive service functions. As an investor and organizer of the development of county towns, the government can use administrative and policy means to concentrate different elements and systematically implement investment and construction in county towns, which will directly affect the scale and function of county towns [45]. In general, county towns are most dependent on the role of the government during the early stages of development and construction. If government finances are adequate, planning strategies are appropriate, and investment guidelines are strong and accurate, it will help county towns to concentrate their population and industries, thus forming a stronger foundation for the development of comprehensive service functions, and vice versa [49].

At the same time, county towns are not independent from neighboring towns, and their comprehensive service functions are also influenced by external environmental forces with certain “spatial attributes”, such as high-level neighboring cities [39,50]. Scholars have discovered two sides of the influence of high-grade cities on county towns [51]. According to the theory of “borrowing scale”, county towns are able to interact with high-level cities and simultaneously improve their status in the nodes of the town network by “scaling up” and form a complete system of service functions driven by the radiation of high-level cities. On the contrary, county towns that do not benefit from the functions of neighboring large cities may be affected by an “echo” or “clustering shadow”, whereby development opportunities are clearly denied by medium- and high-ranking cities, thereby limiting the development of their own comprehensive service functions [31].

3. Data Sources and Research Methodology

3.1. Data Sources

The study unit was 1788 county towns in China (as of 2021, there are 2843 county-level administrative units in China, excluding 977 municipal districts that are not included in the study area and 78 counties in inland China with missing data). The remaining 1788 counties were the study sample. (Hong Kong, China; Macau, China; and Taiwan, China, are not included in the study.) To better compare the spatial and temporal differentiation of service functions in county towns, combining the commonly used economic region division in China, and referring to the research results of Tong et al. [50], the county towns in China were divided into seven geographic regions: northeast, northwest, southwest, north, central, east, and south China (

Table 1. The seven major geographic regions of China and their provinces.

Region	Province
North China	Beijing, Tianjin, Hebei, Shanxi
East China	Shanghai, Jiangsu, Zhejiang, Anhui, Fujian, Shandong
South China	Guangdong, Guangxi Zhuang Autonomous Region, Hainan
Central China	Henan, Hubei, Hunan
Northeast China	Heilongjiang, Jilin, Liaoning, Inner Mongolia
Southwest China	Chongqing, Sichuan, Guizhou, Yunnan, Tibet Autonomous Region
Northwest China	Shaanxi, Manlu, Qinghai, Ningxia Hui Autonomous Region, Xinjiang Uygur Autonomous Region

, Figure 2).

In 2012, China’s “New Urbanization Strategy” was formally proposed. The important position of county towns in the process of urban–rural integration received renewed attention, so the period 2012–2021 was chosen as the study period. The research of Sun Dongqi [31], Li Jiangsu [52], and other scholars, following the principle that small towns should first satisfy the basic life services and social needs services of the residents, then guarantee the support of public services such as schools and hospitals, and then gradually transition to the enhancement of high-end services [53], and in accordance with the “Classification and Coding of National Economy Industries (GB/4754-2011)” [54] and the Gaoder Open Platform website published in the Gaode POI classification code (https://developer.amap.com/api/webservice/download, accessed on 5 April 2023), was used to design the evaluation index system of comprehensive service function of county towns in China (

Table 2. Evaluation index system of comprehensive service function of county towns in China based on POI data.

Target Layer	Standard Layer	Indicator Layer	POI Data Content
A. Comprehensive service functions of county towns in China	B1. Basic Life Services	C1. Shopping services	Shopping places; shopping malls; supermarkets; convenience stores; home appliance and electronic stores; medical and healthcare stores; gas stations; gas filling stations; other energy stations; charging stations; flower, bird, fish, and insect markets; home building materials markets; general markets; cultural goods stores; sports stores; clothing, shoes, hats, and leather goods stores; specialty stores; special places for buying and selling; personal goods/cosmetic stores; automobile sales; motorcycle sales; automobile accessory sales; etc.
		C2. Accommodation services	Accommodation services related to guest houses, hostels, etc.
		C3. Catering services	Catering-related establishments, Chinese restaurants, foreign restaurants, fast food restaurants, casual dining establishments, cafes, tea houses, cold beverage stores, pastry stores, dessert stores, etc.
		C4. Life services	Life service facilities, travel agencies, information and consultation centers, offices, job markets, water supply offices, electric power supply offices, beauty salons, automobile repairs, motorcycle repairs, repair stations, laundromats, funeral facilities, telecommunication offices, ticketing offices, post offices, logistics and courier services, lottery ticket sales outlets, bath and massage establishments, intermediary organizations, etc.
	B2. Basic public services	C5. Science, education and cultural services	Science, education, and cultural venues; museums; exhibition halls; convention and exhibition centers; art museums; libraries; science and technology museums; planetariums; cultural palaces; archives; arts and culture groups; media organizations; schools; scientific research institutes; training institutes; driving schools; etc.
		C6. Sports and leisure services	Sports and leisure service venues, sports venues, golf venues, recreational venues, vacation and health resort venues, leisure venues, theaters, etc.
		C7. Health care services	Healthcare facilities, general hospitals, specialized hospitals, clinics, first aid centers, disease prevention agencies, healthcare sales outlets, animal healthcare establishments, etc.
		C8. Public facility services	Scenic spots related to parks and squares, public facilities, newsstands, public telephones, public restrooms, emergency shelters, etc.
	B3. Production Requirements Service	C9. Traffic services	Transportation services related to airports, long-distance bus stations, train stations, subway stations, light rail stations, bus stops, shuttle bus stops, parking lots, border crossings, roadway ancillary facilities, warning information, toll booths, service zones, red streetlights, road sign information, etc.
		C10. Financial and insurance services	Financial and insurance organizations, banks, ATMs, securities companies, insurance companies, finance companies, etc.
		C11. Commercial residential services	Business and residential services related to industrial parks, buildings, residential areas, etc.
	B4. Social needs services	C12. Government agencies, social organizations services	Government and social organizations, government agencies, foreign institutions, democratic parties, social organizations, public prosecutors and law enforcement agencies, traffic and vehicle management, industry, commerce and taxation agencies, etc.

).

The used data were from two main sources. First, comprehensive service function data of counties were selected from 2012 and 2021 POI data of county towns collected through the open API platform of Gaode Maps. Second, the socio-economic statistics data of the county towns were collected, which are involved in the analysis of the influencing factors of the spatio-temporal pattern of the comprehensive service function of the county towns, mainly from “China County Town Economic Statistics Yearbook (County and City Volume)”, “China County (City) Social and Economic Statistics Yearbook”, and bulletins of social and economic statistics issued by county town statistical bureaus for 2013 and 2022.

3.2. Methodology

3.2.1. Measurement of Comprehensive Service Functions of County Towns in China

The AHP method is a combined qualitative and quantitative decision analysis method for solving complex problems with multiple indicators. This paper refers to the related studies and compares C₁ to C₁₂ in Table 2 two by two to generate the 12 × 12 judgment matrix A, as shown in Equation (1):

$$A=\left[\begin{array}{ccc}1& \cdots & a_{ji}\\ ⋮& \ddots & ⋮\\ a_{ij}& \cdots & 1\end{array}\right]$$

(1)

In the matrix, aij denotes the importance of C_i to C_j and aji denotes the importance of C_j to C_i. According to the 1–9 scale method, aij and aji are judged and assigned values (1, 3, 5, 7, 9 for equally important, slightly important, relatively important, very important, absolutely important, 2, 4, 6, 8 for between the above two adjacent levels, and the inverse for opposite importance).

The results of the assignment are normalized and thus the weights of the indicators are determined, as shown in Equations (2) and (3):

$$G=\sqrt[n]{{\prod }_{i=1}^n{a}_{ij}}$$

(2)

$${\omega }_i=\frac{G_i}{\sum _{i=1}^n{G}_i}$$

(3)

The consistency test refers to the test of each mean or variance calculated from different samples and calculates the inconsistency index CI, where CI is the consistency index, λ is the eigenvalue corresponding to the judgment matrix A, A is the corresponding constructed judgment matrix, and n is the order. The formula is as follows:

In the equation, G is the geometric mean of the ith row of the judgment matrix (the ith service function), and the weight of the geometric mean of the ith service function as a proportion of the sum of the 12 geometric means is the weight ${\omega }_i$ of the ith service function indicator. Finally, the consistency test was conducted and the judgment matrix of this study passed the test. The final weighting results are shown in

Table 3. Results of weighting analysis of each service function.

Service Function	Weight	Service Function	Weight
C1. Shopping service	0.2318	C7. Healthcare services	0.0489
C2. Accommodation service	0.1150	C8. Public facility service	0.0417
C3. Catering service	0.1560	C9. Traffic service	0.0887
C4. Life service	0.1591	C10. Financial and insurance services	0.0224
C5. Science, education and cultural services	0.0393	C11. Commercial residential service	0.0191
C6. Sports and leisure service	0.0255	C12. Government agencies, social organizations	0.0523

.

The service function weights of each county are used as the basis for calculating the comprehensive service function index for each county separately. The formula is as follows:

$$I={\sum }_{i=1}^{12}{F}_i{V}_i$$

(4)

where I is the comprehensive service function index of the county town; i is the 12 service industry types corresponding to Table 2; F_i is the number of facilities (i.e., the number of POIs) for service function; and V_i is the weight of service function.

3.2.2. Getis-Ord Gi* Analysis

The Getis-Ord Gi* index is used to detect hot and cold spots of comprehensive service functions in county towns in China. The cold–hot spot analysis is an effective method for measuring global and local characteristics of cold or hot spot clustering, and the Getis-Ord Gi* index is used to identify high-value clusters and low-value clusters in different spatial locations. The hot and cold spots of county towns’ comprehensive service functions represent the highlands and depressions of their comprehensive service function development in China, respectively. The specific formula used is as follows:

$$G_i^{*}=\frac{\sum _{j=1}^n{w}_{ij}{x}_j-\sum _{j=1}^n{x}_j/n{\sum }_{j=1}^n{w}_{ij}}{\sqrt{\frac{\sum _{j=1}^n{x}_j^2}{n}-{\left(\frac{\sum _{j=1}^n{x}_j}{n}\right)}^2}\sqrt{\frac{\left[n\sum _{j=1}^n{w}_{ij}^2-{\left(\sum _{j=1}^n{w}_{ij}\right)}^2\right]}{n-1}}}$$

(5)

where X_j is the comprehensive service function index of county town j, W_ij is the spatial weight between county town i and county town j; and n is the total number of county towns used to measure the comprehensive service function of the county. The positive value of Gi* is medium–high compared to the critical value, indicating that there is a significant hot spot in the development of comprehensive service functions in county towns, while the negative value of Gi* is medium–low compared to the critical value, indicating that there is a significant cold spot in the development of comprehensive service functions in county towns. The remaining spots are not significant.

3.2.3. Random Forest Model

The random forest model overcame problems such as the possible covariance of the factors influencing the service function of each county town in the data processing process and had obvious and unique advantages in the algorithm, which made the results of the assessment of the degree of importance of the variables more reliable. Accordingly, the random forest model was used to rank the importance of factors affecting the comprehensive service functions of county towns in China and to screen out their core influencing factors. The evaluation indicators include the coefficient of determination (R²) and the root mean square error (RMSE). The IncMSE method was used for the importance ranking. The calculation formula is as follows:

$$R=\frac{{\sum }_{i=1}^n\left(P_i-\overline{P}\right)\left(O_i-\overline{O}\right)}{\sqrt{{\sum }_{i=1}^n{\left(P_i-\overline{P}\right)}^2}\sqrt{{\sum }_{i=1}^n{\left(O_i-\overline{O}\right)}^2}}$$

(6)

$$RMSE=\sqrt{\frac{1}{n}{\sum }_{i=1}^n{\left(O_i-P_i\right)}^2}$$

(7)

$$I={\sum }_{i=1}^N\frac{\left(E_{rror2i}-E_{rorr1i}\right)}{N}$$

(8)

where O_i and P_i are the i-th measured and fitted values, respectively; O and P are the mean values of the measured and fitted values, respectively; I is the RMSE increased by the influencing factors; E_rror1i is the data that do not participate in the decision tree training when selecting the decision tree; and E_rror2i is the data that do not participate in the decision tree training when selecting the decision tree after adding a random interference, and the closer R² is to 1, the closer the RMSE is to 0, and the medium–high is the explanation accuracy of the model. Based on MATLAB 2022 software, 80% of the data are used as training samples and 20% of the data are used as test samples. Furthermore, the optimal number of leaves is 5 and the decision tree is 100 after repeated testing.

3.2.4. Multiscale Geographically Weighted Regression (MGWR)

Using the Multiscale Geographically Weighted Regression (MGWR) model to analyze the spatio-temporal heterogeneity of the factors influencing the comprehensive service functions of county towns in China. MGWR is an application used to calibrate the GWR model. Compared with the classical GWR model, the MGWR model adds spatially smooth variables on the basis of the GWR model, which can more accurately describe the spatial heterogeneity and nonlinear relationships. The formula is as follows:

$$y_i={\beta }_{0\left({\mu }_i{,v}_i\right)}+{\sum }_{j=1}^k{\beta }_{bwj}\left({\mu }_i{,v}_i\right)x_{ij}+{\epsilon }_i$$

(9)

where y_i denotes the regression coefficient of the influence factor; x_ij denotes the influence factor, k is the total number of analyzed spatial units; i denotes the random error term; (μ_i,υ_i) denotes the spatial coordinates of the sample; β₀(μ_i,υ_i) is the intercept at location i; and β_bwj(μ_i,υ_i) is the localized regression coefficient of the j variables at location i. Each MGWR regression coefficient β_bwj is obtained based on local regression and the bandwidth has specificity, which is the biggest difference from classical GWR, where β_bwj has the same bandwidth for all variables. In this study, the quadratic kernel function and Akaik Information Criterion (AIC_C) method were used to determine the optimal bandwidth.

4. Results

4.1. Spatio-Temporal Evolution of Comprehensive Service Functions in County Towns

4.1.1. Overall Evolutionary Characteristics of the Time Series

The overall trend of the level of comprehensive service functions in the county towns in China shows a significant improvement in the period 2012–2021. The average value of the comprehensive service function index for county towns increases from 277.94 in 2012 to 1550.14 in 2021, a nearly five-fold increase, indicating that since China began to implement the New Urbanization Strategy in 2012, the status of small towns in the national urban system has substantially increased, and county towns have also received more opportunities for development. The extreme difference in the comprehensive service function index of county towns has increased from 3160.92 in 2012 to 14,609.82 in 2021. The absolute internal gap is expanding rapidly, and the uneven development of county towns as a whole has become more obvious, presenting the Matthew effect of “the strong are always stronger, and the weak are always weaker”.

By conducting cluster analysis on the comprehensive service function index of county towns in 2012 and 2021, taking into account the principle of taking integers, the thresholds of the indices are determined as 250, 500, 1000, and 2000 in five levels. I > 2000.00 is a high service level type, 1000.00 < I ≤ 2000.00 is a medium–high service level type, 500.00 < I ≤ 1000.00 is a medium service level type, 250.00 < I ≤ 500.00 is a medium–low service level type, and 0 ≤ I ≤ 250.00 is a low service level type. From the viewpoint of the number of county towns within each level of county town comprehensive service function (Figure 3), a typical “pyramid” structure was shown with 1158 county towns with a low service level type, the proportion as high as 64.76%, in 2012. In 2021, an “inverted pyramid” structure is initially formed, with 974 county towns with high- and medium–high service level types, and the number of low service level types is reduced to 206. Over the decade, the number of county towns of all types has been gradually rationalized, but there is still a large share of county towns with a lower than medium level of service.

4.1.2. Characteristics of Spatial Evolution

ArcGIS 10.8 software was used to analyze the spatial visualization of the comprehensive service function index of county towns in China in 2012 and 2021 (Figure 4a,b). The overall performance of the basic spatial pattern of “high in the east and low in the west, high in the south and low in the north” continuously strengthens. In 2012, the average value of the comprehensive service function of county towns was as follows: east China (509.84) > northeast China (323.77) > south China (234.96) > central China (221.30) > north China (218.12) > southwest China (201.53) > northwest China (150.73). The county towns with high service level and medium–high service level are mainly distributed in east China, while county towns with medium service level are mainly distributed in northeast China. The medium–low service level towns are mainly distributed in north, central, and south China, while low service level towns are mainly distributed in southwest and northwest China. In 2021, the average value of the comprehensive service function of county towns was as follows: east China (2871.18) > central China (2178.76) > south China (1702.30) > north China (1324.36) > northeast China (1002.16) > southwest China (981.40) > northwest China (642.61). The high service level towns are mainly distributed in east and central China, while medium–high service level towns are mainly distributed in south and north China. The northeast China is still the main distribution area for medium service level towns, while southwest and northwest China are the clusters of low service level towns. Regarding the changes in the growth rate of the comprehensive service function index of county towns in different regions of China (Figure 4c), the service function of county towns in central China has the fastest growth rate, with an average growth of 14.93, and 52 of China’s 63 county towns with an increase of more than 20 times are distributed here. The regions with growth rates in the second tier are south, north and east China, with an average increase of 7.15, 6.74 and 6.24, respectively. The southwest, northwest and northeast region are in the third growth rate gradient with a relatively slow growth, and the average growth rate was 4.58, 4.42, 3.14, respectively. Eighty-five of all eighty-six negative service function counties in China are located in these three regions.

Getis-Ord Gi* was used to characterize the spatial agglomeration of the comprehensive service function index of county towns in China (Figure 5). In 2012, the hotspot areas of comprehensive service functions of county towns were distributed in east China, Liaoning in northeast China, and Sichuan-Chongqing in southwest China. The cold spot areas of comprehensive service functions of county towns were distributed in the border areas of Guizhou, Hunan, and Guangxi, Shaanxi, Ningxia, and Shanxi, and in the Tibet area of southwest China. In 2021, the hotspot agglomeration effect of comprehensive service functions disappears in the northeast and Sichuan-Chongqing regions. The hotspot area of county towns’ comprehensive service functions in east China gradually expands to south and central China, forming a center of county towns’ comprehensive service functions. The cold spot area of comprehensive service functions is further expanded. Two new cold spot agglomeration areas appeared in the north of the northeast region and the Xinjiang region in the northwest. This is consistent with the spatial pattern of “high in the east, low in the west, high in the south, and low in the north” of comprehensive service functions of county towns in China during the study period.

4.2. Key Influencing Factors and Spatial and Temporal Heterogeneity of Role Intensity

4.2.1. Identification of Key Influencing Factors

Based on the theoretical framework of the comprehensive service function of small towns, combined with China’s socio-economic development and data accessibility, the study takes the county towns’ comprehensive service index as the dependent variable. Then, the influencing factors of the temporal and spatial evolution of comprehensive service functions were analyzed from four aspects, namely, natural endowment and transportation conditions, the initial agglomeration scale of county towns, governmental impetus, and external environmental force. Eleven indicators (

Table 4. Factors influencing the evolution of spatio-temporal differentiation of comprehensive service functions in county towns in China.

Mechanisms of Influence	Pathways of Influence	Factor	Expected Effect
Natural endowment and transportation conditions	Natural endowment	Altitude (X1)	−
		River network density (X2)	+
	Transportation conditions	Transportation accessibility (X3)	+
Initial agglomeration scale of county town	County scale	county town resident population (X4)	+
		Added value of secondary industry (X5)	+
		Urban construction land area (X6)	+
		GDP per capita (X7)	−
Governmental impetus	Construction investment	County fixed asset investment (X8)	+
	Government strength	County general public budget revenue (X9)	+
External environmental force	Proximity effect	Distance to prefecture-level city (X10)	−
		GDP per capita of affiliated prefecture-level cities (X11)	+

) were used as independent variables. Among them, natural endowment and transportation conditions include altitude, river network density, and transportation accessibility as indicators. The initial scale of the county town is characterized by the population size, the added value of the secondary industry, the urban construction land area, and GDP per capita. The government’s impetus is reflected by the fixed asset investment and general public budget revenue. Due to China’s “county under prefecture-level cities” system of administration, the development of county towns is strongly influenced by the vertical influence of prefecture-level cities; therefore, the external environmental force is expressed by the distance to the prefecture-level city and GDP per capita of affiliated prefecture-level cities.

The development of comprehensive service functions in county towns in China is influenced by various factors and tends to cause severe covariance problems, which reduces the explanatory role of each factor. The random forest model does not need to take this into account, and it can evaluate the importance of each influencing factor and filter out the key factors influencing the development of comprehensive service functions in county towns. As shown in

Table 5. Parameters related to random forest modeling results in 2012 and 2021.

	MSE		RMSE		MAE		MAPE		R2
	2012	2021	2012	2021	2012	2021	2012	2021	2012	2021
Training set	12,032.50	167,779.25	109.69	409.61	83.16	297.83	32.12	20.86	0.88	0.94
Prediction set	19,935.40	246,933.93	141.19	469.92	83.22	348.84	53.09	58.79	0.43	0.52

, the random forest model applied to the problem fits well, and the R² of the training sets in the model are all greater than 0.85, with high model accuracy. Due to the extreme internal variance of the sample data, the RMSE value was taken as the minimum value to guarantee model accuracy.

The IncMSE method and the average values for each impact factor in 2012 and 2021 were used to determine the importance ranking of the impact factors. The greater the %IncMSE, the more important it is, as shown in Figure 6: X9 > X5 > X7 = X4 > X1 > X11 = X8 > X3 > X10 > X6 > X2. The top six indicators are taken as key factors affecting the development of comprehensive service functions in Chinese counties and cities. Since X11 and X8 are tied for the sixth place, while X9 and X8 jointly characterize the governmental impetus, and the IncMSE of X9 = 56.25% has a stronger explanatory power of comprehensive service functions in county towns, X11 is retained to exclude X8.

4.2.2. Spatio-Temporal Variation in the Intensity of the Effects of Core Influences

Spatial heterogeneity exists in the intensity of the role of factors influencing the comprehensive service function of county towns in China, but the classical OLS regression analysis model ignores the spatial influence, and the regression results only represent the average level of the role of each influencing factor. Improved spatial regression models such as SEM, SLM, SDM, and MGWR are commonly used to analyze the spatial heterogeneity of the role of influencing factors. Due to the fact that this paper uses cross-sectional data for the study, the SDM model requires continuous panel data for the analysis. Therefore, OLS, SEM, SLM, and MGWR models were compared in order to select the most suitable spatial regression model for this study (

Table 6. Comparison of fitting results of OLS, SAM, SLM, and MGWR models.

	2012			2021
	${\mathbf{R}}^2$	$\mathbf{ADJ}.{\mathbf{R}}^2$	${\mathbf{A}\mathbf{I}\mathbf{C}}_{\mathbf{C}}$	${\mathbf{R}}^2$	$\mathbf{ADJ}.{\mathbf{R}}^2$	${\mathbf{A}\mathbf{I}\mathbf{C}}_{\mathbf{C}}$
OLS	0.528	0.527	3746.208	0.694	0.693	2971.350
SEM	0.626	–	–	0.785	–	–
SLM	0.579	–	–	0.753	–	–
$\mathrm{MGWR}$	0.821	0.782	2788.511	0.923	0.908	1191.138

). The regression results of the MGWR model in 2012 were an ADJ.R² > 0.78 and in 2021 an ADJ.R² > 0.90, so the fitting effect is more satisfactory and it is the optimal model for spatial analysis in this paper.

The MGWR2.2 model was used to analyze the spatio-temporal divergence of the intensity of the six key influencing factors, with each key influencing factor satisfying the 1% significance test. Descriptive statistics of the MGWR model regression results for 2012 and 2021 are shown in

Table 7. Descriptive statistics of the regression results of the MGWR model.

	Mean		STD		Min		Median		Max
	2012	2021	2012	2021	2012	2021	2012	2021	2012	2021
X9	0.145	0.657	0.190	0.500	−0.113	0.012	0.084	0.488	0.740	2.024
X5	0318	0.655	0.001	0.246	0.317	0.191	0.318	0.629	0.321	1.343
X7	0.064	−0.543	0.428	0.361	−0.867	−1.640	0.021	−0.528	2.035	0.578
X4	0.341	0.010	0.243	0.014	−0.226	−0.020	1.199	0.031	1.199	0.031
X1	−0.005	−0.225	0.001	0.071	−0.007	−0.299	−0.005	−0.116	−0.003	−0.116
X11	0.071	0.021	0.165	0.133	−0.452	−0.3360	0.061	0.011	0.630	0.647

. Using the ArcGIS natural breakpoint method, the regression coefficients of the variables processed by the MGWR model were classified into five impact intensity levels. As shown in Figure 7, Figure 8, Figure 9, Figure 10, Figure 11 and Figure 12.

1.

X9 represents the economic strength of county governments. In 2012, the areas of strong influence of X9 on the comprehensive service functions were distributed in northeast China, Beijing-Tianjin-Hebei in north China, the northern part of east China, and the Pearl River Delta in south China. The medium impact areas were distributed in the Guanzhong Urban Agglomeration in northwest China and the Sichuan-Chongqing Urban Agglomeration in southwest China. Most areas in northwest and southwest China were low impact areas. In 2021, the strong impact area of X9 shifted to south and central China, and Yunnan Province in southwest China, while the medium impact area is in the southern part of north China. The low impact areas are distributed in northwest, southwest, northeast, and east China (Figure 7). From 2012 to 2021, the area of high impact intensity shows a trend of shifting from north to south. The pattern of the comprehensive service function of county towns, high in the south and low in the north, has been further consolidated.

Figure 7. Spatial distribution of regression coefficients for county general public budget revenues.

Figure 7. Spatial distribution of regression coefficients for county general public budget revenues.

2.

X5 reflects the industrial agglomeration base of the county itself. In terms of spatial and temporal changes in the intensity of X5 impact, the westernmost parts of northwest China and southwest China were the strong impact zones, south China was a medium impact zone, and other regions belonged to the weak impact zone in 2012. In 2021, the strong impact zone was shifted to Guanzhong urban agglomeration in northwest China, Chuanchuan and Chongqing urban agglomerations in southwest China, and south China. The medium impact zone was distributed in the northern part of east China and the Beijing-Tianjin-Hebei region of north China, and northern part of northeast China. Other areas belonged to the low impact zone. Overall, the high impact area of this indicator shows a trend of shifting from west to east, which explains the increasingly stronger development pattern of county towns in China with high comprehensive service functions in the east and low in the west (Figure 8).

Figure 8. Spatial distribution of regression coefficients of added value of secondary industry.

Figure 8. Spatial distribution of regression coefficients of added value of secondary industry.

3.

X7 represents the economic base of the county itself, and this indicator has a weak influence on the comprehensive service function of county towns. In terms of spatial differentiation of changes in the intensity of the influence, there is always a large area of negative influence (Figure 9). In 2012, the high-value impact area of X7 was distributed in southwest, south, and east China. In 2021, the high-value impact area is mainly located in north China, as well as the Yangtze River Delta region in east China. The intensity of influence of this indicator shifted from high in the south and low in the north to high in the north and low in the south, which is contrary to the trend of the overall comprehensive service function pattern change. This is related to the inflated GDP per capita caused by the exodus of population from the towns in the northeast and northwest and does not reflect the real relationship between the town’s development and its economic strength.

Figure 9. Spatial distribution of regression coefficients for GDP per capita.

Figure 9. Spatial distribution of regression coefficients for GDP per capita.

4.

X4 shows a decreasing trend in the intensity of its impact on the comprehensive service function of county towns in China, reflecting that the population size of county towns is generally small and most of them are in a state of population decline. In 2012, the strong impact area of this indicator is located in east China, the medium impact area is located in northeast, northwest and southwest China, and the low impact area includes central China, south China, and Sichuan and Chongqing urban agglomerations in southwest China. In 2021, most of the provinces in the central China are population return, small and medium-sized county towns will become the main carrying space, the population of county towns will have a certain growth, and the intensity of the comprehensive service functions will be strengthened accordingly. north China, central China, and the eastern part of southwest and northwest China are under a strong influence, while northeast, east China, and the western part of southwest and northwest China are the peripheral areas in terms of the impact of this indicator (Figure 10).

Figure 10. Spatial distribution of regression coefficients for county population.

Figure 10. Spatial distribution of regression coefficients for county population.

5.

X1 has always been an obstacle to the development of comprehensive service functions, reflecting the fact that topography is the natural basis for the development of county towns. In 2012 and 2021, based on the general pattern of China’s topography being low in the east and high in the west, the hindering effect of X1 on the comprehensive service functions always showed an increasing trend from east to west. The area of strong influence was always located in northeast, north and east China. This shows that the gentler the overall topography of the town is, the more conducive it is for the construction of buildings and roads, which in turn favors the development of county town agglomeration and comprehensive service functions (Figure 11).

Figure 11. Spatial distribution of regression coefficients for altitude.

Figure 11. Spatial distribution of regression coefficients for altitude.

6.

X11 represents the economic strength of the neighboring big cities. The influence of this indicator on the comprehensive service function of county towns is generally weak, but with large spatial differences. In 2012, the areas of strong influence of this factor on the comprehensive service functions were concentrated in north China, central China, and the urban agglomerations of Sichuan-Chongqing in southwest China. Most county towns in northwest and southwest China as well as in the south of northeast China were moderately affected, while the northern part of northeast China, western part of northwest China, Yunnan in southwest China, and east China were weakly affected. In 2021, the area of strong influence remained in central China and the urban agglomerations of Sichuan-Chongqing in southwest China. The influence on north China and the southern county towns of northeast China changed from strong to weak, while the northern part of northeast China, the western part of northwest China, Yunnan in southwest China, and east China remained an area of weak influence (Figure 12).

Figure 12. Spatial distribution of regression coefficients of GDP per capita in affiliated prefecture-level cities.

Figure 12. Spatial distribution of regression coefficients of GDP per capita in affiliated prefecture-level cities.

5. Discussion

(1)

Since 2012, the promotion of the New Urbanization Strategy has improved the status of county towns, and the level of comprehensive service functions has also improved rapidly. The average value of the service function index has increased from 277.94 to 1550.14 in 2021. The internal differences in the comprehensive service functions of county towns in China have become larger, and the Matthew Effect is obvious. Not only has the extreme difference in the county towns’ comprehensive service function indices increased from 3160.92 to 14,609.82 but also the number of county towns with high service level and medium–high service level rapidly increased from 61 to 974. This changed the hierarchical structure of county towns from “pyramid type” to “inverted pyramid type”.

(2)

During the period 2012–2021, the comprehensive service function of county towns as a whole shows the basic spatial pattern of “high in the east, low in the west, high in the south and low in the north”, with east China always being a high-level area, northwest and southwest China always being a low-level area, northeast China being a stagnation area, and central, north and south China being a fast-growing area. The distribution of hot and cold spots of comprehensive service functions shows that east China and northeast China are the initial hotspots of a county’s comprehensive service functions. By 2021, the range of hotspots will gradually expand to south, north and central China, while northeast China will change from a hotspot to a cold spot of comprehensive service functions. Southwest and northwest China are always cold spots of comprehensive service functions.

(3)

The random forest model screening results show that county general public budget income, added value of secondary industry, GDP per capita, county town resident population, altitude, and GDP per capita of affiliated prefecture-level cities are the key factors affecting the development of comprehensive service functions of county towns in China. Finally, the regression results of the MGWR model show that there is significant spatio-temporal heterogeneity of six key factors with different coefficient levels in different geographic regions, and these factors influence the spatio-temporal pattern that characterizes the comprehensive service functions of county towns.

6. Conclusions

The enhancement of the comprehensive service function of small towns has been an urgent requirement for the steady advancement of urbanization and urban–rural integration in developing countries, but in the current context of research on the comprehensive service function of towns, few research results have focused on small towns in developing countries [37]. This paper takes a unique type of small town in China, the county town, as an example, and analyzes the spatio-temporal differentiation of the comprehensive service function of county towns in China and the influencing mechanism from a geographic point of view, which can make up for the inadequacy of the international research system of comprehensive service function of towns that pays little attention to the small towns of the developing countries. The findings of the study can also provide a theoretical basis for the improvement of the comprehensive service function of county towns in China.

The research findings of this paper also have policy applications in the following areas: The development of large cities in east China has an obvious “siphon effect” on the peripheral county towns, so county towns need to be actively integrated into the regional city network to avoid the shadow problem of aggregation, and the government should turn the “siphon effect” into a “radiation effect” through scientific and reasonable interventions. The enhancement of the comprehensive service function of county towns in central and north China relies on rapid economic growth, so county towns need to accelerate the enhancement of endogenous development capacity, increase the proportion of secondary and tertiary industries, and upgrade the industrial structure. The development of comprehensive service functions in county towns in south China has benefited from large government investments, and county towns should adapt their comprehensive service function development patterns to local conditions and gradually get rid of their high dependence on the role of the government. The northwest and southwest regions have a strong path dependence on population as well as government strength, so county towns should attract the rural population to gather through the strengthening of service capacity in medical care and education, et cetera, and the government should reasonably regulate the investment plan to ensure the development of weak zones focusing on construction. The main reasons for the slow development of comprehensive service functions in county towns in northeast China are serious population loss and insufficient driving force of the central cities in the region, so slowing down the population loss and enhancing the driving force of the central cities are the issues that need to be focused on.

There are also some limitations to the research in this paper. First, the selection of POI data, which are large and rich in information, makes data acquisition as well as processing more difficult; thus, only 2 years of point-in-time data were selected, ignoring the impact of continuous time period data on the study. In addition, this study characterizes the level of comprehensive service functions of county towns by the number of service facilities in the county towns, so the measurement of “quality by quantity” needs to be further deliberated. Meanwhile, in the analysis of the influencing factors of the spatio-temporal pattern of the comprehensive service function of county towns, some of the representative indicators could not be obtained from the Government Statistical Yearbook, which had an impact on the generalization of the research results. Secondly, there are some shortcomings in the research methodology of this paper: the random forest model may have the problem of too large decision trees when dealing with very large data sets, which may lead to bias in the analysis results. Finally, when this paper analyzes the influencing factors of the spatial and temporal differentiation of the comprehensive service functions of county towns in China, this paper only analyzes the influencing factors separately for each county town. In the future, the correlation between influencing factors and the spatial spillover effect of each influencing factor can be analyzed.