Does the Accessibility of Regional Internal and External Traffic Play the Same Role in Achieving Anti-Poverty Goals?

Abstract

Traffic development can promote the flow of goods and people, which has long been widely considered to have a poverty reduction effect but, in fact, is not unbreakable. The development of traffic is similar to economic and social development, with internal and external characteristics, but few studies have explored the differences between the effects of their poverty reduction. Taking the land traffic of the Chengdu-Chongqing Economic Zone (CCEZ) as an example, this paper represents traffic accessibility at a county level by relying on the average internal and external travel times. Rural poverty was identified by the pentagon of livelihoods to measure the Multidimensional Development Index (MDI). Furthermore, a Geographically Weighted Regression (GWR) model was used to explore the relationship and spatial differentiation characteristics between county traffic accessibility and poverty. The results show that the traffic accessibility of the counties in the CCEZ was quite different. The average internal travel time was between 0.16 and 7 h, and the average external travel time was between 4.2 and 10.6 h. The radiation gradient structure centered on Chengdu municipal districts and the Chongqing main urban area, and the accessibility level needed to be improved. Furthermore, the MDI values of each county in the CCEZ showed the structural characteristics of “large bottom and small top”; additionally, the higher the high-value group of MDI, the stronger the spatial aggregation and the more obvious the characteristics of regional differentiation. Finally, the relationship between traffic accessibility and poverty in counties cannot be generalized. The improvement of external traffic accessibility obviously helped to improve the poverty situation in the CCEZ; the improvement of internal traffic accessibility had a multidimensional impact, but it was mainly due to the occupation or spillover of livelihood capital in rural areas; counties accounting for 82.74% would even reduce the MDI and, thus, aggravate poverty.

1. Introduction

After 2015, the UN still sees global poverty reduction as the core of its agenda [1,2,3]. Indeed, anti-poverty has remained an important subject for all mankind [4,5,6]. While both international organizations and sovereign countries have struggled to reduce poverty, poverty still exists even in the richest countries [7]. From a global perspective, poverty remains persistent, and so, exploring the measures of poverty, drivers, and the logic of evolution and developing poverty reduction strategies [8,9,10] have never lost their popularity.

Although income factors are core and fundamental in poverty [11,12], poverty acts as a rather complex and stubborn social phenomenon [13,14], and its evaluation and recognition methods of poverty gradually tend to be multidimensional and comprehensive [15,16,17]. Among the indicators of quantitative evaluation, the Multidimensional Poverty Index (MPI) has been widely accepted for better echoing the United Nations Millennium Development Goals (MDGs) and has produced a series of studies [18,19,20]. Based on the three main dimensions of health, education, and standard of living, the MPI continues to be expanded and updated. While gaining a stronger representational ability, it also constantly faces questions and challenges. The study led by Masset [21] showed that the MPI was sensitive to small changes in a few metrics and was not necessarily suited to assess the effects of developmental interventions. Furthermore, the MPI was mainly used for families and stability after scaling was deemed controversial [14,18,22,23]. In contrast, the Sustainable Livelihood Approaches (SLAs) of the UK’s Department for International Development (DFID) emphasized multiple interactions between multiple factors, especially applicable to interventions on poverty reduction policies and measures [24,25]. Because countries often adopt multiple measures to reduce poverty [26,27,28,29], the Multidimensional Development Index (MDI) that matches the livelihood framework is more applicable in the geographical identification of poverty.

In order to formulate poverty reduction strategies, scholars from all over the world have systematically analyzed the impact of education, finance, trade, and industry on poverty from different perspectives. While affirming their positive role, researchers have also constantly revised their existing cognition. The research led by Brown et al. [30] suggested that the policy debate on education and poverty needed to be redefined and that the expansion of higher education was unlikely to reduce poverty; Nanivazo et al. [31] considered that due to the rigid national financial system, the entry of foreign banks had widened the wealth gap in Africa by 2.47%; the study led by Sae-Ra et al. [32] showed that export trade in rural poverty reduction was vague and even detrimental to poverty reduction in Asian countries; in turn, Li et al. [33] surveyed data using rural examples and showed that the poverty reduction effect of solar photovoltaic poverty alleviation projects (PPAP) was not effective. Taking Brazil as an example, Medeiros et al. [34] emphasized that infrastructure investment policies should take into account regional heterogeneity to achieve better poverty reduction results.

Transportation plays an important role in infrastructure, and its importance to economic growth has also been widely acknowledged [35,36,37]. In the process of economic development, opportunities are usually highly related to the mobility of people and goods, so an efficient transportation system brings positive multiplier benefits and ultimately reduces the cost of many economic sectors [38]. Rodrigue et al. [38] thought that the economic importance of transportation can be reflected from three main dimensions: core, operation, and geography, which have direct, indirect, and induced impacts on macro and micro economies. The existing research has provided abundant supporting materials. Adams et al. [39] thought that well-designed transportation projects have a great influence on activity sites and real estate values, significantly promoting economic development. Sheffi et al. [40] pointed out that the transportation and logistics cluster can improve the resource-sharing ability and economic efficiency of the industry, which were important reasons for maintaining the regional competitive advantage and economic density. Herzog et al. [41] put forward that the difference in market access caused by traffic led to the differences in employment in different counties, which, in turn, had an important impact on regional income level and economic growth.

Transportation is inseparable from the economy, and the economy determines poverty to a great extent. Therefore, transportation is closely related to rural poverty. Inadequate transportation may lead to social exclusion, which is manifested in the lack of employment opportunities, commodities, and public service resources [42,43,44,45]. A great deal of evidence shows that the disadvantages of transportation, the economy, and society cross each other in a mutually reinforcing way, forming a vicious circle [46,47,48,49]. Moreover, the transportation system may expand the hidden unfairness in vertical resource allocation [50,51,52], and it is more likely to have a sustained negative impact on the poor groups. From another point of view, the improvement of traffic conditions is an important aspect of poverty reduction [53]. The improvement of the coverage, affordability, and accessibility of the transportation system is likely to improve the poverty situation [54]. Adequate road infrastructures may also alleviate poverty by providing social access and economic opportunities [55].

As the largest developing country, China invests in transportation infrastructures concurrent with poverty reduction [56]. According to statistics [57], there were 250 million poor people in China in 1978, but by 2020, all poverty alleviation had been basically achieved despite the great increase of poverty line standards. Over the past 40 years of reform and opening-up, China has contributed more than 70% [48] to the global population’s poverty reduction, thanks to China’s effective governance. In the fight to eliminate poverty, the government has paid special attention to the positive role of transportation in reducing poverty. Along with the poverty alleviation of 250 million people, railway mileage has increased from 51,700 to 146,300 km from 1978 to 2020 and road mileage from 890,200 to 5,198,100 km [58]. In addition, in recent years, China’s development plan has placed more emphasis on the traffic poverty alleviation strategy [59] and promoted the “100 Traffic Poverty Alleviation Key Channel Project” and the “One Million Km Rural Road Project”. It can be predicted that China will still promote development through the transportation “hematopoietic” in the future. Since China will still vigorously promote traffic poverty reduction, it is meaningful to carry out further research and discussion on the topic.

At present, the research on the anti-poverty function of traffic accessibility has achieved some results. According to the traditional view, the improvement of traffic accessibility has a good anti-poverty effect, and a few researchers have questioned the anti-poverty effect of traffic accessibility. Some achievements in the exploration and discovery of this phenomenon are worth mentioning. Rodrigue et al. [38] analyzed the mechanism and thought that the advantages of economic agglomeration could not be easily reversed by improving traffic accessibility; because the process of improving traffic accessibility may be unproductive, it may eventually hinder regional development. Zhang et al. [60] provided evidence of a negative economic spillover caused by transportation infrastructures and considered that the positive impact of transportation on economic development was overestimated. Jiang et al. [61] used panel data analysis to show that for those underdeveloped provinces, high traffic accessibility often led to negative spillover effects of economic growth. Since the improvement of traffic accessibility may have a negative impact on the economy, its anti-poverty effect is debatable. Unfortunately, it is rare for studies to conduct in-depth discussions of the anti-poverty effect of traffic accessibility in specific economic areas alone and to conduct quantitative studies of traffic accessibility and anti-poverty effects, which may not be conducive to our clearer cognition.

Therefore, we designed a complete set of technical processes to quantify the traffic accessibility and poverty situation in counties of the Chengdu-Chongqing Economic Zone (CCEZ) to explore the differences between internal and external traffic accessibility and anti-poverty effects so as to fill the gaps in the existing research. This paper aims to answer three main questions:

(1)

What are the internal and external traffic conditions of the CCEZ?

(2)

What is the difference between the poverty reduction effect of internal and external traffic accessibility?

(3)

How should the CCEZ develop transportation in the future?

The paper was written after exploring and summarizing the abovementioned questions.

2. Data and Methods

2.1. Study Area

The Chengdu-Chongqing Economic Zone (CCEZ) is located in southwest China (101°56′~109°14′ E, 27°39′~33°02′ N), which is the region with the strongest comprehensive economic strength in western China (Figure 1). As a dual-core urban cluster relying on Chengdu and Chongqing, including 15 cities in Sichuan and 31 districts and counties in Chongqing, it is a pilot area for deepening inland opening-up and an important modern industrial base in China. In terms of natural endowment, the land in the economic zone is fertile, with high terrains at the edge and in the east, and low ones in the middle and in the west, featuring altitudes of 76~5700 m. There are four main terrains: plain, platform, mountains, and hills. In recent years, the CCEZ has witnessed rapid economic and social development, forming the aggregate benefit of “1 + 1 > 2”. The construction and development of the economic zone have become a national strategy. It should not only become an important growth pole to drive the development of the surrounding areas but should also improve its own facilities system to provide a demonstration for the coordinated development of urban and rural areas in China.

Data source: The definition of scope is based on the regional development plan of the Chengdu-Chongqing Economic Zone [62], and topography data comes from the Resource and Environment Science and Data Center [63].

2.2. Data Sources

Multiple data about the CCEZ were collected (

Table 1. Description of the data used for this paper.

Data Types	Data Description	Data Sources and References	Data Specific Corresponding Time Period
Road network data	Vector data, including roads of different grades, expressways, and railways, as well as high-speed railways.	Open Street Map [64]	1 October 2020
Baidu point of interest (POI)	Vector data, including expressway intersections, high-speed railway stations, and railway stations.	Baidu Maps [65]	1 October 2020
Digital elevation model (DEM)	Vector data; the Shuttle Radar Topography Mission (SRTM) digital elevation model with a 30 m resolution.	Geospatial Data Cloud [66]	11 February 2000
Land use/cover data	Raster data, obtained by visual interpretation of Landsat TM images with a 30 m resolution.	Resource and Environment Science and Data Center [63]	March–October 2020
Meteorological data	Data recorded in a notepad. Daily precipitation at meteorological monitoring stations (mm) from 1980 to 2015.	Meteorological Information Center [67]	January 1980–December 2015
Socio-economic data	Statistical data, including population, education, medical care, economic income, urbanization rate, energy, and other data of the administrative region.	Sichuan Provincial Bureau of Statistics [68] Chongqing Statistics Bureau [69]	The year of 2020

), including road network data, Baidu points of interest (POI), digital elevation models (DEMs), land use/cover data, meteorological data, and social–economic data. The data needed some pre-processing, including carrying out spatial correction and topology processing on the road network data, adjusting POI projection correction and matching with the road network, repairing DEM outliers, and filling in null values. These three data were used to make a travel cost raster. Moreover, land use/coverage data were reclassified, precipitation data of meteorological stations were calculated year by year using Kriging interpolation, and socio-economic data were collated. These data were used in the MDI calculation.

2.3. Methods

2.3.1. Transport Related Terminology

In the expression and discussion of this paper, some traffic terminology is inevitably used (

Table 2. Introduction of basic traffic terminology in this study.

Terminology	Abbreviate	Definition and Explanation	Unit
Travel Time	TT	The time required to reach the destination from the source point through a certain path.	hour
Internal Traffic	IT	Refers to the traffic within a specific area.	—
External Traffic	ET	Refers to the traffic outside a specific area. It should be noted that the outside is not infinite and will be limited by a larger area.	—
Traffic Accessibility	TA	In the transportation system, an index to measure the convenience of reaching the destination by one or more ways.	—
Average Internal Travel Time	AITT	Refers to the average travel time from the center of a region to the destination within the region.	hour
Average External Travel Time	AETT	Refers to the average travel time from the center of a certain area to the destination outside the area; outside the area, especially, is not arbitrary and will be limited by a larger area.	hour
Internal Accessibility	IA	Measures the convenience of starting from the center of an administrative region and reaching the destination in the administrative region.	—
External Accessibility	EA	Measures the convenience from the center of an administrative region to the destination outside the administrative region.	—
Cost Distance	CD	The least accumulative cost distance for each cell from or to the least-cost source over a cost surface.	min

). Among them, IT and ET are conceptual; TT, TA, IA, EA, AITT, and AETT are all quantifiable. The difference is that A, IA, and EA are dimensionless after quantization, while TT, AITT, and AETT are quantized in units of time. The basic definitions and explanations of these terms in this paper are unchanged, but they will be more flexible in specific expressions.

2.3.2. Overall Technical Route

The entire process of this paper using the data and methods is presented in Figure 2. There are three main steps. In the first step, internal accessibility (IA) and external accessibility (EA) were calculated for each county domain. This step would be done mainly in a Geographical Information System (GIS). The second step involved using the polygon of livelihood to measure the MDI, and the third step was to further explore the relationship between IA, EA, and MDI using the Geographically Weighted Regression (GWR) model. For the key methods and steps in the flowchart, such as obtaining a Travel Time Cost Raster (CR) and computing the Minimum Cost Distance Raster (MCR), the computational MDI for spatial statistics using GWR will be described in detail later.

2.3.3. Measurement of Traffic Accessibility

Travel Time Cost Raster (CR) and Minimum Cost Distance Raster (MCR)

We covered the entire study area with a 100 m × 100 m raster, counting the time required to pass through a certain raster as a cost. The time cost (TC) to pass a single raster is expressed as:

TC = 0.1/V ∗ 60

(1)

where “V” represents the velocity through the raster and the unit is km/h. The unit of TC is in min. Obviously, “V” is an important variable; there are three different cases for the determination of the “V” value: no transport infrastructure, only one kind of transport infrastructure, and multiple transport infrastructures crossing in a single raster. We determined “V” based on the situation and then determined and assigned TC (

Table 3. Assignment table of V and TC values.

On Foot	Slope (°)				Topographic Undulation (m)
	<5°	5–15°	15–25°	>25°	<25	25–50	50–100	>100
V (km/h)	6	4	2	1.5	6	4	2	1.5
TC (min)	1	1.5	3	4	1	1.5	3	4
Transport Infrastructure	High-speed railway	Railway	Expressway	First-class road or G	Second-class Road or S	Third-class Road or X	Fourth-class Road or Y
V (km/h)	250	100	100	80	60	40	30
TC (min)	0.024	0.06	0.06	0.075	0.1	0.15	0.2

Note: Adapted from Technical Standard of Highway Engineering [70]: G (National Road); S (Provincial Road); X (County Road); Y (Township Road). See Table A1 in Appendix A for details of the functions and geometric characteristics of the Expressway, First-class Road, Second-class Road, Third-class Road, and Fourth-class Road.

). If there was no road passing within the raster, the TC was determined based on the slope and topographic undulation calculated with the DEM, and if the raster had different TC values corresponding to slope and topographic undulation, the higher TC value was assigned to the raster. The corresponding TC assignment was used when only one kind of transport infrastructure passed within a single raster. The lowest TC value was assigned for the passage of multiple kinds of transport infrastructure within a single raster.

If there were only open transport infrastructures in our travel path, the Travel Time Cost Raster (CR) of the study area was obtained after assigning the time cost of every cell. Obviously, however, if the path contained closed transport infrastructures, the special circumstances of closed roads needed to be considered. In theory, entering and leaving high-speed railways and ordinary railways can only be done at stations, while entering and leaving highways can only happen at intersections. Therefore, it is necessary for us to conduct the buffer setting: set the buffer on both sides of the closed road at 300 m (Figure A1 in Appendix A), high-speed station and expressway intersection at 600 m, and railway station at 1000 m (Figure A2 in Appendix A). When choosing the closed road passage, the buffer area on both sides shall have the effect of blocking and prohibiting entry; the buffer area at the station or intersection shall have a connection effect and entry is allowed.

In GIS, the travel time raster distinguished from the mode of travel could be attained using buffer analysis and a raster calculator as an open transport infrastructure cost raster (OCR), a high-speed-rail cost raster (HCR), a route-rail cost raster (RCR), and expressway cost raster (ECR). Among them, OCR is the most basic cost raster. In HCR, RCR, and ECR, the main line of closed road was assigned according to Table 3, and the buffer rasters on both sides were set as null; the other rasters were consistent with OCR.

Taking OCR, HCR, and ECR as input rasters, respectively, the center of each county was set as the starting point, and the cost distance calculation was used. Specifically, for each raster, the values of the four cost paths were sorted, where the minimum value was set as the value of the raster in the MCR. The whole process was done in the Arcpy interface.

Calculating Accessibility

The partition statistics for each MCR were used to obtain the average external travel time (AETT) and average internal travel time (AITT) for each county, and traffic accessibility was further calculated using the formulas:

$$I{A}_i=\frac{MAX\left(AIT{T}_i\right)-AIT{T}_i}{MAX\left(AIT{T}_i\right)-MIN\left(AIT{T}_i\right)}$$

(2)

$$E{A}_i=\frac{MAX\left(AET{T}_i\right)-AET{T}_i}{MAX\left(AET{T}_i\right)-MIN\left(AET{T}_i\right)}$$

(3)

where “$I{A}_i$” represents the traffic accessibility within the i-th county; “$AIT{T}_i$” represents the average internal travel time of the i-th county; “$E{A}_i$” represents the external traffic accessibility of the i-th county; “$AET{T}_i$” represents the average external travel time of the i-th county.

2.3.4. Measurement of the Multidimensional Development Index (MDI)

Order Relation Analysis Method

According to the principle of livelihood theory and the situation of social and economic development, livelihood capital was analyzed and the indicators were selected, as shown in

Table 4. Rational assignment of importance.

${\mathit{r}}_{\mathit{j}}$	Meaning
1.0	$X_{j-1}$ and $X_j$ are equally important.
1.2	$X_{j-1}$ is slightly more important than $X_j$.
1.5	$X_{j-1}$ is obviously more important than $X_j$.
1.8	$X_{j-1}$ is significantly more important than $X_j$.
2.0	$X_{j-1}$ is more extremely important than $X_j$.

Note: Adapted from Fu, et al. [71].

, where the weights were obtained with the Order Relation Analysis method [71,72].

Order Relation Analysis is a method used to fully reflect the will of experts. Its analysis steps are as follows:

(1)

Rank the m-th evaluation metrics by the importance relationship and set as {X₁, X₂,…, X_m}; the relation is satisfied:

$$X_1\ge X_2\ge X_3\dots \ge X_m$$

(4)

(2)

According to the judgment, the evaluation index sets the ratio according to the hierarchy relationship, and the satisfaction relation between the j-th index and the j−1-th index is satisfied:

$$\frac{w_{j-1}}{w_j}=r_j, j=m, m-1, \dots , 2$$

(5)

where “$w_{j-1}$” and “$w_j$” are the weights of the j-th index, and “$r_j$” represents the importance ratio. For the assignments in this article, see Table 3.

Poverty Identification Method

Combined with the study measuring poverty led by Liu et al. [73], this study used five livelihood capital dimensions proposed by the SLA using the MDI. For setting the indicators of poverty based on the five dimensions of the sustainable livelihood framework, 11 indicators and 14 sub-indicators were selected, and hierarchy relationship analysis was used to obtain the appropriate weights (

Table 5. Indicator system and weights for multidimensional poverty at the county level.

Dimension	Indicator	Weight	Sub-Indicator	Weight
Financial capital	Net income	1	Per capita net income of permanent rural residents	1
Human capital	Comprehensive labor capacity	0.5	Percentage of rural people aged 15–59	1
	Health level	0.3	Rural per capita living consumption expenditure	0.5
			Number of health technicians per 100 rural residents	0.5
	Educational strength	0.2	Full-time teachers of per 100 students	1
Natural capital	Level of quantity and quality of cultivated land	0.45	Average area of farmland per rural person	0.6
			Average grain production per sown area	0.4
	Level of forest and grass resources	0.25	Average area of forest per rural person	0.6
			Average area of grassland per rural person	0.4
	Level of water resources	0.3	Average precipitation for 26 years (1980–2015)	1
Physical capital	Grade of fixed assets	0.67	Average area of construction land per rural person	1
	Level of electricity usage	0.33	Annual electricity consumption per rural resident	1
Social capital	Level of urbanization	0.5	Rate of urbanization	1
	Level of social support	0.5	Income ratio between rural and urban residents	1

Note: Adapted from Liu and Xu [73].

).

For certain dimensions, the pentagon of livelihoods [73] was used to calculate the MDI (Figure 3). The specific calculation formula for the MDI is as follows:

$$\mathrm{MDI}=\mathrm{a}\ast \mathrm{b}+\mathrm{b}\ast \mathrm{c}+\mathrm{c}\ast \mathrm{d}+\mathrm{d}\ast \mathrm{e}+\mathrm{e}\ast \mathrm{a}+\mathrm{a}\ast \mathrm{c}+\mathrm{c}\ast \mathrm{e}+\mathrm{e}\ast \mathrm{b}+\mathrm{b}\ast \mathrm{d}+\mathrm{d}\ast \mathrm{a}$$

(6)

where a, b, c, d, and e correspond to the edge lengths of different dimensions in the pentagon of livelihoods.

2.3.5. Geographically Weighted Regression (GWR)

The spatial proximity of the observation points of variables makes the observation values of variables correlate with each other, which is the prerequisite for GWR model analysis. Therefore, before constructing the GWR model, the dependent variables of the model should be statistically tested with the spatial autocorrelation test. The measurement method is based on the Global Moran’s I index. Moran’s I index takes into account the relationship between the numerical value and two-dimensional space and can fully reflect the similarity of attribute values of adjacent areas in space. Global Moran’s I is as follows:

$$\mathrm{I}=\frac{n{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^n{w}_{ij}\left(X_i-\overline{X}\right)\left(X_j-\overline{X}\right)}{\left({{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^n{w}_{ij}\right){{\displaystyle \sum }}_{i=1}^n{\left(X_i-\overline{X}\right)}^2}$$

(7)

where “$w_{ij}$” is the spatial weight matrix; $“X_i”$ and $“X_j”$ are the observed values of i-th and j-th, respectively; $“n”$ is the number of spatial units. When I > 0, it means agglomeration in space; when I < 0, it means dispersion in space; the closer I is to 0, the more it tends to random distribution.

Social development has an inseparable relation to the geographical environment; in the field of geoscience, GWR models are commonly used to mine numerical relationships between multiple variables with spatially divergent features. This model can be applied in the study of traffic-related anti-poverty problems. The regression analysis of traffic accessibility and MDI in districts and counties to build a GWR model [74] follows:

$$Y_i={\beta }_0(u_i,v_i)+\sum _n{\beta }_n\left(u_i,v_i\right)X_i+{\epsilon }_i$$

(8)

where “$u_i, v_i$” is the spatial longitude and latitude coordinates of the center point of the i-th county, and “$Y_i$” is the multidimensional poverty index of the i-th county; “${\beta }_0{u}_i, v_i$” is the constant term of the i-th county area; “${\beta }_n{u}_i, v_i$” is the regression coefficient of the i-th county area, which is the estimated value of the continuous function related to the spatial position at the point “i”. “${\beta }_n{u}_i, v_i$” can be positive or negative. When it is positive, it represents a positive effect, and when it is negative, it represents a negative effect.

“$X_i$” is the independent explanatory variables. “${\epsilon }_i$” is the random error, which obeys a random distribution with a constant variance.

Regression coefficients were solved using the GWR model in GIS to explore the spatial divergence features of poverty and traffic accessibility with MDI values as dependent variables and EA and IA as base explanatory variables. The setting of the spatial kernel function and the kernel broadband in the model had a great impact on the fitting results of the GWR model. The kernel type was selected as “ADAPTIVE” and the bandwidth method as “AICC”; that is, the minimum information criterion to determine the range of the kernel and the superiority of fitting was improved when considering the difference of different degrees of freedom of different models.

3. Results

3.1. Traffic Accessibility

3.1.1. Average Travel Time and Pattern of Counties

The AITT and AETT values of the 97 counties in the CCEZ are shown in Figure 4. Combined with the natural breakpoint method and people’s travel habits, they were divided into five levels. The minimum AITT was 0.16 h, while the maximum was 7 h. There were 56 counties with AITT values less than one hour, accounting for 57.73%, mainly in the central part of the CCEZ. There were 30 counties with AITT values between 1 and 3 h, accounting for 30.93%, mainly distributed in the eastern and southern fringes of the CCEZ. There were 11 counties with AITT values between 3 h and 7 h, accounting for 11.34%, mainly distributed in the western edge of the CCEZ. Obviously, the distribution of counties with AITT values of less than three hours was more concentrated, while those with values of more than 3 h were more dispersed. As can also be seen from the violin diagram, in the value interval with larger AIIT values, the phenomenon of discrete distribution was more obvious (the upper and lower spans are large).

The values of AETT ranged from 4.2 to 10.6. The number of counties with AETT values in the range of (4.2, 5.5) and (5.5, 10.6) accounted for 51.54% and 48.46% of the CCEZ, respectively. Obvious differences were shown in the marginal and non-marginal areas of the CCEZ. Counties with AETT values of less than 5.5 h were almost all located in the center, corresponding to counties greater than 5.5 h, forming almost a closed marginal ring. Similarly, the larger the value interval of AEIT, the more discrete the spatial distribution of the counties.

From the linear fit graph of AITT and AETT (the significance level was 0.05, and the Pearson correlation coefficient was 0.823), it can be seen that there was at least a 95% confidence level that AITT and AETT had a significant positive correlation. The scatter plots showed that the vast majority of the observations fell within the 95% predicted band. The knowledge based on statistics could determine that internal and external traffic development were well synchronized (see Figure A3 in Appendix B for other significance levels).

3.1.2. Distribution Pattern of Traffic Accessibility in the Counties

The IA and EA values of the 97 counties in the CCEZ are shown in Figure 5. Using the natural breakpoint method, they were divided into four levels:

(1)

Low accessibility (L, 0.00~0.45). There were seven counties whose IA values belonged to this level (mainly distributed in the west of the CCEZ) and 14 counties whose EA values belonged to this level (distributed along both the eastern and western edges).

(2)

Intermediate accessibility (I, 0.45~0.75). Counties whose IA or EA values belonged to this level were distributed at the edge of the CCEZ in all directions, with obvious dispersion. There were 15 counties with IA level I, mainly distributed in the four corners of the CCEZ, while 25 counties with EA level I were adjacent to the marginal counties, with low accessibility, and were distributed in pieces.

(3)

Upper accessibility (U, 0.75~0.85). Counties whose IA or EA values at this level were obviously adjacent to the municipal districts. The difference is that the group with higher EA values included 23 counties surrounding municipal districts, while the other group included 14 counties with higher IA values, which also surrounded the municipal districts but were obviously further away from the central area.

(4)

Higher accessibility (H, 0.85~1). There were 61 counties and 35 counties whose IA and EA values belonged to level H, respectively, mainly distributed between multiple municipal districts and clustered in the central region.

We compared the levels of EA and IA in each county. The four counties with increased levels were all in the west, while the 51 counties with unchanged values were distributed all over the CCEZ. There were 33 and 9 counties that had dropped one grade and two grades, respectively. In addition to being affected by the level of traffic development, IA was also affected by the area and compactness of the county; hence, some marginal counties could have a higher IA level. However, the influence of area and shape for external transportation was weakened, and remote counties largely could not maintain this level.

3.2. Analysis of Poverty Status

3.2.1. Single-dimensional Livelihood Capital

The score values of each single dimension of the CCEZ were standardized and are shown in Figure 6. From the dimension of financial capital, the high values were concentrated in the counties adjacent to Chengdu City and the west of the main urban area of Chongqing. Secondly, there was a small-scale distribution of counties in the south of Yibin and Meishan City, north of Mianyang City. The low values mainly appeared in the counties located at the edge and not adjacent to the above municipal jurisdiction. The middle-value counties were mainly distributed in the connection direction perpendicular to Chengdu-Chongqing (Nanchong-Suining-Neijiang-Zigong), followed by those in the northeast of the main urban area of Chongqing. The high-value areas of the human capital dimension score appeared in the west of the CCEZ, and the medium- to low-value areas were distributed in each orientation. The scores of the natural capital dimension obviously had advantages in marginal counties, while the scores of counties around the Chengdu-Chongqing link were relatively low. The high scores of material capital and social capital were mainly found around the counties of Chengdu and the main urban areas of Chongqing, and the distribution of low-value counties was very scattered.

3.2.2. Multidimensional Development Index (MDI)

The counties with the lowest and sub-low MDI values appeared in mountainous and hilly areas, such as Mianyang and the Ya’an mountains in the west, eastern Sichuan and the western Chongqing hills in the middle, southeastern Sichuan mountains in the south, and northeast and southeast Chongqing mountains in the west (Figure 7). The counties of the highest value group were concentrated in the core area of the Chengdu Plain. The counties of the second-highest group were concentrated in the western part of the Chengdu Plain and sporadically appeared in the hills of western Chongqing. The group with medium MPI values was mainly located in the transitional interlaced zone between “plain–platform–hill–mountain”, surrounded by the Chengdu Plain and the main urban area of Chongqing. As seen among the different groups, the higher the high-value group of MDI, the stronger the spatial aggregation, and the characteristics of regional differentiation were obvious. The highest value groups and the sub-high groups were only distributed in a corner of the Chengdu Plain, while the lowest value groups and sub-low group counties were widely distributed in each region. There were 29, 41, 16, 7, and 4 counties from the lowest group to the highest group, respectively, showing the quantitative structure characteristics of “large bottom and small top”.

3.3. Spatial Differentiation of Anti-Poverty Effects of Traffic Accessibility at a County Level

On the premise that MDI passes Moran’s I index test (Figure A4 in Appendix B), Figure 8 is the result of the spatial visualization of coefficients of explanatory variables. The system values of EA were all positive, indicating that the improvement of external traffic accessibility promoted MDI and decreased multidimensional poverty in the county. The coefficient value of IA was positive or negative, indicating that the relationship between the increase in internal traffic accessibility and the MDI value was non-uni-dimensional, and the increase in internal traffic accessibility could improve or restrict multidimensional poverty. According to the parameter values of the explanatory variables, the anti-poverty effect of the accessibility of county traffic and land transportation had the following two main characteristics.

On the one hand, the improvement of external traffic accessibility was conducive to the anti-poverty of the whole CCEZ, although the sensitivity of the action was different. Obviously, the improvement of the same degree of external traffic accessibility had a significant/obvious anti-poverty effect in the western counties of the CCEZ but a relatively weak effect in the central region and a slight effect in the eastern region.

On the other hand, the improvement of internal traffic accessibility may have limited or even had negative effects on the anti-poverty effects of individual counties in the CCEZ. As shown by the coefficient of IA, the increase in internal traffic accessibility in 82.47% (80/97) of counties would differently lead to increased poverty. The spatial distribution characteristics showed that the contiguous counties located in the vertical direction of the Chengdu-Chongqing line (Nanchong-Suining-Neijiang-Zigong-Yibin) were most obviously restricted, while the large counties in the northeast and northwest were restricted to a certain extent. Southeast of the economic zone, increased traffic accessibility within the county would help reduce poverty.

4. Discussion

4.1. Explanation for Results

The traffic accessibility of counties in the CCEZ was not balanced. Driven by the dual core of “Chengdu-Chongqing”, the accessibility level was spatially manifested as a radiation gradient structure, with the jurisdiction of Chengdu and the main urban area of Chongqing as the centers. In addition, seeing as it is affected by the natural conditions of the county itself and urban–rural integrated development policies, the distribution was different in different directions of the central city, and the overall traffic accessibility level was characterized by “central > marginal and plain counties > mountainous and hilly counties”. All these are in line with our theories. Such traffic conditions may widely exist in urban agglomerations and economic zones [75,76]. Similarly, the livelihood capital of the county level also presented the characteristics of local concentration and overall dispersion around the city jurisdiction. As megacities, Chengdu and Chongqing show strong centrality, with significant advantages of financial capital, material capital, and social capital, which basically define the distribution pattern of the two axes. However, human capital and natural capital are natural resources and environmental advantages. As a key factor, they are more dependent on their own situation than external influence, so it may also be advantageous for them to be located in marginal counties. After the natural segmentation of MDI obtained from five livelihood capital syntheses, there were structural characteristics of “large bottom and small top”. There were only four counties in the highest value group, distributed in a corner of the Chengdu Plain. This not only confirms the super radiation-driving effect of core cities but also raises new concerns. Because of the existence of a radiation gradient, the difference between counties in economic areas may continue to expand.

The anti-poverty effects of external traffic accessibility and internal traffic accessibility are different. The improvement of external traffic accessibility has a positive anti-poverty effect, although this effect is not balanced in the entire economic zone. The improvement of internal traffic accessibility may have a limited or even negative effect on anti-poverty in a single county in the CCEZ. As can be seen from the coefficient of IA, the improvement in internal traffic accessibility in 82.47% (80/97) of counties will unevenly lead to increased poverty. Thus, we can perhaps understand external and internal traffic with both overall and local thinking. In fact, in addition to the unchangeable geographical location, external traffic accessibility was mainly affected by the overall traffic conditions of the CCEZ. In addition to the shape characteristics of counties, accessibility was mainly affected by the development of internal transportation. In other words, to some extent, the improvement of external traffic accessibility can be regarded as an improvement of the overall traffic conditions in the economic zone, while the improvement of IA only improves the traffic conditions in individual counties. When internal traffic is the main limiting factor of development, increasing investments can promote development; if not, it may occupy other capital or capital spillover, which hinders development.

4.2. Policy Implication

Transportation plays an irreplaceable role in the growth of regional economies. In vast developing countries and regions, there will be large-scale and long-term transportation infrastructure construction in the future. At the same time, poverty is still a chronic social problem in the world, and poverty reduction is an important sustainable development goal. Ideally, we should, therefore, reduce poverty to the greatest extent by investing in transportation. However, some challenges remain. First, the improvement of transportation does not necessarily reduce poverty. The logical chain of evolution in the middle is complex, from the improvement of transportation to the substantial poverty reduction effect. If improving transportation cannot make rural areas benefit from a certain aspect of livelihood capital, such as increasing job opportunities, obtaining a higher income, getting a better education, and accessing medical care, there will be no obvious effect of poverty reduction. Moreover, in the process of investing in transportation infrastructures, if resources such as cultivated lands and forests are occupied or the ecological environment is destroyed, poverty may be aggravated. Second, the improvement of traffic is not conducive to the development of some counties over a certain period of time. With the improvement of traffic, the siphon benefit of central cities will be enhanced, human capital and financial capital will probably flow out, and the gap between urban and rural areas will also widen, thus hindering the development of rural areas in counties.

In view of the fact that improving transportation does not necessarily achieve the goal of poverty reduction, it appears necessary to rethink the strategy of transportation poverty reduction. Regional traffic definitely needs to be developed, but we should also have some supporting plans. First of all, when developing regional transportation, it is necessary to make a reasonable layout, formulate an overall plan, and ensure that the transportation infrastructure plays the greatest role. Secondly, while improving transportation, we should encourage local areas where conditions permit the development of rural tourism, green food processing, and other non-agricultural industries so as to increase rural livelihood capital. In addition, for special areas, it is not necessary to forcibly improve traffic but to consider reducing poverty through relocation to different places.

4.3. Comparison with Results Found in Other Studies

Our research proposes that improving traffic is not a sufficient condition for economic growth and may even have negative effects on this, which is quite consistent with many studies. Asher et al. [77] evaluated the economic impact of India’s large-scale projects aimed at popularizing all-weather rural areas and found that it was far less than expected by policy formulation, with little impact on assets, agricultural investments, or expected consumption; the projects produced only a small change in the employment of rural enterprises. Parinduri et al. [78] conducted a longitudinal survey of Indonesian families and found that transportation infrastructures did not increase the income level in rural areas. Anastasia et al. [79] led case studies to show that the rail transit (blue line) passing through poor communities in the Los Angeles county failed to improve the surrounding economic environment. All these studies show that other supporting conditions are needed to reduce poverty by improving transportation. In addition, some scholars have mentioned the possible negative effects of traffic in their research. Rodrigue et al. [38] pointed out that there is a risk of over-investment in transportation. When economic growth is driven by non-transportation factors, the problem of improper capital allocation may occur. Vaturi et al. [80] carried out a case study of the greater Tel Aviv metropolitan area, which showed that the improvement of traffic accessibility might instead limit the population growth in the marginal areas. Therefore, our research results doubt the effect of poverty reduction by improving traffic accessibility, but supporting evidence can still be found in the existing literature.

4.4. Limitations and Constraints

We have optimized the technical route, including but not limited to using 100 × 100 m grids, grabbing POI stations, buffer analysis, setting different transportation modes, and comparing cost grids, all in order to obtain the most reliable accessibility results. The calculation of MDI covers the main aspects of five livelihood capital dimensions as much as possible. However, there are still some issues to be discussed.

(1)

Considering the idealized transportation process, we assume that the flow of people or goods can pass through the raster unit only with a certain time cost and the cumulative time cost of the route from the source to the destination is regarded as the cost distance, all of which are ideal values. In reality, the connection of transportation modes, changing routes, passing through special areas, congestion and so on may increase the time cost, but this study did not consider these factors. In addition, we assigned speed values to the passing raster cells according to the infrastructure type, which was also a simplified case. The real situation is that the speed v is different in different road sections and road conditions.

(2)

Using MDI to identify poverty is insufficient. For the identification of multidimensional poverty in rural areas, we have not considered vulnerable social backgrounds. Existing research shows that aging [81], unemployment [82], and social assistance [83] can all reflect poverty. However, due to the difficulty in obtaining data, we have not included them in the index system. Therefore, this paper lacks social environment adjustment indicators other than livelihood capital, which may exaggerate or underestimate the poverty level.

5. Conclusions

Considering the example of the CCEZ, this study uses the OSM road network and natural geographic data in 2020 to calculate the traffic accessibility by cost distance and identify rural poverty by establishing a multidimensional poverty identification index system, combined with social and economic statistical data from 2020. The livelihood Pentagon method was also used to calculate the MDI. Then, using the GWR model, this paper has discussed the relationship between the internal and external accessibility of counties and poverty and its spatial differentiation characteristics so as to provide policy suggestions for decision-makers.

The CCEZ has an average internal traffic time of 0.16–7 h and an average external traffic time of 4.2–10.6 h. In space, it shows a radiation gradient structure centered on the jurisdiction of Chengdu and the main urban area of Chongqing; thus, the accessibility level needs to be improved. The results of the GWR model with MDI values as dependent variables and EA and IA as explanatory variables show that the coefficient values of EA are all positive, indicating that the improvement of foreign traffic accessibility will promote the improvement of MDI and the multidimensional poverty level of counties will decrease. The coefficient value of IA is positive or negative, which indicates that the relationship between the improvement of internal traffic accessibility and the MDI value is not one-dimensional, and the improvement of internal traffic accessibility in 82.47% (80/97) of the counties may lead to the aggravation of poverty to different extents.

In a specific economic zone, when promoting the strategy of poverty alleviation by transportation, the key is overall scientific planning, and the difference in the anti-poverty effect of accessibility to external and internal traffic is actually the overall and local difference. Worldwide, there are many regions similar to the CCEZ, and the development of transportation and poverty reduction go hand in hand in time and space. Infrastructure investment should play an active role as much as possible to improve the regional economic situation and achieve the substantial effect of reducing poverty. Therefore, the existence of situations different from conventional thinking should be urgently recognized, and the relationship between developing transportation and reducing poverty should be coordinated. We advocate that in regional transportation planning, priority should be given to the overall situation and the combined benefits of comprehensive transportation should be continuously improved. This means that county transportation planning and investment may need to give way to the overall transportation strategy of the economic zone. On the other hand, it is not obvious that only relying on the development of the county’s internal transportation can reduce poverty. At this time, we can consider changing the production and lifestyle of the region and breaking the bottleneck of livelihood capital, such as by optimizing the production mode, relocating to different places, providing ecological compensation, developing education systems, and ensuring social security to help reduce poverty in individual counties.