The Relative Roles of Socioeconomic Factors and Governance Policies in Urban Traffic Congestion: A Global Perspective

Abstract

There is still doubt about how to accurately distinguish the objective and subjective factors that affect traffic congestion. This study aims to re-examine the role of socioeconomic factors and governance policies in the variation of urban traffic congestion from a global perspective at the macro-social development level. The heterogeneous characteristics of the traffic congestion index under the influence of changes in governance policies and variations of social–economic factors are distinguished through a residual trend model. Different time scenarios are addressed in the study to assess the respective contribution of socioeconomic factors and governance policies to variations of traffic congestion, which is helpful for identifying the most destructive factors and the most effective combination of governance policies. Taking Beijing City as the case, the results show that the social-economic factors and governance policies jointly drive congestion fluctuations, but the contribution of governance policies is greater. In addition, governance policies play an important role in alleviating traffic congestion, but unfortunately, an unreasonable combination of governance policies is also the chief culprit for the rise of the traffic congestion index in certain periods. The findings presented here can make us understand macro-mechanisms in response to urban traffic congestion and provide evidence for formulating regional development policies.

1. Introduction

The modeling process to solve traffic problems is often limited by various constraints, which will make certain traffic problems impossible to overcome completely, one of which is traffic congestion [1,2]. With the rapid development of cities, traffic congestion has spread in major and even small and medium-sized cities around the world, showing global characteristics, thus attributing traffic congestion to the disorderly expansion of cities as well as social and economic development [3]. Obviously, the most fundamental source of traffic congestion is the imbalance between supply and demand [4], but it is still unknown the role played by the level of socioeconomic development in this balance difference [5]. The economic base often determines the superstructure. In the future, the total economic output of all countries, especially developing countries, will certainly continue rising, which will make various factors of the social system more complicated. Therefore, it is one of the key steps to solving urban traffic congestion by analyzing the interactive relationship between traffic congestion and social and economic development. In addition, many traffic paradoxes have been bothering planners and researchers, the most typical of which is that “the more roads are built, the more serious the congestion will be” [6]. It is assumed that the new construction of urban roads will lead to a substantial increase in traffic demand, which will further aggravate the imbalance between traffic supply and demand. However, there is still a lack of quantitative proof in this respect, and any conjecture without data is not persuasive. More importantly, governance policies are usually considered the most effective measure to alleviate traffic congestion, especially by managing traffic demand, so as to improve the match between traffic supply and demand [7,8]. The biggest question is whether all governance policies are effective in different periods, or in other words, which combination of governance policies is more effective, which may also be the focus of our attention.

The research on influencing factors of traffic congestion is the premise and foundation of traffic planning as well as management, which plays an important role in the prevention and elimination of traffic congestion. In the existing literatures, analyzing the impact of traffic congestion on socioeconomic development status attracts the most interest. For example, Lomendra et al. [9] used the convenience sampling method and selected 100 respondents so as to show that traffic congestion has a negative impact on all aspects of society; Jin et al. [10] demonstrate that traffic congestion growth has a negative impact on income growth and employment growth; Sweet et al. [11] used statistics from 88 US cities to estimate the impact of traffic congestion on job growth and productivity growth; Javid et al. [12] established a carpooling structure model to prove that social environment and economic benefits play an important decisive role in promoting carpooling; Weerasinghe et al. [13] investigated a sample of 420 employees who are vulnerable to traffic congestion in Colombo metropolitan area, and concluded that traffic congestion would bring more pressure to commuters. Thomson et al. [14], Shao et al. [15] and Zhu et al. [16] respectively utilized the cost approach, theoretical models and system dynamics models to discuss the socioeconomic effects of traffic congestion and the relationships among various complex systems. On the contrary, relatively few papers have studied the common impact of socioeconomic factors on the level of traffic congestion. Dadashova et al. [5] considered the spatial-temporal autocorrelation of traffic congestion, deployed multiple time series models to explain the complex relationship between performance indicators and socioeconomic factors, and finally determined the most influential factors affecting system performance.

In general, most of the studies analyze the causes of traffic congestion, propagation mechanisms, and solutions from a relatively microscopic perspective, while fewer macro traffic studies have been conducted to address traffic congestion. For example, S. Karthik et al. [17] evaluated traffic congestion index based on congestion costs and indicators using travel time models; Toan et al. [18] employed fuzzy theory to quantify highway congestion levels based on traffic flow speed and density; Bai et al. [19] constructed the congestion matrix of regional traffic network based on the concept of the relative position of road nodes and convolution long-term and short-term memory network method to predict congestion at all locations in the road network in the near future. On the basis of the cause analysis, traffic congestion prediction has also been further developed using linear and nonlinear methods, which include time series, logistic regression, linear regression, nonlinear algorithms as well as Markov, neural network, support vector machine, etc. Tampubolon et al. [20] proposed the traffic prediction method of road network width (RNW) that converts traffic data into images representing the spatial-temporal relationships between RNWs; Bouyahia et al. [21] introduced a two-stage traffic resource scheduling strategies with Markov Random Fields (MRF) to predict the spread of road traffic congestion; Kohan et al. [22] predicted the development of traffic congestion through the analysis of movement trajectories. A review of the existing research literature reveals that there are few studies considering the quantitative statistical relationships between socioeconomic factors and traffic congestion levels, especially the lack of attention to governance policies. Meanwhile, distinguishing between the influences of objective factors and governing policies is also a challenge that needs attention.

The aim of this article is to respond to the above-mentioned challenges by applying a residual-trend model and obtaining spatial panel data so as to address the following research questions:

①

Under the effect of variations in socioeconomic factors and governance policies, what can occur in traffic congestion?

②

How extensive are the impacts of socioeconomic factors and governance policies on the variations of traffic congestion?

③

What are the most drivers for variations of traffic congestion? What is the best combination of governance policies for alleviating traffic congestion?

The answers to these questions will play an important role in guiding the practice of traffic engineering and urban planning. Spatial panel data in 2000~2018 is employed in this study to quantitatively evaluate the influence of socioeconomic factors and governance policies on the variations of traffic congestion in Beijing City over the past 19 years, meanwhile understanding the trends and driving mechanisms of variations in traffic congestion. The results here will provide a scientific basis for further building traffic infrastructures and formulating regional development policies. The remainder of the paper is structured as follows. Section 2 provides an overview of the methodologies, as well as a detailed description of a residual-trend model and a relative role analysis. Additionally, the hierarchical method of “traffic congestion index” (referred to as CI) is described in Section 2 and the distinctiveness between the influence of socioeconomic factors as well as policies is emphasized. In Section 3, the data sources and the results of exploratory analyses of the data are described. In Section 4, Beijing City is taken as the study case to demonstrate the specific operation process of an assessment of influence factors, the contribution difference between social and economic factors and policy factors as well as the temporal volatility of the results. This study is concluded and discussed in Section 4.

2. Materials and Methods

2.1. Trend Analysis

The trend of congestion level will be indicated by the traffic congestion index using linear regression slope, as follows [23]:

$$slop{e}_{CI}=\frac{n{\displaystyle \sum _{i=1}^n(i\times C{I}_i)-({\displaystyle \sum _{i=1}^{n}i})({\displaystyle \sum _{i=1}^{n}C{I}_i)}}}{n{\displaystyle \sum _{i=1}^n{i}^2-{({\displaystyle \sum _{i=1}^{n}i})}^2}}$$

(1)

Slope_CI is the slope of the congestion index regression equation; i is the number of study years from 1 to n; n is the total number of years in the study period; CI is the traffic congestion index. The slope value reflects the trend of the congestion index data in a region during the study period. When the Slope_CI > 0, the CI tends to increase during the study period. Moreover, the larger the Slope_CI, the more obvious the CI increase. When Slope_CI < 0, on the contrary, the CI tends to decrease during the study period.

2.2. Residual Trending Method

In order to distinguish the influences of socioeconomic factors and governance policies on CI variations, the residual-trend method proposed by Evans and Geerken [24] is employed in this study. By establishing the regression model between socioeconomic factors and CI, the annual CI values determined by economic factors are predicted. The impact of governance policies on CI can be expressed by the variation in the residual value, which is the difference between the observed CI value and the predicted CI value of socioeconomic factors [23], as shown in the following.

$$C{I}_{HA}=C{I}_{obs}-C{I}_{CC}$$

(2)

CI_HA is the residual of the traffic congestion index; CI_obs is the actual observed value of the traffic congestion index; CI_CC is the predicted value of the traffic congestion index. The variations in the residuals of the CI over the time series range from 2000 to 2018 will be analyzed: if the residual is greater than zero, the policy can make the CI higher; if the result is negative, the traffic congestion is improved by the influence of the changed policy.

For the regression model of socioeconomic factors for CI, we select the gross domestic product (GDP), vehicle ownership, and population size, which all pass the significance test and correlation test. Pearson’s correlation coefficient (significance) between each socioeconomic factor and traffic congestion index is shown in the order of Equation (3) as follows: −0.609 ** (0.006), −0.668 ** (0.002), −0.673 ** (0.002), −0.615 ** (0.005), −0.437 (0.062) and −0.665 ** (0.002). Then, we establish a six-element linear regression equation between CI and socioeconomic factors for the years 2000 to 2018. The congestion levels of the study area during all the study periods are discussed one by one. The equation is as follows:

$$C{I}_{CC}(t)=a\times GDP+b\times VP+c\times PP+d\times URM+e\times UBM+f\times UWM+g$$

(3)

where, CI_CC(t) is the predicted traffic congestion index influenced by socioeconomic factors in tth year; a is the regression coefficient of traffic congestion level on regional economic level (GDP); b is the regression coefficient of traffic congestion level on regional vehicle ownership (VP); c is the regression coefficient of traffic congestion level on regional population number (PP); d is the regression coefficient of traffic congestion level on urban road mileage (URM); e is the regression coefficient of traffic congestion level on urban bus mileage (UBM); f is the regression coefficient of traffic congestion level on urban railway mileage (UWM); g is a constant. We perform a linear fit between CI and each socioeconomic factor, as shown in Figure 1. Figure 1 shows that there is a certain linear relationship between socioeconomic factors and the traffic congestion index. Although this relationship is large or small, it is sufficient to support our study of the relative role of socioeconomic factors in traffic congestion changes by using linear regression models.

2.3. The Analysis of Relative Role

In this paper, we adopt the idea and method of calculating relative roles proposed by Duanyang [25], so as to identify the relative roles of socioeconomic factors and governance policies in the variations process of CI, as shown in Figure 2. First, through residual analysis, we obtain the CI values caused by socioeconomic factors and governance policies in each period separately.

The linear trend rates of CI_CC and CI_HA in a city can be calculated according to Equation (1), which, respectively, represents the annual trend of CI changes under the influence of socioeconomic factors and governance policies. A positive trend rate represents that socioeconomic factors or governance policies can promote the increase of CI, which has a negative impact on the service level of the transportation system. On the contrary, a negative trend will lead to a decrease in CI, which will have a positive impact on the service level of the transportation system. In order to better evaluate the roles of socioeconomic factors or governance policies on the CI, their roles will be classified into 8 levels according to the trends of CI_CC and CI_HA, i.e., “significantly inhibited”, “moderately inhibited”, “slightly inhibited”, and so on (as shown in

Table 1. Grading of the degree of CI being affected.

Slope (CI)	The Degree of Influence on CI
<−0.5	Extremely inhabited
−0.5~−0.3	Significantly inhibited
−0.3~−0.1	Moderately inhibited
−0.1~0	Slightly inhibited
0~0.1	Slightly promoted
0.1~0.3	Moderate promotion
0.3~0.5	Significantly promoted
≥0.5	Extremely promoted

). In addition, the main driving forces of CI variations can be distinguished according to Table 1, and the relative contribution degree of socioeconomic factors and governance policies to the CI variations can be calculated.

In this study, the impacts of socioeconomic factors variations are measured using the predicted values of CI and the impacts of govern policies variations are measured using the residual values of CI. Then, by combining the trends of predicted CI and residual CI with the trend of observed CI for the years 2002–2018, three evaluation methods are established according to different scenarios, so as to evaluate the relative roles of socioeconomic factors and governance policies in CI variations, as shown in

Table 2. Methods for assessing the relative roles.

Slope (CIobs)		Influencing Factors	Influencing Factors Classification Criteria		Contribution to the Variations of CI		Explanation (SF: Socio-Economic Factors; GP: Governance Policies)
			Slope (CICC)	Slope (CIHA)	Socio-Economic Factors	Governance Policy
<0	Scenario1	CC&HA	<0	<0	$\frac{slope(C{I}_{CC})}{slope(C{I}_{obs})}$	$\frac{slope(C{I}_{HA})}{slope(C{I}_{obs})}$	Both SF and GP-induced CI decrease.
	Scenario2	CC	<0	>0	100	0	SF-induced CI decrease.
	Scenario3	HA	>0	<0	0	100	GP-induced CI decrease.
>0	Scenario1	CC&HA	>0	>0	$\frac{slope(C{I}_{CC})}{slope(C{I}_{obs})}$	$\frac{slope(C{I}_{HA})}{slope(C{I}_{obs})}$	Both SF and GP-induced CI increase.
	Scenario2	CC	>0	<0	100	0	SF-induced CI increase.
	Scenario3	HA	<0	>0	0	100	GP-induced CI increase.

.

3. Study Area and Data

3.1. Study Area

Take Beijing City as a case city. Beijing is not only the capital of China, but also the political, economic and cultural center of China, with a total population of more than 20 million. Therefore, the transportation system plays an important role in the development and operation of the whole city. As one of the earliest cities to introduce public transport and rail transit, Beijing has invested a lot of money in the transportation system in recent decades, with high coverage and high utilization rate in China. At the same time, since 1998, Beijing has implemented a series of traffic control policies, including restrictions on purchases and restrictions on traffic. However, the traffic congestion in Beijing is always at the forefront of the country, which seriously affects urban development and the quality of life of residents. Therefore, by analyzing the changing characteristics of traffic congestion in Beijing as well as the interaction between it and socioeconomic factors, the contribution of different factors to traffic congestion variations and the effectiveness of policies can be demonstrated.

3.2. Study Data and Preparation

3.2.1. Traffic Congestion Index

The traffic congestion index is a variable reflecting the degree of road traffic congestion. The larger the index is, the higher the traffic congestion level. The evaluation methods in different countries have considerable differences [21]. In Beijing, the statistical data of CI is the Traffic Performance Index (TPI) published and implemented, and the relevant data source is the Beijing Traffic Development Annual Report. According to the TPI of the road network, as shown in

Table 3. Traffic congestion index corresponding to travel time.

Traffic Congestion Index	Road Network Operation Status	Travel Time
[0,2]	Unobstructed	We Can drive according to the road speed limit standard
(2,4]	Basically unobstructed	It takes 0.2 to 0.5 times more time than when the traffic is smooth
(4,6]	Minor congestion	It takes 0.5 to 0.8 times more time than when the traffic is smooth
(6,8]	Moderate congestion	It takes 0.8 to 1.1 times more time than when the traffic is smooth
(8,10]	Severe congestion	More than 1.1 times more time than the open traffic

, the degree of congestion can be divided into five levels, with values ranging from 0 to 10, and the corresponding values are as following: [0,2] for smooth, (2,4] for basic smooth, (4,6] for light congestion, (6,8] for moderate congestion, and (8,10] for severe congestion.

Figure 3 shows the historical fluctuation of the traffic congestion index, with the congestion index being greater than 6 until 2008, and reaching an all-time high of 7.73 in 2007. Due to many factors, the congestion index has been unstable, but the average value is obviously lower than that of the first half.

3.2.2. Factors Influencing Traffic Congestion

First of all, the factors affecting traffic congestion include two categories: socioeconomic factors and governance policies. Because the first-order factors are difficult to describe, it is necessary to select the measurable second-order variables to refine them. Among them, for socioeconomic factors, based on the existing comprehensive quantitative analysis of their interaction with traffic development, six quantifiable second-order indicators are selected, as shown in the following

Table 4. The socioeconomic factors influencing on traffic congestion.

First-Order Factors	Second-Order Variables (Measurable Variables)	Variable Description
Socioeconomic factors	GDP	The Gross Domestic Product in a region as of the statistical time window.
	Motor vehicle ownership	The number of motor vehicles in a region, including car, motorcycles, agricultural vehicles and so on, as of the statistical time window.
	Urban road mileage	The total length of roads built in a city as of the statistical time window
	Public transport mileage	The total length of bus lines built in a city as of the time window
	Railroad mileage	Total length of railroad lines built in a city as of the time window
	Urban population	The total number of urban population in a city as of the time window

.

Governance policy is another type of important factor affecting traffic congestion. As shown in

Table 5. Beijing Traffic Management and Control Policy Start and End Times.

Policy Name	Start Time	End Time
Parking permits	1998	2004
Parking fee	2002	continuing to the present
odd-and-even license plate rule	2007	continuing to the present
Vehicle purchase restriction policy	2011	continuing to the present
Staggered rush hour commuting	2014	continuing to the present
Encourage carpooling	2016	continuing to the present

, since 1998, the Beijing Municipal Government has implemented six types of traffic management and control policies. ① Parking permits: It can limit the increase in the number of motor vehicles and effectively reduce traffic demand, so the number of motor vehicles can be used to measure its role in reducing congestion. ② Parking fee: It can reduce the demand for private cars in some areas, thus directly reducing road traffic flow and alleviating traffic congestion. ③ Vehicle purchase restriction policy: similar to parking spaces, but its effect of restricting the growth rate of motor vehicles is more obvious, and its effect for alleviating congestion can be measured by the number of motor vehicles. ④ Staggered commuting during peak hours: It can reduce the traffic flow on the road network during peak hours, ease the traffic pressure, and reasonably guide travelers to avoid gathering on the road network. ⑤ Encourage carpooling: It can maximize the utilization rate of vehicles and improve the efficiency of road use, thus reducing the traffic flow and alleviating traffic congestion. ⑥ Odd-and-even license plate rule: the available time of each car can be controlled by the tail number of the license plate, thus reducing the frequency of private car use and greatly reducing road traffic flow.

To sum up, the factors that affect traffic congestion can be divided into subjective factors and objective factors, as shown in Figure 4. In short, objective factors are external socioeconomic factors that affect traffic congestion, but they exist independently of the managers’ consciousness and do not depend on their will. The subjective factors are mainly governance policies, which are closely related to the subjective consciousness of traffic congestion-related groups, managers and ordinary participants so as to reduce the negative impact of traffic congestion on the participants.

3.2.3. Study Data and Exploratory Analysis

In order to quantitatively present the influence degree of socioeconomic factors, especially governance policies under the traffic congestion influencing system, we obtain the corresponding panel data based on the Beijing City Statistical Yearbook and “annual report of traffic development” within the statistical time window. Next, the missing data for individual years can be supplemented by the average interpolation. The fluctuations value of all second order factors in Table 4 are shown in Figure 5.

According to the curve trend in Figure 5, GDP, the number of motor vehicles, resident population and railway mileage are all on the increase. Among them, the GDP increased nearly nine times, from 316.17 billion yuan in 2000 to 303.20 billion yuan in 2018. The number of motor vehicles increased from 1.5 million in 2000 to more than 6 million in 2018, with an average annual increase of nearly 300,000. As for the population, it is only 13.56 million at the beginning. By the end of the statistical period, this figure is almost twice the original. For the other three second-order factors, the growth trend is different, although the number is also rising in the statistical period. For example, as far as bus mileage is concerned, by 2006, their lengths increased by about 6000 km, but then it decreased, and there are some fluctuations with time. Similarly, urban road mileage, regarding 2008 as the time cut-off point, grew rapidly before 2008, by nearly 4000 km in eight years. However, since then, the growth has been slow, with a downward trend in individual periods.

4. Results

4.1. The Temporal Variation Characteristics of CI

From 2000 to 2018, the CI in Beijing shows a fluctuating downward trend as a whole, with an average annual change rate (slope) of −0.076772 y⁻¹. Historically, traffic congestion in Beijing has been increasing. The different trends of the CI_obs in different periods show that the traffic congestion index has strong time heterogeneity, which may be related to the change of governance policies. Therefore, the study time periods are divided into six, according to the years of implementing management policies, and the following

Table 6. Slope values of CI_obs in different periods.

Time Period	Slope Size	Promoted or Suppressed for CIobs
2000–2002	−0.350000 y−1	Significantly inhibited
2002–2005	0.280000 y−1	Moderately promoted
2005–2007	0.465000 y−1	Significantly promoted
2007–2011	−0.556000 y−1	Extremely inhibited
2011–2014	0.210000 y−1	Moderately promoted
2014–2018	−0.060000 y−1	Slightly inhibited

shows the differences of the CI_obs increase in different time periods. Among them, we divide 2000–2018 into different time periods according to the implementation time of different governing policies (as shown in Table 5); this aims to evaluate the relative roles of different governing policies in traffic congestion. Among the divided six time periods, there are three time periods showing an increasing trend, and the other three time periods showing a decreasing trend. Among them, the decline is the largest in 2007–2011 with a slope of −0.556 y⁻¹, which belongs to a significantly inhibited level, followed by the decreasing trend in the 2000–2002 period (slope = −0.350). In addition, the increasing trend is most pronounced in the period of 2005–2007, with a slope of −0.366 y⁻¹, which is a moderately promoted level, followed by the period from 2002 to 2005 (slope = 0.280), and again by the period of 2011–2014 (slope = 0.210). While from 2014 to 2018, the slope of the congestion index is −0.060 y⁻¹, which indicates that the external variation trend is not obvious during this period.

4.2. The Regression Analysis of CI

In the process of regression analysis, all six factors are employed into the model as important factors. The R-squared in the model is 0.658, which means that the explanatory power of the socioeconomic influences on the independent variables in the linear regression model equation is 65.8%. In addition, the regression sum of squares is 6.058, the degree of freedom is 6, and the residual sum of squares is 3.148. The final regression model obtained is as follows:

$$\begin{array}{l}C{I}_{CC}(t)=0.000321GDP+0.018267VP-0.010011PP+\\ 0.000215URM+0.000124UBM-0.015642UWM-5.790707\end{array}$$

(4)

The predicted values of Beijing’s traffic congestion index from 2000 to 2018 also show a volatility decline, among which the overall decrease of CI_cc is faster in 2006–2010 and 2011–2014, while the overall increase is slower in 2002–2006 (Figure 6). During the study period, the predicted values of the traffic congestion index in Beijing are between 5.277 and 7.065, with the minimum and maximum values appearing in 2006 and 2014, respectively. Overall, the difference value of the predicted Beijing CI from 2000 to 2018 at the beginning and end of the statistical period is 0.923077, with an average change rate of −0.048583 (p < 0.01), which is slightly lower than the observed value overall. Therefore, if we only look at the predicted values at both ends of the statistical time window, during the whole study period, the traffic congestion in Beijing may be due to the alleviation of socioeconomic factors. However, this ignores the influence of internal trends, as seen Figure 6, which shows that the trend of CI in Beijing from 2000 to 2018 also has significant temporal heterogeneity. Therefore, the slope trend analysis model will be adopted below, considering all factors and internal heterogeneity, to analyze the real impact of socioeconomic factors on CI.

As for the residual values, generally speaking, there is no obvious law of the change trend. The mean value of the residuals of the traffic congestion index in Beijing from 2000–2018 is 0.00002 and the standard deviation is 0.41817. In addition, the exploratory statistical indicators of residual values show that the average increase and decrease of the traffic congestion index affected by the governance policies are similar, especially the number of positive residuals and the negative residuals is 10 and 9, respectively. However, it can be seen from Figure 6 that the residual values themselves are more volatile with a large standard deviation. The positive residuals reached a maximum value of 0.84075 in 2010, and the negative residuals reached a minimum value of −0.98922 in 2011, with a range of 1.8.

4.3. The Characteristics of CI_CC and CC_HA Variations

In order to identify in depth the characteristics of the variations in traffic congestion levels in Beijing for 2000–2018 under the influence of socioeconomic factors through regression values and residual analysis, this paper still employs Equation (1) to calculate the slope of regression values in different time periods throughout the time windows. Overall, Beijing’s regression values of CI (CI_CC) decrease obviously in fluctuation, with a slope of −0.077738 y⁻¹ (as shown in

Table 7. Slope values of CI_CC in different periods.

Time Period	Slope Size	Promoted or Suppressed for CICC
2000–2002	0.105290 y−1	Moderately promoted
2002–2005	0.020531 y−1	Slightly promoted
2005–2007	0.160780 y−1	Moderately promoted
2007–2011	−0.319914 y−1	Significantly inhibited
2011–2014	−0.061356 y−1	Slightly inhibited
2014–2018	−0.003677 y−1	Slightly inhibited

); this is consistent with the growth rate of the actual value of CI (CI_obs) calculated from the traffic parameters and the values of both are also very close. However, the overall trend of CI_CC variations in different time periods remains different. From the trend analysis of the regression values, it can be seen can be concluded that the variations trend in different periods is obviously different from those of CI_obs. For example, the most obvious one is the first time period, when CI_obs is in a suppressed state while CI_CC is in a promoted state. In addition, the average slope of the regression value CI_CC during the promotion period is obviously lower than that in the suppression period. The number of time periods in which CI shows an upward and downward trend is consistent with that of CI_obs.

Similarly, continue to calculate the trend characteristics of residual values in the statistical time window. It can be seen from

Table 8. Slope values of CI_HA in different periods.

Time Period	Slope Size	Promoted or Suppressed for CIHA
2000–2002	−0.455290 y−1	Significantly inhibited
2002–2005	0.259469 y−1	Moderately promoted
2005–2007	0.304220 y−1	Promoted significantly
2007–2011	−0.236086 y−1	Moderately inhibited
2011–2014	0.271356 y−1	Moderately promoted
2014–2018	−0.056323 y−1	Slightly inhibited

that the overall variations trend of residual CI (CI_HA) in the study area is on the rise, with an increase of 0.000966 y⁻¹; this is contrary to the growth rate of CI_obs calculated from traffic parameters, which is much smaller. In addition, although the variations trend of the CI_HA in different time periods is different from that of CI_obs, its positive and negative characteristics remain unchanged, which indicates that the residual CI is also increasing in the interval where the actual value of CI is increasing, and vice versa. CI_HA changes most from 2005 to 2007, with a slope of 0.304220 y⁻¹, followed by 2011–2014 and 2002–2005. The decline of CI_HA is mainly concentrated from 2000 to 2002. In addition, as shown in Table 8, the variations of CI_HA also show a downward trend during the years 2007–2011 and 2014–2018. In the past five years, the CI level has shown a downward trend rather than an upward trend, indicating that the latest governance policies have alleviated traffic congestion to some extent.

4.4. Comprehensive Driving Force Analysis of CI Variations

The previous analysis shows that the variations of socioeconomic factors and governance policies in Beijing have a great time heterogeneity in the impact on the CI variations. For the same time period, the influence of these two types of factors on the variations of CI is also quite different. In order to fully explain the variations of traffic congestion index under different combinations of factors, this paper uses the research model given in Table 2 and Figure 2 to demonstrate the effects of a different combination of factors under different periods, as follows:

As can be seen in

Table 9. The analysis results of driving forces.

Time	Slope (CIobs)	Results of the Role of Influencing Factors on the Level of Traffic Congestion		Driving Forces
		CICC	CIHA
2000–2002	<0	Moderate promotion	Significant inhibition	CIHA
2002–2005	>0	Slight promotion	Moderate promotion	CICC&CIHA
2005–2007	>0	Moderate promotion	Apparent promotion	CICC&CIHA
2007–2011	<0	Significant inhibition	Moderate inhibition	CICC&CIHA
2011–2014	>0	Slight inhibition	Moderate promotion	CIHA
2014–2018	<0	Slight inhibition	Slight inhibition	CICC&CIHA
The whole	<0	Slight inhibition	Slight promotion	CICC

, if we take the statistical time window as a whole, the stronger driving force for the variations in the traffic congestion index in Beijing is socioeconomic factors. Unexpectedly, during the 20 years’ development, socioeconomic factors even slightly inhibit the growth of the traffic congestion index (−0.07738 y⁻¹). On the contrary, the governance policies implemented in the whole period lead to an increase of traffic congestion levels (0.000966 y⁻¹). However, CI_HA is selected as the driver of CI variations in all six time periods, while CI_CC does not meet the criteria of being selected as the driver in two time periods (2000–2002 and 2011–2014). In the other four periods, both CI_CC and CI_HA are the most important drivers of CI variations, accounting for 66% of the total. Different policy combinations have fundamentally different impacts on traffic congestion at different socioeconomic levels, which can promote or inhibit traffic congestion, and the degree of inhibition or promotion is also different. If all time periods are regarded as a whole, there will be errors due to interference of extreme value effects.

Based on the average trend of Beijing’s CI variations caused by the socioeconomic factors and governance policies, respectively, it is proved that the average contributions of the socioeconomic factors and governance policies to the CI variations in Beijing are about 17.596% and 82.404%, respectively. In different periods, the contributions of socioeconomic factors to CI variations are between 0% and 58% (Figure 7). In only one period, the contributions of socioeconomic factors exceed 50%. Socioeconomic factors contribute the most to the change of CI in 2007–2011 (58%), followed by 2005–2007 (about 35%) and again in 2002–2005 (about 7%). The contributions of governance policy to the variations in CI range from 42% to 100% (Figure 7). Among them, the contributions of governance policies in five time periods are more than 50%, indicating that the impact of governance policies on CI variations is greater than the socioeconomic factors in these time periods. In 2000–2002 and 2011–2014, the governance policies make the greatest contribution (about 100%) to the variation of CI, followed by 2014–2018 (about 94%) and 2002–2005 (about 93%).

5. Discussion and Conclusions

5.1. Discussion

The research shows that from 2000 to 2018, the level of traffic congestion in Beijing shows a downward trend, but the time heterogeneity is very large. Most of the time, the combination of social factors and governance policies is the main reason for the variations of the traffic congestion index in Beijing. On the one hand, more attention and resources are devoted to the development of transportation, including a large number of traffic infrastructures with more advanced technologies, which can reduce the level of traffic congestion [26]. On the other hand, traffic demand management (TDM) policies can strengthen the match between traffic supply and demand, meanwhile reducing traffic congestion at the source. For example, Tian et al. [27] demonstrate that TDM encourages commuters to alter their departures spatially or temporarily with the goal of alleviating congestion. However, the corresponding time of this study is relatively short, and the role of TDM in long-term development has not been fully proved. At present, most of the research fields are small with a short attention period, and a large amount of subjective data, such as questionnaire data, is mixed. In this paper, binary regression residual analysis and objective panel data have been used to further distinguish and quantify the impact of socioeconomic factors and governance policies over different long periods. The results show that the largest influence of governance policies on the variations of CI in Beijing exceeds 0.4 y⁻¹, with a contribution rate that is mostly above 50%.

Variations in both socioeconomic factors and governance policies are able to contribute to the increase in the traffic congestion index, which even leads to a more pronounced worsening trend in the traffic congestion level in Beijing during some periods. It is worth noting that, in the first half of the statistical window, variations in socioeconomic factors have a positive impact on those in CI. Among them, rapid economic development leads to rising traffic demand, an increasing imbalance between traffic supply and demand, and a lack of experience in the planning as well as management of traffic, which can lead to an increase in the level of traffic congestion [28]. Moreover, the positive contribution of governance policies to variations of CI is more obvious in the time period of 2005–2007, because the contribution of different policy combinations to traffic congestion will be different after reaching a certain socioeconomic level. Furthermore, certain governance policies have a strong guiding effect on public travel behavior, which can change the overall structure of the traffic system and lead to the worsening of congestion instead. Therefore, it can be seen that the complex influence of socioeconomic factors and governance policies plays a decisive role in the time domain distribution of variations in CI.

In this paper, on the scale of time domain, the variations in traffic congestion index during the time periods 2000–2002 and 2011–2014 are mostly influenced by governance policy factors, which contribute 100% to the variations in CI. Chin [7] reviewed and analyzed the effectiveness of two policies that help to control congestion and car ownership. Yang [29] investigated the positive value of congestion charging policies in reducing traffic congestion using real-time fine-scale traffic data in Beijing. In this study, all demand management policies in Beijing are taken into account, and the results show that the contribution of governance policies to variations in traffic congestion in Beijing is mostly above 60% (Figure 7). Although there are differences, the results of all the above studies reflect the important role of governance policies in the variations of traffic congestion in the city.

Based on the above discussion, it can be concluded that in the initial stage of urban expansion and development, socioeconomic factors tend to promote the increase in traffic congestion. Before 2007, Beijing’s GDP as well as the number of motor vehicles and traffic infrastructures both increased, but their contribution to the traffic congestion index is positive, indicating that traffic supply brought by the construction of infrastructure increased too slowly, which could not meet the travel demand brought by the increase in population and the number of motor vehicles. In order to further explain this reason, the slope values of each second-order factor are calculated using a previous model. It can be found that the growth rates of bus mileage and urban road mileage in the second stage show a rapid downward trend compared with that in the first stage, in which the urban road mileage shows the fastest decline, with a decrease of 97.172% in the second half compared to the first half, and followed by bus mileage (76.884%). The growth rates of the other four second-order factors are increasing, and railway mileage shows the biggest growth rate (323.24%), which has increased from 12.283 y⁻¹ in the first half of the period to 51.988 y⁻¹. The GDP growth rate is also over 100%, reaching 146.241%. The growth rate of motor vehicle ownership increases the least, which is only 14.576%. The above data corroborates our judgment, from which some interesting inferences can be drawn. First of all, as far as the paradox that “the more roads are built, the more congestion there will be” is concerned, although it cannot be proved accurately, this paper tells us that road construction has a limited role in solving traffic congestion [30]. Secondly, the development of traditional public transport is not as valuable or effective as that of rail transport in solving traffic congestion. Thirdly, vigorously developing the economy may be the most fundamental solution to traffic congestion. Like the environmental Kuznets curve, the early rapid economic growth will indeed lead to traffic congestion, but with the continuous improvement of economic strength, the level of traffic congestion will begin to fall, thus showing an inverted U-shaped curve. For example, the growing financial strength of the Government enables us to invest more money in traffic planning, management and traffic infrastructures, such as building subways.

Different strategies have contributed to the variations of traffic congestion at different rates, and some traffic strategies can even aggravate traffic congestion to some extent. Although the parking permit has been abandoned, it is the most effective policy to curb traffic congestion, which is also the only policy that can play a role at a ‘significantly inhibited’ level. The implementation of ‘parking fee policy alone’ or ‘parking fee + odd-and-even license plate rule’ cannot reduce traffic congestion in essence, but to a great extent, it has contributed to the increasing trend of the traffic congestion index, which contradicts our traditional cognition. However, these policies, with the help of vehicle purchase restrictions, can moderately inhibit the traffic congestion index, which is confirmed by the corresponding role of a parking permit. After the implementation of staggered commuting, the role of governance policies in CI variations from ‘moderate inhibiting’ to ‘moderate promoting’, suggesting that it causes an increase in traffic congestion index; this is because the mismatch between time and space staggering may be serious. Meanwhile, commuters will psychologically suffer a sense of urgency in certain time periods and sections, which will make travel more focused. Encouraging carpooling is valuable because it can make the role of the policies return to the level of “inhibiting traffic congestion”. Overall, controlling the number of motor vehicles at the source may be the most effective traffic strategy, followed by encouraging carpooling, which is also of obvious value for reducing traffic congestion, but other strategies may be counterproductive.

5.2. Conclusions

In this study, urban panel data and traffic congestion index data are used to discuss the variations of traffic congestion from 2000 to 2018, and the role of socioeconomic variations and governance policies in these variations. The main conclusions are drawn as follows: (1) From 2000 to 2018, the growth of traffic congestion in Beijing shows a downward trend. For all the periods, the average relative role of socioeconomic factors on CI variations is 17.60%, and the average relative role of govern policies is 82.40%. Except for 2007–2011, which is mainly driven by socioeconomic factors, the CI variations in 2000–2002 and 2002–2005 are mainly driven by govern policies; this shows that, in the past 19 years, governance policies are the more important factors that affect the CI variations in Beijing. (2) In terms of the temporal distribution of CI variations, the number of periods of increasing and decreasing traffic congestion accounts for 50%, respectively. The relative roles of socioeconomic factors and governance policies on the increased distribution of traffic congestion are 25.56% and 85.15%, respectively, and the relative roles of socioeconomic factors and governance policies in the reduction distribution of traffic congestion are 66.01% and 34.99%, respectively; these results show that governance policies play an important role in alleviating traffic congestion, but unfortunately, the unreasonable combination of governance policies is also the main culprit that leads to the rise of the traffic congestion in certain periods.

We try to re-examine the role of socioeconomic factors and governance policies in the variation of traffic congestion from the macro-social development level, so as to bring some conclusions for traffic management and control:

(1) In the early stage of urban expansion and development, socioeconomic factors are more inclined to promote the increase in traffic congestion. As time goes by, however, social-economic factors begin to actively inhibit traffic congestion; this is similar to the role of economic development level in environmental pollution. According to the theory of the environmental Kuznets curve, the environmental situation first deteriorates and then gradually improves during process of economic development. Similar to environmental pollution, a key to alleviating or even solving traffic congestion should be to continue to develop the economy. In conclusion, the novel research perspective and conclusion enrich the theoretical and application value of socioeconomic factors in traffic congestion modeling.

(2) Unlike the general traffic congestion model, this paper is not based on basic traffic parameters (such as traffic density, traffic flow, etc.). Herein, a before-after analysis is deployed to understand the statistical relationship between urban traffic congestion and governance policies from a global perspective: ① Not all governance policies are conducive to solving traffic congestion. ② Controlling the number of motor vehicles from the source is perhaps the most effective govern strategy, followed by encouraging carpooling, which is also effective to some extent. ③ Parking charges or odd-and-even license plate rules may be valuable in regulating traffic congestion for a single small area, but they have little or even negative effects from the global perspective of the whole city. Although some conclusions are different from previous studies, the new research perspective has improved the theory between traffic congestion and policies, which is of positive significance to the macro-control of traffic.

In the future, there is still something to be improved. (1) In fact, we take Beijing city as the case, but it does not mean that the model is only suitable for Beijing. Due to the limited availability of data, we did not use other cities to conduct relevant studies. If researchers want to employ it in other types of regions, it is unnecessary to make a huge adjustment to the model, just have to re-fit it based on corresponding data. (2) In the regression of the relationship between socioeconomic factors and traffic congestion, only one method is used, and more different methods should be used to compare and explore the performance differences in the future. In addition, we also need to refine the selection of quantitative indicators and expand the selection range of social and economic influencing factors. (3) The amount of data used in this study is relatively small, although it is sufficient for regression. However, in the future, we would like to expand the amount of data by collecting relevant data from different cities. Although Beijing is very representative, it still needs to form a theoretical framework for similar research in different types in the future.