1. Introduction
The transport infrastructure investment (TII) is significant for economic growth (EG) [
1,
2]. TII serves as a “catalyst” for economic globalization [
3,
4], and high-quality transport infrastructure provides efficient and sustainable economic and trade services and tightens the relations of global economy and trade [
5,
6]. Continuous TII also represents an important impetus of EG in members and partners of the Organization for Economic Cooperation and Development (OECD+) [
3,
7,
8]. Development in OECD+ countries is listed in
Table 1, and the gross domestic product (GDP) was 73.47 trillion dollars in 2019, accounting for 86.77% of global GDP, of which TII was 1.20 trillion. The economy and trade have been connected increasingly tightly in OECD+ countries, forming the most dynamic economic aggregates. In such context, how to promote EG through TII in OECD+ countries has become a system engineering issue and the focus of much research among numerous scholars [
9,
10].
Baron and Kenny [
11] have discussed the effect of TII on EG in the frameworks of the endogenous growth theory and the new economic geography theory, thus, theoretically proving the presence of complex correlation mechanisms between TII and EG [
12]. To estimate the effect of TII, the correlations between TII and EG have been discussed through Granger causality tests and cointegration analysis [
13]. To overcome drawbacks of direct correlation analysis, TII is separated from influencing factors of EG to establish an econometric model of multiple variables [
14]. In terms of research into the data structure of TII, the commonly used methods include cross-section data, time series data, panel data, and spatial panel data (SPD) models. For the research on the spatial spillover effect of TII, the effect on the one hand has significant time delay and, therefore, shows certain differences in short-term and long-term responses [
15]; on the other hand, the spatial spillover effect of TII also demonstrates the regional economic gap [
16], and the promotional effect of TII on EG is highly heterogeneous in countries with different income levels [
12,
17]. Previous research has shown that the economic benefit of TII is highly correlated with the spatial location [
17] and it is not only dependent on the investment scale but also determined by the level of economic development [
11], so, the TII differs greatly in countries with different income levels [
18]. Therefore, studying the dynamic spatial economic relationship between TII and EG and their mechanism of action in OECD+ countries with different income levels is of importance when trying to improve the TII quality in OECD+ countries [
7].
Although scholars have proved the presence of the spatial spillover effect of TII on EG by selecting different data [
19], different models, and different variables, EG is a complex system with a nonlinear structure [
17,
20]. To study the essential features and intrinsic rules of the complex system of EG [
21], it is necessary to determine the relational routes of TII with EG from the perspective of underlying logic [
18]. With the integrated development of spatial econometrics, the existing regression models can further characterize the complex structures between variables based on quantifying influences of spatial correlations [
22]. Especially in the research paradigm with multiple explained and explanatory variables, the structural equation model (SEM) can be used to calculate the effect of latent variables that cannot be observed directly, which provides a new path for studying the complex relationship and mechanism of action between TII and EG in OECD+ countries [
18,
23].
The current research proposed a method integrating the spatial Durbin model (SDM) and the SEM to explore the spillover effect and routes whereby TII influences EG. The effectiveness and applicability of the proposed method were verified taking the panel data of 35 OECD+ countries from 2007 to 2019 as examples. The proposed method based on the SDM and SEM enshrines the following innovations: (1) the SDM was used to empirically estimate the spatial spillover effect of TII on EG and study the spatial correlations between TII and EG in countries with different income levels; (2) the SEM was adopted to identify the multiple mediating effects of TII on EG and discuss the modes of action and key functional routes of TII on EG; (3) compared with other research, the method can deal with the heterogeneity of research conclusions due to the multilevel research objects on the basis of studying the spatial spillover effect and functional routes of TII on EG.
The remainder of the paper is organized as follows:
Section 2 introduces the SDM, SEM, and SDM–SEM hybrid model;
Section 3 discusses the research variables and research objects; the SDM–SEM model is developed in
Section 4, in which the detailed analysis steps are also proposed;
Section 5 covers empirical analysis of the spatial spillover effect and multiple mediating effects of TII on EG; in
Section 6, we draw our conclusions arising from the research; in
Section 7, we propose policy suggestions based thereon.
5. Empirical Analysis Results
5.1. Step 1: Development of the Spatial Econometric Model
Variance inflation factor (VIF) is one of the important methods to measure multicollinearity in multiple linear regression models, which is very suitable for economic and social multicollinearity analysis. VIF of TII, UP, IS, LI, and TR is calculated as 1.81, 1.76, 1.46, 1.44, and 1.09; mean VIF is 1.51 and after VIF screening (VIF > 10), all variables are retained.
In the present research, the explained variable is EG; the core explanatory variable is TII; the control variables include UP, IS, LI, and TR. To estimate the sensitivity of the spatial spillover effect, the SDM was built based on
WD and
WE as follow:
where
,
,
,
,
, and
, separately, represent the economic development level, TII, level of urbanization, industrial structure, scale of labor population, and foreign trade.
and
, separately, denote the inverse distance spatial weight matrices of geology and economy;
,
, and
, separately, refer to the model parameters to be estimated;
and
, separately, represent the effect of spatial characteristics of spatial units and the temporal trend effect of research objects; and
is the error term.
5.2. Step 2: Tests on the Spatial Econometric Model
5.2.1. Prior Tests on Spatial Correlations
According to Equations (4) and (5), the Moran’s I of TII and EG in
WE in each OECD+ country from 2007 to 2019 was calculated, and the results are shown in
Figure 5.
As shown in
Figure 5, Moran’s I is always positive, which indicates that TII and EG have a positive spatial agglomeration effect. Moran’s I of TII increases in fluctuation, suggesting that the degree of spatial agglomeration of variables is enhanced.
According to Equations (6) and (7), the statistics of local Moran’s I could be used as the regression coefficient of scatter plots, in which the abscissa and ordinate, separately, represent variables and the spatial lag vector of variables, as shown in
Figure 6.
According to
Figure 6, the local Moran’s I of TII and EG is mainly distributed in the first and third quadrants, with only a few observed values distributed in the second and fourth quadrants. This indicates that spatial units with high observed values are surrounded by those also having high observed values over the chosen 13-year period.
In summary, TII and EG have a large spatial interaction in each OECD+ country from 2007 to 2019 and show positive spatial correlations.
5.2.2. Post Hoc Tests on Spatial Correlations
To select the econometric model of TII and EG, the spatial panel data model was tested according to
Table 4 and the test results are summarized in
Table 6.
Table 6 shows that the Sargan–Hansen statistic significantly negates the null hypothesis of random effects, implying that the fixed-effect model should be used. LM-Lag, LM-error, and robust LM-error statistics all significantly reject hypotheses of the nonspatial effect, which means that it is reasonable to use the spatial panel data model to determine the spatial effect of TII on EG. The results of LR and Wald tests are significantly positive at the 1% level, separately, rejecting the null hypotheses of “the SDM may be degraded into the spatial error model” and “the SDM may be degraded into the spatial lag model”, indicating that it is appropriate to select the SDM. Comprehensive analysis of the diagnostic test results of the spatial panel data model suggests that the fixed-effect SDM model should be used here.
5.3. Step 3: Analysis of the Spatial Econometric Model
5.3.1. Regression Estimation of the SDM
Analysis of Steps 1 and 2 shows that the fixed-effect SDM model was selected as the benchmark analysis model in the research. OECD+ countries were classified into two types of research objects: high-income and upper-middle-income ones. Based on
WD and
WE, maximum likelihood estimation was adopted for regression estimation of the SDM of the two types of research objects, separately, using Equations (8) and (9). Weighted least squares (WLS) and robust standard errors are sometimes presented as alternative approaches for obtaining reliable standard errors of estimates of regression coefficients in the presence of heteroscedasticity. The use of weighted least squares with robust standard errors is displayed in
Table 7.
Analysis of
Table 7 implies that TII in high-income countries is underestimated by 21.30%, (0.108–0.085)/0.108 = 21.30%, while that in upper-middle-income countries is underestimated by 53.85%, (0.104–0.048)/0.104 = 53.85%, when only considering the geographical distance compared with coefficient estimation of
WE. This finding indicates that ignoring the economic development level may underestimate the contribution of TII to EG. The
WD of the spatial lag of TII is significantly positive, which suggests that geographical distance is a leading factor that affects the spatial spillover effect of transport infrastructure
It is worth noting that TII and the estimated coefficients of the spatial lag of upper-middle-income countries are both lower than those of high-income countries. This indicates that TII is more active in high-income countries and it is easier to drive EG in other countries. Furthermore, the level of urbanization and industrial structure both significantly stimulate EG and have a positive spatial spillover effect.
5.3.2. Estimation of the Spatial Spillover Effect
To estimate the spatial spillover effect, the partial differential equation is adopted to decompose the spatial effect of the SDM. Based on
WD and
WE, the total effect of TII on EG is decomposed into direct and indirect effects. The decomposition results are listed in
Table 8.
Analysis of
Table 7 and
Table 8 shows that the feedback effect of TII in high-income countries is larger under
WE; the short-term feedback effect of TII is almost zero while the long-term feedback effect is larger under
WD in upper-middle-income countries. The result suggests that the level of economic development can magnify the promotional effect of TII on EG in high-income countries to a greater extent, while geographical distance is the main factor that enables TII to promote EG in upper-middle-income countries.
By analyzing
Table 8, it shows that when taking high-income countries as an example, the gross economic output in the local area increases by 0.077% in the short term under
WD and the value gradually increases to 0.275% per 1% increase in TII while keeping other conditions unchanged. In addition, the gross economic output in surrounding areas grows by 1.275%, which gradually reduces and then stabilizes at −0.081%, suggesting the simultaneous reduction in the total effect. Under
WE, the gross economic output in the local area may increase by 0.105% in the short term and then grow to 0.129%; the gross economic output in surrounding areas increases by 0.027% and then constantly increases to 0.17%, indicating a simultaneous increase in the total effect. Furthermore, the total response of urbanization, industrial structure, labor input, and trade significantly increases in high-income countries under
WE.
5.3.3. Robustness Tests
To diminish the deviation arising from specific data on the model estimation, the method of shortening the time window was used to exclude statistical data in 2007. Then, panel data from 35 OECD+ countries (2008 to 2019) were used for regression of the spatial Durbin panel model, as shown in
Table 9.
Comparative analysis of
Table 7 and
Table 9 indicates that there is no significant difference in signs and significance of coefficients, so it is inferred that the results of the SDM for the effect of TII on EG are relatively robust.
5.3.4. Analysis of the Spatial Spillover Effect
On the one hand, the network effect of transport infrastructure is conducive to promoting spatial agglomeration and decentralization of economic activities and enhancing the geographical contacts and economic ties between regions in OECD+ countries. On the other hand, in OECD+ countries with similar geographical distances and levels of economic development, whether the transport networks are developed or not directly influences the strength of spatial economic ties between them, thus, affecting the level of economic cooperation and development.
- (1).
Characteristics of the spatial spillover effect in high-income countries
In high-income OECD+ countries, the economic development level magnifies the spatial spillover effect of TII on EG. The spatial networking and density increase in transport infrastructure incur a space–time compression effect, which enhances the frequency and breadth of connections between high-income OECD+ countries and promotes coordinated development in neighboring regions. As a result, the zone of influence gradually extends beyond traditional geographical boundaries and coalesces, thus, forming a benign cycle between TII and EG.
- (2).
Characteristics of the spatial spillover effect in upper-middle-income countries
In upper-middle-income OECD+ countries, geographical distance enhances the negative spatial spillover effect of TII on EG. That is, increasing TII in a country is conducive to improving EG in the country and its competitiveness, while this intensifies competition between upper-middle-income countries, thus, occupying economic resources of other upper-middle-income OECD+ countries. Meanwhile, the spatial spillover effect of TII in upper-middle-income OECD+ countries attenuates with distance according to the new economic geography theory. As the distance constantly increases, the number of regions with spatial relations reduces continuously and the spatial spillover effect of TII gradually decreases and almost disappears. It is worth noting that in Bulgaria, Albania, Azerbaijan, Belarus, Georgia, Montenegro, and the Russian Federation, as OECD+ “economies in transition” due to their specific socioeconomic characteristics and the closeness of their connections with OECD countries, the spatial spillover effect of transport infrastructure investment is affected to different degrees.
5.4. Step 4: Development of the Functional Route Model
Results of the SDM show that the TII exhibits more significant economic benefits in high-income OECD+ countries. Therefore, to further explore functional routes of economic benefits, only high-income OECD+ countries with significant economic benefits are selected as research objects below. As shown in
Table 3, the development of ME and CE is not balanced in these countries. In addition, in
Table 7, the value of IS is the largest and much larger than 1. The explained and explanatory variables in the SDM are, separately, transformed into dependent and independent variables in the SEM, with ME, IS, and CE as mediators. Moreover, the service industry is taken to represent the direction of development and the transformation of the economy. Driven by high-quality development of the service industry, the service supply is constantly released, which boosts the upgrading of CE. Considering this, the hypotheses in
Table 10 are proposed.
The functional route model of TII on EG is built, as shown in
Figure 7.
Various function routes in each subsection in the multiple mediation model are connected to form a direct functional route and four indirect functional routes of TII to EG (
Table 11).
5.5. Step 5: Tests of the Function Route Model
5.5.1. Fitness Tests of the Multiple Mediation Model
Amos software was used to conduct fitness tests and estimate coefficients of functional routes, calculate the overall fitness of the multiple mediation model, and test the degree of matching between the presumptive model and the actual data. The initial model was modified and optimized according to the MI parameters and the final fitting results are listed in
Table 12.
Analysis of
Table 12 reveals that the CMIN/DF value for testing the overall fitness of the model is 0.008, which is smaller than 3; the
p-value is 0.928, which is greater than 0.05; RMR and RMSEA are both less than 0.05; and GFI, AGFI, IFI, CFI, NFI, and TLI are all greater than 0.9. These results indicate that all test statistics meet the required fitness standards, that is, the results of the model are a good fit to the data.
Arrows were used to represent the influencing routes of mediators and the standardized estimated coefficient was adopted to show the degree of influence of different variables on EG. The functional routes of TII on EG are plotted according to
Figure 7, as illustrated in
Figure 8.
Analysis of
Figure 8 shows that the route coefficients are all significant at the 1% statistical level, which means that hypotheses H1, H2, H2a, H2b, H3, H3a, H3b, H4, H4a, H4b, and H3-H4 are all accepted. To evaluate the relative importance of different independent variables to dependent variables, we calculate the standardized route coefficients by the method of maximum likelihood. Therein, the standardized route coefficients of TII → IS and TII → ME are, separately, 0.962 and 0.915, which indicate that TII mainly promotes EG indirectly through IS upgrading and stimulation of foreign trade. In addition, there is a significant route coefficient between TII and EG, which is consistent with the estimation results of the SDM, so the direct effect of TII on EG is further verified. Both the SDM and SEM models show the relationship between IS, CE and ME, and EG. It is reasonable to promote economic growth directly or indirectly through industrial structure, merchandize export, and consumer expenditure.
5.5.2. Bootstrap Tests on Function Routes
At first, the estimates of each function route were calculated according to the standardized estimated coefficient and, then, the rate of contribution of each complete function route was calculated based on the absolute value of the estimates. Then, 2000 rounds of repeated random sampling were simulated in Amos software and the bootstrap 95% confidence interval after error correction was outputted. The maximum likelihood estimation was adopted as the estimation method and bootstrap tests were conducted on the functional routes; results are listed in
Table 13.
According to
Table 13, the bootstrap 95% confidence interval of the direct and indirect effects does not include 0 and
p < 0.001. ME, IS, and CE have a significant mediating effect, while IS and CE have significant chained mediating effects. The result indicates that TII not only directly affects EG, but also can act on EG indirectly through IS, CE, and ME, and the indirect effects are greater than the direct effect. Suggesting that, when increasing direct investment, it is also necessary to pay close attention to the mediating effects of other modes on EG.
5.6. Step 6: Analysis of the Functional Route Model
To further explore the influencing mechanism of functional routes of TII on EG, functional routes FR1 to FR5 were expounded in detail based on bootstrap tests on the function route model.
FR1: TII → EG, in which the direct effect contributes 4.09%.
TII, as an investment element, can directly affect EG and amplifies its effect on EG through the multiplier effect. To be specific, improving the existing transport infrastructure and increasing the sharing and efficiency of transport infrastructure can expand demand and directly drive supply to the economy, thus, promoting EG.
FR2: TII → ME → EG, in which the indirect effect contributes 7.37%.
Enlarging the effective investment in transport infrastructure can maximize the positional advantages of transport infrastructure and narrow the gap in the development of transport infrastructure in various countries. This stimulates the potential impetus for foreign trade and becomes an important condition for expanding the export marginal effect and improving export comparative advantages of various countries. This also drives down logistics costs across the whole of society, unblocking the cycling of commodities, services, and production elements, and strengthening the vitality of economic development.
FR3: TII → IS → EG, in which the indirect effect contributes 67.25%.
The upgrading of IS is a key route for the effect of TII on EG. TII can exert an indirect effect on EG by influencing the upgrading of IS. Through the integration of elements, the favorable transport infrastructure lowers the trade cost and expands the market range of the service industry, thus, changing the IS. The adjustment of IS further changes resource allocation in the market and, thus, leads to EG.
FR4: TII → CE → EG, in which the indirect effect contributes 2.97%.
Thorough transport infrastructure provides convenience for residents’ travel and affords people access to a larger scale of socioeconomic activities. Meanwhile, improving the quality of transport infrastructure can ensure travel safety and enhance personal willingness towards CE, changing the thinking behind consumption and consumer psychology, finally driving EG.
FR5: TII → IS → CE → EG, in which the indirect effect contributes 18.42%.
The high-quality development of the service industry promotes the upgrading of CE. The constant development of the service industry towards higher quality multiple categories provides diverse consumption choices to residents and enables the transformation of CE from traditional subsistence consumption to more spiritual consumption.