Transportation infrastructure has been prioritized by both central and local Chinese governments since the eighth Five-Year plan (1991-1995) with the realization of significant role played in promoting economic development. Since then, transportation infrastructure continues to be an essential part of China’s regional development policy. The total length of railway in operation has been increased from 57.8 thousand kilometers to 103.1 thousand kilometers since 1991 until 2013, while the length of highway has increased significantly from 1041.1 thousand kilometers (in 1991) to 4356.2 (in 2013) thousand kilometers (National Bureau of Statistics of China, 2013). China has made significant investment in transportation infrastructure development over the recent three decades, and the average growth rate was over 10% per year, since 1978. During the 2008-2009, China stimulated the economy by using 40% of the US$586 billion economic stimulus package devoted to infrastructure development. From the significant investments made to develop infrastructure, China achieves a substantial growth in her economic output. It is thus imperative to answer whether and to what extent the infrastructure investments contribute to the economic growth of the Chinese economy  .
There is an abundant of international empirical evidence showing an affirmative answer with a wide range of elasticity estimates    . The variety could be attributed to varied econometric specifications with or without accounting for the time and spatial effects, the definitions and measures of public infrastructure, the estimation methodology as well as research contexts in terms of study period and geographical scales. The strand of literature (see Table 1) analyzes the effects of public infrastructure on private output, and has been brought to the limelight by the seminal paper of Aschauer  . The elasticity results in his
Table 1. Empirical evidence on the estimates of output elasticity of public/transport infrastructure.
paper range from 0.25 to 0.56, and the different types of public inputs are termed as the “core” infrastructure such as streets, highways, mass transits, and airports. These results are found to be consistent with other studies  -  which are carried out at both national and regional levels.
Using Chinese provincial level data, few studies have examined the contribution of the aggregate public infrastructure to the productive performance  , the spatial spillover effects of transport infrastructure   as well as the poverty reduction effect  . Demurger  measures the transport endowment using the overall network density (incorporating road, railway and waterway) based on a panel data from 24 Chinese provinces (excluding municipalities) during 1985-1998, and shows that transport facilities are a key differentiating factor in explaining the growth gap. Hong, Chu and Wang  construct a provincial-level comprehensive index based on quantity and quality of railway, roadway, airport and seaport to show that the output elasticity of land transport (including roadway and railway) ranges from 0.554 to 2.757. The role of China’s bullet trains to facilitate market integration and mitigate the cost of megacity growth is confirmed by Zheng and Kahn  .
The present paper, with a regional focus on China, carries out a study using city-level annual data from 2000-2012 to gain a deep understanding of the heterogeneous impacts of varied transportation modes on city economic performance. Compared with existing studies, we explore the growth effects across varied-sized Chinese cities of both inter-city and intra-city transport networks. Highways and railways represent inter-city infrastructure and the public road network represents intra-city infrastructure. Accounting for possible unobserved heterogeneity, we use the random-parameters model  to shed light on the effect of transportation-related characteristics (including transportation freight and passenger volumes, public transit transportation, paved road, and highway congestion) on a city’s economic growth. By so doing, we answer the question that to what degree transport infrastructure and which type of transport infrastructure matters for which specific city in China.
The present paper is structured in five sections. Section 2 describes the data, and the methodology is discussed in Section 3. The estimated results and discussions on the estimated parameters and elastic values are found in Section 4. The summary and conclusions are presented in Section 5.
China consists of 34 provincial administrative units including 23 provinces, 5 autonomous regions, 4 municipalities, and 2 special economic zones. Subordinate to provinces are prefectures and each prefecture has at least one core city, some rural counties, and several county-level cities. The current number of Chinese prefecture cities is 289, however, due to the unavailability of consistent data across all the cities, only 237 prefecture cities are considered in the present study. Thus, the analysis is carried out using data from prefecture-level city, which includes both the urban and rural administrative areas.
The prefecture-level data during the period of 2000-2012 are collected from a number of sources including China City Statistical Yearbook (various years 2000-2013), CEIC, China Data Online and Wind Financial Database. We include the gross city product (GCP), total population, price indices, and physical measures of transport infrastructure-related characteristics, investment, and employment in the model. Specifically, the gross city product per capita is defined as the ratio of the gross city product over the city total population. The labor participation rate is measured as the ratio of the number of employees over the total population. We use the fixed asset investment as a proxy for physical capital. The inter-city infrastructure development is measured by the freight or passenger volumes for both the highway and railway. And the intra-city infrastructure development is defined using the area of paved roads within the city and the public transportation unit per ten thousand people. We also control for the urbanization rate measured as the number of urban population over the city total population as well as the congestion rate defined as the total number of vehicles over the area of paved roads to avoid the potential missing variables biasedness. The descriptive statistics of the significant variables used in the final model are presented in Table 2.
3. Methodological Approach
To examine the economic impacts of highway and railway across the selected cities in China, a methodological procedure that accounts for unobserved heterogeneity across cities will be appropriate. In the past, a number of statistical methods have been used to carry out this type of investigation including ordinary least square regression models, and fixed-effects model    . However, in recent years, a new methodological approach, a random-parameters regression model has been applied for the first time in economic impact analysis of transportation infrastructure expenditure to capture unobserved heterogeneity across observations and also heterogeneity across observations and time. This new me-
Table 2. Descriptive statistics of selected variables.
thod has been shown to be more statistically robust compared to the previous statistical methods (ordinary least squares regression, fixed- and random-effects models). Furthermore, the random-parameters regression model is able to account for unobserved heterogeneity across observations compared to the previous statistical methods. Thus in the present paper, we will follow the random-parameters regression model as derived and applied to investigate the economic impacts of transportation infrastructure expenditures starting with the following equation:
where is the gross city’s product (GCP) per capita (in 2010 USD) for city k at year t, is a vector of the independent variables (highway freight volume, railway freight volume, the area of paved roads, highway passenger volumes, fixed asset investment, urbanization rate, public transportation unit per ten thousand people, industrial sector’s contribution to gross city product, highway congestion rate, and labor participation rate) for city k in time t, is a vector of estimable parameters, and are normally distributed random disturbances.
The estimation of Equation (1) by the ordinary least square approach has two distinct issues. First, it is possible that higher GCP generates higher freight volumes in highway and railway, while it is expected also that higher freight volumes in highway and railway would promote growth in city outputs. Thus, the gross product and freight volumes could be endogenous and violates the fundamental assumption underpinning the ordinary least squares estimation, resulting in biased coefficient estimates. This concern is resolved in the present study by adopting instrumental variable procedure whereby highway and railway freight volumes are regressed against exogenous variables and the predicted values are used as variables in the estimation of Equation (1). The second issue with the estimation is that each of the cities will produce 13 observations from 2000-2012, and these 13 observations are likely to share unobserved effects resulting in serially correlated data, thus, violating one of the OLS assumptions of no serial correlation. This issue can be resolved by allowing the constant term to vary across observations   . Therefore, the use of the random-parameters model allows all estimable parameters to be fixed for each individual city but to vary across cities1.
To include random parameters in Equation (1), the city-specific estimable parameter is written as,
where is a parameter estimated for city k, is fixed across city, and is a randomly distributed term (for each city k) that can take on an extensive variety of distributions including the log-normal, beta, normal, and so on. Equation (1) can be estimated, with such random parameters (since varies across cities according to the random term as shown in Equation (2)), with maximum likelihood techniques. However, the maximum likelihood estimation of random- parameters regression model is computationally complex. Simulation-based likelihood methods are proven to be more appropriate, and an approach that employs Halton draws has more efficient distribution of draws than purely random draws  . Thus, for the present study’s estimation of the random-parameters model, we use Halton draws in NLOGIT 5.
To interpret estimation findings, the elasticity of gross city product per capita (GCPPC) with respect to each independent variable is defined as,
where and are the log-linearized forms of per capita gross city product and control variables for the kth city.
4. Estimation Results
The estimated results are illustrated in Table 32 and the detailed estimation for each specific city3 is reported in Tables 4(a)-(d). Turning to specific variables, highway freight volume is found to be statistically significant with a lognormal distribution and the expected positive sign, indicating that an increase in highway freight volume increases gross city product per capita (GCPPC). The average elasticity for highway freight volume across cities is 0.016 (as shown in Table 3), showing that a 1% increase in highway freight volume would increase a city’s gross product per capita, on average, by 0.016% and the impact varies across the selected cities in China. And it can be observed that the computed elasticity values vary significantly across different tier of cities4 shown in Tables 4(a)-(d). For example, highway freight volume generates the highest impact in the city of Shenzhen (Table 4(a)) and the lowest impact value in the cities of Zhangzhou in Fujian province (Table 4(c)) and Ji’an in Jiangxi Province (Table 4(d)). The average elasticity for highway freight volume is 0.03, 0.01, 0.01 and 0.02 for the 1st, 2nd, 3rd and 4th tier cities in China. This indicates the growth impacts of highway are larger in the first and the fourth tier cities.
The parameter for railway freight volume is found to be statistically significant with a positive impact on the GCPPC. The average is 0.005 and ranges from 0.0001 to 0.0307 across the selected cities. In Table 4(c) and Table 4(d), it shows that a 1% increase in railway freight volume increases GCPPC by 0.0001% in the city of An Qing in Anhui province, Rizhao in Shandong province, Zhuzhou in Hunan province, Mianyang in Sichuan province, Zibo in Shandong province and Chaozhou in Guangdong province. In comparison, the economic growth will rise by 0.031 in the city of Qinhuangdao in Hebei province. It appears that
Table 3. Random-parameters model estimation results (All random parameters are normally distributed).
Note: Value in parenthesis is the standard deviation of parameter distribution for parameter estimate and t-statistic.
Table 4. (a) Elasticities of highway and railway freight volumes in Tier 1 cities; (b) Elasticities of highway and railway freight volumes in Tier 2 cities; (c) Elasticities of highway and railway freight volumes in Tier 3 cities; (d) Elasticities of highway and railway freight volumes in Tier 4 cities.
while railway significantly contributes to city’s economic development, though on average, it generates a smaller economic impact compared to highway. This result is consistent in direction with previous studies   suggesting that for short distances and having flexibility of time in mind, highway is relatively a more effective way to transport freight; thus, the lower economic impact from railways compared to highways. In addition, the railway impact in the 1st tier of Chinese cities has been the lowest (0.0065) compared with the 2nd tier cities (0.0079). The contribution of railway is found to be approximately the same in the 3rd (0.0072) and the 4th tier cities (0.0073).
Highway passenger volume, which indicates the number of people commuting from one place to another along the highway network, is found to produce a statistically significant random parameter. The average elasticity is 0.017 with respect to GCPPC, with values ranging from 0.0001 to 0.1415 across cities. It can be observed that highway passenger volume significantly impacts a city’s gross product, and if this variable is ignored in the economic impacts analysis of transportation at the city’s level, the impacts from the other transportation variables would be upward biased.
The area of paved roads produces a positive and statistically significant effect, indicating that intra-city infrastructure can also be considered as an important factor in determining a city’s gross product, ostensibly by providing mobility and accessibility resulting in economic productivity. A 1% increase in paved road area in a city would increase GCPPC by an average of 0.007%, and this impact varies significantly from 0.00001% to 0.0233% across the selected 237 cities in China. Similarly, the number of public transportation units per person in a city, an alternative measure of intra-city infrastructure, is found to be statistically significant and the sign is positive as well. A 1% increase in public transportation unit per person would increase gross city product per capita, on average, by 0.021%, and the impact varies from 0.012% to 0.034% across cities. From the computed impact value, the result indicates that an increase in public transportation units per person would facilitate mobility and would improve accessibility, thus enhancing economic activity in a city.
With regard to non-transportation related variables, estimation results presented in Table 3 show that an increase in the labor participation rate (percentage of employable people in a city) increases the gross city product per capita. A 1% increase in labor participation rate would increase, on average, the GCPPC by 3.11% and the impact varies from 2.55% to 3.62% across cities, suggesting that labor participation rate, on average, has an elastic relationship with economic output in a city. The elasticity for fixed asset investment is found to vary from 0.007 to 0.103 across cities and a 1% increase in total fixed asset investments would increase gross city product per capita, on average, by 0.051% across cities. Urbanization rate, considered as the ratio of a city’s urban population to the city’s total population, is found to produce a statistically significant random parameter, and the average elasticity was 0.313, and varies from 0.002 to 1.042 across cities. This implies that in China the ongoing urbanization process significantly boosts local economic growth. The city’s congestion rate, measured as the ratio of the number of buses and taxies divided by the city’s area of paved roads at year-end, is also found to be statistically significant with a negative impact on a city’s gross product. A 1% increase in highway congestion rate would reduce GCPPC, on average, by 0.031%. Industrial sector’s contribution to a city’s gross product per capita results in statistically significant random parameter. A 1% increase in industrial sector’s contribution would increase, on average, a city’s gross product by 0.127%, and the impact varies from 0.103% to 0.149% across cities. Finally, we show that a 1% contribution from the service sector would increase a city’s gross product by 0.362%, and the economic impact varies from 0.148% to 0.51% across cities. The result indicates that the service sector’s elasticity is relatively higher than that of the industrial’s sector. This finding is also consistent with previous studies  -  . The preceding studies conclude that recent trends show that the service sector continues to be more innovative and productive compared to the industrial sector in many countries.
The present paper takes a renewed look at the relationship between transport and its effects on a city’s economic growth considering the differentiated transportation modes and the varied local economic conditions. In the past, cross-city analyses of this topic, especially in China, did not receive adequate attention due to data limitations and the absence of a methodological framework that could account for unobserved heterogeneity across cities. This paper shows the first attempt to use a multi-city data base to estimate a random-parameters model to account for unobserved heterogeneity across cities and we are practically capable to answer that which means of transportation matters more in which city of China.
From policy perspective, the results also provide clear evidence showing that urbanization plays a significant positive role in growing the city, which are echoed in Chen et al.  . And we show that key transportation measures, which have not been considered in past economic impact studies, including highway and railway freight volumes, highway passenger volumes, congestion rate, public transportation, fixed asset investment clearly influence a city’s gross product. However, the magnitude of the influence and the resulting impact on a city’s growth with respect to transportation varies considerably across cities. Among highway and rail freight volumes, it is found that highway freight volume on average has a much larger effect on a city’s economic output compared to railway freight volume. Among the cities, Shenzhen benefits most from its development of highway network while rail development is the growth engine for Qinhuangdao.
The findings of this paper generate rich policy implications in transportation infrastructure evaluations across Chinese cities. At the national level, the differences in elasticity values can enable the development of effective expenditure strategies for assigning weights to each mode in a multi-modal decision making process. At the regional policy level, the elasticity values estimated for highways and railways can be adopted to influence the distribution of transportation investment between inter- and intra-city transport networks.
1Simple fixed and random effects models are also estimated, in addition to a finite mixture model. However, likelihood ratio tests clearly indicate that a full random-parameters approach provides a superior statistical fit to the data.
2Detailed estimated parameters for the 237 cities are available upon requests.
4The subdivisions of the 1st-, 2nd-, 3rd-, and 4th-tier cities are based on the definitions given by the Institute of Finance and Trade Economics, Chinese Academy of Social Sciences.
 Melo, P.C., Graham, D.J. and Brage-Ardao, R. (2013) The Productivity of Transport Infrastructure Investment: A Meta-Analysis of Empirical Evidence. Regional Science and Urban Economics, 43, 695-706.
 Deng, T. (2013) Impacts of Transport Infrastructure on Productivity and Economic Growth: Recent Advances and Research Challenges. Transport Reviews, 33, 686-699.
 Vijverberg, W.P., Fu, F.C. and Vijverberg, C.P.C. (2011) Public Infrastructure as a Determinant of Productive Performance in China. Journal of Productivity Analysis, 36, 91-111.
 Yu, N., De Jong, M., Storm, S. and Mi, J. (2013) Spatial Spillover Effects of Transport Infrastructure: Evidence from Chinese Regions. Journal of Transport Geography, 28, 56-66.
 Demurger, S. (2001) Infrastructure Development and Economic Growth: An Explanation for Regional Disparities in China. Journal of Comparative Economics, 29, 95-117.
 Zheng, S. and Kahn, M.E. (2013) China’s Bullet Trains Facilitate Market Integration and Mitigate the Cost of Megacity Growth. Proceedings of the National Academy of Sciences, 110, 1248-1253.
 Agbelie, B.R.D.K. (2014) An Empirical Analysis of Three Econometric Frameworks for Evaluating Economic Impacts of Transportation Infrastructure Expenditures across Countries. Transport Policy, 35, 304-310.
 Brun, J.F., Combes, J.L. and Renard, M.F. (2002) Are There Spillover Effects between Coastal and Noncoastal Regions in China. China Economic Review, 13, 161-169.
 Del Bo, C.F. and Florio, M. (2012) Infrastructure and Growth in a Spatial Framework: Evidence from the EU Regions. European Planning Studies, 20, 1393-1414.
 Chen, Y., Salike, N., Luan, F.S. and He, M. (2006) Heterogeneous Effects of Inter- and Intra-City Transportation Infrastructure on Economic Growth: Evidence from Chinese Cities. Cambridge Journal of Regions, Economy and Society, 9, 571-587.