Socioeconomic impact assessments of high-Speed rail: A meta-Analysis

ABSTRACT

The interest in the socioeconomic impact assessments of the high-speed rail (HSR) system is burgeoning as many countries are considering investing in such a system. While extensive studies have provided a wide range of evaluations of HSR on the socioeconomic impact, primarily from an ex-post perspective, it remains unclear how reliable the evaluation outcomes were, given the variety of data, method and research framework being adopted. This paper provides a comprehensive review of the socioeconomic impact studies of high-speed rail (HSR), based on 242 academic publications. In particular, we examined to what extent the impact outcome is influenced by various modelling factors, such as data, model, and research design, using meta-analysis based on 368 empirical estimates derived from 45 empirical studies. Our analysis confirms that the factors of research design (e.g. variables, data, and modelling method) do have various influences on the empirical estimates of HSR project appraisals. In the end, the paper discusses the implications for future infrastructure project appraisals.

KEYWORDS:

• High-Speed rail (HSR)
• meta-analysis
• socioeconomic impact
• assessment approaches
• reliability and robustness

1. Introduction

High-Speed Rail (HSR) has received increasing attention worldwide. While HSR has experienced substantial development in many countries, such as France, Germany, Italy, Spain, Japan, and China, the development of HSR in other countries, such as the U.K., United States, India, and Malaysia has undergone substantial debate as it remains unclear whether the gargantuan cost of developing an HSR may be offset by the benefits the system is expected to generate.

One critical requirement of the decision-making of infrastructure investment is that a thorough socioeconomic impact assessment has to be implemented. The objective is to assess whether the project has the potential to generate positive impacts on society and the economy so that the validity and merit of the investment can be justified. The socioeconomic impact of HSR is phenomenal as the system may influence various aspects, such as the economy, the environment, tourism, housing, and the labour market. In addition, the effect may vary considerably depending on the scale of assessment and the area of interest.

Given the increasing interest in HSR planning and investment worldwide, extensive studies have assessed the socioeconomic impacts of HSR from different perspectives (Andersson, Shyr, & Fu, 2010; Cartenì, Pariota, & Henke, 2017; Chen, Xue, Rose, & Haynes, 2016; Diao, 2018; Pagliara, Mauriello, & Garofalo, 2017; Wang, Liu, Sun, & Liu, 2016). Although most of these studies evaluated the impact of HSR from an ex-post perspective, research findings are generally inconsistent. For example, Chen and Haynes (2017) revealed that regional economic disparity in China has decreased after the introduction of HSR. Conversely, Sun and Mansury (2016) showed that the regional economic gaps in terms of household income were enlarged due to the introduction of HSR. In addition, it is also unclear how reliable and robust the results of the empirical studies are because each assessment was conducted based on a specific analytical framework and assumptions. As a result, the implications for planning and policy of these studies could be questionable as these outcomes could hardly be validated.

While a few pioneering studies have provided reviews of the impact assessment of transport systems on related topics, such as multimodal transportation (Roberts, Melecky, Bougna, & Xu, 2020), land and housing impact (Bešinović, 2020; Wu, Zheng, Wang, & Du, 2020), and competition of HSR with aviation (Zhang, Wan, & Yang, 2019), there is still a lack of a comprehensive and systematic review of socioeconomic impact assessment of the HSR system.

Hence, the objective of this paper is to fill this research gap by addressing the following research questions: 1) how the socioeconomic impacts of HSR were evaluated in the literature and 2) how the empirically estimated outcome is influenced by various factors, such as the geographic scale of analysis, data, modelling method, control variables, etc.

This study contributes to the literature in the following aspects. Firstly, we provide a systematic review of the socioeconomic impact assessment of HSR projects based on 242 empirical studies published between 1997 and 2020. To our limited understanding, this is the first attempt to provide a comprehensive and in-depth knowledge of the literature. Such an understanding will help us identify the research gaps in this field and directions for future research. Secondly, we examine the sensitivity of the evaluation outcomes to the influence of various contributing factors through a meta-analysis. The research findings guide scholars to improve the impact evaluation of infrastructure in future impact assessment. Thirdly, this study provides important implications for planners and policymakers to reconsider the value of transportation infrastructure investment by understanding HSR’s effects from various angles.

The rest of the paper is organised as follows. Section 2 provides a detailed review of the studies that examined the socioeconomic impacts of HSR from various perspectives. Section 3 introduces the data and the meta-analysis method. Section 4 discusses the results, whereas section 5 discusses limitations and future research directions.

2. Literature review

The review of the socioeconomic impacts of HSR focuses on the following seven aspects: change in accessibility, the effects on other transportation modes, environment, tourism, housing and land use, labour market and industry, and economic performance. These aspects are identified and selected based on two considerations. First, they were identified based on the frequency of keywords of the reviewed articles. Second, these aspects were selected to provide comprehensive coverage of the various aspects of socioeconomic activities, based on the classification of other relevant studies (Chen, Haynes, Zhou, & Dai, 2019). In addition, it should be noted that tourism impact and housing impact were reviewed separately rather than included in the industrial impact since their focuses were different. Specifically, tourism impact focuses on effects on the tourism demand or destination choices; housing impact concentrates on the impact on the hedonic value of houses which refers to people’s subject experience on the housing price rather than the housing industry. Conversely, the industrial impact emphasises industrial structures, geographic distribution, and labour activities across regions.

The conceptual framework of the socioeconomic impact of HSR investment is illustrated in Figure 1. The development of HSR generates a direct effect on travel demand through an improvement in transportation accessibility and increased substitution and complementarity with other transport modes, such as aviation and highways. The travel effects may further affect other aspects such as the environment, tourism, housing and land use, labour market, and economic performance, which may eventually generate an impact on the economy and society.

Figure 1. The conceptual framework of the socioeconomic impact of HSR investment

2.1 Change of accessibility

Accessibility refers to the capacity of being reached or accessed by various transportation modes (Cao, Liu, Wang, & Li, 2013). It can be represented by the travel time to a place, the daily accessible population, and the economic opportunities in that place. Change of accessibility is generally considered a key outcome of HSR development. The evaluation of accessibility changes of HSR mainly addresses the following two questions: to what extent the accessibility varies after the introduction of HSR, and whether the change in accessibility is balanced among different regions. Most studies suggest that HSR improves intercity accessibility due to travel time savings. For example, Wang et al. (2016) revealed that accessibility is likely to improve by 9.6% in the Jiangsu Province in China by 2030 when 95% of its counties are served by HSR. Conversely, other studies indicated that the accessibility impact of HSR is limited in certain cities/regions, such as only in big cities and the cities being served directly by an HSR. For example, Sánchez-Mateos and Givoni (2012) showed that the accessibility of some cities in the U.K. was not improved by HSR because its positive impacts on accessibility were restricted to major cities with HSR stations; other cities without a direct connection to HSR received minimal benefits. The authors also suggested that a better approach to evaluate the impact of HSR on accessibility requires covering a wider geographic area beyond the cities that are served by HSR. Similarly, Chen and Wei (2013) argued that the development of the Hangzhou East HSR station has a limited impact on urban accessibility because the new station is located in a suburban area that is relatively far from the city centre. As a result, they indicated that the location of HSR stations should be determined more cautiously to maximise accessibility benefits.

The issue of whether the changes in accessibility are balanced has been evaluated at various geographic scales. For instance, Jin, Xu, Lu, and Huang (2013) revealed that the change in accessibility was quite different among various stations after the opening of the Shanghai – Nanjing HSR line. Cities located in economically developed areas tend to benefit the most from accessibility improvements than those without HSR, or those are located in less developed regions (Zhu et al., 2016). Conversely, Wang et al. (2016) indicated that although central cities generally have higher accessibility than peripheral cities, changes in accessibility were found to be more significant in peripheral cities after the introduction of HSR. In addition, accessibility changes were also uneven at different geographic levels. For example, Gutiérrez (2001) revealed that after the opening of the Madrid – Barcelona – French HSR line, accessibility inequalities were found to have increased at the national level but decreased at the corridor levels. The evidence suggests that the results of accessibility assessments can be sensitive to the geographic scope of the analysis. Hence it deserves further attention in future assessments.

Depending on the specific focus, accessibility is often measured using indicators, such as weighted average travel time (WATT), daily accessibility (DA), and potential accessibility (PA) (Cao et al., 2013; Wang et al., 2016). Given that each indicator represents different aspects of accessibility, the outcomes of accessibility analyses can vary even when the same system is analysed. For example, Gutiérrez (2001) evaluated the accessibility impact of the Madrid – Barcelona – French HSR line using the aforementioned indicators. Surprisingly, the evaluation of the outcomes of accessibility changes was quite different: the areas with the greatest accessibility changes were measured by daily accessibility, whereas the changes were small if measured by potential accessibility. The results were found to be more dispersed when measured by weighted average travel time. To identify the common factors of different indicators, Martín and Reggiani (2007) developed a comprehensive accessibility indicator using the principal component analysis method. In addition, accessibility was measured using the shortest path method (Moyano, Martínez, & Coronado, 2018; Shaw, Fang, Lu, & Tao, 2014). Specifically, the shortest paths were calculated by travel time, travel distance, or travel cost based on live timetables and ticket prices. The change of accessibility thus can be displayed by comparing the average shortest paths at different times.

2.2 Modal impact

The socioeconomic impact of HSR has also been extensively examined from the perspective of modal competition, with a comparative analysis between HSR and other modes such as air, conventional rail, bus, and private vehicle. The impacts of HSR on aviation have received the most attention, given that HSR and short-haul air services are highly competitive. In general, the literature revealed a mixed impact of HSR on air transportation.

Some studies indicated that HSR has a negative impact on air transportation due to its substitutional effect on air, which results in reductions in air travel demand and airlines’ seat capacity (Jiménez & Betancor, 2012; Wan, Ha, Yoshida, & Zhang, 2016; Zhang, Yang, & Wang, 2017). However, the substitutional effects of HSR on air travel vary due to the influence of factors, such as routes, travel distance, city type, and countries (Chai et al., 2018; Chen, 2017; Wan et al., 2016). Castillo-Manzano, Pozo-Barajas, and Trapero (2015) found that only part of HSR passenger demand was substituted from air travel. Some studies, such as (Behrens & Pels, 2012; Danapour, Nickkar, Jeihani, & Khaksar, 2018; Li & Sheng, 2016) indicated that the major factors that influence HSR and air transportation competition include price, travel time, service frequency, and travel distance.

The development of HSR was also found to have positive impacts on air transportation. The competition can promote the expansion of air transport networks to cover more fringe markets (Jiang & Zhang, 2016), which then provides more options for passengers. In addition, the cooperation between HSR and aviation through strategies such as enhancing modal integration by seamless intermodal transfer is also likely to facilitate intercity travel experience and promote the growth of air transportation services (Xia & Zhang, 2017). The substitution between HSR and air transportation can also generate positive impacts under disruptive events such as transportation system failures or natural disasters (Chen & Wang, 2019). The development of HSR has also been found to have a positive outcome on air transportation in terms of reducing congestion at bigger airports (Takebayashi, 2016).

2.3 Environmental impact

The evaluations of the environmental impacts of HSR primarily focused on two key issues. The first issue is about how the development of HSR affects the environment in terms of CO₂ emissions. For instance, Chen et al. (2016) showed that the development of HSR in China has a substantial effect on generating CO₂ emissions due to increased demand. Similarly, Chang and Kendall (2011) indicated that although its negative effect on climate change could be offset after an extended period, HSR contributes to CO₂ emissions more substantially during its construction. In particular, the production of construction materials comprises 80% of HSR’s CO₂ emissions. Yue et al. (2015) showed that HSR operation also generates negative environmental impacts because the production of electricity for the operation of high-speed trains would increase the emission of greenhouse gas. In sum, it should be noted that a comprehensive environmental impact assessment of HSR should include emissions that occurred during both the construction and operation phases.

The second issue being investigated extensively concerns in what circumstances HSR can be environmentally beneficial. For instance, it is generally recognisd that the operation of HSR can reduce CO₂ emissions by substituting the demand from other transportation modes that have higher emission rates, such as short-haul air and automobiles. However, the effect of HSR on CO₂ emission reduction has been found limited (Åkerman, 2011; Bueno, Hoyos, & Capellán-Pérez, 2017; Dalkic, Balaban, Tuydes-Yaman, & Celikkol-Kocak, 2017; Robertson, 2016; Westin & Kågeson, 2012). For example, Dalkic et al. (2017) argued that in the case of Turkey, the effect of HSR in reducing CO₂ emission is limited because it has attracted passengers from buses rather than air travel and cars, and the latter two have higher CO₂ emissions than the former. In addition, mode access to HSR is also essential. For instance, HSR stations with access by public transit are more environment-friendly than those that could only be accessed by private cars. In sum, whether or not the environmental benefits of HSR can be achieved depends on two critical factors: the degree of modal shift from other modes and the type of energies used for the electricity generation for HSR (Åkerman, 2011; Westin & Kågeson, 2012).

2.4 Tourism impact

Most studies investigating the impacts of HSR on tourism have found that the opening of HSR promotes tourism growth (Cartenì et al., 2017; Pagliara et al., 2017). However, some studies found that the impact of HSR on tourism was negligible in small cities (Albalate, Campos, & Jiménez, 2017), while other studies showed that HSR contributes to changes in the spatial structure of tourism (Huang, Xi, & Ge, 2017; Wang, Huang, Zou, & Yan, 2012; Wang, Liu, & Wang, 2018). For example, Wang et al. (2012) revealed that HSR causes a redistribution of the tourist demand, as well as cooperation and competition among tourist markets on a wider geographic range. Huang et al. (2017) showed that HSR increased the hierarchical differences in regional tourism.

Different research methods have been adopted to evaluate the impact of HSR on tourism. The choice destination approach, which is often conducted based on survey data, focuses on the estimation of how tourists’ travel choice is influenced by HSR service. For example, Pagliara, La Pietra, Gomez, and Manuel Vassallo (2015) discovered that more tourists rode HSR to visit cities near Madrid after the opening of the Spanish HSR. In addition, studies have also found that the impact of HSR on tourism may also vary by different HSR systems. For instance, Delaplace, Pagliara, Perrin, and Mermet (2014) showed that the presence of HSR had a positive and significant effect on tourists’ willingness to visit Paris, while it had no significant impact on tourists’ desire to visit Rome. The authors suspected that such a phenomenon might be because the French HSR system had been deployed for a long time while the Italian HSR system was relatively new. Hence, most travellers were still unaware of the system at the time when the survey was conducted.

Using gravity models is another key approach of tourism impact assessment of HSR. Based on the economic geography theory, most studies focus on examining the spatial effects of HSR, such as agglomeration and the diffusion of the tourism market. For instance, Huang et al. (2017) found that tourists tend to concentrate at core cities after the opening of HSR, which conforms with the core–periphery theory. HSR may, as a result, also enlarge the imbalances in regional development of tourism. For instance, Wang, Liu, et al. (2018) found that cities with HSR tend to attract more tourists than cities without it.

The impact of HSR on the tourism sector has also been examined using econometric analyses. Tourism output is often represented using variables such as the number of tourists, the number of overnights stays, and tourism revenue, whereas the effects of HSR are often measured by either a dummy variable indicating if HSR exists in the study area or a dummy variable indicating if tourists’ choice of destination is influenced by HSR. Continuous variables, such as HSR ridership and the number of HSR destinations, have also been adopted to reflect the effect of HSR (Albalate & Fageda, 2016; Castillo-Manzano, Castro-Nuño, López-Valpuesta, Pedregal-Tercero, & Garrido-Michó, 2018). The results are found to vary substantially given the different characteristics of research design, such as city size, trip distance, the degree of HSR development, the origin of tourists, and geographic scale. For instance, Albalate et al. (2017) showed that the effects of HSR on tourism are more significant in larger cities when other factors, such as population and hotel prices, were controlled. Cartenì et al. (2017) compared tourists’ preferences to HSR services with different distance range and revealed that people prefer to utilise HSR services when the trip is more than 400 km. Castillo-Manzano et al. (2018) found that HSR’s effects are more significant on domestic tourism than foreign tourism because air transportation is still the main transportation mode for international trips, but HSR increases the number of foreign tourists in cities with or near HSR stations. A similar result was also confirmed in the case of China (Chen et al., 2019)

In sum, although many studies have investigated the impact of HSR on tourism demand, the results are inconsistent, which is partially due to different regions and systems being evaluated. Another important reason, as mentioned earlier, is due to different research designs. As Chen and Haynes (2015b) indicated, given that the effects of HSR on tourism can be measured at different levels, the research findings are expected to be dissimilar.

2.5 Housing impact

Given that HSR has unique characteristics, such as a higher speed of operation and lower pollution than conventional diesel engines, it is expected that the housing impacts of HSR will be different from that of the conventional rail system. It is assumed that the development of HSR is likely to generate a positive impact on housing prices due to the improvement in accessibility. However, the magnitude of impacts could vary among different cities, spatial ranges, and HSR locations. For instance, Andersson et al. (2010) showed that the impact of HSR on housing prices in Taiwan was minor due to high ticket prices and entrenched residential location patterns. Similarly, Chen and Haynes (2015a) indicated that HSR’s impact on housing values is significant in small and medium cities in China. The authors further suggested that spatial dependence should be considered; otherwise, the estimation may be biased and inefficient. Conversely, Geng, Bao, and Liang (2015) revealed that due to negative effects from HSR such as noise and pollution, the system might negatively impact housing prices near a station. Moreover, Diao, Zhu, and Zhu (2017) indicated that the location of HSR stations should be considered when measuring the impacts of HSR on housing values.

The hedonic price model is often used to assess the impact of HSR on housing values. The distance from a property to an HSR station is considered a critical accessibility attribute. The impacts of HSR on housing values are evaluated based on the identification of the statistical relationship between variations in housing prices and that of the distance to a station. Moreover, the functional forms of the hedonic price model can be set as either linear or nonlinear. The most common functional forms in hedonic price estimations include log-linear, semi-logarithmic, and Box–Cox-transformed forms (Andersson et al., 2010). Conversely, to capture the nonlinear relationship between housing prices and distance to HSR stations, the linear model can also be revised using a spline function of distance to HSR stations while holding other variables linear (Diao et al., 2017; Geng et al., 2015).

2.6 Labour market and industry

The impact evaluation of HSR on the labour market (e.g. employment change, labour mobility, firm location, and industry agglomeration/diffusion) has also received increasing attention. Some studies, such as Lin (2017), found that the opening of HSR service is associated with an increased employment rate. Other studies also found that the development of HSR can promote labour migration and facilitate labour mobility due to the improvement of accessibility (Guirao, Campa, & Casado-Sanz, 2018; Guirao, Lara-Galera, & Campa, 2017). In addition, the opening of an HSR service was also found to have an association with the change of firm location, as the improved accessibility would make certain locations more attractive to businesses (Willigers & van Wee, 2011), although other studies argued that the effect of HSR on the location choice of firms is not deterministic (Beckerich, Benoit-Bazin, & Delaplace, 2017).

The impact of HSR on the economic performance of industries was also evaluated. For instance, Shao, Tian, and Yang (2017) investigated the effects of HSR on the agglomeration of the service sector, in which they found that the effect was more significant in production service than consumer service and public service-related industries. Dai, Xu, and Wang (2018) investigated three types of industries, which showed that HSR generated both an agglomeration and a diffusion effect on the secondary industry and tertiary industry, although the agglomeration effect is more significant in cities with HSR stations while the diffusion effect is more evident in cities without HSR stations.

The evaluation of HSR’s impacts on the labour market and industry is primarily conducted using econometric analysis. Labour mobility is generally measured in the number of non-resident persons (Guirao et al., 2017; Guirao et al., 2018). Industrial output is often measured in employment rate or employment density. The location quotient is often used to measure industrial agglomeration (Dai et al., 2018; Shao et al., 2017). In terms of HSR measurements, dummy variables that represent whether or not there is an HSR service in the study area are often adopted. In addition, other studies, such as Shao et al. (2017), measured the effect of HSR using the number of trains as a proxy.

2.7 Economic performance impacts

Economic performance is a concept that includes economic growth, productivity growth, and consumer welfare (Dedrick, Gurbaxani, & Kraemer, 2003). Economic growth is generally measured in GDP, gross output and employment changes, whereas productivity is often defined as a ratio between the output volume and the input volume and it measures the efficiency to use resources. Consumer welfare addresses the benefits obtained by the consumption of goods and services. Most studies on the economic performance impacts of HSR mainly explore two issues: how the development of HSR promotes economic growth, and how its effect on economic growth varies among different regions. Several studies revealed that HSR investment has a positive effect on promoting economic growth given the improvement of intercity accessibility (Chen et al., 2016; Chen & Silva, 2015; Diao, 2018; Li, Strauss, Shunxiang, & Lui, 2018). However, other scholars, such as de Rus and Nombela (2007) argued that the effect of HSR on economic growth depends on the transportation demand, and they further indicated that a positive economic effect from HSR is difficult to achieve when the number of passengers is below 8–10 million for a line with a distance of 500 km. Moreover, some studies revealed that the economic impacts of HSR vary depending on other factors, such as different lines (Ke, Chen, Hong, & Hsiao, 2017; Martín & Nombela, 2007), the distance of service (Meng, Lin, & Zhu, 2018), industries (Circella, 2015), and different stages of economic development (Cheng, Loo, & Vickerman, 2015). Nash (2015) indicated that the economic success of HSR projects is influenced by several factors, such as construction costs, travel time saving, traffic demand, and the level of congestion of the existing transportation network.

Different research methods, such as cost–benefit analysis (CBA), econometric models, and simulation analysis, were often adopted for the economic impact assessment of HSR. For instance, de Rus and Inglada (1997) and de Rus (2011) found that the economic impacts of the Spanish HSR system could be positive with a higher transport demand using CBA. Similarly, CBA has also been applied to analyse HSR in Taiwan (Cheng, 2010) and many other systems (Button, 2012; Nash, 2015). Although CBA has been widely adopted for large infrastructure project evaluations, Hansjürgens (2004) indicated that the method also has several limitations. For instance, benefits such as quality improvement and social welfare are hard to weigh up against the costs. The method also has a limitation of quantifying long-term economic benefits because future benefits need to be discounted to the current value using a discount factor which should be selected cautiously. Different discount factors might lead to various evaluation results. In addition, CBA tends to evaluate the direct benefits such as operation revenues, time savings, reduction of traffic congestion and accidents, and environmental impact (de Rus & Inglada, 1997), regardless of follow-on economic activities.

Instead, econometric analysis methods have been adopted as a complement to the CBA for the long-term economic impact assessment of HSR projects. The econometric analysis approach captures both direct benefits like time-saving and indirect/induced benefits generated by follow-on economic activities such as tourism and labour migration after new infrastructure investment. In econometric models, the economic output is often measured using macroeconomic indicators, such as GDP, GDP per capita, income, or employment. The economic impacts of HSR can be generally estimated using econometric models, through identifying the statistical relationship between the economic output variables and HSR variables.

Specifically, the difference-in-differences (DID) method is one of the most frequently adopted econometric approaches for HSR assessment (Ahlfeldt & Feddersen, 2018; Diao, 2018; Hernández & Jiménez, 2014; Li & Xu, 2018; Sun & Mansury, 2016). One critical step in DID analysis is to determine control groups and treatment groups. The rationale of the DID method is to compare the economic output between cities/regions that have HSR and those counterparts without it to identify whether the influence of HSR on economic outputs varies significantly among different groups. It has become a common practice that cities/regions with an HSR project or service and those without are considered the treatment group and control group, respectively. Given that large infrastructure projects, such as HSR, often have a strong spillover effect, the development of HSR may affect not only the city/region it is located in; it also influences neighbouring cities/regions even though they do not have an HSR station. As a result, it is possible that the control group may not be chosen appropriately.

Another limitation of DID is that the specification of the intervention for the treatment group can be problematic since the identification of the change of a “policy” (which, in this case, the development of HSR) in different periods is required. For instance, some studies, such as Diao (2018), considered the year when HSR is in operation as the intervention in their DID model, whereas other works, such as Ahlfeldt and Feddersen (2018) considered the year of HSR construction as the intervention instead. Unfortunately, it remains unclear to what extent the outcome of assessments may be different due to these various considerations.

Moreover, in the absence of treatment, it is generally assumed that the difference between the “treatment” and “control” groups is constant over time in a DID setting. However, the assumption may not be reasonable since many evaluations were conducted based on a dataset with a long-term period. As a result, the assumption is less likely to hold, which could lead to biased estimation.

To control for the influence of unobserved differences between the control group and treatment group, an advanced version of DID, known as the propensity score matching – difference-in-differences (PSM-DID) was also adapted (Jia, Zhou, & Qin, 2017; Meng et al., 2018). The method has an advantage over the basic DID in that the treatment effect is predicted based on a randomised estimation.¹ Hence, it is reasonable to assume that the outcomes of the socioeconomic impact assessment of HSR can be dissimilar due to the different methods being adopted.

Overall, the literature review suggests that although extensive studies have investigated the socioeconomic impact of HSR, there is a lack of systematic review of the outcomes of these studies, although the results were found to vary substantially, and in some cases, with contradictory findings. It is also unclear whether such variations reflect the genuine outcomes of different HSR systems due to the different objectives and transport policy settings of HSR projects, or if they reflect an inconsistency among various statistical models. Moreover, the variations of estimated outcomes may be essentially influenced by the intrinsic characteristics of the assessment framework, such as data, method, and assumptions. Hence, it becomes imperative to examine to what extent the evaluation outcomes may be influenced by various factors so that sound implications can be developed for future evaluation and decision-making.

3. Data and methodology

3.1 Data

To understand how the socioeconomic impacts of HSR were evaluated in previous studies, we reviewed the HSR-related academic papers published until October 2020. The search was conducted through “Google Scholar” and “ScienceDirect,” using keywords such as “high speed rail”, “impact,” and the related terms. In addition, to avoid omitting important studies, we also conducted a manual search with a focus on relevant journals², such as Transportation Research Part A, B, C, D, E, and F, Transport Reviews, Journal of Transport Geography, Urban Studies, Cities, Environmental Planning A, B and C, Journal of Economic Geography, Journal of Regional Science, Journal of Benefit–Cost Analysis, and several railway specific journals. The process of review is illustrated in Figure 2. The initial search returned 446 articles. Given that our focus is on the empirical studies of the socioeconomic impacts of HSR, the samples were refined. Eventually, 242 papers were found the most relevant.

Figure 2. Data selection process

The publications were classified in terms of year of publication, country of focus, area of evaluation, and research method, respectively. As illustrated in Figure 3(a), the number of publications related to socioeconomic impact evaluations of HSR has increased substantially since 1997. In terms of the country of focus, Figure 3(b) shows that most evaluations focus on China and Spain, which is not surprising given that both countries have developed a large-scale HSR system in recent decades. We also note a high correlation between the HSR length and the number of publications in our review database divided by countries, which is approximate 0.95.³ In addition, extensive studies have also attempted to evaluate the impact of HSR from an international comparative perspective. The topics of evaluation are summarized in Figure 3(c). The modal impact of HSR has received the most attention, which accounts for 28% of the total publications. The evaluations of economic, tourism, and accessibility impacts are also extensive, which account for 25%, 11.6%, and 10.7%, respectively. Figure 3(d) summarizes the distribution of evaluations by research method. Almost 46% of the samples evaluated the impact of HSR using econometric models, including OLS regression model, difference-in-difference model, discrete choice model, Poisson regression model, spatial econometric models such as geographically weighted regression model, spatial lag model, spatial error model, and so on. In addition, other research methods, such as the indicator approach, cost–benefit method, and simulation method, have also been frequently adopted.

Figure 3. The Comparison of the Empirical Studies on Socioeconomic Impact Evaluations of HSR

3.2 Meta-Analysis

A meta-analysis was adopted to examine to what extent different empirical evaluation outcomes may be influenced by modelling characteristics. Meta-analysis is a statistical extrapolative technique, meaning it extracts information from previous studies to identify the influences of various data characteristics and analytical features on the output estimates (Fang, 2015; Graves, 2019; Park, Huang, & Newman, 2016). The basic structure of meta-analysis is formulated as equation 1: (1) $C_j=\alpha +\sum _{i=1}^n{b}_i\cdot HS{R}_{ij}+\sum _{k=1}^m{\gamma }_k\cdot X_{kj}+{ϵ}_j$(1) where $C_j$ refers to the output estimate, $b_i$ is the coefficient associated with the HSR variables, and ${\gamma }_k$ represents the coefficients associated with variable $X_{kj}$, representing different modelling characteristics (also called factors) in the jth empirical model. ${ϵ}_j$ is the error term corresponding to the jth observation; n represents the total number of HSR variables; and m represents the total number of factors.

To achieve a reasonable comparison, we focus on empirical work that evaluates HSR’s economic impact using econometric models only. Moreover, we focus on those models where dependent variables were taken log transformation. In that case, the coefficients of HSR variables could be explained as the percent change of economic output brought by a 1% change of the HSR variable, which is comparable across papers. The estimates of HSR variables that represent HSR’s impact on economic outputs were collected from 45 papers (summarized in Appendix 3). In total, 368 estimates were obtained from various studies with a mean value of 0.149, a median value of 0.06, a minimum value of −3.487, and a maximum value of 6.155. Moreover, the standard deviations (SD) of these estimates, representing the variation of estimated HSR’s impact on economic outputs, were also analysed. These SD estimates were treated as an outcome variable and were regressed on various independent variables that capture model and data characteristics. It should be noted that some papers reported t-value, standard error, or p-value instead of SD. Hence, we used this information and converted the outcome variable into a consistent format, measured in SD. In total, 315 observations of SD were collected with a mean value of 2.88, ranging between 0 and 42.12. The distributions of output estimates and their SD were summarized in Appendix4.

As summarized in Appendix 5, the explanatory variables adopted in the meta-analysis are classified into the following five categories:

Dependent variable: various dummy variables were introduced to represent different dependent variables used to measure socioeconomic outputs, such as population, GDP, personal income, and employment. In addition, other dependent variables such as housing price, number of tourists, tourism revenue, night lighting were also introduced to examine whether the output estimates vary given the different types of dependent variables being adopted. It should be noted that for categorial variables, the reference category should be removed due to the consideration of avoiding collinearity. GDP is treated as the reference of the dependent variable group. The estimated values of the remaining categories represent the variation compared with other categories and the reference group.

HSR variable: the dummy variables in this category represent different types of indicators that were used to quantify the effect of HSR among various studies. HSR dummy was the most frequently used variable in previous studies to represent the effects of HSR. Hence, it was selected as the reference case of HSR variable group. The variable of “hsrafter” was introduced to represent the interaction term of “HSR×After,” which has often been adopted to capture the change before and after the implementation of HSR as well as the difference between cities/regions with and without it. In addition, the effects of HSR were also measured using other indicators, such as frequency, the number of services, and the level of accessibility.

Control variable: variables in this group capture various types of control variables that were adopted in the evaluation. For example, the variable of the year was often introduced to control for the time-fixed effect. The variables, such as province, county, and city (base case), were often introduced to control the regional-fixed effect. Other variables, such as dummy variables representing different industrial sectors, were widely used to control the sectorial effect. In addition, dummy variables that represent other transport modes, such as air and highway, were also introduced to control the corresponding effects.

Data characteristics: these variables were introduced to examine to what extent the output estimates of HSR impact evaluations vary given the consideration of date characteristics, such as data type, periods, source, and format. For instance, the dummy variable of “panel” was introduced to represent the fact that whether the output was estimated based on the panel dataset. To examine the variation of estimated outcome concerning the geographic unit of data analysis, three dummies representing county-level, city-level, and country-level, were introduced to control for the different scales of data adopted in different studies. The “city-level” was treated as the reference case, which hence was left out of the meta-analysis model. The basic assumption was that output estimates are likely to vary when the geographic unit of analysis is different. The variable “data length” was adopted to indicate how many years the panel data covers. The variable “time difference” was used to represent the time difference between the year of data collection and the year that HSR started to be in operation. In addition, to examine whether the impact of HSR differs among different countries, country-specific variables were also introduced.

Model specification: variables in this group examine if modelling structure influences the output estimates. In particular, the following modelling types were examined: difference-in-difference (DID), discrete choice, Poisson, and spatial econometrics, and the basic OLS model (which was treated as the reference case). Additionally, the variable “iv” was introduced to examine whether the output coefficient estimated from a model with an instrumental variable differs from other regression models. In addition, the function type of models, such as log-linear, log-dummy, and log–log (reference case), was introduced to compare whether the variation of estimates is also influenced by such modelling specifications.

The meta-analysis was conducted separately in six groups to assure robustness and consistency. Group 1–5 examine the influences of the corresponding five groups of explanatory variables on the output estimate, respectively, whereas Group 6 examines the influence of the aggregate factors of all the variables. Moreover, the output estimates representing HSR’s impact on economic output and its SD are adopted as dependent variables of the meta-analysis.

4. Results

The detailed results of the meta-analysis are summarized in Appendix 6. Model 1 shows the influence of dependent variables on the estimated output. The impact of HSR on the economy, in general, is more substantial if the output variable reflects socioeconomic performance at the individual level rather than at the macroeconomic level. For example, the impact of HSR was found relatively higher if the dependent variable is measured in personal income, employment, or population rather than GDP. Moreover, the impact of HSR was found to be relatively higher if the dependent variable is measured in Gross regional product (GRP), which suggests that the impact of HSR is found relatively higher at the regional level than at the national level. Such an outcome may reflect the fact that the distributional effect of HSR can be uneven within a country. As a result, the effects are attenuated if the analysis covers the entire country rather than some regions with HSR services. In addition, the estimation was found to be relatively higher if the dependent variable is measured in total factor productivity (TFP) rather than GDP. In other words, HSR assessment using TFP as the dependent variable is likely to find a relatively higher coefficient than that using GDP. The result suggests that the impact of HSR on production efficiency as a consequence of economic efficiency improvement was found more significant than that on GDP.

The analysis also shows that the variable representing the brightness of night light is statistically significant (at a 5% level), which indicates that the output estimated value of HSR’s impacts is 21.7% higher if it is measured by night lighting instead of other measures, such as GDP. Conversely, the impact of HSR was found to be relatively lower if the dependent variable is measured in proportions of secondary industries rather than other types of dependent variables. It implies that the impact of HSR on secondary industries is lower than on other output measures, such as GDP.

Model 2 displays the influence of HSR variables on the output estimate of impact evaluations. It shows that the estimated output was found to be relatively higher if the effect of HSR was measured by accessibility indicators like WATT rather than other variables, such as the HSR dummy generated based on the year of service availability. Such a result suggests that the impact of HSR is likely to be greater if the key variable is able to capture the quality change (e.g. accessibility, frequency, cost) brought by HSR, rather than a spatial and temporal change (e.g. which is often captured through an HSR dummy variable) of the system. The analysis also shows that the estimated output was likely to be relatively higher if the effect of HSR was measured by the average share of HSR trips out of all trips. Such a result suggests that the impact of HSR is likely to be greater when compared with other travel modes. In addition, the estimated output coefficient value is likely to be lower when both the regional and temporal effects of HSR are measured simultaneously by “HSR×After” rather than other forms of HSR variable, although it is not statistically significant in our model.

Model 3 summarizes the influences of control variables on the output estimate in HSR’s impact assessment studies. One of the key findings is that some of the control variables such as year, regional fixed effect, sociodemographic characteristics, geographic location, and other transport modes have statistically significant influences on the estimated outcomes of impact evaluations. For instance, the output estimate of HSR’s impact was found to be relatively higher if the assessment was conducted at the province level, if sociodemographic variables such as age and wage were controlled for, or if geographic location like distance to core city was controlled. Conversely, the control of the temporal effect using a year dummy variable is likely to be associated with a 26.8% lower value of the outcome, ceteris paribus. This could be explained by the fact that the evaluation outcome tends to be more accurate if the temporal effect of the socioeconomic output growth is controlled. Moreover, the output estimate of HSR’s impact is likely to be lower if highway features such as the length, numbers, or density of highways are controlled for in the modelling framework, ceteris paribus, because in such a way, the independent effects of HSR could be separated from that of other competitive transport modes like the highway.

Model 4 reveals the statistical association between data characteristic variables and the output estimate of HSR’s impact. The key findings are summarized in the following aspects. First, the output estimate tends to be dissimilar given the different geographic units of data analysis. In general, a relatively larger impact outcome is likely to be observed if the geographic unit of assessment is at a province-level rather than at a city-level. In other words, A study that adopted province as the unit of analysis is likely to observe a relatively larger impact than that based on a unit of analysis at the city-level. Such a result suggests that the assessment with a larger geographic unit is likely to capture spillover effects of HSR, which could hardly be measured at the city-level of analysis that may only consider the neighbouring areas within the urban scale. This may also reflect the fact that the agglomeration effect of HSR, which refers to the phenomenon that HSR promotes the concentration of economic activity at a larger geographic scale, is more likely to be captured at a more aggregate level of assessment. The reason for such an outcome, as explained by Giuliano, Kang, and Yuan (2019), could be that large areas have the benefits of scale, labour diversity, extensive transport and communication network. In addition, a relatively large scale of area is likely to capture the agglomeration activities among multiple centres, which lower costs of land, wage, and congestion while still preserve agglomeration benefits.

The output estimates also show spatial heterogeneity, which means that the estimation outcome is likely to change among different types of areas. In particular, the impact of HSR was found to be higher in urban areas when compared with rural areas. Third, a relatively higher impact output is likely to be observed if the time difference between data collection and the start of the operation of HSR is higher, which implies that the impact of HSR is enlarged with the continuous operation of HSR. It also confirms the time frame of the analysis (i.e. whether HSR has been there for a long time or is relatively new) is an important factor that influences the impact of HSR.

Model 5 shows the influence of model specification on the output estimate. The results show that compared to the result using basic models, such as OLS, the estimated outcome is likely to be higher if the following types of models are adopted in the assessment: DID, discrete choice, and models with fixed effects. Conversely, the estimated outcome is found to be generally lower than average if the assessment adopted either the propensity score matching - difference in differences (PSM-DID) model or spatial econometrics method. The results, in general, suggest that a relatively lower estimate is likely to be achieved if the issue of spatial heteroscedasticity is addressed in the assessment. The analysis also reveals that a higher output estimate is likely to be observed when the output variable and HSR variable are estimated in a log-linear format than the log–log format. This may reflect the different meanings of output coefficients being estimated from different modelling structures.⁴

Model 6 shows the influences on the output estimates of the socioeconomic impacts of HSR, with control of all the key variables from the five categories of factors. Some variables were omitted due to multicollinearity. Only those variables that are statistically significant are displayed. Although the values of the estimated coefficients of each factor vary slightly, the results are generally consistent with those estimated in separate models in terms of sign and significance. This confirms that the results of the meta-analysis are robust. Overall, the R-squared value of Model 6 is 0.571, suggesting that 57.1% of the influences on the variation of the output estimate in HSR’s impact assessment are explained.

Moreover, the meta-analysis results using standard deviation (SD) as the output estimates are summarized in Appendix 7. The SD models show how the selection of variables and the model specification affect the variability of the output estimates of HSR’s impact. It helps us to understand how the economic effects of HSR fluctuate under different model settings. Overall, the analysis reveals the following major findings:

Firstly, Model 7 shows that the variability of estimated effects was found higher if the output variable is a human-related metric, such as employment and population, rather than an economic metric (such as GDP). Such a pattern suggests that the effects of HSR might be found more varied if the human-related metrics were adopted. Secondly, Model 8 displays that the variance of estimation is likely to be higher if the effect of HSR is measured by HSR*After (often used in DID models) and distance variables than the HSR dummy variable. This could be because the DID parameter and distance variable may better reflect the quality change of intercity travel by HSR than the simple dummy variable approach, which may provide a wide range of impact outcomes given the variability of the corresponding characteristics. Thirdly, Model 9 suggests that the introduction of key control variables, such as alternative transport mode, population density, neighbourhood factors, and other geographic and temporal factors, may also reduce the variability of estimated outcomes. In other words, a modelling framework with considerations of all the key control variables is likely to generate a more robust estimate result.

In addition, Model 10 reveals that the variations of the estimated HSR effects are also relatively lower in European cases (such as Spain, Italy, Germany, and France) than that of the cases in Asian countries (such as China, Japan, Korea), which suggests that there is also a systematic difference in the socioeconomic impact assessment outcomes of various HSR systems globally. The impact of European systems, in general, was found to have less variation than that in Asia, which may also reflect the levels of maturity of the system.

In sum, the comparison of the estimates based on mean estimates and SD reveals that the models with higher estimated HSR effects tend to be associated with higher variability of the estimates. We also observed that previous empirical assessments of HSR tend to primarily report the results of output mean-estimates, whereas SD was less often reported directly, although some studies may report either T-statistics, standard error, or p-value instead.

5. Conclusion

Despite a rising number of empirical studies attempting to understand the socioeconomic impact of different HSR systems, the evaluation outcomes remain unsatisfactory as the research outcomes are quite heterogeneous. While such a difference may reflect the various impacts of different systems and varying specific areas, it is likely that other factors related to the modelling designs may also cause a major difference. As a result, there is a lack of systematic understanding of how the issues were investigated in the field and how reliable and robust the empirical results are, given the adoptions of different modelling frameworks with varying considerations.

This paper tends to fill this gap by conducting a comprehensive overview of empirical studies focusing on socioeconomic impact assessments of HSR. Our analysis reveals that previous studies mainly attempted to evaluate the impact of the system from the aspects, such as accessibility change, modal impact, environmental impact, tourism impact, housing or land impact, labour market impact, and economic performance impact. In particular, most assessments were evaluated from the aspects of accessibility, economic impact, and modal impact. In terms of the regional focus, most of the empirical assessments evaluated the HSR systems in China and Spain, which is not surprising given that both countries have developed a large-scale HSR system. Hence, there are more pressing desires to understand the impact from an ex-post perspective in these countries than others.

In terms of research methods, econometric analysis was found to be the most frequently adopted method. However, as revealed in Table 1, the outcomes of impact estimates were found to be heterogeneous and are influenced by various factors, including the specifications of the variables, data characteristics, and models. To summarize, the meta-analysis provides the following major research findings.

Table 1. Estimated Sign of Various Variables

Model Group	Dependent variable	Sign	HSR Variable	Sign	Control Variable	Sign	Data Variable	Sign	Model Specification	Sign
Output Estimates	Population	+	Distance to HSR station	-	Year	-	Province level	+	DID	+
	Personal income	+	Average share of HSR distance	+	Province	+	Urban	+	PSM-DID	-
	Employment	+	Average share of HSR travel time	+	Age	+	Time difference	+	Discrete choice model	+
	Willingness to travel	+	Weighted average travel time	+	Distance to city centre	+			Spatial econometrics	-
	Night lighting	+	HSR frequency	+	Wage	+			Fixed effect	+
	TFP	+			Highway	-			Loglinear	+
	GRP	+
	Proportion of secondary industry	-
Standard Deviation	Population	+	HSR*After	+	Distance to city centre	+	County level	+	PSM-DID	-
	Employment	+	Average share of HSR distance	+	Population density	-	Province level	-	IV	+
	Number of tourists	-	Distance to HSR stations	+	Neighbourhood features	-	Data length	+	Discrete choice modelSpatial econometrics	+
					Rail density	-	Europe	-	Spatial econometrics	-
					Air	-			Loglinear	+
					Highway	+

Notes: The table was developed based on the results from Model 1 to 12 (summarized in Appendix 6 and 7). Only estimates with a statistical significance were reported. The reference cases for the five groups are GDP, HSR dummy, no control variable, city level, OLS, respectively.

First, the study confirms that both the estimated result of HSR’s socioeconomic impact and the corresponding SD tend to be higher if the dependent variable is measured using variables, such as population, employment than the base case, which adopted GDP as the dependent variable. In addition, the impact was also found to be relatively higher when the output is measured in personal income GRP and TFP.

Second, our analysis revealed that a larger impact of HSR is likely to be found if the effect of HSR was measured using quality indicators representing HSR service (e.g. accessibility change) rather than a simple dummy variable, which has been excessively used to capture the time difference before and after the implementation of the HSR system. Similarly, the variation of the outcome may also be relatively higher, as reflected in the SD result, suggesting that these variables may provide a wider range of distributions of impact outcomes.

Third, the study confirms that the impact outcome is likely to vary substantially if key control variables, such as the temporal trend (year dummy), regions of the focused area, geographic location, and competing transport modes are not controlled. These results suggest a more careful approach to empirical modelling development is the key to guarantee the robustness of the estimation outcomes.

Fourth, the estimated result of HSR’s socioeconomic impact was found relatively higher if the evaluation is based on a panel dataset. Conversely, the impact estimates are often lower if the geographic unit of data analysis is smaller (e.g. at the city level instead of a provincial level). In addition, the impact estimates were found higher if the HSR being evaluated has a relatively long period. These findings suggest that different evaluation outcomes may reflect various degrees of agglomeration/spillover effects and temporal effects of the HSR captured in the assessment. Hence, future analysis should be selective in terms of identifying the appropriate geographic unit for the impact assessment.

Lastly, the model specification was also found to have a significant influence on the output estimate. In particular, a relatively higher impact is likely to be generated when research methods, such as DID and discrete choice modelling, were adopted. Conversely, a relatively lower output estimate is likely to be found in spatial econometrics models when the issue of spatial dependence is controlled.

These findings also explain why the evaluation outcomes are different in some studies. For example, Chen and Haynes (2017) used GDP to measure the economic performance at the national level, while Sun and Mansury (2016) used household income instead and conducted the analysis at the regional level. Our study reveals that the impact of HSR was relatively higher if the economic outcome is measured in personal income rather than GDP. Moreover, the effects of HSR were found to be somewhat higher at the regional level than at the national level. Therefore, the different findings of Sun and Mansury (2016) and Chen and Haynes (2017) may reflect the different modelling characteristics, such as the selected variables and the geographic scope of analysis.

In sum, the research findings provide detailed guidance and implications for scholars and practitioners to improve their impact evaluation framework for HSR in future assessment endeavours. The evaluation of the socioeconomic impacts of HSR should be conducted more cautiously, with a thorough consideration of data, scale, and modelling specification. Hopefully, these aforementioned key findings may provide some guidance for researchers to develop appropriate empirical models for future assessment. Specifically, the following recommendations can be drawn from this paper. First, although the mean estimation is essential to reflect the level of HSR’s socioeconomic impacts, other metrics, such as standard deviation and confidence interval, can also be valuable as they provide additional information to understand the robustness of the estimates and the distribution of the impact outcomes. Hence, we suggest that scholars consider providing information on the estimation outcomes, such as standard deviations or confidence intervals, in their future impact assessment of HSR. In addition, robust tests of models should be added to help readers understand the extent of the robustness of the analytical results.

Second, given that the impact assessment results were found sensitive to the specifications of geographic and temporal scales, future work on socioeconomic impact assessment of HSR should consider choosing an appropriate scale of analysis if necessary. In particular, a larger geographic unit of data (e.g. at the province level) should be considered if the objective of assessment involves capturing the spatial spillover effects and agglomeration effects that the system generates. In addition, it is important to obtain a longer period of data so that the temporal impacts could be captured more accurately.

Third, we suggest that future impact assessment of HSR should adopt relevant indicators or variables that reflect a quality improvement of the services per se, such as accessibility, travel time, and frequency. Such an assessment could tell more meaningful insights than those analyses based on a simple dummy approach.

Fourth, in terms of evaluation methods, given that spatial heteroscedasticity can be a common issue in mega infrastructure impact assessment due to the scale of these projects involved, we suggest that advanced statistical methods, such as PSM-DID and spatial econometrics models shall be considered as a norm for the assessment, as long as the relevant data is available.

In addition, we also suggest that future research of the socioeconomic impact assessment of HSR could be further expanded in terms of enhancing the linkage between the theoretical foundation and empirical modelling framework, the focus on the social impact generated by the HSR system, such as social equity and travel behaviour changes, and the substitution between HSR and other transportation modes, such as conventional rail and buses, so that the impact of HSR on the change in demand and environment could be further understood.

One should note that our evaluation also has a few limitations, which deserve further research. For instance, the research findings were mainly derived from the selected publications. Hence, the issue of variable selection bias may be unavoidable. As the literature keeps expanding, a continuous effort is needed to expand the database and analysis further. In addition, the analysis can be further expanded by using bibliometric and scientometric analysis and knowledge graph analysis so that the knowledge structure of the field could be further understood.

Acknowledgement

An earlier version of the paper was presented at the 15th biennial NECTAR conference “Towards Human Scale Cities - Open and Happy”, hosted at the University of Helsinki during 5–7 June 2019. The authors want to thank the valuable comments received from both the conference participants and the two anonymous reviewers. Views presented in this study do not represent any organisations, and authors are responsible for any errors or mistakes.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

1 This is due to the fact that the bias caused by confounding variables can be reduced in PSM. The confounding variables could be found in an estimate of the treatment effect obtained from a simple comparison of the outcomes among units that received the treatment and those that did not.

2 The full list of key journals surveyed for the review is summarized in Appendix 1.

3 The length of HSR lines by countries is summarized in Appendix 2.

4 The value estimated from a log-log structure is essentially interpreted as elasticity, whereas the structure of the log-linear models is interpreted as the percent change of the output as the unit change of the independent variable.

Appendix

Appendix 1. Key Journal List for Literature Review

ID	Journal List	ID	Journal List
1	Journal of Transport Geography	21	Annals of Tourism Research
2	Transportation Research Part A, B, C, D, E, F	22	Economic Modelling
3	Cities	23	Journal of Air Transport Management
4	Transport Policy	24	Journal of Urban Economics
5	Transport Reviews	25	The American Economic Review
6	Sustainability	26	Research in Transportation Business & Management
7	Urban Studies	27	Research in Transportation Economics
8	Environmental Planning A, B, C	28	Current Issues in Tourism
9	Journal of Economic Geography	29	International Regional Science Review
10	Journal of Regional Science	30	Tourism Economics
11	Journal of Benefit-Cost Analysis	31	The Journal of Transport and Land Use
12	Applied Economics Letters	32	Journal of Geographical Sciences
13	Journal of Advanced Transportation	33	Tourism Management
14	Transportation Research Record	34	Advances in Civil Engineering
15	The Annals of Regional Science	35	Land Use Policy
16	Research in Transportation Economics	36	International Journal of Rail Transportation
17	Journal of Planning Literature	37	Journal of Rail Transport Planning & Management
18	Journal of transport economics and policy
19	Transportation Planning and Technology
20	Applied Economics Letters

Source: Authors’ collection.

Appendix 2. Summary of HSR and its Socioeconomic Impact Publications by Country

Area	Country	Length (km)					Publication
		In operation	Under construction	Planned	Long-term planning	Total
Asia Pacific	China	35742	5250	1071	257	42320	112
Asia Pacific	Japan	3041	402	194		3637	7
Asia Pacific	South Korea	893		49		942	6
Asia Pacific	India			508	4126	4634	4
Asia Pacific	Australia				1749	1749	2
Europe	Spain	3330	1293	676		5299	35
Europe	Italy	921	327			1248	9
Europe	Germany	1571	147	81	210	2009	2
Europe	France	2734			1725	4459	4
Europe	Netherlands	90				90	4
Europe	Sweden		11	150	589	750	2
Europe	UK	113	230	320		663	9
North American	US	735	763	1659	449	3606	9
Middle-East	Turkey	594	1652	789	4384	7419	2
Middle-East	Iran		1336	117	1651	3104	1
Africa	Egypt			910	300	1210	1

Data Source: UIC Passenger Department. Updated on 27th February 2020.

In operation: HSR is in operation; Under construction: HSR is still under construction; Planned: HSR has been approved while the construction is still yet started; 4. Long-term planning: the HSR was planned but not eventually approved by the government.

The publication column represents the number of publications included in our literature review database.

Appendix 3. Studies Included in the Meta-Analysis

Authors	Journal	Country	Focus
Sun and Mansury (2016)	Transportation Research Record	China	economic impact
Ke et al. (2017)	China Economic Review	China	economic impact
Li and Xu (2018)	Journal of Regional Science	Japan	economic impact
Jia et al. (2017)	Transportation Research Part A: Policy and Practice	China	economic impact
Meng et al. (2018)	Transport Policy	China	economic impact
Ahlfeldt and Feddersen (2018)	Journal of Economic Geography	Germany	economic impact
Chen and Haynes (2017)	Journal of Transport Geography	China	economic impact
Wetwitoo and Kato (2017)	Case Studies on Transport Policy	Japan	economic impact
Jiang and Kim (2016)	International Journal of Urban Sciences	International	economic impact
Carbo, Graham, Anupriya, & Melo, and C (2019)	Journal of Applied Statistics	Spain	economic impact
Deng, Gan, Perl, and Wang (2020)	Cities	China	economic impact
Yu, Lin, Tang, and Zhong (2019)	Journal of Regional Science	China	economic impact
Jin, Lin, Shi, Lee, and Li (2020)	Transportation Research Part A: Policy and Practice	China	economic impact
Li, Su, Xie, Zhu, and Wang (2020)	Sustainability	China	economic impact
Li, Chen, and Wang (2020)	Transport Policy	China	economic impact
Okamoto and Sato (2019)	Social Science Research Network	Japan	economic impact
Yao, Zhang, Wang, and Ou (2019)	Applied Economics	China	economic impact
Wang, Meng, and Hou (2020)	Research in Transportation Economics	China	economic impact
Gao, Song, Sun, and Zang (2020)	Review of Development Economics	China	economic impact
Chen (2019)	Journal of Business and Public Administration	China	economic impact
Zhang (2019)	Modern Economy	China	economic impact
Chong, Chen, and Qin (2019)	Journal of Transport and Land Use	China	economic impact
Zheng, Long, Chang, and Ye (2019)	Transport Policy	China	economic impact
Chen and Haynes (2015a)	Journal of Transport Geography	China	housing impact
Andersson et al. (2010)	Journal of Transport Geography	China	housing impact
Diao et al. (2017)	Urban Studies	China	housing impact
Pagliara (2019)	International Journal of Real Estate & Land Planning	Italy	housing impact
Wang, Ye, and Chen (2019)	Sustainability	China	housing impact
Wang, Niu, and Qian (2018)	Sustainability	China	housing impact
Lin (2017)	Journal of Urban Economics	China	labour and industry
Heuermann and Schmieder (2019)	Journal of Economic Geography	Germany	labour and industry
Shi, Lin, McLaughlin, Qi, and Jin (2020)	Transportation Research Part A: Policy and Practice	China	labour and industry
Campa, López-Lambas, and Guirao (2016)	Journal of Transport Geography	International	tourism impacts
Cartenì et al. (2017)	Transportation Research Part A: Policy and Practice	Italy	tourism impacts
Albalate et al. (2017)	Annals of Tourism Research	Spain	tourism impacts
Chen and Haynes (2015b)	Applied Economics Letters	China	tourism impacts
Pagliara et al. (2015)	Transport Policy	Spain	tourism impacts
Pagliara et al. (2017)	Transportation Research Part A: Policy and Practice	Italy	tourism impacts
Delaplace et al. (2014)	Procedia - Social and Behavioral Sciences	International	tourism impacts
Campa, Pagliara, López-Lambas, Arce, and Guirao (2019)	Sustainability	Spain	tourism impacts
Pagliara and Mauriello (2019)	International Journal of Humanities and Social Sciences	Italy	tourism impacts
Gao, Su, and Wang (2019)	Social Science Research Network	China	tourism impacts
Li, Yang, and Cui (2019)	International Journal of Tourism Research	China	tourism impacts
Pagliara, Mauriello, and Di Martino (2019)	Tourism: An International Interdisciplinary Journal	Italy	tourism impacts
Yang and Li (2020)	Current Issues in Tourism	China	tourism impacts

Source: Authors’ collection.

Appendix 4. Distribution of Output Estimates and Standard Deviations

Appendix 5. Explanatory Variables Used in Meta-Analysis

Category	Variable	N	Share%	Mean	Definition
Dependent variable group	GDP (reference case)	105	28.53	0.285	1 if GDP is used as dependent variable, 0 otherwise
	Population	14	3.80	0.038	1 if population is used as dependent variable, 0 otherwise
	Personal income	7	1.90	0.019	1 if personal income is used as dependent variable, 0 otherwise
	Employment	18	4.89	0.049	1 if employment is used as dependent variable, 0 otherwise
	Housing price	18	4.89	0.049	1 if housing price is used as dependent variable, 0 otherwise
	Number of tourists	49	13.32	0.133	1 if number of tourists is used as dependent variable, 0 otherwise
	Tourism revenue	13	3.53	0.035	1 if tourism revenue is used as dependent variable, 0 otherwise
	Overnight stays	10	2.72	0.027	1 if number of overnight stays is used as dependent variable, 0 otherwise
	Willingness to visit the city	5	1.36	0.014	1 if willingness to visit the city is used as dependent variable, 0 otherwise
	Average stays	8	2.17	0.022	1 if number of average stays is used as dependent variable, 0 otherwise
	Night lighting	22	5.98	0.060	1 if night lighting is used as dependent variable, 0 otherwise
	Total factor productivity	4	1.09	0.011	1 if total factor productivity is used as dependent variable, 0 otherwise
	Gross regional product	36	9.78	0.098	1 if gross regional product is used as dependent variable, 0 otherwise
	Gross value added	4	1.09	0.011	1 if gross value added is used as dependent variable, 0 otherwise
	Proportions of secondary industries	2	0.54	0.005	1 if proportions of secondary industries is used as dependent variable, 0 otherwise
HSR variable group	HSR dummy (reference case)	120	32.60	0.326	1 if HSR dummy variable is used to measure HSR effects, 0 otherwise
	HSR*After	138	37.50	0.375	1 if HSR*After is used to measure HSR effects, 0 otherwise
	Choice be influenced by HSR	16	4.35	0.043	1 if choice be influenced by HSR is used to measure HSR effects, 0 otherwise
	Distance to HSR stations	23	6.25	0.063	1 if distance to HSR stations is used to measure HSR effects, 0 otherwise
	Number of HSR stations	5	1.36	0.014	1 if number of HSR stations is used to measure HSR effects, 0 otherwise
	Number of HSR Lines	7	1.90	0.019	1 if number of HSR lines is used to measure HSR effects, 0 otherwise
	Number of HSR destinations	1	0.27	0.003	1 if number of HSR destinations is used to measure HSR effects, 0 otherwise
	Av. share of HSR distance	7	1.90	0.019	1 if average share of HSR distance is used to measure HSR effects, 0 otherwise
	Av. share of HSR travel time	9	2.45	0.024	1 if average share of HSR travel time is used to measure HSR effects, 0 otherwise
	Weighted average travel time	10	2.72	0.027	1 if weighted average travel time is used to measure HSR effects, 0 otherwise
	Potential accessibility	9	2.45	0.024	1 if potential accessibility is used to measure HSR effects, 0 otherwise
	Daily accessibility	9	2.45	0.024	1 if daily accessibility is used to measure HSR effects, 0 otherwise
	HSR frequency	8	2.17	0.022	1 if HSR frequency is used to measure HSR effects, 0 otherwise
	connectivity	6	1.63	0.016	1 if connectivity of HSR network is used to measure HSR effects, 0 otherwise
control variable group	year	256	69.57	0.696	1 if year is included as a control variable, 0 otherwise
	city (reference case)	152	41.30	0.413	1 if city is included as a control variable, 0 otherwise
	county	22	5.98	0.060	1 if county is included as a control variable, 0 otherwise
	province	9	2.45	0.024	1 if province is included as a control variable, 0 otherwise
	age	8	2.17	0.022	1 if age is included as a control variable, 0 otherwise
	education	37	10.05	0.101	1 if education is included as a control variable, 0 otherwise
	distance to city centre	32	8.70	0.087	1 if distance to core city is included as a control variable, 0 otherwise
	sector	50	13.59	0.136	1 if sector is included as a control variable, 0 otherwise
	population density	169	45.92	0.459	1 if population is included as a control variable, 0 otherwise
	unemployment rate	3	0.82	0.008	1 if unemployment rate is included as a control variable, 0 otherwise
	wage	67	18.21	0.182	1 if wage is included as a control variable, 0 otherwise
	neighbourhood features	27	7.34	0.073	1 if neighbourhood features is included as a control variable, 0 otherwise
	air	110	29.89	0.299	1 if air is included as a control variable, 0 otherwise
	rail	30	8.15	0.082	1 if rail density is included as a control variable, 0 otherwise
	highway	33	8.97	0.090	1 if highway is included as a control variable, 0 otherwise
data characteristics group	city level (reference case)	238	64.67	0.647	1 if the data analysis unit is city, 0 otherwise
	county level	49	13.32	0.133	1 if the data analysis unit is county, 0 otherwise
	province level	77	20.92	0.209	1 if the data analysis unit is province, 0 otherwise
	regional	292	79.35	0.793	1 if the analysis area is regional, 0 otherwise
	urban	12	3.26	0.033	1 if the analysis area is urban, 0 otherwise
	citysize	21	5.72	0.057	1 if the analysis area is categorized into cities in different sizes, 0 otherwise
	panel data	319	86.68	0.867	1 if the data is a panel, 0 otherwise
	secondary	360	97.83	0.978	1 if the data source is secondary, 0 otherwise
	time difference	368	100	15.33	Time difference between data collection and HSR open year
	data length	368	100	15.30	Years the data covers
	europe	84	22.83	0.228	1 if the observations are in European countries, 0 otherwise
model specification group	OLS (reference case)	64	17.39	0.174	1 if OLS model is adopted, 0 otherwise
	DID	210	57.07	0.571	1 if difference in difference model is adopted, 0 otherwise
	PSM-DID	61	16.58	0.166	1 if PSM-DID method is adopted, 0 otherwise
	discrete choice model	5	1.36	0.014	1 if discrete choice model is adopted, 0 otherwise
	spatial econometric	16	4.35	0.043	1 if spatial econometric model is adopted, 0 otherwise
	poisson	10	2.72	0.022	1 if Poisson model is adopted, 0 otherwise
	loglog (reference case)	24	6.52	0.065	1 if the function type is log-log, 0 otherwise
	logdummy	298	80.98	0.810	1 if the function type is log-dummy, 0 otherwise
	loglinear	46	12.50	0.125	1 if the function type is log-linear, 0 otherwise
	synthetic control group	12	3.26	0.033	1 if synthetic control method is adopted, 0 otherwise
	instrumental variable	24	6.52	0.065	1 if instrumental variable is used, 0 otherwise
	fixed effect	283	76.90	0.769	1 if fixed effect model is adopted, 0 otherwise

Note: Some variable groups with the reference case have a share summation of less than 100%. This is due to the existence of a few rare cases in which the category of the variable it represents either has a multicollinearity issue with other variables or is unrepresentative. Hence, no separate variable case was created.

Source: Authors’ summary.

Appendix 6. Results of the Meta-Analysis (1)

	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
Variable	Dependent Variable		HSR Variable		Control Variable		Data Characteristic		Model Specification		All
Population	0.358***	(2.71)
Personal income	2.863***	(15.64)									0.904***	(3.14)
Employment	0.281**	(2.38)									0.306***	(2.60)
Housing price	0.157	(1.33)
Number of tourists	0.106	(1.37)
Tourism revenue	0.027	(0.20)
Overnight stays	0.060	(0.39)
Willingness to visit the city	0.367*	(1.71)
Average stays	−0.093	(−0.54)
Night lighting	0.217**	(2.01)
Total factor productivity	1.117***	(4.66)									1.857***	(5.72)
Gross regional product	0.350***	(3.99)
Gross value added	0.027	(0.11)
Proportions of secondary industries	−2.213***	(−6.56)									−2.192***	(−7.17)
HSR*After			−0.004	(−0.05)
Choice be influenced by HSR			0.063	(0.43)
Distance to HSR stations			−0.254**	(−1.99)
Number of HSR stations			−0.034	(−0.14)
Number of HSR Lines			−0.103	(−0.48)
Number of HSR destinations			−0.021	(−0.04)
Av. share of HSR distance			0.585***	(2.72)
Av. share of HSR travel time			0.520***	(2.73)
Weighted average travel time			1.920***	(10.64)
Potential accessibility			0.015	(0.08)
Daily accessibility			0.012	(0.06)
HSR frequency			0.501**	(2.51)							−0.587**	(−2.54)
connectivity			−0.005	(−0.02)
year					−0.268***	(−3.93)					−0.250***	(−3.83)
province					0.506**	(2.08)					0.496***	(2.93)
county					0.020	(0.17)
age					2.437***	(11.49)					1.395***	(5.41)
education					−0.176	(−0.99)
distance to core city					0.302***	(3.09)					0.328***	(3.96)
sector					0.114	(1.25)					−0.178**	(−2.26)
population density					−0.092	(−1.18)
unemployment rate					−0.005	(−0.02)
Wage					0.177**	(2.35)
neighbourhood features					−0.087	(−0.80)
air					0.064	(0.67)
rail					0.135	(0.83)
highway					−0.327***	(−2.64)
panel data							−0.023	(−0.19)
secondary							−0.174	(−0.84)
county level							−0.085	(−0.87)
province level							0.215***	(2.60)
regional							−0.029	(−0.36)
urban							1.557***	(9.07)
citysize							−0.057	(−0.43)
time difference							0.007**	(2.55)
europe							−0.141	(−1.42)
DID									0.218**	(2.21)
PSM-DID									−0.395***	(−4.32)
synthetic control group									−0.219	(−1.22)
instrumental variable									−0.029	(−0.22)
discrete choice model									0.509*	(1.83)
spatial econometric									−0.284*	(−1.65)
fixed effect									0.186*	(1.72)	0.148**	(2.03)
poisson									0.062	(0.28)
logdummy									−0.154	(−1.11)
loglinear									0.642***	(4.10)	0.484***	(6.03)
Constant	−0.008	(−0.21)	0.078*	(1.73)	0.260***	(4.27)	0.207	(0.83)	0.004	(0.03)	0.074	(1.37)
No. of Obs.	368		368		368		368		368		368
VIF	1.05		1.06		1.88		1.38		1.74		1.81
R^2			0.285		0.427		0.253		0.180		0.571

Notes: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively. T-statistic is shown in parentheses. The reference cases for Model 1–5 are GDP, HSR dummy, no control variable, city level, OLS, respectively. The dependent variable of the meta-analysis is the estimated coefficient of HSR’s impact on the economic outcomes.

Appendix 7. Results of the Meta-Analysis (2)

	Model 7		Model 8		Model 9		Model 10		Model 11		Model 12
Variable	Dependent Variable		HSR Variable		Control Variable		Data Characteristic		Model Specification		All
Population	10.572***	(7.01)									0.323	(0.13)
Personal income	0.023	(0.00)
Employment	6.333***	(4.34)									0.744	(0.42)
Housing price	−1.357	(−0.99)
Number of tourists	−1.764**	(−2.00)									−0.056	(−0.07)
Tourism revenue	−1.320	(−0.88)
Overnight stays	−2.190	(−1.29)
Willingness to visit the city	−0.160	(−0.05)
Average stays	−2.667	(−1.41)
Night lighting	−0.566	(−0.45)
Total factor productivity	2.200	(0.84)
Gross regional product	0.115	(0.11)
Gross value added	−2.596	(−0.99)
HSR*After			1.704**	(2.30)							0.385	(0.42)
Choice be influenced by HSR			−0.970	(−0.62)
Distance to HSR stations			4.914***	(3.84)							6.187***	(3.60)
Number of HSR stations			−1.417	(−0.56)
Number of HSR destinations			−1.693	(−0.31)
Av. share of HSR distance			3.777*	(1.76)							2.919	(1.59)
Av. share of HSR travel time			2.280	(1.20)
Weighted average travel time			−1.704	(−0.61)
Potential accessibility			−1.583	(−0.83)
Daily accessibility			−1.603	(−0.84)
HSR frequency			1.766	(0.77)
connectivity			−0.518	(−0.22)
year					0.374	(0.49)					−1.275*	(−1.74)
province					−4.262	(−1.39)
county					−1.870	(−1.54)
age					2.449	(0.49)
education					2.908	(1.44)
distance to core city					8.368***	(8.37)					5.033***	(4.93)
sector					−0.192	(−0.19)
population density					−1.675*	(−1.93)					−0.274	(−0.34)
unemployment rate					0.033	(0.01)
wage					1.069	(1.36)
neighbourhood features					−2.588**	(−2.33)					−1.068	(−1.11)
air					−1.952*	(−1.90)					0.140	(0.13)
rail					−4.333**	(−2.54)					−1.024	(−0.79)
highway					4.268***	(3.41)					0.927	(0.78)
panel data							−1.448	(−1.28)
secondary							−5.884**	(−1.99)
county level							3.215***	(3.69)			−0.323	(−0.29)
province level							−2.761***	(−3.84)			−1.375	(−1.21)
regional							−0.924	(−1.26)
urban							0.261	(0.13)
citysize							2.001	(1.64)
data length							0.262***	(10.05)			0.209***	(3.83)
europe							−2.583***	(−3.04)			−0.130	(−0.12)
DID									0.022	(0.30)
PSM-DID									−0.187**	(−2.56)	1.775	(1.56)
synthetic control group									−0.079	(−0.61)
instrumental variable									0.178*	(1.85)	4.680***	(4.35)
discrete choice model									0.590**	(2.27)	2.986	(1.07)
spatial econometric									−0.460**	(−2.25)	0.731	(0.31)
fixed effect									0.137	(1.55)
poisson									0.188	(1.09)
logdummy									−0.073	(−0.70)
loglinear									0.278**	(2.35)	−1.603	(−1.15)
Constant	2.722***	(6.01)	1.877***	(3.85)	3.213***	(4.32)	7.254**	(2.11)	0.028	(0.29)	−0.503	(−0.52)
No. of Obs.	315		315		315		315		315		315
VIF	1.05		1.07		1.96		1.39		1.74		2.78
R^2			0.285		0.427		0.253		0.180		0.571

Notes: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively. T-statistic is shown in parentheses. The reference cases for Model 6–12 are GDP, HSR dummy, no control variable, city level, OLS, respectively. The dependent variable of the meta-analysis is the value of the standard deviation of the estimated coefficients of HSR’s impact on the economic outcomes.