Unpacking the Effects of Heterogeneous Incentive Policies on Sea–Rail Intermodal Transport: Evidence from China

Abstract

Sea–rail intermodal transport offers high efficiency and environmental benefits, yet its development in China remains limited. Existing studies have mainly assessed the macro-level benefits of sea–rail intermodal transport policies, but rigorous evidence on whether incentive policies work and how their effects differ across policy types remains scarce, which limits evidence-based policy design and efficient allocation between subsidies and capacity expansion. To address this gap, a dual-policy identification framework was established that combines a multi-period difference-in-differences model with event study analysis and used station–month data from China to assess the independent effects, underlying mechanisms, and spatiotemporal heterogeneity of railway freight price subsidies and freight train expansion on container throughput. The results indicate that both policies significantly increased container throughput. Railway freight price subsidies exhibited stronger and more persistent effects with a certain lag, whereas freight train expansion produced rapid but short-lived responses. The impacts of both policies were more pronounced in short-distance transport, but weakened or even turned negative over longer distances. Moreover, the number of participating entities served as a key mediating pathway, while information sharing positively moderates policy impacts. This study makes theoretical contributions to the identification of heterogeneity, mechanism analysis, and spatiotemporal characterization of SRIT incentive policy effects, while offering refined and actionable guidance for SRIT policy optimization.

1. Introduction

Global decarbonization goals have intensified pressure on the freight sector to reduce emissions [1]. Sea–rail intermodal transport (SRIT) offers an efficient modal integration, combining the low cost and high capacity of maritime transport with the low-emission, high-efficiency advantages of rail [2,3]. By facilitating a modal shift from road to rail, SRIT alleviates highway congestion, lowers carbon intensity, and improves safety, accessibility, and hinterland connectivity [4,5,6,7]. Since the 1960s, Europe and North America have advanced SRIT through national strategies and sustained infrastructure investment, leading gateways such as Los Angeles, Long Beach, and Hamburg now exhibit high adoption [8]. Consequently, SRIT is widely regarded as a central pathway toward cleaner and more efficient freight systems [9]. Recent international assessments indicate that, in Europe, policy mixes centered on subsidies and regulatory support most frequently show positive ex-post modal-shift performance [10], whereas North American intermodal development has hinged on scale economies and network resilience that accrue over longer horizons [11].

The development of SRIT relies on effective policy instruments [12]. In China, uptake remains nascent. Containerized freight is still road-dominated, with higher unit costs and carbon intensity [13]. Recognizing SRIT’s role in the transport sector’s green transition, national planning has introduced a suite of multimodal pilot programs and plans, notably the 14th Five-Year Comprehensive Transportation Plan [14,15,16]. These initiatives helped SRIT container throughput surpass 10 million twenty-foot equivalent units (TEUs) in 2023, a year-on-year increase of 11.7% [8]. Nevertheless, SRIT accounts for only 2–3% of total port throughput, far below the 20–40% observed in developed economies [17,18]. This gap reflects binding constraints arising from uncertainty about economic viability and service accessibility, which limit shipper participation and sustain a persistent preference for road transport [17]. In practice, shippers respond primarily to total logistics cost and expected commercial opportunities [7,19]. Accordingly, optimizing freight-rate structures, lowering end-to-end logistics cost, and extending freight-train service coverage could alleviate perceived risks, increase adoption, and support sustained SRIT development [13].

To address the cost and coverage constraints identified above, policy responses in China have prioritized two incentive instruments. The first, railway freight price subsidies (RFPS), lowers rail charges and improves the price attractiveness of SRIT for shippers [20]. The second, freight train expansion (FTE), optimizes train services and schedules to enhance accessibility and timeliness between ports and inland nodes, thereby enlarging potential markets and commercial opportunities. These instruments aim to provide more competitive modal choices, mitigate path dependence on road transport, and shift container flows toward SRIT. Nevertheless, rigorous empirical evidence remains limited on whether these policies materially promote SRIT and on how their effects differ across time and distance. This evidence gap constrains policy optimization and resource allocation and may reduce the efficiency of high-cost public investments.

Recent studies have examined the value of SRIT policies from multiple perspectives. For example, Li et al. [8] provided evidence that such policies reduce transport-sector carbon emissions, while Liu and Jia [21] highlighted that SRIT development optimizes domestic logistics networks and improves the efficiency of international trade flows. However, most prior work evaluates aggregate benefits or the macro policy environment, with limited attention to whether specific incentive instruments, such as RFPS and FTE policies, effectively promote SRIT. Distinctions across policy types are often overlooked, and systematic assessment of spatiotemporal heterogeneity across implementation periods and transport distances remains scarce. Given the high costs of policy implementation and the persistently low SRIT share in China, rigorous evaluation focused on temporal dynamics and distance-based heterogeneity is needed to inform differentiated, targeted policy portfolios for governments and port–rail operators, thereby maximizing policy returns and advancing sustainable development.

After clarifying the overall effectiveness of incentive policies in promoting SRIT development, it is necessary to further examine the underlying mechanisms that explain how these policies take effect, to identify the key factors for enhancing policy effectiveness. Shippers, as the primary decision makers in mode choice, shape SRIT adoption and operational performance through their behavioral intentions and actual participation. Prior research identifies limited shipper participation as a central bottleneck to SRIT expansion [17]. Thus, the ultimate impact of policy interventions is realized through shippers’ transport decisions. However, the literature has rarely adopted a shipper-response perspective to test the mediating role of shipper market participation and to position it within the policy impact pathway [13]. Introducing this perspective deepens the theoretical understanding of policy transmission and provides guidance for optimizing policy design and improving multimodal system performance.

A further issue is whether RFPS and FTE effects on SRIT container throughput are conditioned by external factors. Information sharing (IS), a core supply-chain mechanism for improving coordination and reducing uncertainty [22], is a plausible moderator. The realized effects of RFPS and FTE depend on the capacity of ports and railways to coordinate service plans and resource allocation in real time [23]. When IS is inadequate, mismatches arise between train operations and container allocation, which in turn cause yard congestion and greater empty-container movements. As a consequence, operating efficiency declines and shippers’ willingness to choose rail diminishes, thereby weakening policy effectiveness. The literature has largely overlooked this moderating role of IS, limiting explanations for divergent outcomes across regions and periods. Treating IS as a moderating variable helps delineate the boundary conditions of policy transmission and clarifies how policy performance responds to different information environments.

Given the limited evaluation of policy-type heterogeneity and spatiotemporal variations in prior research, and the unclear transmission mechanisms involving shipper participation and information sharing, we propose the following three research questions:

RQ1: Can both RFPS and FTE policies significantly increase SRIT container throughput, and what are the differences in their effect intensity?

RQ2: Do the effects of the above policies exhibit dynamic and spatial heterogeneity over time and across transportation distances?

RQ3: Through which mechanisms and boundary conditions do the above policies influence SRIT container throughput?

To address these questions, the analysis uses high-frequency station–month panel data from China and a multi-period difference-in-differences (MPDID) design combined with an event study. Mechanisms are examined from the perspective of shipper behavioral responses, with the number of participating entities (NPE) specified as a mediating pathway and IS incorporated as a moderating variable. Heterogeneity is assessed across time horizons and transport distances to characterize the spatiotemporal applicability of the two policy instruments.

This study makes three main contributions: (1) A comparative, instrument-specific evaluation of RFPS and FTE is provided, recognizing distinct effects on SRIT container throughput. (2) NPE is identified as a key mediating pathway, and IS is established as a boundary condition for policy effectiveness, offering a behaviorally grounded account of how incentives translate into outcomes. (3) Temporal dynamics and distance-based heterogeneity in policy effects are characterized, supplying evidence to support differentiated policy design and targeted implementation strategies. Collectively, our research advances the literature through methodological and theoretical innovations by introducing a unified empirical framework that combines MPDID with mediation and moderation analysis. In parallel, we provide the first comparative evaluation of heterogeneous SRIT policy instruments and leverage high-frequency data to capture policy dynamics, thereby addressing key gaps in prior research.

This paper is organized as follows. Section 2 formulates the hypotheses, develops the methodological model, and describes the materials; Section 3 reports the empirical results; Section 4 discusses the main findings, theoretical and practical implications; Section 5 concludes and outlines directions for future research.

2. Materials and Methods

2.1. Conceptual Framework and Hypotheses

This section articulates the conceptual framework linking policy instruments to SRIT outcomes and states three sets of testable hypotheses. The analysis first considers the direct effects of railway freight price subsidies (RFPS) and freight train expansion (FTE) on container throughput, then posits a participation-based mediation pathway and a boundary condition related to information sharing (IS).

2.1.1. The Impact of Policies on SRIT Container Throughput

The RFPS policy, a common government economic instrument, aims to reduce the rail transport cost within SRIT, thereby improving the economic viability of the transport solution, increasing shipper adoption, and ultimately promoting container throughput. Specifically, RFPS directly reduces the rail transport expenses borne by shippers. For shippers who aim to maximize cost-effectiveness [7], this creates stronger incentives to choose the SRIT mode. For example, evidence from Zhou et al. [20] indicated that subsidies for rail transport significantly increase the proportion of rail in optimal transport routes, underscoring the cost-driven effect of price subsidies on shipper decision-making.

Complementing the price channel, FTE policy can improve the accessibility, timeliness, and reliability of rail transport by adding new freight routes and increasing train frequency [24]. Specifically, expanding routes connects more inland regions directly with coastal ports [25], while increased frequency shortens waiting times and reduces overall transport cycles, offering more frequent and cost-effective options for time-sensitive cargo [26]. Additionally, scheduled services with fixed departure times enhance predictability, aiding shippers in optimizing production and inventory management, reducing uncertainty, and lowering costs [27]. Therefore, by optimizing the transport network, improving efficiency to achieve economies of scale, the FTE policy is expected to significantly increase the freight volume of SRIT. Accordingly, the following hypotheses are stated:

H1a. 

The RFPS policy will significantly increase SRIT container throughput.

H1b. 

The FTE policy will significantly increase SRIT container throughput.

2.1.2. Mediating Role in SRIT Policy Effects

The number of participating entities (NPE) is used to measure the scale of independent market participants adopting the SRIT transport mode. A higher value indicates that more logistics firms and cargo owners are actively involved in SRIT services. The RFPS policy directly reduces the rail segment transport cost, which helps alleviate the price sensitivity of small and medium-sized shippers. As a result, some entities that previously relied on road transport due to cost concerns are able to shift to SRIT [28]. For large-scale shipping enterprises, although mode switching may require upfront investments (e.g., contract adjustments and system integration) [21], it can reduce the trial-and-error costs and perceived risks associated with the transition to rail transport. Once the rail-based scheme proves reliable and timely, shippers who were previously hesitant are more likely to adopt SRIT. Therefore, RFPS can effectively expand the shipper base and improve both market diversity and penetration. FTE attracts new SRIT participants by expanding service coverage and enhancing transport predictability. New routes integrate previously uncovered inland cities into the rail system, while expanded train networks allow shippers to track departure and arrival times, mitigating coordination and delivery concerns [29]. This enhances SRIT accessibility and reliability, increasing its attractiveness and encouraging broader adoption.

Increased NPE consolidates a sufficient and stable container source, integrating scattered cargo demand into organized throughput that supports sustained operation of fully loaded, regularly scheduled freight trains [30]. This shifts from “temporary assembly” to “fixed schedules” improves operational regularity, deliverable capacity, reduces empty loads and dispatching costs, forming a virtuous cycle [31]. Therefore, NPE expansion enhances SRIT’s cargo base, continuously promoting container throughput growth. Based on this analysis, we propose the following hypothesis:

H2a. 

The RFPS policy significantly increases SRIT container throughput by expanding the NPE.

H2b. 

The FTE policy significantly increases SRIT container throughput by expanding the NPE.

2.1.3. The Moderating Mechanism on Policy Performance

Information sharing (IS) is a key supply chain mechanism that enhances coordination, reduces uncertainty, and optimizes resource allocation [32]. In SRIT, the level of IS between ports and railways affects policy effectiveness. For RFPS, while it stimulates demand by lowering costs, effective implementation requires adequate operational response. High IS in areas like train schedules, capacity, and forecasts enables coordinated resource management, reducing risks such as congestion and underutilization [4]. In contrast, poor IS reduces operational efficiency; even with subsidies, lack of execution capacity can suppress transport efficiency and shipper willingness, limiting the container throughput growth. Similarly, when expanding service coverage, FTE also increases transport rhythm and resource coordination complexity. Insufficient IS on train status, cargo in transit, and handling plans can lead to scheduling delays and resource mismatches, causing shipment delays and efficiency losses that weaken the policy’s effectiveness [33]. Thus, high IS is crucial for FTE’s effective implementation and for translating its incentives into actual throughput growth. We proposed two hypotheses:

H3a. 

IS positively moderates the incentive effect of the RFPS policy on SRIT container throughput.

H3b. 

IS positively moderates the incentive effect of the FTE policy on SRIT container throughput.

To visually illustrate the relationships among the hypotheses proposed in this study, the theoretical research model is presented as Figure 1.

2.2. Empirical Methodology and Model Specification

We estimate policy effects using a multi-period difference-in-differences design combined with an event study. Mediation and moderation are incorporated to test the market-participation pathway and the information-sharing boundary condition. Our methodological contribution is to implement a unified empirical pipeline that combines multi-period DID with an event-study decomposition and mechanism analysis (mediation via market participation and moderation via IS) to map instrument-specific effects, pathways, and boundary conditions within one coherent identification framework.

2.2.1. The MPDID Model

This study adopts the Multi-Period Difference-in-Differences (MPDID) model [34] to evaluate the impact of the RFPS and FTE policies on SRIT container throughput. This method is suitable for scenarios where policies are implemented in stages at different time points, and it can effectively identify both average treatment effects and dynamic response paths. Compared with the traditional Two-Way Fixed Effects approach, MPDID avoids estimation bias caused by weighting errors under treatment effect heterogeneity. It thus provides more robust causal inference [35] and enhances the reliability and explanatory power of policy effect identification. The model specification of this study is as follows:

$$\ln \left(C{T}_{it}\right)=\alpha +\beta DI{D}_{it}+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(1)

where $\ln \left(C{T}_{it}\right)$ denotes the natural logarithm of container throughput measured in TEUs for city $i$ in month $t$, which reflects the actual performance of SRIT operations; $\alpha$ is the sample-wide baseline constant; $DI{D}_{it}$ is the key explanatory variable that equals 1 if site $i$ is exposed to the RFPS or FTE policy in month $t$, and 0 otherwise; $\beta$ is the average treatment effect of the policy indicator ($DI{D}_{it}$) on the log outcome; $X_{it}$ denotes the vector of continuous control variables for city $i$ in month $t$, included to absorb non-policy fluctuations in SRIT container throughput. All controls enter the model in natural logarithms and in an additive linear form; $\gamma$ is the coefficient vector on the control variables $X_{it}$; ${\lambda }_i$, ${\delta }_t$ and ${ϵ}_{it}$ represent city fixed effects, month fixed effects, and the stochastic error term, respectively.

2.2.2. Mediation Effect Model

In order to reveal the paths through which the RFPS policy and the FTE policy promote the growth of SRIT container throughputs, this paper draws on the research paradigm of Huang and Yi [36], and introduces the intermediary variables and constructs the following model based on Equation (1) of the aforementioned benchmark regression model:

$$M_{it}=\alpha +\xi DI{D}_{it}+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(2)

$$\ln ({\mathrm{CT}}_{it})={\alpha }^{\prime }+{\beta }^{\prime }DI{D}_{it}+\zeta M_{it}+X_{it}{\gamma }^{\prime }+{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(3)

where $M_{it}$ denotes the mediating variable, representing NPE; $\alpha$ and ${\alpha }^{\prime }$ denote equation specific baseline constants; $\xi$ denotes the marginal impact coefficient of the policy on the mediating variable $M_{it}$; ${\beta }^{\prime }$ is the direct effect of $DI{D}_{it}$ on $\ln ({\mathrm{CT}}_{it})$ conditional on $M_{it}$; $\zeta$ denotes the marginal impact coefficient of $M_{it}$ on $\ln ({\mathrm{CT}}_{it})$; the rest of the symbols are consistent with Equation (1). $\gamma$ and ${\gamma }^{\prime }$ are the associated coefficient vectors in Equations (2) and (3). The benchmark regression has confirmed that both policies significantly promote SRIT container throughputs, thus the next step is to focus on the significance and sign change of $\xi$ and $\zeta$.

2.2.3. Moderating Effect Model

To further examine the moderating effect of IS in the policy impact mechanism, this study incorporates an interaction term into the benchmark MPDID model, following the approach of Guo et al. [37], and constructs the extended model as follows:

$$\ln ({\mathrm{CT}}_{it})=\alpha +{\beta }_{1}DI{D}_{it}+{\beta }_{2}I{S}_{it}+{\beta }_3(DI{D}_{it}\times I{S}_{it})+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(4)

where ${\beta }_1$ denotes the marginal effect of the policy indicator on $\ln ({\mathrm{CT}}_{it})$ at the baseline level of information sharing; ${\beta }_2$ denotes the marginal effect of information sharing on $\ln ({\mathrm{CT}}_{it})$ when the policy is not in effect; $I{S}_{it}$ represents the level of IS between the port and the railway in city $i$ at time $t$; The interaction term $DI{D}_{it}\times I{S}_{it}$ is the core moderating variable, used to assess whether IS enhances or weakens the impact of the policy on SRIT container throughput. If the coefficient ${\beta }_3$ of the interaction term is significantly positive, it indicates that IS plays a positive moderating role in policy implementation and helps amplify the policy effect.

2.3. Variables Selection

2.3.1. Dependent Variable

This study adopts container throughput (CT) as the dependent variable to reflect the actual operational performance of SRIT. The measurement unit is the Twenty-foot Equivalent Unit (TEU) [38]. In this study, the natural logarithm of the throughput variable is taken to mitigate heteroscedasticity and improve the distributional characteristics of the data, bringing it closer to a normal distribution. This approach enhances the robustness and explanatory power of the parameter estimates [39,40].

2.3.2. Independent Variables

The independent variable in this study is whether a given site implemented the RFPS policy or the FTE policy in a given month. Since 2022, to promote the development of containerized SRIT, multiple provincial and municipal governments in China have successively introduced RFPS policies, offering financial support for containers transported via SRIT [19]. In September 2022, Guangdong Province took the lead by launching a pilot subsidy program covering several service routes at Guangzhou Port. Beginning in February 2023, Hunan, Guangxi, and Sichuan followed suit by implementing their own local subsidy schemes.

In addition, ports and railway departments across various regions have jointly advanced the layout and service optimization of freight train networks. By launching new routes and increasing dispatch frequencies, they have gradually improved the SRIT train system [4]. Between 2022 and 2024, the FTE policy was continuously implemented, and the capacity of freight train services steadily improved. In August 2022, Jiangmenbei Station in Guangdong Province took the lead by implementing a train expansion plan, establishing an inland port and launching dedicated freight services, marking the beginning of service extension toward inland regions. Subsequently, stations such as Nanxiong, Changshabei, and Dulaying were connected through newly opened or extended freight lines, gradually forming a multi-node intermodal transport network centered on the Pearl River Delta and extending to the southwestern and central regions.

This study treats the initiation dates of the RFPS and FTE policies as the starting points of policy intervention, and constructs site-month level dummy variables to indicate their implementation status. The timing of policy implementation is determined based on official announcements from local governments, information released by port operators, and publicly available sources. Using the change in implementation status at each site as the basis for identification, we model the two policy variables separately. Observations before the implementation are defined as the control period, while those after implementation serve as the treatment period, allowing us to estimate the impact of each policy on container throughput.

2.3.3. Control Variables

To improve the accuracy and robustness of policy effect estimation, this study introduces four control variables: total import and export volume (TIE), industrial added value (IAV), permanent population of the destination city (POP), and local fiscal revenue (LFR). These variables reflect key dimensions such as trade intensity, industrial structure, population size, and fiscal capacity. They are used to effectively control for structural differences in economic fundamentals and logistics demand across regions, thereby allowing for a clearer identification of the net effects of policy interventions on freight volume [41].

2.3.4. Moderating Variables

This study treats IS as a moderating variable and examines its role in the relationship between policy implementation and container throughput. IS measures the breadth and quality of data exchange between ports and railways in SRIT operations. It reflects the level of coordination between the two parties in transport information, operation planning, and resource allocation. This study collected IS data by distributing questionnaires to relevant personnel at Guangzhou Port. To ensure the reliability of the scale, the IS measurement items were adopted from the study by Li and Lin [22]. We assessed all items using a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The detailed items of the scale are listed in Appendix A,

Table A1. Information Sharing Scale.

Construct	Items	References
Information sharing	1. We give the rail site advance notice of the shipper’s changing transportation needs.	[22]
	2. The rail site shares with us proprietary information on freight trains.
	3. The rail site shares with us the core business process of sea-rail transportation.

. We conducted a field survey at Guangzhou Port in China. First, the original English scale was translated into Chinese. The translation process followed the procedures established in previous studies [42]. After that, five experts reviewed the translated scale to ensure its accuracy, clarity, and comprehensibility in language. Finally, five experts in the field of transportation were invited to evaluate the scale. They revised items that were vague or inaccurate in expression, and finalized a version that could effectively reflect IS in SRIT. Based on this scale, we designed a questionnaire covering the period from June 2022 to June 2024. The survey focused on IS between Nansha Port in Guangzhou and 22 inland railway stations. Through coordination with the responsible officials at Guangzhou Port, we invited 10 information management personnel who were directly involved in SRIT operations to participate in the survey. Each of them independently rated the level of IS for each site in each observed month, resulting in a total of 550 rating records. For items with inconsistent ratings, the ten raters discussed and resolved differences through consultation. External experts were invited to arbitrate when necessary, ensuring consistency and credibility of the final results. The monthly IS level for each site was calculated as the average score of the three scale items [22].

The definitions and descriptions of all variables used in this study are summarized in

Table 1. Variable definition.

Variable Type	Variable	Definition
Dependent Variable	CT	Monthly volume of containers (in thousand TEU) transported from Nansha Port South Station to each inland destination; log-transformed.
Independent Variables	RFPS	Dummy variable indicating whether a RFPS policy is implemented at the station in a given month (1 = subsidized; 0 = not subsidized).
	FTE	Dummy variable indicating whether the station is affected by the FTE policy in a given month (1 = affected; 0 = not affected).
Control Variables	TIE	Total monthly import and export volume (in hundred million RMB) of the destination city, used as a proxy for regional economic activity; log-transformed.
	IAV	Monthly industrial added value (in hundred million RMB) of the destination city, capturing the scale of manufacturing output and industrial freight demand; log-transformed.
	POP	Total permanent population (in person) of the destination city, reflecting regional consumption potential and logistics demand; log-transformed.
	LFR	Total monthly local fiscal revenue (in hundred million RMB) of the destination city, reflecting the local government’s capacity to support infrastructure and freight policy implementation; log-transformed.
Mediation Variable	NPE	Total number of independent market participants involved in SRIT, serving as a measure of the extent of diversified shipper participation; log-transformed.
Moderator Variable	IS	Information sharing score. Measures the degree of information sharing between ports and railways in SRIT; log-transformed.

.

2.4. Data and Description

This study used SRIT waybill data from Guangzhou Nansha Port and the national 95306 railway waybill system for June 2022–June 2024. The data were organized at the station–month level and recorded shipments from Nansha Port South Station to inland cities. Each record included the destination station, month, container throughput, and policy exposure. Policy information for RFPS and FTE was compiled from Port Authority circulars, local government subsidy announcements, and operators’ tariff notices and service schedules. Destination-city covariates (trade intensity, route distance, rail freight price, transit time, truck rate) were assembled from the sources listed in the variables subsection. Throughput was measured in twenty-foot equivalent units (TEUs); prices were in CNY per TEU; distances were in kilometers; transit time was in calendar days.

We constructed a balanced station–month panel by linking the port production system to 95306 using encrypted container identifiers. The linkage yielded monthly origin–destination flows, from which we computed SRIT container throughput and extracted consignor/consignee information to assemble participation measures at the destination station. For each destination station and month, we coded RFPS and FTE indicators according to the documented effective month; under staggered adoption the indicator took value 1 from the start month onward. We matched destination-city controls, removed exact duplicates, enforced consistent identifiers and data types, and addressed missing values in continuous variables via within-station linear interpolation with forward and backward imputation to preserve time-series continuity and minimize sample loss [43]. To mitigate distributional skew, harmonize scales, and facilitate elasticity interpretation, we applied natural-log transformations to all continuous control variables [44]. The information-sharing (IS) moderator was derived from field surveys and structured focus-group discussions with port and railway stakeholders and was aggregated to the station level before monthly alignment. Finally, we conducted quality checks by verifying policy effective dates across independent sources, screening for impossible values, and confirming temporal alignment between administrative documents and observed flow changes. The resulting panel was used in the multi-period DID and event-study analyses; descriptive statistics are reported in

Table 2. Descriptive Statistics.

Variables	Obs	Mean	SD	Min	Median	Max
CT	550	0.331	0.692	0.002	0.077	4.434
RFPS	550	0.782	0.413	0.000	1.000	1.000
FTE	550	0.316	0.465	0.000	0.000	1.000
TIE	550	3.983	1.575	0.182	4.034	7.122
IAV	550	6.936	0.689	4.554	6.981	7.695
POP	550	8.638	0.621	6.640	8.714	9.554
LFR	550	8.457	0.612	6.641	8.624	9.166
NPE	550	2.296	0.390	1.006	2.302	3.844
IS	550	3.848	0.387	3.010	3.924	4.380

Notes: Obs: Observations, SD: Standard Deviation.

.

3. Results

3.1. Baseline Regression Results

The regression results for railway freight price subsidy (RFPS) and freight train expansion (FTE) are reported in

Table 3. Benchmark Regression Results.

Variables	RFPS		FTE
	(1)	(2)	(3)	(4)
	lnCT	lnCT	lnCT	lnCT
RFPS	0.228 ***	0.230 ***
	(−3.76)	(−3.85)
FTE			0.142 *	0.106 **
			(−1.75)	(−2.22)
Controls	No	Yes	No	Yes
Constant	0.203 ***	7.938 ***	−2.394 ***	11.752 ***
	(−14.76)	(−4.44)	(−25.81)	(−3.69)
City F.E.	Yes	Yes	Yes	Yes
Month F.E.	Yes	Yes	Yes	Yes
Group Number	22	22	22	22
Observations	550	550	550	550
R-squared	0.809	0.812	0.630	0.684

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. Under the MPDID specification, we estimate models without and with controls: Columns (1) and (3) present baseline estimates, and Columns (2) and (4) add controls. Adding controls improves model fit without altering signs or inference [8]. Given the log specification of the outcome, coefficients map to percentage level differences relative to the station’s pre-policy reference month. In Column (2), the RFPS coefficient is 0.230 (p < 0.01), implying about 25.9%1 higher post-policy monthly throughput relative to the station’s pre-policy reference month, supporting H1a. In Column (4), the FTE coefficient is 0.106 (p < 0.05), implying around 11.2% higher post-policy monthly throughput relative to the station’s pre-policy reference month, supporting H1b. Both instruments significantly raise SRIT container throughput in the study sample, with RFPS exhibiting a larger effect than FTE.

3.2. Robustness Analysis

The benchmark regression results show that both the RFPS policy and the FTE policy can significantly increase the container throughput of SRIT. To ensure the robustness of the empirical findings, this study conducts a series of additional tests to address potential biases in the DID identification strategy [45].

3.2.1. Parallel Trend Test

In empirical applications of MPDID models, testing the parallel trend assumption is a prerequisite for credible causal inference [46]. If the treated and untreated units exhibit divergent trends before intervention, the DID estimates may be severely biased [47].

In this study, the SRIT policy was implemented across nearly all railway stations, resulting in the absence of a clearly defined untreated group. We therefore adopt a self-controlled event-study design [35], in which each treated unit serves as its own temporal control, and the treatment effect is identified relative to its pre-intervention period [48]. This within-unit comparison framework allows us to evaluate policy impacts in a statistically valid manner and serves as a viable alternative to traditional cross-sectional control group designs.

Specifically, this study constructs a series of event-time dummy variables, taking the month before the policy implementation (pre_1) as the reference period. The variables pre_x and post_y represent the x-th month before and the y-th month after the policy implementation, respectively. This setting allows us to estimate the dynamic effects of each time point relative to the baseline period. To test the parallel trend assumption, we examine the coefficients of the pre-treatment periods. If these coefficients are not statistically significant, it indicates that the treatment group did not exhibit significant trend changes prior to the intervention, thereby supporting the validity of the parallel trend assumption between the treatment and control groups before the policy was implemented.

The regression specification is as follows:

$$\ln \left(C{T}_{it}\right)={\displaystyle \sum _{k\ne -1}{\theta }_k}{D}_k+X_{it}\gamma +{\lambda }_i+{\delta }_t+{\epsilon }_{it}$$

(5)

where $D_k$ is a dummy variable for event time $k$, ${\theta }_k$ measures the treatment effect at each relative time point. This specification allows us to visually inspect pre-trends and observe post-treatment dynamics over time. All other symbols are consistent with those used in the benchmark regression.

Figure 2 presents the parallel trend test results for the RFPS policy (a) and the FTE policy (b). The red dashed line marks the policy implementation month (event time = 0), separating the pre- and post-treatment periods. In both panels, the coefficients of the pre-treatment periods (pre_x) are statistically insignificant and their confidence intervals include zero, indicating that the container throughput trajectories between the treatment and control groups were similar prior to policy implementation. This supports the validity of the parallel trend assumption. After implementation, most post-treatment coefficients are positive and statistically significant, suggesting that both policies increased SRIT container throughput. RFPS shows an effect that appears at impact and remains significant through post_1–post_3. FTE displays an immediate effect in the current period, weakens at post_1 while remaining significant, and becomes statistically insignificant by post_2–post_3. Exact coefficients and significance levels are reported in

Table 4. Event Study Regression Results.

Variables	RFPS	FTE
	(1)	(2)
	lnCT	lnCT
pre_5	0.01
	(−1.19)
pre_4	−0.002	−0.233
	(−0.19)	(−1.62)
pre_3	0.014	−0.065
	(−1.59)	(−0.47)
pre_2	0.005	−0.114
	(−0.69)	(−0.93)
current	0.013 *	0.326 ***
	(−1.75)	(−2.66)
post_1	0.036 ***	0.281 **
	(−4.99)	(−2.31)
post_2	0.033 ***	0.116
	(−4.60)	(−0.95)
post_3	0.042 ***	0.017
	(−6.30)	(−0.14)
post_4	−0.004
	(−0.71)
Controls	YES	YES
City F.E.	YES	YES
Month F.E.	YES	YES
Observations	550	550
R-squared	0.997	0.900

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. These patterns confirm that both policies significantly raise SRIT container throughput, consistent with H1a and H1b.

3.2.2. Placebo Test

This study adopts a placebo test as a counterfactual analysis tool for the RFPS and FTE policies, aiming to rule out the influence of sample selection bias and unobserved confounding factors. This method ensures that the observed effects are attributable to the policy itself rather than model misspecification or random factors [49]. For the RFPS policy, we randomly disrupt the treatment and control groups, set fictitious policy intervention dates, and construct placebo policy variables. This process is repeated 500 times to generate the kernel density distribution of the estimated coefficients [50]. Figure 3 presents that the horizontal red dashed line marks the significance threshold of the p-value, distinguishing significant from non-significant placebo estimates. The placebo coefficients are centered around zero, whereas the actual RFPS estimate (grey dashed line) lies well outside this distribution. This indicates that the observed policy effect is unlikely to be driven by random noise or omitted heterogeneity, validating the causal inference that underlies our answer to RQ1.

To examine whether the estimated effect of the FTE policy is influenced by other potential factors, this study advances the actual implementation date by one and two periods, respectively, and constructs two corresponding lead dummy variables. The regression model is re-estimated under the same fixed effects structure [8]. This test is designed to assess whether container throughput had already changed significantly before the policy officially took effect, thereby identifying whether the estimated policy effect may be driven by pre-existing trends or external shocks. The corresponding regression model is specified as follows:

$$\ln \left(C{T}_{it}\right)=\alpha +\beta DID\text{\_}Lea{d}_{it}^{\left(l\right)}+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(6)

where $DID\text{\_}Lea{d}_{it}^{\left(l\right)}$ denotes the advance $l$-period-constructed dummy treatment variable, which in this study is set to $l$ = 1 and 2, corresponding to $DID\text{\_}Lea{d}_{it}^{(1)}$ and $DID\text{\_}Lea{d}_{it}^{(2)}$, respectively. All other symbols are consistent with those used in the benchmark regression.

As shown in

Table 5. Lead-Time Placebo Test Results for the FTE Policy.

Variables	Lead 1 Period (Placebo)	Lead 2 Period (Placebo)
	(1)	(2)
	lnCT	lnCT
FTE_Lead1	0.158
	(1.60)
FTE_Lead2		0.270
		(1.57)
Controls	Yes	Yes
City F.E.	Yes	Yes
Month F.E.	Yes	Yes
Group Number	22	22
Observations	550	550
R-squared	0.684	0.686

Notes: t-statistics in parentheses.

, the estimated coefficients are statistically insignificant in both the one-period lead model in Column (1) and the two-period lead model in Column (2). This indicates that container throughput did not experience significant changes prior to the official implementation of the FTE policy. These findings suggest that the policy effect began to materialize only after its formal introduction, thereby ruling out the influence of market anticipation or policy endogeneity. This result further confirms the robustness of the benchmark regression estimates.

3.2.3. Lagged Dependent Variable Test

To avoid the problem of endogeneity due to bidirectional causation and to explore the possible time lag in the effect of the independent variables on the dependent variable, this study applies both first- and second-order lags to all explanatory variables [51]. The regression model corresponds to the following mathematical expression:

$$\ln \left(C{T}_{it}\right)={\alpha }_0+{\beta }_{l}DI{D}_{i,t-l}+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(7)

where $DI{D}_{i,t-l}$ denotes the policy treatment dummy variable for city $i$ in period $t-l$, with l = 1 or 2 representing one- and two-period lags, respectively. All other symbols are defined as in the benchmark regression.

Table 6. Lagged impact assessment for policy effect persistence.

Variables	RFPS		FTE
	1-Period Lagged	2-Period Lagged	1-Period Lagged	2-Period Lagged
	(1)	(2)	(3)	(4)
	lnCT	lnCT	lnCT	lnCT
RFPS	0.235 ***	0.347 ***
	(−3.70)	(−6.75)
FTE			0.096 *	0.138 ***
			(−1.97)	(−2.95)
Controls	Yes	Yes	Yes	Yes
Constant	6.486 ***	7.452 ***	38.523 ***	40.108 ***
	(−3.45)	(−3.57)	(−3.01)	(−3.21)
City F.E.	Yes	Yes	Yes	Yes
Month F.E.	Yes	Yes	Yes	Yes
Group Number	22	22	22	22
Observations	528	506	528	506
R-squared	0.813	0.826	0.690	0.696

Notes: t-statistics in parentheses; *** p < 0.01, * p < 0.1.

illustrates the lagged specifications. In Column (1) RFPS, 1-period lag model and Column (2) RFPS, 2-period lag model, the coefficients are 0.235 and 0.347 (both p < 0.01), corresponding to approximately 26.5% and 41.5% higher monthly throughput relative to the pre-policy reference month. In Column (3) FTE, 1-period lag model and Column (4) FTE, 2-period lag model, the coefficients are 0.096 (p < 0.10) and 0.138 (p < 0.01), corresponding to about 10.1% and 14.8% increases. These results confirm robustness, indicate stronger persistence for subsidies than for service expansion, with FTE showing only modest carryover at one to two months. The results are consistent with H1a/H1b regarding policy effectiveness.

3.2.4. Excluding COVID Sample

The COVID-19 pandemic from 2019 to 2022 is widely regarded as an exogenous event that significantly impacted the macroeconomy. As the sample in this study includes the pandemic year (2022), failure to control for its effect may lead to endogeneity bias caused by omitted variables [52]. To address this concern, we exclude the 2022 observations and re-estimate the model as a sample sensitivity test.

Table 7. Robustness regression results for samples excluding epidemic years.

Variables	RFPS	FTE
	(1)	(2)
	lnCT	lnCT
RFPS	0.166 ***
	(−2.98)
FTE		0.074 **
		(−2.11)
Constant	9.200 **	−1.592 ***
	(−2.27)	(−4.20)
Controls	Yes	Yes
City F.E.	Yes	Yes
Month F.E.	Yes	Yes
Group Number	22	22
Observations	396	396
R-squared	0.850	0.664

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05.

shows that the RFPS coefficient is 0.166 (p < 0.01), corresponding to roughly an 18.1% increase in monthly throughput relative to the station’s pre-policy reference month, and the FTE coefficient is 0.074 (p < 0.05), corresponding to roughly an 7.7% increase. Both effects remain statistically significant and exhibit same effect directions as the benchmark results, indicating that the findings are not driven by the pandemic year. These findings further support the robustness of H1a and H1b.

3.3. Heterogeneity Analysis

To examine the heterogeneous effects of transportation distance, this study follows the approach of Credit and Lehnert [53] by using the sample median of 1128 km as the threshold to divide the shipments into two groups: “long-distance” and “short-distance”. Separate regressions are then conducted for the two groups to identify whether the policy impacts on container throughput differ by transportation distance.

Specifically, short-distance transport in our sample primarily encompasses intra-regional movements within a 1–2-day transit time, characterized by direct rail services, relatively low transshipment requirements, and often characterized by frequent schedules and fast turnaround times, offering time-sensitive advantages over road transport. These routes typically connect ports with their immediate economic hinterlands, such as Guangzhou-Changsha (885 km), where daily service frequency and quick turnaround are feasible. Long-distance transport involves routes requiring 3–5 days transit time, multiple railway administrations, and complex coordination across provincial boundaries. Representative routes include Guangzhou-Guiyang (1698 km), where service consolidation and network effects become critical for operational viability. The grouping indicator variable is defined as:

$$G_i=\left\{\begin{array}{ll}1,\hfill & {\mathrm{if} \mathrm{distance}}_i>\mathrm{median}(\mathrm{distance})\hfill \\ 0,\hfill & {\mathrm{if} \mathrm{distance}}_i\le \mathrm{median}\left(\mathrm{distance}\right)\hfill \end{array}\right.$$

The heterogeneity test model is:

$$\ln (C{T}_{it})=\alpha +{\beta }_{1}DI{D}_{it}{G}_i+{\beta }_{2}DI{D}_{it}(1-G_i)+X_{it}\gamma +{\lambda }_i+{\delta }_t+{ϵ}_{it}$$

(8)

where $DI{D}_{it}{G}_i$ denotes the treatment effect under the long-distance grouping ($G_i=1$), and $DI{D}_{it}(1-G_i)$ denotes the treatment effect under the short-distance grouping ($G_i=0$), which allows for a comparison between ${\beta }_1$ and ${\beta }_2$, and a direct observation of the difference in policy effects between the two groups. All other symbols remain consistent with those defined in the benchmark regression model.

Table 8. Heterogeneous effects results.

Variables	RFPS		FTE
	Long Distance	Short Distance	Long Distance	Short Distance
	(1)	(2)	(3)	(4)
	lnCT	lnCT	lnCT	lnCT
RFPS	−0.069 ***	0.143 ***
	(−22.79)	(−3.35)
FTE			−0.112	0.144 ***
			(−0.71)	(−5.15)
Control variables	Yes	Yes	Yes	Yes
Constant	4.094 ***	1.259 **	27.597 ***	−1.267
	(−3.95)	(−2.32)	(−3.50)	(−0.61)
City F.E.	Yes	Yes	Yes	Yes
Month F.E.	Yes	Yes	Yes	Yes
Group Number	11	11	11	11
Observations	275	275	275	275
R-squared	0.349	0.815	0.367	0.838

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05.

reports the regression results of RFPS and FTE under different transportation distances. In the short-distance subsample, the RFPS coefficient is 0.143 (p < 0.01), corresponding to an increase of approximately 15.4% in monthly throughput, while the effect turns significantly negative in the long-distance group (${\beta }_1$ = −0.069, p < 0.01), corresponding to a decrease of approximately 6.7% in monthly throughput, indicating that this policy is more effective in promoting short-distance transport. In contrast, the FTE policy also exhibits a significant positive impact in the short-distance group (${\beta }_2$ = 0.144, p < 0.01), which corresponding to approximately 15.5%, but shows a negative and insignificant coefficient in the long-distance group (${\beta }_1$ = −0.112). These results suggest that both policies present clear spatial heterogeneity across transportation distances.

Operationally, these magnitudes imply that both instruments deliver sizable gains on short corridors (around 15%), warranting priority in train-plan allocation, yard capacity, and customer engagement, whereas long-haul corridors do not benefit reliably without complementary improvements in schedule reliability, transfer efficiency, and information coordination. Policymakers should therefore target incentives and service commitments on short-distance networks and pair long-distance subsidies with explicit reliability and visibility requirements. These heterogeneous effects directly address RQ2 on spatial heterogeneity and, taken together with the baseline results, are consistent with H1a and H1b by confirming statistically significant improvements in SRIT throughput overall and clarifying the distance ranges where policy impacts are most pronounced.

3.4. Mediation Analysis

The results of the mediation analysis are presented in

Table 9. Results of mediation mechanisms.

Variables	RFPS		FTE
	(1)	(2)	(3)	(4)
	lnNPE	lnCT	lnNPE	lnCT
RFPS	0.207 **	0.256 ***
	(2.54)	(3.91)
FTE			0.219 **	0.123 **
			(2.16)	(2.26)
lnNPE		0.137 **		0.131 **
		(2.28)		(2.31)
Controls	Yes	Yes	Yes	Yes
City F.E.	Yes	Yes	Yes	Yes
Month F.E.	Yes	Yes	Yes	Yes
Observations	550	550	550	550
R-squared	0.714	0.945	0.786	0.952

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05.

. Columns (1) and (3) show that RFPS and FTE both have statistically significant effects on lnNPE, indicating that both policies effectively expand NPE. Columns (2) and (4) reveal that NPE significantly increases SRIT container throughput. In addition, compared with the results in Table 3, the absolute values of the policy coefficients increase, suggesting that RFPS and FTE enhance container throughput by increasing the number of shippers. This confirms the mediating role of market participation and supports H2a and H2b.

From an implementation perspective, these magnitudes indicate that expanding the shipper base is a material transmission channel through which both RFPS and FTE raise SRIT volumes. For operators, this supports prioritizing targeted shipper acquisition, onboarding programs with simplified booking and documentation, and partnerships with consolidators to scale multi-shipper flows. For policymakers, incentive design should reward participation growth (e.g., performance clauses tied to the number of active shippers) and couple financial support with measures that reduce entry frictions.

3.5. Moderation Analysis

Building on the confirmed positive effects of the policies on SRIT container throughput, this study further introduces the variable of IS and includes interaction terms to test its moderating mechanism.

Table 10. Results of moderation mechanisms.

Variables	RFPS	FTE
	(1)	(2)
	lnCT	lnCT
RFPS	0.198 ***
	(−3.12)
FTE		0.092 **
		(−2.01)
RFPS × IS	0.087 *
	(−1.89)
FTE × IS		0.135 **
		(−2.47)
Controls	Yes	Yes
Constant	7.902 ***	11.610 ***
	(−4.41)	(−3.56)
City F.E.	Yes	Yes
Month F.E.	Yes	Yes
Observations	550	550
R-squared	0.824	0.702

Notes: t-statistics in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

presents the regression results. Column (1) shows that the regression coefficient of the main effect for RFPS is 0.198 and significantly positive (p < 0.01). The coefficient of the interaction term RFPS × IS is 0.087 and statistically significant at the 10% level, implying that a one-unit increase in the IS index strengthens the RFPS effect by approximately 9.1% on monthly throughput, supporting H3a. In Column (2), the regression coefficient of the main effect for FTE is 0.092 (p < 0.05), and the interaction term FTE × IS has a coefficient of 0.135 (p < 0.05), indicating that a one-unit increase in IS enhances the FTE effect by approximately 14.5%, supporting H3b. These findings directly address RQ3 by identifying IS as a key boundary condition that amplifies policy effectiveness, with regions possessing higher information-sharing capabilities realizing substantially greater benefits from both subsidy and infrastructure interventions.

4. Discussion

4.1. Main Findings

This study constructs a dual-policy identification framework to identify the independent effects, transmission mechanisms, and duration differences of two types of incentive policies, as well as their spatial heterogeneity with respect to transport distance. The results show that both railway freight price subsidies (RFPS) and freight train expansion (FTE) significantly increase container throughput. Further analysis reveals that RFPS has a stronger positive effect than FTE, suggesting that direct financial subsidies can more effectively stimulate transport demand by reducing transport costs compared to infrastructure or service expansion. This can be attributed to the fact that infrastructure-related policies are constrained by longer implementation cycles and more complex procedures, which limit their ability to be rapidly converted into transport benefits. This finding supports the argument of Liu and Jia [21] that price-based policies exert a stronger influence on promoting modal shift.

In addition, the impacts of RFPS and FTE policies on container throughput in the post-implementation time evolution show differentiated dynamic characteristics. The RFPS policy produced initial effects at the time of implementation and continued to strengthen in the subsequent period. Although the effect declined slightly in the later stage of observation, it generally showed a trend of gradual enhancement over time. This outcome may result from the delayed response of shippers in receiving subsidy information, assessing cost–benefit relations, and adjusting transport decisions, especially when the transmission of information is inefficient or when coordination among multiple parties is required [54]. In contrast, the FTE policy produced a significant positive effect in the early stage of implementation, but the effect diminished rapidly and became statistically insignificant. This result indicates that operational policies have stronger short-term responsiveness but lack sustained driving force in the long term. One possible explanation is that the IS mechanism remains underdeveloped, or the regional freight foundation is relatively weak, which prevents newly added services from continuously translating into growth in container throughput [55]. These findings highlight the necessity of incorporating a temporal dimension into policy evaluation, as different types of policies exhibit significant differences in the sustainability of their incentive effects [56].

This study further finds that the RFPS and FTE policies exhibit significant spatial heterogeneity under different transport distance conditions. The RFPS policy shows a clear positive incentive effect in short-distance transport, effectively increasing container throughput. However, in long-distance transport, its incentive effect turns negative. This may be attributed to the fact that subsidies substantially reduce unit transport costs in short-distance scenarios, thereby enhancing the price competitiveness of SRIT. In contrast, long-distance transport is constrained by limited-service capacity, efficiency bottlenecks, and differences in shipper preferences, which weaken the sustainability of the incentive effect. The FTE policy also demonstrates a significant positive effect in short-distance transportation scenarios, indicating that this policy more easily stimulates freight demand in such regions. The high frequency of freight trains and strong cargo agglomeration facilitate the rapid release of policy incentives and lead to improved actual loading rates. In comparison, the effect of this policy in long-distance transport is not significant and even shows a negative sign. This may be due to the complexity of the transport network, higher difficulty in node coordination, and frequent delays in transshipment and train marshalling, which reduce shippers’ willingness to adopt railway options [57]. This finding is supported by Zhang and Zhong [58], who observed that the incentive effect of transport policies is more pronounced over short distances.

Mechanism analysis shows that both RFPS and FTE can increase container throughput by expanding the number of independent market participants involved in SRIT. RFPS reduces the entry barriers for small and medium-sized shippers, thereby generating a scale effect [59], while FTE attracts more shippers by improving service accessibility and operational efficiency. The expansion of market participants enhances market activity and resource matching efficiency, which in turn promotes container throughput [60]. We also find that IS strengthens the impact of policies on container throughput and plays a moderating role in their effectiveness. This finding is consistent with prior supply chain research that views information sharing (IS) as a core mechanism for improving cross-organizational coordination and reducing demand and capacity uncertainty [22]. High-level IS between ports and railways helps coordinate capacity in advance and reduces the capacity pressure caused by RFPS [33]. In addition, IS promotes coordination between ports and railways in terms of service expansion, timetable adjustment, and loading and unloading plans, which effectively mitigates the scheduling uncertainty and resource mismatch problems caused by FTE [61]. This helps improve the efficiency of SRIT freight, reduce empty load rates, and enhance shippers’ willingness to adopt SRIT, thereby further strengthening the positive effects of the policies on container throughput. However, these mechanisms face practical barriers including long-term road transport contracts that create switching costs, limited rail terminal access in industrial zones, and technical incompatibilities between port and railway IT systems that impede real-time IS [62,63]. These constraints suggest that realizing the full potential of identified mechanisms requires complementary interventions addressing infrastructure and institutional coordination.

Our findings both align with and diverge from international patterns. European experiences demonstrate that combined subsidy-infrastructure approaches achieve higher modal shares, though in markedly different operational contexts [10]. Their emphasis on direct financial incentives aligns with our finding that RFPS generates stronger and more sustained effects than infrastructure expansion. The U.S. focus on infrastructure-led development through double-stack technology required decades to materialize benefits [64]. By contrast, the FTE policy yields immediate but short-lived increases, potentially reflecting our limited observation window relative to the multi-year maturation typical of corridor and terminal investments. Overall, these international contrasts highlight that while the principle of price incentives dominating service expansion may be generalizable, specific implementation strategies must account for local institutional and market conditions.

4.2. Theoretical Implications

These findings reinforce instrument-choice theory in transport policy by showing that price-based instruments dominate service-based adjustments in both magnitude and persistence, a pattern consistent with generalized logistics-cost models and dynamic treatment effects documented in recent SRIT and freight modal-shift studies. First, this study is the first to identify and compare the effects of economic policy instruments (i.e., RFPS) and infrastructure-oriented policy instruments (i.e., FTE) on the development of SRIT, thereby advancing the theoretical understanding of incentive policy effects. Unlike existing studies that predominantly treat SRIT policies as homogeneous interventions and focus on their macro-level benefits for environmental improvement or energy efficiency, we shift the analytical focus to the developmental impacts of specific incentive policy tools on SRIT. By constructing a dual-policy identification framework, this study uncovers significant differences between two types of incentive policies in terms of effect magnitude, persistence, and spatiotemporal dynamics. This integrated analysis that combines policy type, temporal dimension, and spatial variation not only expands policy evaluation from the macro-level question of “whether it is effective” to a deeper inquiry into “how it becomes effective” and “under what conditions it is most effective,” but also provides a robust theoretical and methodological foundation for incorporating policy heterogeneity into transport policy evaluation frameworks.

Second, this study deepens the understanding of the internal mechanisms through which SRIT incentive policies promote container throughput by identifying the mediating variable (NPE) and the moderating variable (IS). Although existing studies have emphasized the role of shippers, few have explored how policies drive freight growth through behavioral adjustments of market participants [13]. This study finds that NPE plays a significant mediating role in the policy impact, and IS can enhance the effect of policies on container throughput. These findings not only advance the understanding of how shipper behavior influences incentive policy effectiveness in the context of SRIT, but also further establish IS as a critical boundary condition, identifying the circumstances under which policy impacts can be maximized.

Finally, this study is the first to investigate the differentiated effects of SRIT incentive policies from the perspective of spatiotemporal heterogeneity, providing new insights into an aspect that has received limited attention in the existing literature. Most prior studies rely on annual or quarterly data and adopt static analytical frameworks, which limit their ability to capture short-term policy responses and heterogeneity across transport distances [65]. To overcome this limitation, we utilize high-frequency, waybill-level monthly data to reveal the dynamic evolution of the impacts of two incentive policies (i.e., RFPS and FTE) on container throughput in SRIT, as well as their distance sensitivity. Our study extends the understanding of policy effectiveness by revealing that uniform incentive policies may not produce consistent impacts across different spatial scales. We also transcend the theoretical limitations of conventional static policy evaluations, thereby enriching the theoretical discourse on spatiotemporal heterogeneity in SRIT policy research.

4.3. Practical Implications

Our empirical findings provide guidance for resource allocation and operational decisions. RFPS shows a more significant effect in stimulating transport demand, indicating that in situations with limited resources and short-term goals, fiscal incentive policies can more directly reduce shipper costs and rapidly release transport demand. Therefore, policymakers should give priority to subsidy policies and increase relevant investment, making such policies a core tool for promoting the optimization of the transport structure. In contrast, FTE should serve as a strategic supplement to enhance long-term transport capacity and lay a solid foundation for future development.

The spatiotemporal heterogeneity of policy effects demands differentiated implementation strategies. RFPS shows a lagged but persistent effect, suggesting that policymakers should maintain longer-term subsidy commitments to realize full benefits. To accelerate policy transmission, authorities should establish unified policy communication platforms and conduct targeted industry engagement to ensure timely dissemination of policy information. Meanwhile, given the immediate but short-lived effects of FTE, railway operators should prioritize short-distance corridor improvements to sustain initial gains. For corridors where both policies demonstrate stronger effects, policymakers should concentrate subsidies and increase train frequency to maximize impact. Conversely, long-distance routes require that structural barriers such as service reliability, transshipment efficiency, and interprovincial coordination be addressed before financial incentives are deployed.

In addition, our mechanism analysis provides specific operational guidance. The mediating role of market participation indicates that the number of participating entities is a critical channel linking policy implementation to SRIT throughput, underscoring the need to place shipper demand at the center of policy design. RFPS enhances throughput primarily by expanding participation through price reductions, indicating high price sensitivity among shippers; governments and railway operators should continuously optimize the cost structure to sustain the viability of subsidy programs while fostering a more open and competitive market environment. FTE promotes throughput growth by improving service quality and accessibility, which implies that operators should improve train punctuality, broaden service coverage, and further optimize intermodal connection systems to attract and retain a larger shipper base. To lower entry barriers for small and medium-sized shippers, establishing consolidation centers at key nodes and developing unified booking systems is recommended. Moreover, the significant moderating effect of information sharing justifies investment in digital platforms for port–rail coordination, together with enhanced system integration, data interoperability, and real-time information exchange, enabling better capacity planning and resource allocation.

4.4. Limitations and Future Research

While this study provides valuable insights into the differential effects of SRIT incentive policies, several limitations warrant acknowledgment and point toward promising avenues for future research.

First, while this study measures policy intervention effects through container throughput changes and reveals significant spatial heterogeneity arising from transport distances, transport costs, as a critical factor influencing shippers’ mode choice decisions, are also expected to affect the actual effectiveness of SRIT policies. Future research could develop comprehensive cost accounting frameworks to quantify the relationship between unit transport costs and specific distance brackets deepening the throughput-based analysis with detailed cost–benefit assessments. Such assessments would help policymakers determine the optimal allocation between subsidy inputs and infrastructure investments across different transport distances.

Second, the empirical evidence derived from a representative Chinese port provides robust insights under a state-led transport system, offering valuable implications for developing countries with similar institutional contexts. However, the applicability of these findings remains uncertain for countries with different governance structures or market-oriented logistics systems. Future research could incorporate cross-country and multi-port comparative analyses to assess the consistency of our findings under different governance and regulatory environments.

The rapid development of artificial intelligence and related digital technologies may exert significant influence on the evolution of SRIT. Future research could investigate how SRIT development interacts with broader trends such as logistics digitalization, tariff integration, and the application of new monitoring and big data technologies. These factors are likely to reshape information transparency, coordination efficiency, and cost structures within multimodal transport chains, thereby affecting the long-term effectiveness and scalability of SRIT policies. Incorporating these dimensions into future evaluations would contribute to a more comprehensive understanding of policy impacts under evolving technological and institutional contexts.

5. Conclusions

This study advances the understanding of SRIT policy effectiveness by constructing a dual-policy identification framework that systematically compares economic instruments (RFPS) and infrastructure-oriented instruments (FTE). Using high-frequency waybill-level data from a major Chinese port, we reveal that RFPS achieves stronger and more persistent effects on container throughput compared to FTE, with both policies showing greater effectiveness in short-distance transport while exhibiting diminished returns in long-distance scenarios. The mechanism analysis identifies market participation expansion as a critical mediator and information sharing (IS) as a key moderator that amplifies policy impacts. This study moves beyond the conventional view of policies as homogeneous interventions by identifying the heterogeneous effects of different policy types, their spatiotemporal variations, and underlying mechanisms, thereby extending the depth and breadth of transport policy evaluation and overcoming the limitations of static assessment frameworks. These findings provide critical guidance for optimizing resource allocation in modal shift initiatives, suggesting that policymakers should prioritize distance-differentiated subsidy strategies for immediate impact while treating infrastructure expansion as a complementary long-term measure, particularly relevant for countries pursuing rail freight development under carbon neutrality goals. This study also has room for further development and improvement. Future research could extend throughput-based policy evaluation by integrating cost–benefit assessments, cross-country comparisons, and the impacts of digital technologies, thereby offering a more comprehensive understanding of SRIT policy effectiveness across diverse governance and institutional contexts.