Investigating the impact of spatial dependence and heterogeneity on airport relationships: Empirical evidence from China

The study investigates how spatial dependence and spatial heterogeneity shape inter-airport relationships in China. It addresses three questions: (1) whether spatial dependence between airports changes with geographic distance and how this alters competition versus complementarity; (2) how to construct indicators that capture spatial heterogeneity across airports; and (3) how relationships vary across airport hierarchies. The context is a complex Chinese airport system where proximity, traffic flows, ownership/management, partnerships, and regulation create competitive and cooperative dynamics, often embedded in hub-and-spoke structures. Understanding these mechanisms is important for accurate modeling of airport interactions and for informing infrastructure planning and policy.

Prior work has explored airport relationships through competition, complementarity, market structure, and regional development, employing methods such as discrete choice models and spatial econometrics. Studies indicate distance-dependent competition/cooperation effects and spillovers, with spatial heterogeneity influencing efficiency and performance. Spatial econometric models (SAR, SEM, SDM) capture spatial autocorrelation, spillovers, and heterogeneity. However, gaps remain regarding the mechanisms by which spatial dependence varies with distance and how heterogeneity across airport hierarchies affects relationships. Existing models often rely on a single weight matrix and static views, potentially masking dynamic and heterogeneous effects. This study advances the literature by explicitly modeling distance thresholds for spatial dependence, constructing an economic (asymmetric) distance matrix to capture heterogeneity, and examining relationships across airport hierarchies.

- Design: Spatial panel econometric analysis of 34 major Chinese airports (2007–2019), producing a balanced panel of 442 observations. Dependent variable is passenger throughput (PA); controls include GDP per capita (GDP), aviation employment (EM), foreign exchange tourism income (FTI), and crude oil price (CP). All variables are log-transformed. - Models: General spatial panel framework with spatially lagged dependent variable (δ), spatial lags of exogenous variables (ρ), and spatially correlated errors (λ). Special cases considered: SAR (δ only), SEM (λ only), SDM (δ and spatially lagged X). Endogeneity from spatial lag handled via quasi-maximum likelihood (QML) following Lee (2004), with individual fixed effects. - Empirical specification: ln(PA_it) = δ W ln(PA_it) + β1 ln(GDP_it) + β2 ln(EM_it) + β3 ln(FTI_it) + β4 ln(CP_it) + ρ1 W ln(GDP_it) + ρ2 W ln(EM_it) + ρ3 W ln(FTI_it) + μ_i + ε_it. - Spatial weights: 1) Geographic inverse-distance matrix W_g: w_ij = d_ij^(-1) (i ≠ j), using great-circle distances. 2) Asymmetric economic distance matrix W_e: base weights reflect similarity in average passenger traffic across airports, scaled by relative traffic to capture directional influence and hub impacts; produces W_e ≠ W_e^T. 3) Distance-threshold matrices: to assess attenuation, W is modified by setting w_ij = d_ij^(-2) for d_ij ≤ threshold d, and 0 otherwise; thresholds from 3600 km down to 100 km in 100 km steps. - Airport hierarchy construction: A comprehensive strength index is built using entropy-weight TOPSIS across multiple indicators (standardized; weights via entropy method), then airports are classified into three tiers: hub (index > 0.5), regional trunk (0.2–0.4), and local branch (< 0.2). Relationships are re-estimated within and across these groups using SDM with W_e. - Data sources: Passenger traffic from CAAC Airport Production Statistics; GDP per capita, aviation employment, and international tourism receipts from China Economic Net; crude oil prices as shipping cost proxy. Panel unit root testing (Harris–Tzavalis) rejects joint non-stationarity (z = −2.102, p = 0.018). - Model selection and diagnostics: Hausman tests favor fixed effects; LM and robust LM tests support inclusion of spatial lags; LR/Wald tests generally favor SDM over SAR/SEM. Goodness-of-fit assessed via R² and log-likelihood.

- Spatial clustering: Moran’s I indicates significant positive spatial correlation in passenger traffic across years, with stronger correlation under the economic distance matrix than geographic distance, implying heterogeneity matters for spatial interaction. - Overall geographic-weight results (W_g/W_a): In full-sample SDM/SAR, δ > 0 and significant (e.g., SDM δ ≈ 0.643), indicating complementarity: a 1% increase in neighboring airports’ average traffic is associated with a ~0.643% increase in an airport’s traffic. Residuals show spatial autocorrelation (SEM λ significant), signaling unobserved spatially patterned factors. - Distance attenuation: Using thresholded distance matrices, δ decreases as the threshold distance shrinks. For 1500–3600 km, δ remains positive (complementarity). Below 1400 km, δ becomes negative and grows more negative as distance decreases, reaching its most negative at 100 km. Thus, complementarity dominates at longer distances, while competition dominates at short distances (<1400 km). - Economic-distance results (W_e): Spatial autoregressive coefficients are larger and positive across SDM/SAR/SEM, reinforcing complementarity and a hub-and-spoke structure (e.g., SDM δ = 0.703***; SAR δ = 0.784***; SEM δ = 0.946***). Tests again favor SDM with fixed effects. - Hierarchy-specific relationships (SDM with W_e): - Within-group: Hubs compete (δ = −0.286***). Trunk and branch groups each show complementarity (δ = 0.536*** and 0.533***, respectively). - Between-group: Complementarity is significant between hub–trunk (δ = 0.276***), hub–branch (δ = 0.467***), and trunk–branch (δ = 0.485***). The hub–trunk complementarity is weakest, suggesting some latent competition. - Robustness: Re-estimation with inverse distance weights confirms the distance threshold at ~1400 km for sign change and similar hierarchy patterns; results are robust across specifications.

The findings directly address the research questions by showing that spatial dependence varies systematically with distance, shifting from complementarity at long range to competition at short range. This clarifies a distance-based attenuation boundary around 1400 km for Chinese airports. Incorporating spatial heterogeneity through an asymmetric economic distance matrix captures directional, hub-driven influences and unobservable heterogeneity, revealing stronger complementarity when network hierarchy is considered. Disaggregating by airport tiers uncovers masked dynamics: hubs predominantly compete (notably through substitutable international transfer functions), whereas trunk and branch airports complement each other and also complement hubs. Importantly, when hubs are absent from the modeled network, overall complementarity weakens, underlining the central coordinating role of hubs in China’s hub-and-spoke system. These insights have practical relevance for network design, capacity planning, and coordinated development policies, especially in multi-airport regions where ground transport improvements and high-speed rail increase substitutability at short distances.

This study contributes an integrated spatial econometric framework that jointly considers distance-dependent spatial dependence and heterogeneity via asymmetric economic weights. Empirically, it shows: (1) clear clustering of airport passenger traffic; (2) a distance threshold (~1400 km) delineating complementary (long-distance) from competitive (short-distance) relationships; (3) strong network-wide complementarity under economic-distance weighting consistent with a hub-and-spoke structure; and (4) heterogeneity across tiers, with competition among hubs but complementarity between and within non-hub groups. Policy implications include accounting for short-range competition in infrastructure and service planning, prioritizing hub development and integration to enhance network efficiency, and pursuing coordinated investment strategies that leverage complementarity across airport tiers. Future research should extend the airport sample beyond major nodes, test additional spatial/temporal modeling approaches, and examine transferability to other countries and network contexts to strengthen external validity.

- Generalizability: Results are specific to China’s spatial, economic, and infrastructural context and may not directly generalize to other regions. - Sample scope: Focusing on 34 major airports may omit dynamics present in smaller or remote airports, potentially biasing system-wide inferences. - Unobserved factors: Despite spatial econometric controls, some unobservable influences may remain, and the chosen weight matrices and specification may not fully capture all heterogeneity or dynamic adjustments.