Spatio-temporal changes in the causal interactions among Sustainable Development Goals in China

The SDGs are integrated and indivisible, balancing economic, social, and environmental dimensions. Actions toward one goal can create synergies or trade-offs with others. Prior work shows interactions vary by income, development level, population, and region, but strategies to prevent beneficial synergies from being offset by harmful trade-offs are underexplored. Synergies and trade-offs also change over time, and simple counts may mask offsetting effects when relationships switch direction. Policymakers need quantitative, spatially and temporally resolved evidence on causal interactions, closed loops (virtuous/vicious cycles), and shifts between active and passive roles of SDGs. Focusing on China, which has extensive subnational SDG data and multiple policies to accelerate SDGs but still faces gaps, the study asks: (i) What are the offsetting effects and spatial variations of shifts between synergies and trade-offs among SDGs in China over 2000–2020? (ii) What are the spatio-temporal characteristics of closed loops in China’s SDG networks? (iii) What are the spatial and temporal dynamics of active–passive relations among SDGs in China?

Existing studies assess SDG interactions using qualitative expert mapping and document analysis, and quantitative correlations (Pearson, Spearman), multiple factor analysis, and autoregressive correlations. Findings highlight that interlinkages vary by country income, development level, population, and region. Some work identifies dominant goals, frequent interlinkages, and levers/obstacles, and explores causation via expert elicitation, Granger causality, systems models, statistical structure learning, and physics-inspired approaches. Unified SDG databases have been proposed to harmonize assessments. However, most analyses are undirected or lack causality, rely heavily on expert judgment, and seldom address temporal changes or offsetting effects when synergies convert to trade-offs (and vice versa). Prior work on closed loops and active/passive roles is largely conceptual and not spatio-temporally resolved, leaving a gap for data-driven causal analysis at fine spatial and temporal scales.

Data sources: Provincial statistical data for 31 mainland provinces (excluding Hong Kong, Macau, Taiwan) from 2000–2020 were compiled from the National Bureau of Statistics and yearbooks. A provincial SDG indicator system referencing the UN SDG indicator framework, Sustainable Development Report 2021, and literature yielded 86 indicators covering 80 targets and 16 goals; SDG14 (life below water) was excluded due to insufficient data for most provinces. Indicators were classified as positive, negative, or moderate, and normalized to a 0–100 scale (0 worst, 100 best) using min–max or mid-point proximity transformations; bounds were set at the 2.5th percentiles, with censoring beyond bounds. Optimal values for moderate indicators were literature-based. Multicollinearity was screened via variance inflation factor; 65 indicators were retained for analysis. Causal inference via GTWR: To account for spatio-temporal non-stationarity, spatio-temporal geographically weighted regression (GTWR) was applied. For each of the 65 indicators, a regression was run with that indicator as dependent variable and the remaining indicators as predictors, using 651 observations (31 provinces × 21 years). Augmented Dickey-Fuller and spatial autocorrelation tests assessed non-stationarity; model fit was evaluated using adjusted R², with values >0.8 considered credible. The GTWR coefficients provided directed, location- and time-specific causal intensities. Coefficient thresholds defined interaction types: >0.1 synergy (positive effect), <−0.1 trade-off (negative effect), and between −0.1 and 0.1 non-significant. Network construction and analysis: For each province-year (2000–2020), directed weighted networks were built separately for synergies and trade-offs from the GTWR coefficient matrices (total 1302 networks) using NetworkX. Nodes represent indicators; directed edges represent causal effects; edge weights are coefficient magnitudes. Edge betweenness centrality was computed, and the top 300 edges formed sub-networks for loop detection. Simple cycles of length 3 to 5 were identified using a simple cycle algorithm; due to computational constraints, a 3-hour time limit per province-year was set. Indicators appearing in critical loops ≥3 times were counted to identify high-frequency nodes. Active–passive relations: For each network, indicator-level out-degree and in-degree were computed; goal-level values were obtained by averaging across constituent indicators. The active ratio was defined as out-degree divided by in-degree (>1 active; <1 passive). Interaction degree was the sum of out-degree and in-degree. The overall change metric for each SDG between two years was defined as the Euclidean distance in the plane of active ratio (x) and interaction degree (y); SDGs with the largest change were tagged as typical goals of active–passive changes for each province in synergy (Table S2) and trade-off (Table S3) networks.

• Offsetting effects: Although the total numbers of synergistic and trade-off pairs remained broadly stable nationally from 2000 to 2020, substantial interchanges occurred between types. Approximately 27% of trade-off indicator pairs turned into synergies and about 25% of synergy pairs turned into trade-offs. Around 50% of indicator-pair interactions were non-significant over the period, underscoring that relationships are not binary. • Notable transitions: The effect of SDG3 (health) on SDG15 (life on land) shifted from trade-off to synergy, while its effect on SDG4 (education) shifted from synergy to trade-off, producing offsetting national effects. Regional patterns showed stronger offsets in Northeastern and Central provinces. In Eastern coastal regions, SDG6 (clean water) effects shifted from trade-off to synergy with SDG2 (zero hunger) but from synergy to trade-off with SDG4. • Closed loops: Provinces exhibited closed loops (virtuous cycles) in synergy networks but none in trade-off networks. A typical virtuous cycle in Southeast China linked SDG1.1.1 (eradicate extreme poverty), SDG15.5.1 (Red List Index), and SDG15.1.1 (forest cover rate), indicating mutual reinforcement between poverty alleviation and terrestrial ecosystem improvements. Loops tended to recur in neighboring provinces and consecutive years. • High-frequency indicators in virtuous cycles: SDG12.2.2 (domestic material consumption), SDG6.5.1 (water resources management), SDG4.1.2 (education completion rate), SDG4.a.1 (educational facilities), SDG4.b.1 (educational investment), and SDG15.5.1 (Red List Index) appeared frequently in critical loops. Prioritizing these can multiply positive systemic effects. • Active–passive dynamics (national): In synergy networks, ≈19.1% of indicators moved from passive to active and ≈19.3% from active to passive. In trade-off networks, ≈20.1% moved from passive to active and ≈16.5% from active to passive. SDG1 (no poverty) and SDG7 (affordable and clean energy) shifted from passive to active in both networks, indicating growing influence. SDG5 (gender equality) and SDG16 (peace, justice, and strong institutions) shifted from active to passive in both networks, indicating increased susceptibility. • Spatial dynamics (provincial): In synergy networks, SDG13 (climate action) became a typical passive-to-active goal in Southwest/Southern China; SDG4 became passive-to-active in Northeast China. Several synergy relations became more passive, notably for SDG16, SDG5, and SDG11 in many provinces. In trade-off networks, SDG1 was typically passive-to-active in Northern provinces (with decreasing trade-off intensity), while SDG7 became passive-to-active in Southern provinces (with increasing trade-off intensity). SDG11 and SDG16 commonly shifted from active to passive in both network types, acting as dependent nodes that may slow trade-off propagation.

This study advances SDG interaction analysis by deriving directed, spatio-temporally varying causal networks using GTWR rather than undirected correlations or expert-only maps. The results show that while headline counts of synergies and trade-offs may look stable, underlying relationships switch substantially, causing offsets when benefits in one link are counteracted by emergent trade-offs elsewhere (e.g., SDG3’s shifting effects on SDG15 vs SDG4). Plausible mechanisms include increased adoption of green lifestyles improving ecosystems, enhanced water quality management aiding food security, and competing investments or shocks (e.g., health vs education budgets, COVID-19 impacts on education and mobility). The presence of virtuous cycles exclusively in synergy networks suggests concrete entry points for policy: targeting high-frequency loop indicators (resource efficiency, water management, education investment and facilities, biodiversity conservation) can magnify positive effects across goals. Shared cycles across adjacent provinces indicate opportunities for coordinated, regionally integrated policies. Active–passive analyses reveal evolving leverage points: SDG13 and SDG4 becoming more active in several regions suggests that climate action and education can drive broader progress if supported. Conversely, SDG1 and SDG7 becoming more active in trade-off networks warns that pursuing poverty eradication and clean energy without addressing cross-goal trade-offs could hinder overall SDG advancement. SDG11 and SDG16’s increasing passivity signals their dependence on broader systemic progress and their role in mediating trade-off propagation.

The study evidences strong offsetting between evolving synergies and trade-offs among SDG indicators in China over 2000–2020. It identifies virtuous cycles and high-frequency indicators that can be prioritized to multiply positive systemic effects and mitigate offsets. Policy design should be adaptive, emphasizing promotion of active synergistic goals (e.g., quality education, climate regulation) and mitigation of active trade-offs (e.g., from poverty eradication and clean energy expansion) by anticipating cross-goal impacts. Coordinated regional policies exploiting shared virtuous cycles can reduce costs and accelerate progress. Recommended measures include shifting toward service-oriented consumption, strengthening unified water resources management, and sustained, high-quality education development and funding, alongside biodiversity protection. Future work will update SDG data, assess COVID-19 impacts, and extend analyses to city-level SDG networks and simulations to further support achieving the 2030 Agenda in China.

• Scope and data coverage: SDG14 (life below water) was excluded due to insufficient provincial data; Hong Kong, Macau, and Taiwan were not included. The indicator set was reduced to 65 after multicollinearity checks, which may omit relevant dimensions. • Methodological thresholds: Interaction classification relied on coefficient thresholds (±0.1) and adjusted R² credibility cutoffs, which may affect sensitivity to weaker links. Directed causality is inferred from GTWR associations and may still be subject to unobserved confounders. • Network detection constraints: Loop detection was limited to simple cycles of length 3–5 within sub-networks (top 300 betweenness edges) and capped by a 3-hour time limit per province-year, potentially missing longer or lower-centrality cycles. • Generalizability: Findings are specific to China’s provincial context and the 2000–2020 period; temporal changes, policy environments, and data availability may limit transferability to other scales or countries. • Data transparency: While source data are from official statistics, the compiled datasets are not publicly available due to confidentiality, limiting external reproducibility.