Analysis

[Link]

Decoupling of SDGs followed by re-coupling as

sustainable development progresses
Xutong Wu 1, Bojie Fu 1,2 ✉, Shuai Wang 1
, Shuang Song 1
, Yingjie Li 3,4
, Zhenci Xu5,
Yongping Wei 6 and Jianguo Liu 3

Understanding the complex interactions among the Sustainable Development Goals (SDGs) is key to achieving all of the SDGs
and ‘leaving no one behind’. However, research about dynamic changes of SDG interactions is limited, and how they change as
sustainable development progresses remains elusive. Here, we used a correlational network approach and a global SDG data-
base of 166 countries to analyse the evolution of SDG interactions along a progression of sustainable development measured by
the SDG Index. SDG interactions showed nonlinear changes as the SDG Index increased: SDGs were both more positively and
more negatively connected at low and high sustainable development levels, but they were clustered into more isolated positive
connection groups at middle levels. The identification of a process of decoupling followed by re-coupling along the SDG Index
strengthens our understanding of sustainable development and may help to suggest action priorities to achieve as many SDGs
as possible by 2030.

T
o tackle the most pressing issues facing humanity, such gies and trade-offs between SDGs, previous studies have identified
as climate change, poverty, inequality and quality educa- the frequency of SDG interactions and the importance of individual
tion, the United Nations adopted 17 ambitious Sustainable SDG goals or targets at different scales4,5,21,22, as well as their dif-
Development Goals (SDGs) to stimulate actions in critically impor- ferences across regions7,12. Comparisons among different groups of
tant areas for people, the planet and prosperity1. The 17 SDGs are countries have shown that SDG interactions vary with a country’s
integrated and indivisible, balancing the economic, social and envi- socioeconomic characteristics, such as income, region and popula-
ronmental dimensions of sustainable development1. They cover all tion composition2,9,17.
aspects of human life and interact in complex ways2. Actions for one Although previous studies have helped policymakers and ana-
goal may reinforce or offset the actions for another3,4, resulting in lysts grasp the complex and systemic nature of SDGs23, research
synergies and trade-offs among the SDGs. For example, using coal about dynamic changes of SDG interactions, that is, how SDG inter-
to improve energy access (SDG 7) will accelerate climate change actions change as sustainable development progresses, is limited. By
(SDG 13) and disrupt health (SDG 3) through air pollution3. Given revealing dynamic changes of SDG interactions along sustainable
the ‘leave no one behind’ objective of the 2030 agenda3,4, under- development levels, we can determine the critical transformative
standing interactions among the SDGs is crucial when designing stages of sustainable development, identify the hurdles and oppor-
appropriate and efficient policies to implement them5,6. tunities of sustainable development for countries at different levels
Using systems thinking and analysis to assess the complex SDG and find specific action priorities for countries at different levels
interactions is at the forefront of sustainability research7. Multiple based on a better understanding of the sustainable development
studies qualitatively scored and assessed SDG interactions by expert process. To fill this knowledge gap, this study addressed three major
expertise3,8,9 or text mining applied to official documents and the questions with a correlational network approach (Fig. 1a). First, did
wording of SDG targets10,11, while other studies used pairwise cor- SDG interactions change along sustainable development levels and,
relations between the official indicator data for each SDG to quan- if so, how? Second, which SDGs were more related to others, and
titively analyse relationships between SDGs4,5,12. Network analysis, how did the connections change along sustainable development lev-
which has been widely used in studies of complex systems (for exam- els? Third, which groups of SDGs tended to be achieved together,
ple, health13, ecosystems14 and societies15,16), is a holistic approach to and how did the compositions of these groups change along sustain-
explore the characteristics of SDG interactions17 and their changes18. able development levels?
It provides clear visualization and conceptualization of interactions To address these questions, we used SDG data of 166 coun-
between variables and well-developed notions to characterize those tries (Fig. 1b and Supplementary Table 1) from the Sustainable
interactions7. An array of network centrality measures (for example, Development Report 2020 prepared by the Sustainable Development
degree centrality, betweenness centrality, eigenvector centrality and Solutions Network and the Bertelsmann Stiftung24, which calcu-
closeness centrality) can measure the importance of SDG goals or lated scores for each of the 17 goals and the SDG Index (reflect-
targets in the interaction network2,7,12,19, while network community ing the overall sustainable development level) for each country, to
detection can reveal the strongly connected groups of SDG goals or build correlational networks along an SDG Index gradient. SDG
targets in the interaction network5,7,20. By characterizing the syner- interactions can be analysed at both goal and target levels2,12,19,21,22.

1
State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University, Beijing, China. 2State
Key Laboratory of Urban and Regional Ecology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, China. 3Center for
Systems Integration and Sustainability, Department of Fisheries and Wildlife, Michigan State University, East Lansing, MI, USA. 4Environmental Science
and Policy Program, Michigan State University, East Lansing, MI, USA. 5Department of Geography, The University of Hong Kong, Hong Kong, China.
6
School of Earth and Environmental Sciences, The University of Queensland, Brisbane, Queensland, Australia. ✉e-mail: bfu@[Link]

Nature Sustainability | [Link]/natsustain

Analysis NaTUre SUsTaInabIlITy

Network metric Definition Meaning in the SDG network Low value High value

Connectivity Proportion of positive or negative High connectivity of synergy 3 3

links to all possible links in the networks indicates that more 4 4
network, weighted by the strength SDGs can be achieved 2 2
of the links. simultaneously. 5 5
High connectivity of trade-off
networks indicates that more 1 1
SDGs cannot be achieved
simultaneously 6 6
9 9
7 7
8 8

Modularity A module represents a group of In a highly modular network, 3 3

nodes that are highly connected SDGs can be divided into isolated 4 4
among them and loosely groups according to their 2 2
connected to others. connections, while in a less
Modularity represents the modular network, the interactions 5 5
strength of the partition of a of all SDGs are closer.
1 1
network into modules.
6 6

9 9
7 7
8 8

SDG Index score

NA
<50
50–54
55–59
60–64
65–69
70–74
75–79
80–85

Fig. 1 | Metrics and countries used in the network analysis. a, Network metrics used in this study. b, SDG Index score of 166 countries from Sustainable
Development Report 202024. The index score signifies a country’s position between the worst (0) and the best or target (100) outcomes across the 17 SDGs.

As previous studies suggested that maintaining flexibility on tar- size was set at 50 (Methods), resulting in a total of 117 windows
gets while remaining focused on goals may offer more opportu- (that is, countries 1–50, 2–51, …, 117–166) and 117 correlational
nities to avoid SDG conflicts and achieve overall sustainability2, networks (Supplementary Fig. 1). In the networks, each node repre-
we chose the goal level for analysis. Causal relationships of SDGs sented an individual SDG, and pairwise SDGs that were significantly
have been explored qualitatively or quantitively in previous stud- (P < 0.05) correlated were connected by a link, where the strength of
ies8,12,22,25; however, due to data constraints, interactions between each link indicates the Pearson correlation coefficient. The obtained
pairwise SDGs were represented as correlations between advance- networks with 17 nodes were weighted and undirected (directional-
ments towards each SDG in this study. We sorted the 166 coun- ity can be estimated only if the direction of causality is known2,12,22).
tries along the Sustainable Development Solutions Network’s SDG Because correlations can be positive or negative, we built separate
Index gradient from the minimum to the maximum score and used networks for synergies (positive correlations, meaning a pair of
a moving-window approach18 to analyse the effect of an increasing SDGs improve or deteriorate together) and trade-offs (negative cor-
SDG Index on interactions among the SDGs. The moving-window relations, meaning one SDG improves while the other deteriorates).

Nature Sustainability | [Link]/natsustain

NaTUre SUsTaInabIlITy Analysis
Two key network metrics, connectivity and modularity, were The connectivity of most SDGs was relatively low in the trade-off
selected to reflect the characteristics of the SDG interaction net- networks, but SDG 12 (responsible consumption and production)
works and analyse their changes (Fig. 1a). Connectivity is the pro- and SDG 13 (climate action) played dominant roles in the negative
portion of present links to all possible links in the network weighted interactions among SDGs (Fig. 3b and Supplementary Fig. 3b). This
by the strengths of the links, reflecting the number and strength of means that actions to meet these goals impair the ability to address
the correlations18. A module represents a group of nodes that are many other SDGs (and vice versa). Connectivity of these two SDGs
highly connected among themselves and loosely connected to oth- declined at first but then increased to a higher value as the SDG
ers, and modularity represents the strength of the partition of a Index increased. SDG 15 (life on land) and SDG 14 (life below
network into modules, indicating the degree of network compart- water) were also relatively dominant SDGs (the connectivity of SDG
mentalization18. To determine whether and how SDG interactions 15 exceeded that of SDG 13 at one point) when the SDG Index was
changed along the SDG Index, we fitted generalized additive models low, but they became less important in the trade-off networks when
to analyse the effect of the SDG Index and other possible variables the SDG Index was high. On the contrary, connectivity of SDG 17
on the network metrics (Methods). After evaluating the changes of (partnerships for the goals) remained low when the SDG Index was
these metrics along the SDG Index at the network level, we further low but was in the top three when the SDG Index was high.
calculated the weighted node degree (that is, the average strength
of connection to other nodes18) of each node in each network to Disruption then reunion of SDG synergy modules. Changes
examine changes in connections of individual SDGs and identified in the module composition of SDG synergy networks along the
the module composition of synergy networks to examine changes SDG Index (Fig. 4 and Supplementary Fig. 4) also reflected a pro-
of groups of SDGs tending to be achieved together (Methods). On cess of decoupling followed by re-coupling of SDGs. According to
the basis of the findings, this study provides new insights about their positive interactions, the 17 SDGs clustered into 3 modules
dynamic changes of SDG interactions along sustainable develop- at a low SDG Index (score = 54), mainly reflecting the social and
ment levels, which will be useful for identifying action priorities for economic SDGs (L1: SDGs 1–9, 11 and 16), environmental SDGs
countries at different levels of sustainable development. (L2: SDGs 12–15 and 10) and partnerships for the goals (L3: SDG
17). As the level of sustainable development improved, these mod-
Results ules were disrupted, and the number of modules increased as the
Nonlinear changes of SDG interactions. All the metrics of the modules became smaller, and the modularity increased. There were
SDG interaction networks changed significantly along the SDG five modules at the middle SDG Index level (score = 66): module
Index (Fig. 2, Supplementary Fig. 2 and Supplementary Table 2) L1 broke up into two modules and SDG 10 (M4) became isolated
while controlling for the effects of other factors such as gross from module L2. These modules reunited when the SDG Index was
national income (GNI) per capita, population density and pre- high (score = 78): modules M1, M2 and M4 and SDG 15 from mod-
cipitation (Methods). The nonlinear changes in the SDG inter- ule M3 made up a module of 13 SDGs (H1), leaving only SDG 14,
actions suggested a process of decoupling of SDGs followed by SDG 17 and a pair of SDGs (12 and 13) isolated from these con-
re-coupling along sustainable development levels. Connectivity of nected SDGs.
both synergy and trade-off networks showed an overall trend from
decreasing to increasing as sustainable development progressed: Discussion
SDGs became less connected when sustainable development lev- Understanding dynamic changes in SDG interactions as sustainable
els increased from low to middle (a decoupling of SDGs), but development progresses is crucial for developing appropriate and
they became more connected than the initial state when sustain- integrative policies for countries at different development stages.
able development reached high levels (a re-coupling of SDGs). The Our study provided new insights into the sustainable develop-
overall trend of the synergy networks’ modularity was the oppo- ment process and identified an interesting pattern. The 17 SDGs
site to that of connectivity; it increased at first but then decreased first experienced a decoupling followed by a re-coupling process
along the SDG Index, reaching a peak when the SDG Index was as sustainable development level increased: SDGs were both more
about 69. This indicates that the 17 SDGs divided into more isolated positively and negatively connected at low and high sustainable
positive connection groups at the mid-level of sustainable devel- development levels but clustered into more isolated positive con-
opment. However, the trend of the trade-off networks’ modularity nection groups at middle levels.
depended on the selection of moving-window size (Supplementary The observed nonlinear changes of SDG interactions along sus-
Fig. 6 and Supplementary Table 3) and was not considered in tainable development levels partially coincide with the findings of
our study. other studies from distinct perspectives but unify them into the gen-
eral context of sustainable development. Previous studies employ-
Highly connected SDGs in the sustainable development pro- ing concepts such as the Kuznets curve, environmental Kuznets
cess. The connectivity of most SDGs first declined then increased curve, and gender Kuznets curve empirically proved that the rela-
to a value higher than the original value along the SDG Index in tionships between economic development and income inequality26,
the synergy networks (Fig. 3a and Supplementary Fig. 3a). SDG between economic development and environmental degradation27
3 (good health and well-being), SDG 9 (industry, innovation and (for example, CO2 emission28, air pollution29, water pollution30 and
infrastructure), SDG 16 (peace, justice and strong institutions) and deforestation31) and between economic development and gender
SDG 6 (clean water and sanitation) were the relatively dominant inequality32 form inverted U-shaped curves. These studies indicate
SDGs along the index, meaning that many other SDGs will be dis- that improvements in economic SDGs would impair social and
proportionately improved as progress is made towards these four environmental SDGs in the initial stages of economic development
goals (and vice versa). SDG 4 (quality education) was a dominant but then would improve simultaneously with these other SDGs after
SDG when the SDG Index was low but became less important in the a turning point. These studies explained the observed decoupling
synergy networks when the SDG Index was high, and SDG 1 (no followed by re-coupling of SDGs in synergy networks and the dis-
poverty) and SDG 7 (affordable and clean energy) showed similar ruption followed by reunion of the SDG synergy modules. They also
trends. SDG 8 (decent work and economic growth) was less impor- explained why SDGs 10, 14 and 15 were highly negatively connected
tant when the SDG Index was low but became one of the top five goals at low SDG Index levels and SDGs 5 and 8 became more posi-
dominant nodes when the SDG Index was high. A similar trend was tively connected goals at high SDG Index levels. As for the trade-off
observed for SDG 5 (gender equality). networks, the total number of negative interactions first increased

Nature Sustainability | [Link]/natsustain

Analysis NaTUre SUsTaInabIlITy

a Revealing the highly connected SDGs can help to identify the

0.15 hurdles and opportunities facing policymakers when trying to
implement successful SDG policies2. Our results showed that SDGs
3, 6, 9 and 16 remained highly positively connected while SDGs 12
Partial effect on connectivity

0.10 and 13 remained highly negatively connected throughout the SDG

Index progression. Some studies combining expert judgement and
scientific literature reviews or using causal analysis have explored
0.05 the causal relations of these and other SDGs8,12,22. Health (SDG 3)
is both a key enabler and a critical outcome of sustainable develop-
ment8: progress on gender equality (SDG 5), clean water (SDG 6),
0 clean energy (SDG 7) and many other goals are likely to support it
by improving reproductive health33, reducing the spread of infec-
tious disease34 and improving respiratory health35; good health is
–0.05 also a strong enabling factor for effective poverty reduction (SDG 1)
and a prerequisite for economic growth (SDG 8)8. In addition to
55 60 65 70 75
health, progress on SDG 6 will also enable and drive progress on
SDG Index score
other SDGs, such as hunger (SDG 2), education (SDG 4), gender
b equality (SDG 5) and environmental protection (SDG 15)36. Reliable
and sustainable infrastructure (SDG 9) enhances resilient rural and
0.10 urban livelihoods (SDG 11), agricultural productivity (SDG 2) and
water availability (SDG 6); moreover, science, technology and inno-
Partial effect on connectivity

vation are essential for health (SDG 3), renewable energy (SDG 7)
and climate action (SDG 13)37. Having effective governance sys-
0.05
tems and institutions is key to an effective, efficient and coher-
ent approach to implementation of many, if not all, SDGs8, which
explains the dominant role of SDG 16 in the synergy networks.
0 Most trade-offs among SDGs can be linked to the unsustainable
development paradigm that focuses on economic growth to gener-
ate human welfare at the expense of environmental sustainability4.
Previous studies found that higher levels of gross domestic product
–0.05
and human development index contributed to the improvement of
55 60 65 70 75
health and nutritional status but also caused larger environmental
SDG Index score
and material footprints and higher greenhouse gas emissions4,38,39,
which are barriers to achieving responsible consumption and pro-
c duction (SDG 12) and climate action (SDG 13)2. This relationship
0.4
explains the increasing level of conflicts between these two SDGs
and other goals at a high SDG Index level.
Our study deepens the understanding of the sustainable devel-
Partial effect on modularity

0 opment process and provides new insights into achieving as many

SDGs as possible by 2030. Learning from the changes in relation-
ships among economic, social and environmental SDGs, we found
–0.4
that sustainable transformation that minimizes trade-offs and facil-
itates new synergies is necessary, especially for the countries that
Synergy networks
are currently at the middle sustainable development level, near the
Trade-off networks turning point of SDG interactions. This transformation calls for
–0.8 the collaboration of scientists, policymakers, non-governmental
95% confidence intervals
organizations and other actors and the pursuit of broader policy
mixes40–42. All countries should pay more attention to the SDGs
55 60 65 70 75
that dominate the synergy and trade-off networks throughout the
SDG Index score
sustainable development progression. Our results indicate that
actions taken to ensure healthy lives and promote well-being,
ensure availability and sustainable management of water and sanita-
Fig. 2 | Effects of SDG Index score on the structure of both positive and tion, promote inclusive and sustainable industrialization and foster
negative correlation networks of SDGs. a,c, Synergy networks (blue). innovation, and build effective, accountable and inclusive institu-
b, Trade-off networks (red). The graphs show the partial effects of the tions will have simultaneous compounded positive effects on other
SDG Index score on connectivity (a,b) and modularity (c) while controlling SDGs. However, simply promoting sustainable consumption and
for the other variables listed in Supplementary Table 2. Shaded areas production and climate actions might hinder the overall progress
represent the 95% confidence intervals of the fitted generalized additive towards the 2030 agenda. In addition to focusing on SDGs that are
models. All effects are significant (P < 0.001). dominant throughout the process of sustainable development, we
need to contextualize and prioritize SDGs by different sustainable
development levels. For countries with a low SDG Index (for exam-
before decreasing along the SDG Index (Supplementary Fig. 8, net- ple, most African countries), alleviating poverty and ensuring inclu-
work density, that is, the proportion of present links to all possible sive and equitable quality education and access to affordable and
links in the network), but because we excluded non-significant cor- clean energy should be considered as high priorities because they
relations with low correlation coefficients, the connectivity showed can positively affect other SDGs. But care must be taken that these
a first declining then rising trend. goals are achieved without compromising marine and terrestrial

Nature Sustainability | [Link]/natsustain

NaTUre SUsTaInabIlITy Analysis
a b
4
4 SDG
SDG 3 good health and well-being
SDG 4 quality education
3 3 SDG 5 gender equality
SDG 6 clean water and sanitation
Connectivity of SDG

SDG 8 decent work and economic growth

2 2 SDG 9 industry, innovation and infrastructure

SDG 10 reduced inequalities
SDG 12 responsible consumption and production
1 SDG 13 climate action
1
SDG 14 life below water
SDG 15 life on land
0 SDG 16 peace, justice and strong institutions
0 SDG 17 partnerships for the goals

55 60 65 70 75 55 60 65 70 75
SDG Index score SDG Index score

Fig. 3 | Connectivity of individual SDGs along the SDG Index. a, Synergy networks. b, Trade-off networks. Only SDGs with a maximum weighted node
degree >1.5 in synergy networks and SDGs with a maximum weighted node degree >1.0 in trade-off networks are shown in the panels.

a
SDG 1
SDG 2 SDG 1
SDG 1 SDG 3 SDG 2
SDG 2 SDG 4 SDG 3
SDG 3 M1 SDG 4
SDG 7
SDG 4 SDG 8 SDG 5
SDG 5 SDG 9 SDG 6
SDG 6 L1 H1 SDG 7
SDG 16
SDG 7 SDG 8
SDG 8 SDG 5 SDG 9
SDG 9 M2 SDG 6 SDG 10
SDG 11 SDG 11 SDG 11
SDG 16 SDG 15
SDG 12 SDG 16
SDG 10 SDG 13
M3
SDG 12 SDG 14 SDG 12
H2
SDG 13 L2 SDG 15 SDG 13
SDG 14
SDG 15 M4 SDG 10 H3 SDG 14 SDG
SDG 1 No poverty
SDG 17 L3 M5 SDG 17 H4 SDG 17
SDG 2 Zero hunger
Low SDG Index Middle SDG Index High SDG Index
SDG 3 Good health and well-being
SDG 4 Quality education
b SDG 5 Gender equality
SDG Index score = 54 SDG Index score = 66 SDG Index score = 78
Modularity = 0.284 Modularity = 0.460 Modularity = 0.079 SDG 6 Clean water and sanitation
SDG 7 Affordable and clean energy
SDG 10 SDG 12 SDG 17 SDG 14 SDG 8 Decent work and economic growth
SDG 12
SDG 13 SDG 13 SDG 9 Industry, innovation and infrastructure
SDG 14 SDG 15 SDG 17
SDG 14
SDG 10 Reduced inequalities
SDG 15 SDG 10
SDG 2 SDG 8 SDG 11 Sustainable cities and communities
SDG 15
SDG 8 SDG 1 SDG 12 Responsible consumption and production
SDG 4 SDG 5 SDG 7 SDG 13
SDG 6
SDG 13 Climate action
SDG 16 SDG 3 SDG 8
SDG 5 SDG 9 SDG 6 SDG 11
SDG 5 SDG 16 SDG 9 SDG 12 SDG 14 Life below water
SDG 6
SDG 17 SDG 9 SDG 3
SDG 4 SDG 1 SDG 15 Life on land
SDG 16 SDG 7 SDG 7 SDG 11 SDG 4 SDG 2
SDG 3 SDG 16 Peace, justice and strong institutions
SDG 1
SDG 11 SDG 2 SDG 10 SDG 17 Partnerships for the goals

Fig. 4 | Module composition of synergy networks and their changes at different SDG Index score levels. a, Changes in module composition of the
synergy networks at different SDG Index score levels. b, Modules of the synergy networks at different SDG Index score levels. Different background colours
represent different modules. Black lines represent SDG interactions in the same module; red lines represent SDG interactions in different modules.

ecosystems because SDGs 14 and 15 are highly negatively con- to adopt environmentally friendly practices for better environ-
nected at this development level. Payments for ecosystem services, mental outcomes, human welfare and social equality, provide an
which directly incentivize landowners and other resource stewards innovative economic intervention to mitigate trade-offs between

Nature Sustainability | [Link]/natsustain

Analysis NaTUre SUsTaInabIlITy

environmental and development goals43,44. For countries with a negative value represents a trade-off, and the absolute value of the correlation
high SDG Index (for example, most Organisation for Economic coefficient represents the strength of the interaction. To analyse the effect of an
increasing SDG Index on interactions among the SDGs, we used a moving-window
Co-operation and Development (OECD) countries), progress in approach18. In total, 166 countries were sorted along the SDG Index gradient from
achieving gender equality, promoting sustainable economic growth the minimum to the maximum SDG Index score. To examine the influences of
and reducing inequality can also promote other SDGs. These coun- moving-window size, we tried using different moving-window sizes (from 30 to
tries should especially focus on developing new governance and 80) to compare the trends and turning points of the network metrics, which were
the main focuses of this study. Under all moving-window sizes, the connectivity
technologies to increase resilience to climate change and decrease
of both the synergy and trade-off networks showed an overall trend of declining
emissions to mitigate the negative connections between SDGs 12 then rising while the modularity of the synergy networks showed an overall trend
and 13 and the other SDGs. Policies in the European Union, such of rising then declining. The turning points of the trends of these metrics fell
as the EU Emissions Trading System (ETS), the Effort Sharing within an SDG Index range of 67.9–71.2 (Supplementary Fig. 6 and Supplementary
Regulation for non-ETS sectors, the land-use, land-use change and Table 3). However, the modularity trend of the trade-off networks depended on
the selection of the moving-window size and thus was excluded from this study.
forestry regulation, and the Renewable Energy Directive, are good The relative ranges of the three network metrics (calculated as the proportion of
examples for the development of such policy mixes45,46. the range of the network metric under one moving-window size to the maximum
Although this study provides important and interesting find- range of the network metric under all moving-window sizes) all exceeded 60%
ings, there are some limitations in the data and methodology. when the moving-window size was 50 (Supplementary Table 3). Compared
First, the identified synergies and trade-offs were based on data with other moving-window sizes, the size 50 can avoid an overemphasis of the
change of one network metric (Supplementary Table 3). Therefore, we set the
from Sustainable Development Report 2020, which used only 115 moving-window size at 50, resulting in a total of 117 windows, each containing
of the United Nations’ 231 indicators, primarily because most of 50 countries. We calculated the mean SDG Index score for each window, resulting
the other indicators are either hard to quantify or lack data24. As in an SDG Index gradient ranging from 53.7 to 77.8.
more data become available in the future, our approach can easily For each window, we calculated the Pearson correlation between each pair of
be applied to an updated SDG goals or targets database to provide a SDGs and used the significant correlation coefficients (P < 0.05) for additional
analyses. To examine the influences of different selection criteria of correlation
more comprehensive and detailed picture. Second, the SDG inter- coefficients, we also built networks using correlations with an absolute coefficient
actions were analysed by using correlation coefficients as proxies, >0.5 and correlations with an absolute coefficient >0.3. We then repeated the
but they do not imply causality10. The observed synergies between analyses of network metrics for comparison. The network metrics showed similar
two SDGs could be independently related to another process linked trends under different selection criteria of correlation coefficients (Supplementary
with SDGs4. Nevertheless, the causal relations of some SDGs were Fig. 7), which reinforced our conclusion. Some SDGs (for example, SDG 14) were
missing for some countries because of a lack of indicators. The missing SDGs were
explained in our discussion on the basis of existing studies that used dropped individually for each pairwise correlation by using the ‘[Link].
expert knowledge or causal analysis. Several studies have started to observation’ mode18. To facilitate the interpretation of the results, the networks
explicitly address causation in SDG interaction networks by using for positive (synergies) and negative (trade-offs) correlations were calculated
approaches such as Granger causality analysis23,25. Additional data separately, while network modules were calculated only for synergy networks.
and the development of methods of analysis will enable us to move
from correlation to causality and build directed and weighted net- Network analyses. The Pearson correlations for each window were converted
works to analyse dynamic changes of SDG interactions2,23,25. Future to a network graph object and analysed by the R package igraph48. In the
research can further investigate the complex mechanisms behind network, the nodes represent the 17 interactive SDGs, and links between nodes
represent positive/negative correlations between two nodes and their weights
the trade-offs and synergies among SDGs47 and find solutions to (Supplementary Fig. 1). We calculated connectivity and modularity (Fig. 1) for
address conflicts among them9. the synergy and trade-off networks, respectively. Connectivity was calculated
In conclusion, this study revealed changes in SDG interactions as the proportion of present links to all possible links in the network, weighted
as sustainable development progresses. The identification of a pro- by the absolute value of the correlation coefficient18. Modularity was calculated
by the ‘cluster walktrap’ algorithm in igraph, which separates densely connected
cess of decoupling followed by re-coupling along the SDG Index
subgraphs via random walks using correlation coefficients as weights18.
strengthens our understanding of sustainable development and may Previous studies have found that SDG interactions vary with a country’s
help to suggest specific action priorities to achieve as many SDGs as income and region, along with the gender, age and location of its population2,17.
possible by 2030. This study proves the necessity of research about To determine whether and how the SDG Index affects SDG interactions, we also
dynamic changes in SDG interactions and may also begin to lay a collected the GNI per capita, precipitation, urbanization, population density,
percentage of females in the population and percentage of the population aged
foundation for analysing such dynamic changes at different scales. 0 to 14 years from the World Bank’s World Development Indicators. For each of
these variables, we used the average value of all the countries in each window. To
Methods avoid multicollinearity, we then calculated each variable’s variance inflation factor
Data sources. The overall SDG Index scores and scores on the individual SDGs and excluded variables for which it was >10. Only SDG Index, GNI per capita,
of 166 countries were collected from Sustainable Development Report 202024, population density and precipitation remained after this exclusion process. We
which describes each country’s progress towards achieving the SDGs. The report fitted generalized additive models to analyse the effect of these four variables on
is available for each year from 2017 to 2021, and each report provides the scores the network metrics using the mgcv package49 and smoothed the fitted response
of each country for that year. However, due to changes in the indicators as well as by setting the k attribute of the generalized additive models to avoid unexpected
some refinements in the methodology, SDG scores cannot be compared among the wiggliness of the curve18 (Supplementary Table 2).
different years24. Because the trends of network metrics along the SDG Index are To test whether the effect of SDG Index on network metrics differed from
similar when calculated using data from the different years (Supplementary Fig. 5) random expectations, we compared our results with the network metrics for 100
and the number of countries is greatest in Sustainable Development Report 2020, we randomizations of the dataset along the SDG Index gradient. In the randomizations,
selected that report for use as our dataset. the individual SDG scores of each country were maintained, but the overall SDG
The scores can be interpreted as a percentage of optimal performance. To Index scores assigned to each country were randomized18. This procedure allowed
generate comparable scores and rankings, the same basket of indicators is used for us to test whether the observed changes are related to the SDG Index or occur by
all countries. A total of 115 indicators, 85 global indicators and 30 indicators added chance. Through the comparison, we found clear differences in the trends of the
specifically for OECD countries was used (Supplementary Table 4). Most of the observed network versus the 100 randomizations (Supplementary Fig. 8). To assess
data used in this report come from international organizations such as the World the effect of excluding non-significant correlations, we repeated the analyses of
Bank, the Food and Agriculture Organization, the World Health Organization, network metrics for synergy and trade-off networks with the raw correlations. There
the United Nations Children’s Fund, OECD and others, which have extensive and were no major differences between the approaches except for an opposite trend
rigorous data-validation processes. To calculate the SDG Index, this report first of density (the proportion of present links to all possible links in the network) in
censored extreme values from the distribution of each indicator, then rescaled the trade-off networks when using the raw correlations (Supplementary Fig. 7).
data to ensure comparability across indicators and finally aggregated the indicators Besides the network metrics, we used the weighted node degree (the average
within and across SDGs. strength of connection to other nodes, calculated as the product of the degree of a
node and the mean of the absolute correlation coefficients of all connections18)
Interactions among SDGs. We used Pearson correlation coefficients to represent to calculate the connectivity of individual SDGs in the interaction networks.
the interactions among SDGs: a positive value represents a synergy whereas a We calculated this value for each node in the networks to identify the most

Nature Sustainability | [Link]/natsustain

NaTUre SUsTaInabIlITy Analysis
connected node and the change in connectivity of each node along the SDG Index. 22. Zhou, X., Moinuddin, M. & Li, Y. SDG Interlinkages Analysis &
To compare the module composition of the synergy networks at different levels of Visualisation Tool Version 4.0 (Institute for Global Environmental Strategies,
the SDG Index, we used the cluster walktrap algorithm in igraph to identify the 2021); [Link]
module composition of each network (Supplementary Fig. 4). We compared the 23. Ospina-Forero, L., Castañeda, G. & Guerrero, O. A. Estimating networks of
composition of the network modules at the low (SDG Index score = 54), middle Sustainable Development Goals. Information & Management [Link]
(SDG Index score = 66) and high (SDG Index score = 78) SDG Index levels. Note org/10.1016/[Link].2020.103342 (2020).
that the existence and composition of the modules in a network is independent 24. Sachs, J. et al. Sustainable Development Report 2020: The Sustainable
from the network’s modularity value, which means that modules can be identified Development Goals and COVID-19 (Cambridge Univ. Press, 2020).
even if the modularity value is low18. 25. Dörgő, G., Sebestyén, V. & Abonyi, J. Evaluating the interconnectedness of
the Sustainable Development Goals based on the causality analysis of
Data availability sustainability indicators. Sustainability 10, 3766 (2018).
All of the data used in this paper can be obtained from the Sustainable Development 26. Kuznets, S. Economic Growth and Income Inequality (Routledge, 2019).
Report ([Link] and the World Bank World 27. Dinda, S. Environmental Kuznets curve hypothesis: a survey. Ecol. Econ. 49,
Development Indicators ([Link] 431–455 (2004).
e=world-development-indicators). 28. Wang, H. et al. China’s CO2 peak before 2030 implied from characteristics
and growth of cities. Nat. Sustain. 2, 748–754 (2019).
29. Grossman, G. M. & Krueger, A. B. Economic growth and the environment.
Code availability Q. J. Econ. 110, 353–377 (1995).
All computer code used in conducting the analyses summarized in this paper is 30. Paudel, K. P., Zapata, H. & Susanto, D. An empirical test of environmental
available from the corresponding author upon reasonable request. Kuznets curve for water pollution. Environ. Resour. Econ. 31, 325–348 (2005).
31. Panayotou, T. Empirical Tests and Policy Analysis of Environmental
Received: 29 September 2021; Accepted: 22 February 2022; Degradation at Different Stages of Economic Development (International
Published: xx xx xxxx Labour Organization, 1993).
32. Eastin, J. & Prakash, A. Economic development and gender equality: is there
a gender Kuznets curve? World Pol. 65, 156 (2013).
References 33. Wang, G.-z. Reproductive Health and Gender Equality: Method, Measurement,
1. Transforming Our World: The 2030 Agenda for Sustainable Development and Implications (Routledge, 2016).
(United Nations, 2015). 34. Progress on Sanitation and Drinking Water: 2015 Update and MDG
2. Lusseau, D. & Mancini, F. Income-based variation in Sustainable Assessment (World Health Organization, 2015).
Development Goal interaction networks. Nat. Sustain. 2, 242–247 (2019). 35. WHO Guidelines for Indoor Air Quality: Household Fuel Combustion (World
3. Nilsson, M., Griggs, D. & Visbeck, M. Policy: map the interactions between Health Organization, 2014).
Sustainable Development Goals. Nature 534, 320–322 (2016). 36. Progress on Drinking Water, Sanitation and Hygiene: 2017 Update and SDG
4. Pradhan, P., Costa, L., Rybski, D., Lucht, W. & Kropp, J. P. A systematic study Baselines (World Health Organization, 2017).
of Sustainable Development Goal (SDG) interactions. Earth’s Future 5, 37. Mantlana, K. B. & Maoela, M. A. Mapping the interlinkages between
1169–1179 (2017). Sustainable Development Goal 9 and other Sustainable Development Goals: a
5. Sebestyén, V., Bulla, M., Rédey, Á. & Abonyi, J. Network model-based preliminary exploration. Bus. Strategy Dev. 3, 344–355 (2020).
analysis of the goals, targets and indicators of sustainable development for 38. Costa, L., Rybski, D. & Kropp, J. P. A human development framework for
strategic environmental assessment. J. Environ. Manage. 238, 126–135 (2019). CO2 reductions. PLoS ONE 6, e29262 (2011).
6. ISDG Integrated Simulation Tool: Policy Coherence and Integration to Achieve 39. Omri, A., Nguyen, D. K. & Rault, C. Causal interactions between CO2
the Sustainable Development Goals (Millennium Institute, 2018); [Link] emissions, FDI, and economic growth: evidence from dynamic
[Link]/isdg simultaneous-equation models. Econ. Model. 42, 382–389 (2014).
7. Bali Swain, R. & Ranganathan, S. Modeling interlinkages between sustainable 40. Morrison, T. H. et al. Save reefs to rescue all ecosystems. Nature 573,
development goals using network analysis. World Dev. 138, 105136 (2021). 333–336 (2019).
8. Griggs, D., Nilsson, M., Stevance, A. & McCollum, D. A Guide to SDG 41. Gunningham, N., Grabosky, P. & Sinclair, D. Smart Regulation: Designing
Interactions: From Science to Implementation (International Council for Environment Policy (Clarendon Press, 1998).
Science, 2017). 42. Axsen, J., Plötz, P. & Wolinetz, M. Crafting strong, integrated policy
9. Nilsson, M. et al. Mapping interactions between the sustainable development mixes for deep CO2 mitigation in road transport. Nat. Clim. Change 10,
goals: lessons learned and ways forward. Sustain. Sci. 13, 1489–1503 (2018). 809–818 (2020).
10. Le Blanc, D. Towards integration at last? The sustainable development goals 43. Wunder, S. et al. From principles to practice in paying for nature’s services.
as a network of targets. Sustain. Dev. 23, 176–187 (2015). Nat. Sustain. 1, 145–150 (2018).
11. Pham‐Truffert, M., Metz, F., Fischer, M., Rueff, H. & Messerli, P. Interactions 44. Salzman, J., Bennett, G., Carroll, N., Goldstein, A. & Jenkins, M. The global
among Sustainable Development Goals: knowledge for identifying multipliers status and trends of payments for ecosystem services. Nat. Sustain. 1,
and virtuous cycles. Sustain. Dev. 28, 1236–1250 (2020). 136–144 (2018).
12. Zhou, X., Moinuddin, M. & Xu, Z. Sustainable Development Goals 45. Rogge, K. S., Kern, F. & Howlett, M. Conceptual and empirical advances
Interlinkages and Network Analysis: A Practical Tool for SDG Integration and in analysing policy mixes for energy transitions. Energy Res. Soc. Sci. 33,
Policy Coherence (Institute for Global Environmental Strategies, 2017). 1–10 (2017).
13. Barabási, A.-L., Gulbahce, N. & Loscalzo, J. Network medicine: a 46. Skjærseth, J. B. Towards a European Green Deal: the evolution of EU climate
network-based approach to human disease. Nat. Rev. Genet. 12, 56–68 (2011). and energy policy mixes. Int. Environ. Agreem. 21, 25–41 (2021).
14. Saavedra, S., Stouffer, D. B., Uzzi, B. & Bascompte, J. Strong contributors 47. Fuso Nerini, F. et al. Mapping synergies and trade-offs between energy and
to network persistence are the most vulnerable to extinction. Nature 478, the Sustainable Development Goals. Nat. Energy 3, 10–15 (2018).
233–235 (2011). 48. Csardi, G. & Nepusz, T. The igraph software package for complex network
15. Bond, R. Complex networks: network healing after loss. Nat. Hum. Behav. 1, research. InterJournal [Link] (2006).
0087 (2017). 49. Wood, S. N. Generalized Additive Models: An Introduction with R (CRC
16. Bodin, Ö. et al. Improving network approaches to the study of complex Press, 2017).
social–ecological interdependencies. Nat. Sustain. 2, 551–559 (2019).
17. Warchold, A., Pradhan, P. & Kropp, J. P. Variations in sustainable
development goal interactions: population, regional, and income Acknowledgements
disaggregation. Sustain. Dev. 29, 285–299 (2020). This research was financially supported by the National Natural Science Foundation of
18. Felipe-Lucia, M. R. et al. Land-use intensity alters networks between China (42041007, B.F. and S.W.), the National Key Research and Development Program
biodiversity, ecosystem functions, and services. Proc. Natl Acad. Sci. USA 117, of China (2017YFA0604701, B.F. and S.W.), the China National Postdoctoral Program
28140–28149 (2020). for Innovative Talents (BX2021042, X.W.), the China Postdoctoral Science Foundation
19. Allen, C., Metternicht, G. & Wiedmann, T. Prioritising SDG targets: assessing (2021M700458, X.W.) and the US National Science Foundation (1924111, J.L.). We thank
baselines, gaps and interlinkages. Sustain. Sci. 14, 421–438 (2019). M. R. Felipe-Lucia et al. for sharing the R script for network analysis in their publication
20. Weitz, N., Carlsen, H., Nilsson, M. & Skånberg, K. Towards systemic and ([Link]/cgi/doi/10.1073/pnas.2016210117).
contextual priority setting for implementing the 2030 Agenda. Sustain. Sci.
13, 531–548 (2018).
21. Miola, A., Borchardt, S., Neher, F. & Buscaglia, D. Interlinkages and Policy Author contributions
Coherence for the Sustainable Development Goals Implementation: An B.F. and X.W. designed the research. X.W., S.W. and S.S. performed the data analysis.
Operational Method to Identify Trade-offs and Co-benefits in a Systemic Way X.W., B.F., S.W., S.S., Y.L., Z.X., Y.W. and J.L. contributed to the interpretation
(Publications Office of the European Union, 2019). and writing.

Nature Sustainability | [Link]/natsustain

Analysis NaTUre SUsTaInabIlITy

Competing interests Peer review information Nature Sustainability thanks Mustafa Moinuddin, Tiffany
The authors declare no competing interests. Morrison and the other, anonymous, reviewer(s) for their contribution to the peer review
of this work.

Additional information Reprints and permissions information is available at [Link]/reprints.

Supplementary information The online version contains supplementary material Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
available at [Link] published maps and institutional affiliations.
Correspondence and requests for materials should be addressed to Bojie Fu. © The Author(s), under exclusive licence to Springer Nature Limited 2022

Nature Sustainability | [Link]/natsustain