Global transitions to sustainable development: A cross-country clustering of SDG progress

2.1. Data preparation

No official mapping of SDG indicators to the six transitions existed at the time of this study. Therefore, as the first step in preparing data, the SDG indicators with sufficient data (a minimum of two data points for over half of the countries in the world) were mapped on the six transitions. Data were mainly sourced from the Global SDG Indicator Database (maintained by UN Statistics Division), spanning 2000–2025 across 288 countries/territories. Two principles guided the mapping: first, indicators were mapped based on their direct and indirect relevance to the transition to increase the coverage of the SDGs. Mapping of indicators from several goals (in some cases up to 8) into a single transition area reflects the cross-cutting nature of the SDGs and captures the trade-offs under each transition. Second, some indicators were mapped on multiple transitions to reflect inherent interlinkages of the SDGs. ¹ 63 out of 128 SDG indicators with sufficient data, covering all 17 SDGs and 53 SDG targets, were mapped on the 6 transitions (Annex 1).

A second data availability assessment was conducted to include only countries with at least two data points for a minimum of 30% of indicators under all transitions. As a result, 184 countries were included in the analysis. For selected countries, and for indicators with sufficient data, missing data for 2015 or 2025 were filled using a combination of linear interpolation and constant imputation (last observation carried forward or next observation carried backward). In addition, moving averages were applied to indicators with high seasonality (e.g., deaths/missing/affected from disasters, economic loss from disasters, international support for clean and renewable energy) to offset the impact of extreme events.

2.2 Correlation analysis

Many SDG indicators comprise multiple data series that capture different dimensions or variable types, raising the challenge of selecting a single series that can reliably represent the indicator as a whole. In addition, since numerous SDG indicators exhibit high intercorrelations, including all of them in clustering procedures would introduce redundancy and reduce computational efficiency without improving analytical value. Two complementary correlation analyses were conducted to manage these complexities inherent in the SDG indicator framework. First, a within-indicator correlation analysis was performed to identify the series most strongly correlated with the others under each indicator. Second, a correlation analysis was performed within each transition to detect and remove redundant indicators (those whose patterns were already well captured by one or more other indicators).

2.3 Imputation and dimension reduction

The number of indicators varies between 6 and 16 across transitions, which makes an efficient and comparable clustering challenging. A Principal Component Analysis (PCA) was applied to reduce clustering variables to a fixed number of independent variables across transitions. Four principal components, together explaining over 70% of the total variation in each transition, were selected as clustering variables.

One challenge in analyzing SDG indicators is the large number of missing values. Large, especially nonrandom, missingness substantially reduces the effective sample size and introduces bias in estimates, or forces inefficient imputation and model assumptions when implementing PCA.

Three approaches were examined in treating missing values:

Mean imputation by indicators: Missing values are replaced by the average of the indicator across countries.

2.4 Clustering

To identify the most suitable clustering strategy for this dataset, multiple methods that differ in their underlying assumptions, optimization procedures, and sensitivity to data characteristics such as dimensionality and completeness were considered. Eventually, two clustering techniques were compared for their efficiency: K-means (a distance-minimizing partitioning algorithm) and Self-Organizing Maps (SOM) (an unsupervised neural network–based and structure-learning approach). Each technique was applied in combination with three approaches for handling the missing values. Their performance was examined using two evaluation metrics to determine which method provides the most efficient partitioning structure for the SDG data.

The two clustering techniques were applied in combination with three approaches for treating missing data (six variants in total) to cluster 184 countries in each transition in 2015 and 2025. Two complementary metrics were used to select the most efficient clustering across all transitions and years: Silhouette score and Davies–Bouldin (DB) index (see Annex 3 for more details). The Silhouette score measures how similar an object is to its own cluster compared to other clusters. Values closer to 1 on average indicate better clustering. The DB index, on the other hand, minimizes the ratio of within-cluster dispersion to between-cluster separation, where values closer to 0 indicate better clustering results. Table 1 presents the results of the two indices for six clustering approaches under the Education transition for the year 2025 as an example. After examining results across all six transitions for both 2015 and 2025 data, the SOM clustering technique combined with MICE imputation method was identified as the most efficient clustering.

OPEN IN VIEWER

Table 2.

Pearson correlations between countries’ GDP per capita and transition scores (2015 and 2025).

	2015	2025
Transition	ρ (rho)	ρ (rho)
Digital connectivity	0.88***	0.91***
Education	0.83***	0.85***
Food systems	0.81***	0.78***
Welfare and social protection	0.74***	0.77***
Energy	0.51***	0.53***
Environment	0.40***	0.24*

Significance: ***p < 0.001, **p < 0.01, *p < 0.05.

OPEN IN VIEWER

Table 3.

Cross-Transition drivers of positive shifts – OLS regression results.

Dependent		Digital			Food	Welfare & Social	GDP per
Variable	Environment	Connectivity	Education	Energy	Systems	Protection	capita (ln)
Environment	−0.13***	0.06*	0.02	0.02	−0.04	−0.02	−0.05*
Digital Connectivity	0.04	−0.47***	0.26***	−0.01	−0.01	0.06	0.09**
Education	−0.03	−0.01	−0.07*	0.04	0.01	0.05	0
Energy	−0.04	0.08***	0.00	−0.03	−0.09**	0.05	−0.02
Food Systems	−0.00	0	−0.03	−0.04	−0.06*	0.06*	−0.00
Welfare & Social Protection	0.05	0.07*	−0.01	−0.01	0.01	−0.09	−0.01

Note: Coefficients are from OLS models predicting change (2015–2025) using 2015 transition scores and ln GDP per capita. Significance: ***p < 0.001, **p < 0.01, *p < 0.05.

The final step was to assign cluster labels. To ensure comparability across clusters and avoid label-switching between years, average cluster values for 2015 and 2025 for each indicator were combined and standardized (Annex 3). The average standard value for each cluster was then used as the “overall performance score” to assign one of the four cluster labels (lagging, emerging, performing, and achieving). Not all transition-year clustering outputs resulted in four distinct clusters with significantly different overall performance scores. In several cases, multiple clusters with similar performance scores receive the same label, although they represent a different typology of performance based on indicators and as a result, some transition-years have less than four clusters.

To examine the association between national income and performance across different transitions, Pearson correlation was used to assess the relationship between countries’ GDP per capita and their standardized transition scores for each domain in 2015 and 2025 separately. The standardized scores were derived by standardizing each indicator across all countries (pooling both years) to obtain z-scores, which were then multiplied by their respective directionality (ensuring that higher scores reflect better performance) and averaged by country-year to produce a single score per transition for each country.

To understand the determinants of progress across transitions, a multiple regression model was specified as follows:

Where D ( B i ) = B i , 2025 − B i , 2015 is the change in a country's standardized score for transition B from 2015 to 2025, where positive values indicate improvement. The variable B i , 2015 denotes the initial performance in the transition, while A i , 2015 captures initial scores in all other transitions when, A ≠ B and ln ( GD P i , 2015 ) controls for initial economic status. The model was estimated separately for each of the six transitions (food systems; energy; digital connectivity; education; welfare and social protection; and environment). All models were estimated using ordinary least squares (OLS) with standard errors that are robust to unequal variances, ensuring reliable statistical inference.

Over the past decade, the world has made a small transition towards greater food security, with only around 10% of countries moving into better-performing groups since 2015 (Figure 1). The shift is mild and mainly driven by a slight improvement in nutrition outcomes. Drastic positive shifts are only seen in rare cases (Bahamas, Botswana) (Figure 2). On the contrary, some dimensions of food system have seen increased risks linked to changing lifestyles (such as rising alcohol consumption and the increasing prevalence of overweight among children under 5), female health and nutrition (such as high rates of anaemia among women), economic shocks (such as high food price anomalies), and environmental sustainability (such as increasing risks to livestock genetic diversity). This underscores that while food availability and nutrition have improved, the nexus between the food system and health, lifestyle, economy, and environment continue to pose significant threats to countries. Across continents, Latin America and the Caribbean experienced the sharpest shifts in food system performance. Bolivia, Brazil, and Uruguay moved into less food-secure clusters, while Argentina, Belize, and Peru advanced to better-performing groups, highlighting divergent trajectories within the region.

3.2 Energy

Since 2015, the world has witnessed a minimal energy transition, with fewer than 8% of countries moving to clusters featuring a more sustainable and accessible energy systems (Figure 3). Nevertheless, seven countries (Bangladesh, Botswana, Cambodia, Eswatini, Honduras, Kiribati, Nauru) have demonstrated notable improvement, shifting from lagging in 2015 to performing by 2025 (Figure 4). On the other hand, reliance on fossil fuel has remained high, the total greenhouse gas (GHG) emissions continue to rise (and this is considerably so for the two largest countries in the world, China and India) and although capacity for producing renewable energy has increased in emerging cluster, the share of renewables out of the total energy consumption has remained low. The disappearance of the emerging cluster also underscores a widening gap in energy sustainability. Of the twelve countries classified as emerging in 2015, five (China, India, Lao, Pakistan, South Africa) have fallen into the lagging group in 2025. This shift is especially consequential because China and India alone account for roughly 35% of the global population, amplifying the implications of their regression for the global energy transition and, ultimately, the world's ability to deliver on the SDGs. This is also reflected in their performance on environmental transition (below).

3.3 Digital connectivity

The world has experienced a significant digital transformation since the adoption of the SDGs, marking the most positive shift among all six transitions. A decade ago, digital exclusion was a global norm, with 72% of countries lagging across the board or struggling with basic connectivity. By 2025, more than two-thirds of those left-behind countries had repositioned themselves to a more digitally connected status (Figure 5). In fact, some countries (such as Cape Verde, Uzbekistan, and Viet Nam) made stunning leaps, moving from a lagging position to new digital players (Figure 6). Three regions (Africa, Asia-Pacific, and Latin America) have driven this global surge in digital connectivity, accounting for most of the countries making transitions. Yet, a large part of sub-Saharan Africa (Chad, Ethiopia, Mauritania, Mozambique, Niger, Sierra Leone, Somalia, Sudan, Zimbabwe) has remained disconnected, and the digital divide between developed and developing countries is now reducing to a divide between Africa and the rest of the world.

With this transition also comes narrowing the connectivity gap, only to confront a new and more complex one; The gap between being online and truly leveraging digital tools for inclusive, sustainable growth. Many countries are now digitally connected, but struggling to turn that into better development outcomes, such as enhancing research and development, and harnessing digital technology for equal access to quality services (in sectors such as education, health, and finance).

3.4 Education

During the past 10 years, only 21% of countries have transitioned to a more advanced position on education (Figure 7). The majority of the achievers in 2015 are still at the head of the class today. But the good news is that one-third of those lagging in 2015 have managed to catch up with the performing in 2025, the second largest positive shift by left-behind groups after the digital connectivity transition. ² With nearly half of these countries located in Latin America and the Caribbean, this region is at the forefront of the global education transition (Figure 8).

Progress is more evident beyond direct education indicators, where fewer children go hungry, teen pregnancy rates are declining, and the Internet is reaching more homes than ever. These are vital gains that make it easier for children to learn and thrive. Despite a more supportive education environment, progress on a critical pillar remains stalled: governments are underinvesting in education. Additionally, high school completion rates have not translated into improved employment prospects for youth or enhanced research and innovation capacity in many countries.

3.5 Welfare and social protection

The positive shift in welfare and social protection has been relatively high, with 23% of countries moving into better-performing groups over the past decade (Figure 9). Sixteen countries, ³ many of them upper-middle-income economies from the Balkans, managed to leapfrog progress, moving up more than one position in the transition clusters (Figure 10). Yet the global progress masks growing challenges: as economies advance with lower poverty and broader access to services, income inequality is deepening, and labour share from GDP continues to shrink. In addition, within the economies that have improved their overall performance, concerning trends persist. Rates of youth not in employment, education, or training remain high in better-performing clusters, highlighting limited opportunities for young people, while at the same time compliance with labour rights is low. These trends underscore that, despite the global success in reducing extreme poverty and expanding access to basic services, persistent gaps in equity, decent work, and education-employment mismatch remain unaddressed.

3.6 Environment

Over the past decade, among six transitions, the world experienced the most severe regression in environment. In 2015, forty-seven countries were classified as environmental leaders (“achieving” cluster), characterized by low disaster impacts, strong conservation of marine and forest resources, and effective air pollution management (Figure 11). Yet by 2025, none remained in this top-performing category. This negative shift is mainly associated with increases in GHG emissions, impacts from natural disasters, resource-intensive production, and a larger share of GDP allocated to fossil fuel subsidies. Six countries with the most significant negative shifts are Azerbaijan, Liberia, Panama, Saint Lucia, Thailand, and Venezuela. Mali and Ukraine, by contrast, managed to make a significant shift from lagging in 2015 to performing in 2025 (Figure 12).

Geographically, pockets of positive shift are evident in the African continent, with thirty-two countries, primarily low-income countries in Sub-Saharan Africa (such as Burkina Faso, Ethiopia, Ghana, and Mali), that managed to advance from “emerging” to “performing” clusters. The opposite has occurred in many countries in South America, with Bolivia being an exception. The positive progress in Sub-Saharan countries is largely driven by advancing environmental conservation practices.

The two largest economies in the world, the United States and China, have made a negative shift from emerging cluster in 2015 to lagging in 2025, a group categorized by the highest levels of air pollution, food loss, fossil fuel subsidies, and GHG emissions. This is particularly consequential because the environmental trajectories of these two economies carry disproportionate weight. Together, they shape global consumption and production, trade, and energy dynamics, thereby influencing the world's overall ability to meet climate and sustainability targets.

3.7 The role of national income and cross-transition interdependencies

Beyond the transition-level patterns described above, the analysis sheds light on the interdependencies among transitions and with economic development. The correlation analysis (Table 2) shows that GDP per capita is strongly associated with digital connectivity, education, food systems, welfare and social protection, with digital connectivity showing the strongest correlation in both 2015 and 2025. Nevertheless, energy and environment transitions stand out as exceptions, showing consistently weaker correlations with income level in both years. Notably, the correlation for environment weakened substantially from 2015 to 2025, while the correlation for energy remained moderate.

Results from regression models (Table 3) reveal patterns in how performance levels in 2015 impact subsequent shifts across transition areas over 2015 and 2025, and how different transitions are interconnected. The first finding is that, with the exception of energy and welfare and social protection, countries that started from a lower performance level in 2015 tended to experience more pronounced positive shifts across the remaining transition areas over the decade. This suggests a degree of catch-up dynamics at play for the four transitions.

National income emerges as a particularly nuanced driver. While higher-income countries were more likely to achieve larger positive shifts in digital connectivity, the reverse held true for the environment transition, where on average, countries with lower income recorded comparatively greater shifts. This divergence cautions against treating income as a uniformly determining factor, and instead highlights how its influence depends on the nature of the transition, or is mediated by other development gains, as explained below.

Education itself presents a distinctive profile. Positive shifts in this area are not directly associated with 2015 performance in any other transition. But together with national income, it may act as a background enabler through advancing digital connectivity. Digital connectivity appears to be a key catalyst for the indirect effects of income and education. Stronger performance in digital connectivity may amplify the benefits of higher national income and stronger educational achievements, channeling their influence into positive shifts in environment, energy, and welfare and social protection. And eventually, generating positive outcomes for food systems through gains in welfare and social protection. These results position education as a strong priority for investment that drives transformation, and digital connectivity as a potential leverage point in policy that can generate compounding benefits across multiple transition areas simultaneously.

OPEN IN VIEWER

Figure 8.

Map of education performance (2015 and 2025).

OPEN IN VIEWER

Figure 9.

Performance on welfare and social protection (number of countries by cluster, 2015 and 2025).

OPEN IN VIEWER

Figure 10.

Map of welfare and social protection performance (2015 and 2025).

OPEN IN VIEWER

Figure 11.

Performance on environment (number of countries by cluster, 2015 and 2025).

OPEN IN VIEWER

Figure 12.

Map of environment performance (2015 and 2025).

OPEN IN VIEWER

Figure 13.

Mapping interlinkages between SDGs and transitions.