Addressing Interconnected Global Crises Through Coordinated Economic Development

Between 2000 and 2020 the mean low-income country gained 10 years of life expectancy (52.6 to 62.4), 26 points of secondary enrolment (21% to 47%), 24 points of electricity access (15% to 39%), 5 points of clean cooking (8% to 13%), 14 points of water and sanitation (36% to 50%) and 16 points of internet use (0.1% to 15.7%). The mean high-income country was already near the ceiling on all but digital use (21% to 88%). The gaps remain wide in 2020: electricity 40% against 100%, clean cooking 13% against 98%, internet 16% against 88%.

Progress has slowed in some systems and accelerated in others. Mean progress in life expectancy fell from 3.5 to 2.0 years a decade, in secondary enrolment from 10.9 to 4.6 points and in income from 28% to 12%, partly as the largest gaps closed first; progress in power rose from 5.8 to 7.3 points, in clean cooking from 5.8 to 6.3 and in digital use from 24 to 30 points. The systems with the largest remaining gaps in the poorest countries, clean cooking, power and digital use, are also the ones still accelerating.

These are separate institutions with separate budgets, which is the point of the framework. If progress in one system were independent of the others, sector planning would be efficient. The next section shows it is not: the systems are linked, in measurable and mostly positive ways, so the order and combination of investments matter.

Nine of the 42 ordered pairs are significant at the 5% level. The strongest are water and sanitation to education (0.46, t = 3.2) and to power (0.46, t = 3.5), education to power (0.33, t = 2.6), digital use to income (0.35, t = 2.6), water and sanitation to clean cooking (0.25, t = 2.9), income to clean cooking (0.23, t = 2.9), power to income (0.19, t = 2.2) and clean cooking to digital use (0.14, t = 2.5). The only significant negative is health to digital use (−0.18, t = −3.2), possibly a compositional effect of countries where life expectancy rose fast through disease control rather than infrastructure.

On average, water and sanitation has the largest outgoing effect (0.24), then power (0.11), income (0.10), clean cooking (0.08), education (0.06), digital use (0.06) and health (0.01). The systems most moved by others are power (incoming 0.22) and education (0.19), then income (0.12) and clean cooking (0.09). Health is almost isolated from the rest in both directions: its progress follows from disease-specific programmes and income more than from the infrastructure systems measured here, and it does not itself predict their progress.

Two mechanisms are plausible and cannot be separated here. Water and sanitation may be a marker of state capacity and public investment that also builds schools and power networks; or it may relieve constraints (time, health, women's labour) that let households invest in education and cooking fuel. Digital use may drive income through productivity, and income may pay for cleaner cooking. The coefficients say how development has co-evolved, not what an intervention would cause, and the calculator in section 6 is an illustration of the co-evolution, not a policy simulation.

Across 42 regressions some coefficients will be significant by chance, so the permutation test shuffles each source system's earlier progress across countries, keeping everything else fixed, and counts the significant pairs: the null mean is 2.0, the 95th percentile 5 and the maximum over 300 shuffles 7. The observed count is 9. The mean coefficient, +0.095, is 7 standard deviations above the permutation distribution (standard deviation 0.013). The matrix as a whole is therefore unlikely to be noise.

Four other windows reproduce it: 2002–2008–2015, 2005–2012–2019, 2000–2008–2016 and 2004–2012–2020. The correlation of their 42 coefficients with the main matrix is 0.73, 0.57, 0.77 and 0.65; mean coefficients are 0.07–0.09 and 10–13 pairs are significant in each. Individual cells move between windows; the overall pattern of a positive mean spillover and a handful of strong pairs does not.

These checks address chance and window choice. They do not address confounding: countries with strong states, stable politics or aid inflows may progress faster in everything, and the controls (own momentum, starting level, income) reduce but cannot remove that. The honest reading is that the co-movement is robust and its direction is suggestive, with temporal precedence as the only identification. Policy conclusions should be hedged accordingly.

For each of 124 countries with all six social systems (health, education, power, clean cooking, water and sanitation, digital), each system is converted to a percentile rank in 2020; the average and the standard deviation of the six define level and imbalance. The median country has an imbalance of 9 percentile points and the 90th percentile country 18. It is highest in the middle of the development range (upper-middle-income mean imbalance 14, against 10 for high-income, 9 for lower-middle and 7 for low-income countries), as some systems race ahead: The Gambia, Egypt, Algeria, Sri Lanka and South Africa are the most lopsided; Benin, Uruguay, Liberia, Chad, Myanmar and India the most even (several of these at low levels across the board).

The regression for life expectancy against its 2015–19 trend in 2021 has a coefficient of +0.67 on the average level (standard error 0.26), −0.33 on imbalance (0.09) and −0.59 on log income (0.25), with R² of 0.13; for composite resilience in 2020 they are +0.57, −0.16 (0.09) and −0.60. The imbalance effect is therefore well determined for survival (t ≈ −3.6) and borderline for composite resilience. For growth lost to shortfalls, growth volatility and the disaster death rate, the imbalance coefficients are 0.01, 0.01 and −0.06 (all within 0.10 standard errors of zero), so balance does not matter for those.

A mean-versus-weakest-link alternative favours the weakest link for survival (coefficient +0.94, standard error 0.32, against −0.38 for the mean), but the weakest link and the mean are correlated 0.96, so the models cannot be separated with 122 countries. The pattern fits the idea that a health shock exploits the weakest system, such as water, power or access, but it is one episode. It supports the framework's second principle: coordination means raising the laggards, not only the average.

The weakest of six systems is education in 29 countries, power in 29, clean cooking in 21, health in 18, water and sanitation in 14 and digital use in 13. By people the picture is different: digital use is the weakest system for 31% of the 5.9 billion (large countries such as India), health for 21%, power for 17%, education for 16%, clean cooking for 7% and water and sanitation for 6%. Counting countries and counting people give different answers, which is why the table in section 7 reports both the weakest and the highest-leverage bottleneck.

Leverage multiplies the shortfall (how far a system lags the country's own average, in percentile points) by the system's mean outgoing spillover from section 2. On that measure water and sanitation is the bottleneck for 47 countries and 29% of the people, power for 27 (21%), clean cooking for 22 (7%), education for 18 (11%), digital use for 8 countries with 28% of the people and health for 2 (4%). Health is almost never the priority because its measured spillovers are small, although it is the end-goal of the other systems; that is a statement about leverage, not about importance.

The calculation gives a starting point for sequencing, not a prescription. A country where water and sanitation is the bottleneck has a lagging system that, in the past, was followed by faster progress in education, power and clean cooking. Where the bottleneck is digital use, which is a leading system for income, the case is for connectivity as a development input. The calculator in section 6 lets the reader push any system in any country and see the induced progress.

The calculator applies the standardised spillover matrix. A push of z standard deviations to the chosen system induces, one decade later, progress of M(i,j) × z standard deviations in each other system j; further rounds feed the induced progress back through the matrix. Standard deviations are converted to natural units with the spread of each system's 2010–20 progress (health 2.7 years, education 10.0 points, power 11.9, clean cooking 8.3, water and sanitation 5.6, digital use 15.5 and income 22 log points). Levels are capped at 100 and the percentiles recomputed against the other countries.

The balance effect uses section 4's regression: the change in the average level and in imbalance (in standard deviations of their cross-country spread) is multiplied by the coefficients +0.67 and −0.33, and by the spread of the resilience outcome (about 1.5 years of life expectancy), to give an indicative effect on survival resilience in an episode like 2021. It is an illustration: it assumes the estimated association is causal and linear, which it is not known to be.

Reading the output: pushing water and sanitation by one standard deviation (about 6 points of access) in a typical country induces roughly 0.46 standard deviations of progress in education and in power, 0.25 in clean cooking and 0.14 in income, a total of about 1.5 standard deviations of induced progress (0.13 in digital use and 0.02 in health); pushing health induces almost nothing. The differences between systems are the framework's practical content: it says where a given effort has been associated with the largest co-evolution.

Plan across systems, not within them. Progress in water and sanitation, power and income has been followed by faster progress in education, power and clean cooking; health progress has not been. Development plans that treat sectors as independent miss the spillovers the data show. A national plan should carry a spillover map and name the leading systems: the first steps with the largest co-evolution.

Find the bottleneck, not the average. The weakest-link and balance results suggest that survival resilience suffered where systems were lopsided. A diagnostic that reports the six systems' percentiles, the imbalance, the weakest system and the highest-leverage bottleneck, as the atlas does for 124 countries, is a better planning tool than a composite index. The bottleneck differs by country and, when weighted by people, by size: for large countries digital use is the weak link, for 47 countries water and sanitation is the lever.

Coordinate finance and delivery, and test. The coordination dividend, if it exists, is in sequencing and joint delivery: one utility connecting schools, clinics and homes; one investment plan covering power, water and connectivity. Whether coordination causes the co-movement is unproven; pilots with a comparison group, such as joint water, power and school programmes, should test it. The framework supplies hypotheses with measured priors, not a guarantee.

Sample: 160 countries with at least one million people (2022–23) and a World Bank income group; 124 have all six social systems in 2020. Indicators: life expectancy; gross secondary enrolment (capped at 100); electricity access; access to clean cooking fuels and technologies; the mean of basic drinking-water and basic sanitation access; internet users per 100 people; log GDP per person (PPP, constant 2021 international $). Values are the observation nearest to 2000, 2010 and 2020 within ±2 years (±1 for internet and income). Progress is the difference between years (income in log points × 100).

Spillover regressions: for each ordered pair (source i, target j) across countries, the standardised 2010–2020 progress of j is regressed on the standardised 2000–2010 progress of i, the standardised 2000–2010 progress of j, the standardised 2010 level of j and (unless j is income) standardised log income in 2010; the coefficient on the source is the spillover. Targets (other than health and income) with a 2010 level of 92% or more are excluded because of the ceiling. Permutation test: the source progress is shuffled across countries 300 times and the number of pairs with |t| ≥ 2 and the mean coefficient recorded. Alternative windows: (2002, 2008, 2015), (2005, 2012, 2019), (2000, 2008, 2016) and (2004, 2012, 2020). Balance: each system's 2020 level converted to a percentile among the 124 countries; level = mean, imbalance = standard deviation; outcomes regressed (standardised) on level, imbalance and log income.

Bottleneck: shortfall = the system's percentile below the country's average (zero if above) times the system's mean outgoing spillover (zero if negative); the largest is the highest-leverage bottleneck. The calculator applies the matrix to the chosen push, adds second and third rounds, converts to natural units with the 2010–2020 progress spread, caps levels and recomputes percentiles. Caveats: observational, with no instruments; confounding by state capacity and overall growth; 46–152 countries per cell; multiple comparisons (addressed by the permutation test); ceilings; indicator coverage and quality (secondary enrolment is gross, internet is use not quality); imbalance and survival are one episode; the calculator is illustrative.