Convergence Without Catch-Up: Population Weighting, Regionalisation and the Determinants of Global Diabetes Growth, 2003–2022
Keywords:
Diabetes Prevalence, Global Health, Population Weighting, β -Convergence, Kitagawa Decomposition, Who Global Health Observatory, Measurement DisagreementAbstract
Diabetes prevalence is one of the few major non-communicable disease burdens still rising in most countries. How fast it is rising, and whether the world's countries are converging or diverging depends on analytic choices that are rarely stated. Using WHO Global Health Observatory estimates for 196 countries over 2003–2022, together with World Bank covariates, we show that the headline number is weighting-dependent in a way that reverses over time. The unweighted country mean rises from 10.01% to 13.80% (+3.79 pp); population weighting raises the 2022 level to 14.61% and raises the estimated annual trend by 42%, from +0.194 to +0.275 percentage points per year. The epidemic is also becoming more regionalised — the between-region share of cross-country variance rose from 0.340 to 0.416 (+0.0052/yr), and the high-to-low regional gap widened from 7.8 to 14.6 percentage points. In a multivariate model of the 20-year change across 179 countries, the dominant correlates are the change in obesity (+0.266 pp per pp, p < 0.001), catch-up from a high initial level (−2.13, p = 0.002) and baseline health spending (−1.91, p < 0.001); the full specification explains 40% of the variance, against 0.1% for initial prevalence alone. The convergence evidence is specification- dependent in an instructive way: countries show significant relative convergence in the Barro–Sala-i-Martin growth form (λ = 1.1– 2.3%/yr, all p < 0.001) while showing no absolute convergence and widening absolute dispersion (SD 5.51 to 6.44). A rate-versus- composition decomposition shows that population ageing accounts for 43% of Europe's crude rise but only 6% of Africa's, and the male – female gap reverses sign across regions in a manner that itself depends on weighting. Finally, WHO/NCD-RisC and IDF estimates, though strongly correlated (r = 0.71), disagree on the global level by about 3.6 percentage points. We argue that unweighted country-mean summaries of global diabetes, which remain common, systematically understate both the level and the pace of the epidemic and misplace its centre of gravity.