Wealth and weight revisited: A Bayesian model averaging approach to the Obesity Kuznets Curve

Thu, 08 Oct 2026 00:00:00 +0000

The question

Does obesity eventually decline as a country becomes wealthier? The familiar Obesity Kuznets Curve suggests a rise followed by a fall. This paper asks whether that simple pattern survives when researchers consider many possible explanations instead of choosing a single statistical model.

The study covers 84 countries during 2000–2020. It combines age-standardized adult obesity estimates from the World Health Organization with economic, social, and environmental indicators from the World Bank. Obesity prevalence is the share of adults aged 18 or older with a body mass index of at least 30 kg/m². Income is GDP per person in constant 2015 US dollars.

What the analysis finds

Bayesian model averaging combines evidence across alternative models, weighting each by its support in the data. Country and year fixed effects remain in every model, while 61 other potential explanatory variables enter the model-selection process.

The main results support an inverted-N relationship: the estimated income component first falls, then rises, then falls again. All three income terms have a posterior inclusion probability of 1.00 under both prior specifications. The sign pattern also appears in the two-way fixed-effects estimates.

Income term	Fixed prior: posterior mean	Robust prior: posterior mean
Log GDP per person	−189.358	−187.464
Squared log GDP per person	21.620	21.405
Cubed log GDP per person	−0.804	−0.796

These are the Table 2 coefficients, estimated using 1,195 country-year observations. The terms must be interpreted together; a coefficient on its own is not the effect of a one-dollar increase in income. The models also identify robust associations with indicators such as forest area, renewable energy use, population density, and agricultural structure.

Reading the infographic

The image draws the income component directly from the published Table 2 posterior means for both priors. Each curve is centered at GDP per person of 10,000 constant 2015 US dollars, so the vertical axis measures a difference in percentage points, not the predicted level of obesity. The horizontal axis is logarithmic and spans the observed income range reported in Table 7. This centering changes neither the shape nor the turning points.

The curves show posterior means only. The paper reports marginal coefficient standard deviations, but a valid uncertainty band for the combined curve would require their joint posterior or covariance. The infographic therefore adds no band. It is an original reconstruction of the panel estimates, not a reproduction of the descriptive pooled curves in Figures 1–2.

What the evidence can tell us

The results describe associations, not causal effects. National averages cannot establish how an individual person’s weight responds to income. The low-income end of the curve relies on fewer countries and less reliable obesity surveillance. A check using predictors lagged by five years preserves the income terms and their signs, but cannot eliminate time-varying confounding or reverse causality.

The practical message is to look beyond a single income threshold. Economic development and obesity are linked in several stages, and higher income alone does not guarantee lower obesity.

Read the evidence

Table 2: compare the two Bayesian prior specifications and posterior inclusion probabilities.
Tables 3–4: examine the evidence for two turning points and distinguish within-country panel estimates from pooled cross-country descriptions.
Table 11: inspect the robustness check using predictors lagged by five years.

The open-access published article includes the methods, tables, and appendix. Its data statement says that the full study dataset is available from the corresponding author on reasonable request.

Publication and image sources

Restrepo Gómez, K., Du, H., & Mendez, C. (2026). “Wealth and weight revisited: A Bayesian model averaging approach to the Obesity Kuznets Curve.” The European Journal of Health Economics. Published October 8, 2026. DOI: 10.1007/s10198-026-01998-1. Online first; volume, issue, and pages have not yet been assigned.

The infographic is an original plot of numerical results from Table 2 (PDF page 5), using the income range in Table 7 (PDF page 13). The article is open access under CC BY-NC-ND 4.0. The publisher’s figures are not altered or reproduced here.

Model uncertainty | Carlos Mendez