A fattier diet,
a fatter country.
Smaller than it looks.

Backlog idea #80 names the standard nutrition-transition claim: as national diets shift toward more fat (and sugar) and away from staple carbohydrates, obesity rises. The idea names sugar specifically; no OWID/FAO grapher slug publishes sugar supply per capita as its own series (checked six plausible slugs, all 404) — substituted the two macronutrient shares that ARE published this way and that do carry sugar inside them, dietary energy share from fat and from carbohydrates, both FAO Food Balance Sheets re-served by Our World in Data, keyless, each file already joined by OWID itself to World Bank GDP per capita (PPP). See fetch_dietobesity_573.py for the wall and the substitute.

The naive relationship is real and large. OLS, obesity prevalence on fat share of daily dietary energy, 183 countries, 2022: slope +0.836 obesity-prevalence points per 1-point rise in fat share, 95% CI [+0.645, +1.027], p<0.001, R²=0.292 — excludes zero, agreeing under HC3-robust SE (CI [+0.652, +1.020]), a distribution-free Spearman rank correlation (ρ=0.548, p<0.001), and a 4,000-draw bootstrap (CI [+0.662, +1.027], 100% of resampled slopes positive). The complementary macronutrient runs the mirror image, as it must when fat, carbohydrate, and protein shares sum to 100%: carb share alone, slope -0.750, CI [-0.914, -0.586] — excludes zero, negative.

Neither obvious confound moves it much on its own. GDP per capita is the first suspect: richer countries eat more fat (controlling for nothing, log(GDP) alone predicts fat share with slope +5.01 points per log-GDP-point, CI [+4.22, +5.80], R²=0.47) and richer countries are more obese (log(GDP) alone predicts obesity, slope +4.43, CI [+2.89, +5.96]) for reasons that have nothing to do with fat causing obesity directly — more sedentary work, more calorie-dense processed food, less subsistence labor. Controlling for it: slope +0.781, CI [+0.516, +1.045] — barely moved from naive. Separately, 8 small Pacific nations (French Polynesia, Kiribati, Marshall Islands, Micronesia (country), Nauru, Samoa, Tonga, and Tuvalu) are this sample's most extreme outliers — Tonga and Nauru both post obesity prevalence above 70%, a genetic-and-food-system story specific to the region, not a general fat/obesity mechanism. Excluding them alone: slope +0.704, CI [+0.539, +0.869] — also barely moved.

Doing both at once is where the shrinkage shows up. GDP-controlled AND Pacific-excluded, the strictest cut this run fits: slope +0.338, CI [+0.100, +0.577], p=0.006 — still excludes zero, but at 40% of the naive slope's size. The two confounds barely overlap individually (each alone leaves the estimate close to the naive figure) but compound when applied together, because the Pacific nations are disproportionately responsible for whatever of the naive slope GDP alone does not explain. The same strictest cut on carb share agrees in mirror image: slope -0.319, CI [-0.535, -0.103].

Named: Burundi posts the lowest fat share in the sample (8.2%) at a 4.2% obesity rate; Slovakia posts the highest fat share (46.0%) at 29.0% obesity. Vietnam posts this sample's lowest obesity rate (2.1%); the United States sits at 41.3% obesity on a 41.3% fat share. Read plainly: a fattier national diet does track a fatter population at every cut this run fit, including the strictest one — but the naive cross-country slope oversells how much of that is fat itself rather than the income level and the small set of extreme-outlier nations riding along with it.