Thanks for the link! After Atwater’s time, there were several experiments that allowed scientists to adjust Atwater’s original fat, protein, and carbohydrate calorie values. However, when food manufacturers label a product, they simply count up how many grams of fat, protein, and carbohydrate are in their product and use these adjusted averages. They don’t weigh them differently depending on the food.

For example, they don’t account for the easily available fat in an almond cookie vs. the more-difficult-to-access fat in an almond. So the problem is not really how we arrive at our calorie-value averages. The problem is the assumption that a gram of fat in food A always has the same available calories as a gram of fat in food B. Often it does, but as the nut study showed, the error can be 20%-25%. (And this is before we account for the additional 20% of measurement error that the FDA allows.)