NAASO 2000

NAASO 2000

Prof. Richard Forshee presented a poster, "Nutrition Research and the Ecological Fallacy," at the Annual Meeting of the North American Association for the Study of Obesity held in Long Beach, CA, October 29th through November 2nd, 2000.

Nutrition Research and the Ecological Inference Problem: Determining the Causes of Obesity and Poor Nutrition

Abstract:

Nutrition researchers often want to make individual level inferences based on trends in aggregate level data. For example, some people have pointed to the rise in the consumption of soft drinks and the concomitant increase in obesity rates to suggest that consuming soft drinks causes obesity. Similar arguments have been made to suggest that soft drinks is displacing milk or that added sugars cause obesity. As plausible as these inferences may seem, they are not valid. They are examples of the well-known statistical problem called the ecological fallacy, which has been widely recognized since William Robinson’s seminal 1950 article on the subject. It is not possible to determine the relationship between soft drinks and obesity based only on trends in aggregate level data.

It is difficult, some say impossible, to make sound inferences about individual behavior based on aggregate level trends. Some of the potential problems are relatively obvious. The observed relationship may be a spurious correlation, like the one that purports to show that when ministers' salaries are low the price of vodka rises.

But the heart of the problem is that we do not know which individuals are drinking more soft drinks and which ones are gaining weight. In fact, as I demonstrate with a hypothetical example, we could observe these aggregate level trends in soft drinks consumption and obesity even if the individuals who drink more soft drinks lose weight and the individuals who drink less soft drinks gain weight.

Nutrition researchers must be skeptical of individual level inferences drawn from aggregate level data. Inferences about individual behavior, such as eating choices and obesity, are best tested with individual level data. The best research designs for this type of question are cross-sectional survey analysis and panel studies that track the same individuals over time. Many of the most important nutrition questions incorporate an element of time, and the Departments of Agriculture and Health and Human Services should give serious consideration to developing a major longitudinal panel study to track the effects of diet over the course of an individual’s life.