Many of the measurement tools researchers use in studies relating nutrition and disease just don't cut the mustard, says Raymond Carroll, distinguished professor of statistics at Texas A&M University.

Carroll, who serves on both the statistics and nutrition faculties, says the problem in nutritional epidemiology studies is sometimes the links between diet and disease are so subtle (if the links are there at all) that the methods researchers use to measure diet are simply not good enough to provide an accurate picture of a person's dietary habits.

"The big statistical problem," says Carroll, "is how to measure a person's diet because you can't remember precisely what you ate three days ago, much less for the last year or 10 years. Researchers have all these measures, which are supposed to be measures of what people have eaten, but they clearly have problems because it's very hard to recall what you eat."

One of these methods of measuring food intake is the diary in which participants write down everything they eat for a few days, off and on for several months. The problem with these diaries is that after a few days, people often change their dietary habits to make it look better on paper or they lie about what they eat, excluding from their report ice cream, candy bars or other fattening, unhealthy foods.

The most common measurement tool, though, is the food frequency questionnaire in which participants bubble in their answers like they would on a standardized exam. Food frequency questionnaires, so popular because they're cheap and easy to do, ask participants questions like, "How often do you eat pasta at lunch?" and "When you eat pasta, do you eat it with marinara sauce?" The problem with this, says Carroll, is that food frequency questionnaires only establish a person's usual eating habits as they recall them, not exactly what they ate.

Carroll set out to determine whether or not these questionnaires were any good at measuring diet. His research group obtained nutritional (in this case, protein) intake data based on urine measurements. The advantage to using this data is that it's a biologically based measure of nutrient intake not reliant on self-reporting.

"There are no recall problems," says Carroll, "and no problems of people not wanting to report the ice cream or the candy bar. You can't hide them because they show up in your blood or in your urine and we can measure it."

Carroll found food frequency questionnaires were many times less precise than anyone ever thought and developed statistical models to understand the properties of the questionnaires. Of particular interest was determining if some large studies are actually large enough to find a link between, for instance, fat intake and breast cancer.

"You think that 100,000 or 400,000 people is a big study," says Carroll, "but these questionnaires are so bad that that's just not large enough."

Carroll found that many of those studies should be increased by factors of about six before they can detect the subtle relationships between nutrition and disease.

"The big ones, like heart disease and fat intake, are so strong that the questionnaires are fine, but for things that are harder, like cancer, the questionnaires just aren't good enough. There may not be links between diet and breast cancer, but it's almost impossible to tell that with a frequency questionnaire.

"You can't measure things willy-nilly and hope to find anything. The whole point of statistics is to try to measure things better so you can get more definitive answers."

Contact: Raymond Carroll, (979) 845-3141.

Office of University Relations
Texas A&M University
(979) 845-4641

Kriewald Leslie

College of Science
517 Blocker
TAMU 3257 | 979-845-7361
Site Policies
Contact Webmaster
Social Media