ARTICLE NO. FC970531 EDITORIAL What Are the Variances of Food Composition Data?
Questions I am frequently asked are: ''What is the variance of the composition of this nutrient in this food or diet?'' ''What ranges of composition can be expected in the content of this nutrient in this food?'' Most of the time I have to say, ''I do not know.''
The food composition table values for most foods and nutrients give estimates of the mean concentrations, but rarely estimates of variances for those concentrations. The 95% confidence limits are almost never given. In most cases no one has a clue as to the actual distribution of nutrient composition in foods. For most foods and nutrients the number of samples analyzed is quite low, usually less than 10. It is very rare that analysts have developed a probability-based sampling plan for the selection of the foods that were analyzed. The low numbers of samples and the lack of adequate sampling plans are primary reasons for the general lack of information on the variance.
This lack of information has led to the rather strange phenomenon that many end users of food composition data seem to think that there are no variances in food composition or that such variances are not important. But one of the more important uses of food composition data is to compute the consumption of nutrients by various populations. Nutrient consumption is computed by multiplying the consumption times the composition. The variance in nutrient consumption is obviously a function of the variance of both the amounts of foods that are being consumed and the composition of those foods. Food consumption data are usually acquired using detailed probabilitybased sampling plans, where sometimes thousands of consumers are polled as to their food consumptions. Thus, we know a great deal about variances in the consumption of food and almost nothing about the variances in the food compositions. Obviously, we are missing a critical component of the computation of variances of the mean consumption of nutrients.
The lack of knowledge is an even greater problem when the consumption of nutrients at the upper and lower percentiles of intakes is being estimated. When we have poor estimations of the composition variances, we have very poor estimations of the levels of nutrient consumption at the upper and lower percentiles of intakes. For most nutrients and most populations, the groups at risk are those in the upper or lower percentiles of nutrient consumption. Thus, we have the problem that the regions of greatest risk are the regions of potentially greatest error. We do not even know the impact that the variance in the food composition will have on our estimate of risk at the upper and lower percentiles of nutrient consumption. This is not good science.
We need more data on composition variances; then we can make more accurate estimates of nutrient intakes and of the potential risks of inadequate or excessive intake of nutrients.