to conclude that a causal relationship exists between an independent and a dependent variable. Because of subject heterogeneity, alternative explanations may exist for the observed experimental effects, which lowers internal validity. For example, research on the dietary treatment of hyperactive children has shown that the home environment can affect the results of a study. Children from an unsupportive home environment show much less improvement with dietary treatment than children from a more supportive home environment (Rumsey and Rapoport, 1983).

Finally, the external validity of experimental findings must be considered. For example, the behavioral effects of a nutritional variable, observed in male college students, and tested in a laboratory during a single test session, may have little to do with the behavior of the general public in its everyday lives. Researchers face a dilemma in trying to choose between a controlled but artificial laboratory setting and a "real" or naturalistic setting that may be full of confounding factors. "Quasi-natural" studies, which could capture the advantages of both the laboratory and the real world, should be considered (Kanarek and Orthen-Gambill, 1986).

Three primary strategies have been used in research on diet and behavior. Correlational studies have been employed for generating hypotheses about diet-behavior relationships. The major objective of these studies is to define a link between dietary intake and behavior, with the specific expectation that statistical associations will be derived between the two variables. This type of research can provide important insights for experimental evaluation of the diet-behavior relation.

There are, however, several conditions that must be met before the validity of a correlational study can be accepted. First, reliable and valid measures of dietary intake must be made. One of the most widely used approaches for assessing dietary intake is the 24-hour recall in which subjects are asked to record everything that they have consumed during the preceding day. However, because there are wide day-to-day variations in an individual's food intake, a 24-hour record may not provide an accurate determination of average food intake. As a result, it has been suggested that a minimum of seven 24-hour recalls be used (Anderson and Hrboticky, 1986).

Second, proper subject sampling techniques must be used. In general, the larger the number of subjects, the better. If the number

of subjects is too small, the probability of observing a significant relation between a dietary variable and behavior is reduced, and a false negative association may be assumed. On the other hand, correlation studies using large numbers of subjects risk the possibility of false positive associations. For example, when correlations are made between several dietary variables and behavior, the chance of achieving statistically significant results increases with the number of subjects and with the number of correlations made. In addition, when large numbers of subjects are used, small correlations can become statistically significant, making it necessary for the researcher to decide on the clinical importance of such results (Anderson and Hrboticky, 1986).

A common method of subject selection used in studies of diet and behavior involves the placement of media advertisements. Although this method is convenient, such sampling increases the probability of including self-selected members of the general population. For example, if a researcher wants to test the hypothesis that sugar influences hyperactivity in children and advertises the study as such, the subjects may be derived predominantly from families in which parents believe such an association exists.

Finally, cause and effect relations cannot be established from correlational data. For example, positive correlations have been reported between sugar intake and hyperactive behavior in children. These results have been interpreted by some (especially in the popular media) as demonstrating that sugar causes hyperactivity. However, it is just as possible that high levels of activity increase sugar intake or that a third unidentified variable influences both sugar intake and hyperactivity.

In contrast to correlational studies, experimental studies have the potential of identifying causal links between diet and behavior. The manipulation of a specific dietary component (the independent variable) may alter the occurrence of a behavioral measure or cognitive function (the dependent variable). Two major paradigms have been used in these studies. In dietary replacement studies, the behavioral effects of two diets, one containing the food component of interest (e.g. food additives) and the other not containing that food component, are compared over some period of time (e.g. two to three weeks). One obvious advantage of this method is that chronic dietary effects can be assessed. However, because making two diets equivalent in all factors except the food component being studied is difficult, it may be impossible to use appropriate double-blind techniques. Another limitation