Design and analysis of multiple choice feeding preference data

Traditional analyses of feeding experiments that test consumer preference for an array of foods suffer from several defects. We have modified the experimental design to incorporate into a multivariate analysis the variance due to autogenic change in control replicates. Our design allows the multiple foods to be physically paired with their control counterparts. This physical proximity of the multiple food choices in control/experimental pairs ensures that the variance attributable to external environmental factors jointly affects all combinations within each replicate. Our variance term, therefore, is not a contrived estimate as is the case for the random pairing strategy proposed by previous studies. The statistical analysis then proceeds using standard multivariate statistical tests. We conducted a multiple choice feeding experiment using our experimental design and utilized a Monte Carlo analysis to compare our results with those obtained from an experimental design that employed the random pairing strategy. Our experimental design allowed detection of moderate differences among feeding means when the random design did not.