BU-810-W~ SOME PROBLEMS IN ANIMAL EXPERIMENTATION by

A statistical consultant is faced with many unsolved problems in advis-ing on the design and analysis of animal experiments. Some of the problems arise because of inadequate or inappropriate statistical procedures. Statis-tical textbooks may treat situations which do not hold in animal experimenta-tion. Teachers of statistics classes omit many published techniques that could be useful to the animal scientist. One noteable example is the treat-ment of a statistical analysis for a Latin square design. An analysis of the type given by C. P. Cox (1958) would be more suitable than textbook analyses for a portion of animal experiments using a Latin square design. Repeated measures designs are rarely discussed and little understood. There is much more to repeated measures designs than construction or using linear model theory. The nature of the treatment effect and its duration are vitally important in making intelligent use of such designs. Also, statistical optimality may not be the only usable criterion in selecting a repeated measures design. Two particular examples are discussed with one raising problems of conduct of the experiment and appropriateness of the model. An example in which the nature of the experiment dictates the type of experiment design with no alternatives is given. The nature of the treatments dictate the dam as the whole plot unit and the sex within a litter as the split plot treatment. Some aspects of analyzing this experiment are discussed. In the last section of the paper a variety of problems encountered are discussed briefly. Among them are the unconditional analyses for unbalanced classifications, covariance, multivariate analyses, measurements, outliers, data bases, animal breeding and variance components, recommended levels, Waller-Duncan procedure, response plateaus, and lactation curves