Power of test considerations for beef cattle experiments: A review

ABSTRACT The use of power tests in the planning and design of beef cattle experiments provides critical information on sample sizes necessary to detect treatment differences at a predetermined significance (α) level. Retrospective power analysis provides additional information about previous experiments that may be helpful in designing subsequent investigations. However, in retrospective power analyses, power is inversely related to observed significance level. Benefits of prospective and retrospective power analyses in beef cattle experiments are similar to those for other species; however, because of differences in the methods and conditions involved, considerations for the use of power test procedures are specific for beef cattle research. Retrospective power analyses were conducted on 78 published experiments and on two unpublished experiments. Experiments were compiled into categories that represented group (or pen) feeding, individual feeding, and metabolism studies. Estimated power in pen feeding experiments using randomized block designs (RBD, n = 30) was less than 0.80 for ADG and feed efficiency (FE), but not different from 0.80 for completely random designs (CRD, n = 4). Furthermore, estimated power was less for ADG than for FE in both design types. For individual feeding experiments using RBD (n = 4), power was not different from 0.80 for either ADG or FE; however, for CRD (n = 18), power was less than 0.80 for both ADG and FE. Power was similar for ADG and FE for both RBD and CRD in individual feeding experiments. In metabolism experiments, estimated power for nitrogen retention was less than 0.80 for Latin square designs (n = 20) but not for CRD (n = 4). Comparisons of power between experimental design types were likely influenced by the number of experiments involved. These results indicate that retrospective power in beef cattle experiments is affected by design type, and response variable measured.

Key Words: Beef Cattle, Experimental Design, Statistics

© 2004, by the American Society of Animal Science. All rights reserved.

J. Anim. Sci. 2004. 82(E. Suppl.):E214-E222

Implications

Prospective and retrospective power analyses are fundamentally different, and researchers should be aware of this to avoid misinterpretation of comparative power values. Differences in retrospective power in beef cattle experiments are affected by experimental design, the number of experiments in each design type, and the response variable under investigation. These data show a higher estimate of retrospective power for feed efficiency than for average daily gain in pen feeding experiments. Low estimated retrospective power for a given experiment, or for a group of experiments, may indicate inappropriate experimental design or inaccurate experimental technique, leading to high estimates of experimental error. Nonetheless, a properly designed and conducted experiment can produce a low retrospective estimate of power. Our results show that even in a population of experiments for which true power was relatively high, individual experiments could produce low estimates of power because estimated power is a random variable. Finally, estimated power may be low because true treatment effects are in fact small.