Statistical and epidemiological methodology for sheep research: The needs, the problems, the solutions

There are many rewards for researchers who involve a statistician from the very inception of their research project. Benefits include optimized study designs which result in the use of fewer experimental animals, more effective answers to the questions of interest and an increased probability of successful publication. However, many veterinary research projects do not involve statisticians at all, or only do so at a late stage; this is reflected by many errors in project design and data analysis among published papers of veterinary interest. This situation is partly due to the absence of communication between clinical researchers and statisticians, perhaps because neither party understands how the other works. A solution might be to educate each group on what the other has to offer, perhaps by some form of interdisciplinary workshops. Clinical research with sheep is often constrained to use small samples. This places additional strains on traditional Frequentist statistical analyses: the results tend to lack statistical power, so there is a high likelihood that they will not be statistically significant. In addition, it is difficult to combine the results of more than one study. Many of these difficulties could be overcome if the analysis used a Bayesian approach. This would require changes in the way that researchers and journal referees think, but once this inertia has been overcome Bayesian statistics may well become the methods of choice. © 2008 Elsevier B.V. All rights reserved.