Findings, implications and lessons learned

4. Experimental design is often an art in compromise between that which is statistically desirable and that which is experimentally possible. The biometrician has to be firm when he/she thinks that an experiment that the researcher proposes is not viable and should tell the researcher so. It should always be born in mind that an experiment that may be 'too small' when carried out alone can be replicated. The researcher also needs to be clear of the experimental objectives and what comparisons make sense biologically.

5. An important feature of this experiment is the large reduction in residual variance brought about by introduction of the pre-treatment covariate milk yield. When it is known that there are likely to be large individual variations in the primary response variable being measured it is important to consider either blocking for pre-experimental information or using this information as a covariate. Such variables in animal experiments can be milk yield, body weight, or, in animal health, a blood measurement of particular interest such as haemoglobin or packed cell volume.

6. The researcher chose Option 1 ending up with six goats receiving diet A + D and nine each of B and C. Unfortunately the technician made another mistake in the allocation of feeds and the experiment was terminated.