Refinement of Feeding Experiments for Milk Production 247 



2. The cows in a group should be of approximately equal individual 

 production. 



3. The cows in a group should be as nearly as possible of the same 

 age. There is no indication that cows of any particular age are more 

 variable than those of any other age, but for other reasons it seems desirable 

 to exclude heifers. 



4. Cows in a group should be in approximately the same stage of 

 lactation. There is apparently no increase in variability caused by 

 mixing cows in different stages of lactation, so long as the experimental 

 period is completed before the rapid decrease in production begins. 

 According to Woll (19 12), this rapid decrease begins at about the eighth 

 month, and hence, if a period of about five months is to be used, the cows 

 should have freshened not more than three months before the beginning 

 of the period. 



5. Only healthy, vigorous, normal, mature cows should be included. 



6. Groups should be made as large as possible without sacrificing more 

 important factors. 



In planning a feeding trial the experimenter should generally have some 

 idea as to the difi^erence in production that may be expected with the rations 

 used. In order to get results that may be considered reasonably certain, 

 the percentage difference in total production should be about three times 

 the normal variation due to individuality. Knowing the probable 

 coefficient of individual variation under the conditions of selection used, 

 the experimenter can determine how many animals he must include in a 

 group in order to get significant results. If this nimiber of animals is not 

 available, repetition of the experiment may be necessary. 



In some cases the experimenter may not have much of an idea as to 

 what differences in production he may expect. Even in this case a con- 

 sideration of the coefficient of normal variation will be useful, if only in 

 preventing him from conducting an experiment from which significant 

 results could not be obtained. 



In any case a statistical analysis of the results should be made before 

 any interpretation is placed on them. The relation of the percentage 

 difference in total production obtained by the experiment, to the coefficient 

 of variability to be expected with the groups as selected, gives the best 

 evidence of the trustworthiness of the results. 



357 



