1922. j 



Trials of New Varieties of Cereals. 



have been an actual error of about 8.3 per cent, in the com- 

 parison between the two races clue, not to racial characters, but 

 to soil and other conditions external to the plant. In other 

 words, the experiment would have been (as most variety trials 

 in the past have been) a combined trial of soil fertility and t ic 

 productivity, and it would have boon imp<^sil)!o to disentangle 

 the two. When, however, comparison was made of the w< i'^ht^ 

 grown on the 250 pairs of adjacent plots, it was found that the 

 probable error of the average difference between all adjacent 

 plots taken in pairs was reduced to less than half of 1 per cent. 

 If therefore 250 plots of each of two races had been planted in 

 alternating strips of plots across the field, it may be assumed 

 that whatever was the difference in the total weight of grain of 

 the two races, the probable error of this difference due to soil 

 conditions would have been only about half of 1 per cent, and 

 that if the actual difference amounted to 2 per cent., viz., four 

 times the probable error, it might safely have been regarded as 

 significant. 



It becomes a problem in statistical inference to estimate the 

 number of comparisons necessary with plots of any specified area 

 in order that the probable error under any fairly normal external 

 conditions shall not exceed, say, 1 per cent., and also to find the 

 best way of splitting up any available area into such plots. 



Consideration of the above results, and more especially of the 

 contribution by " Student," suggested to the writer a method 

 of yield testing for barley which after some early preliminary 

 experiments was commenced at Warminster m 1920 for field 

 tests of two different races. 



The method appears after two years' full (rial to be practical, 

 and also economical of area, and furthermore to be adapted to 

 reduce the probable error of yield trials of cereals to a very small 

 percentage of the observed weight of the crop. 



This method may be called the " half-drill-strip " method and 

 will be described in Part II of this paper with a summary of the 

 results so far obtained. It is in course of (rial by the National 

 Institute of Agricultural Botany this year for oats and barley. 

 Whilst it presents distinct advantages there are s'ill difficulties 

 to overcome and it can probably be further improved on. 



(To be concluded.) 



