846 



Trials of New Varieties of Cereals. 



T.Tuly, 



about 10 per cent., and comparisons at this stage may afford the plant breeder 

 some slight but very uncertain indication of racial productivity. 



3. A chessboard of plots — each planted with the same number of seeds with 

 equal soil space per seed. This method has been followed by the writer since 

 1910 in nursery cultivations at Warminster, and has been adopted also by the 

 Irish Department of Agriculture and the Cambridge Plant-breeding Station. 



Where the plots are 16 sq. ft. at seeding time reduced to 9 sq. ft. for 

 harvesting by eliminating margins ; the plot of each race repeated 20 times 

 (giving about 2,000 plants of each race); the whole area protected from 

 birds ; the produce of each plot weighed ; the water-content determined and 

 the dry weight computed ; the probable error of the difference between the 

 total dry weights of any two races is found to be about 4 per cent. By this 

 method eight different races can be compared to this degree of accuracy on 

 less than ten perches of ground, when half a pound of seed of each race is 

 available. Estimates of comparative yield based on these nursery chessboards 

 are generally, but not always quite closely, confirmed by field trials of the 

 same races on similar soils in the same season. The results obtained on 12 

 such chessboards since 1910 are now being summarised for publication. This 

 method undoubtedly provides a means of sifting out those races which are 

 worth carrying on to the field trial stage. 



Proceeding now to systematic attempts to estimate the magni- 

 tude of the probable error to be expected in field trials, Wood and 

 Stratton, in the paper already referred to, from a very large 

 number of published results estimate that the probable error 

 of single plots of an} r size over 1 /40th of an acre is generally 

 about 5 per cent, of the produce of the plot. This gives a 

 probable error of 7 per cent, as that which may be expected 

 when comparing any two adjacent single plots. 



This is fully confirmed by Mercer and Hall in the paper already 

 referred to, and in addition there is described in this paper the 

 most elaborate experiment of which there is any record in this 

 country with the object of determining the number of plots 

 required to be averaged to produce a prescribed probable error. 



An acre of wheat of one variety selected for apparent 

 uniformity was divided at harvest into 500 equal rectangular plots 

 and the grain and straw from each plot was weighed. This was 

 a great undertaking — only possible at such a station as Eotham- 

 sted. The mass of figures was very thoroughly investigated on 

 statistical lines, not only by Mercer and Hall, but also by 

 " Student " in an Appendix in which a still more elaborate 

 statistical method was devised and shown quite clearly to lead 

 to a further reduction in the probable error. 



In this experiment the actual difference in the weight of grain 

 between the two half-acres on the east and west of the acre was 

 8.3 per cent. If, therefore, this acre of ground had been used 

 for a trial of two varieties — one single half-acre on the east and 

 one single half-acre on the west side — there would obviously 



