56 MISSOURI AGR. EXP. STA. RESEARCH BULLETIN 49 



its value in ordinary variety testing would not be definitely estab- 

 lished. The practice involves not only the assumption that the yields 

 of the check plots are a fair indication of the productivity of the in- 

 tervening plots for the check variety, but the further assumption that 

 different varieties respond similarly to differing growing conditions. 

 Adjustment of yields should therefore give better results in such ex- 

 periments as those cited above than it could be expected to give in 

 actual variety tests. 



This point is well illustrated by observations reported by Salmon". 

 Two varieties of barley, Gatami and Odessa, were grown side by side 

 in fiftieth-acre plots in five distributed portions of a field. Gatami gave 

 an average yield of 18.3 bushels per acre, with quite uniform yields in 

 the five plots, as evidenced by their probable error of 0.68 bushel, 

 while Odessa yielded 13.3 bushels per acre in the first plot, 6.35 bushels 

 per acre in the second, and a negligible yield in the other three. Ob- 

 viously the adjustment of the yield of either of these varieties on the 

 basis of the other variety as a check, would enormously increase 

 rather than decrease the experimental error. As Salmon points out, 

 an error similar in kind though less in degree may occur commonly 

 in variety tests, when the yields of varieties are determined by dif- 

 ferent limiting factors. And if this is generally the case, adjustment 

 by check yields will be of doubtful value, even if it were found to 

 eliminate variability completely in uniform plot tests. 



There is a growing tendency, consequently, to discontinue the use 

 of check plots for adjusting yields in variety tests, and to use them 

 only to measure soil variability and to indicate the degree of error 

 in yield determinations of the tested varieties. Adjustment of yields 

 has never been as common in preliminary tests as in tests on larger 

 plots, principally because of the great amount of computation neces- 

 sary in adjusting the yields of ten or twenty replicate rod-rows of a 

 large number of varieties, and because the yield of a single rod-row, 

 exposed to varying competition and materially affected by small me- 

 chanical errors, is at best a very unreliable measure of productivity on 

 which to base the adjustment of the yields of several other plots. 



Experimental Results. It would of course be very desirable to 

 use check plots for reducing plot variability, if the method could be 

 relied on, because of the economy of the practice. The only certain 

 method of reducing plot variability is by means of replication, and it 

 may be considered a fairly general rule that the variability of plots 

 on a given field, as measured by the standard deviation or the prob- 

 able error, will in general be reduced by replication in proportion to 

 the square root of the number of replications. In other words, the 



