22 SCIENCE AND FRUIT GROWING 



With the leaf -size, the probable error, as determined from 

 measurements of these same trees in the same season, was i 16-7 

 per cent, for one tree, or 6-3 per cent, for the mean of six trees, 

 and independent observations showed that, of this latter error, 

 2-2 units were due to errors consequent on having to select certain 

 leaves, for that was the probable error between repetitions of the 

 measurements of the same plots. 



It will be noticed that the errors of the measurements of leaf- 

 size, length of shoots, and crops (in a good cropping season), 

 do not differ much in magnitude 6-3, 8-3, and 7-1 per plot and 

 as the length of shoots shows much greater variation under 

 differences of treatment than the leaf -size, a measurement of 

 the former is generally preferable to that of the latter. 



It does not seem possible to combine the evidence derived 

 from the growth measurements of a tree with that based on its 

 cropping into one general expression, and these features must be 

 treated separately; indeed, they are to a considerable extent 

 opposed to each other, for heavy bearing acts as a check on wood- 

 formation, and vice versa. 



In all experiments with plants, one individual is liable to behave 

 anomalously owing to accidental circumstances, such as insect 

 attacks, etc., and if the number of plants in a plot is small, as in 

 experiments with trees, such exceptional trees must be eliminated 

 in dealing with the results. It is objectionable, of course, to have 

 to exercise any selection in dealing with experimental results, but 

 it cannot always be avoided, though it is rarely needed. Care, 

 however, must always be exercised in examining the results of the 

 check plots before accepting them ; if the behaviour of the trees 

 in these, where presumably the treatment is of a normal character, 

 is not such as general experience would lead one to expect if the 

 trees, for instance, are sickly and show little growth it is probable 

 that some adverse conditions have prevailed, and it would not 

 be safe to accept these trees as standards of comparison for the 

 others. It has been a matter of observation at Woburn that the 

 normal plots often appear to be more subject to accidental varia- 

 tions, and to show greater differences between themselves, than 

 plots where some special treatment has been adopted. This is 

 not altogether surprising ; most forms of special treatment tend 

 to make the plants behave specially well or specially badly, and 

 in such cases, chance conditions would have less effect than in 

 the normal plots, where there was no special stimulus in one 

 direction or the other. This source of error can be guarded 

 against only by multiplying the number of these normal plots, 



