10 



were not used, nor those of more than 75 per cent attack, for the reason 

 given above. (N.B. 75 per cent apparent attack already equals 

 89 per cent true attack.) It should be unnecessary to remark that the 

 percentages indicated are apparent, not true. These figures have been 

 correllated to the number of worms required to produce the same 

 apparent percentage. Six worms maximum, five and four were 

 tested. The best result was obtained for four, r=0-9906+0-0013, 

 which indicates practical identity of the two series. 



Another result is also so suggestive as to be probably based on 

 fact. It will be noticed that the loss expressed as per mille gives 

 practically the same figure as the number of worms required to produce 

 the infection. The totals of the two columns are nearly the same, and 

 the individual pairs of figures correspond closely to each other. This 

 gives a possibility of completing the loss figures given on Table IV, 

 remembering that the percentages now under consideration are of the 

 select population and not of the general total population. The two 

 losses are combined as loss on total population, which is the only fair 

 way of quoting in the last column of Table IV. 



In examining any tables on which the percentage attacked bolls, 

 or the reverse, percentage sound bolls are used, whether in connection 

 with time or any other factor, it is useful to remember that all points 

 on such a table do not stand with the same firmness as established 

 facts. 



The size of the sample to be used is of very great consequence, 

 if errors are to be avoided and false conclusions are to be evaded. It 

 is obvious, in order to obtain satisfactory evidence when the infestation 

 is one or two per cent only, that samples must contain several multiples 

 of the numbers required to produce one or two per cent infestation. 

 In our work we have early recognised this, and most of our samples of 

 green bolls have been made up of at least one thousand bolls. Even so, 

 it is probable that the low end of the infestation tables are likely to 

 have a somewhat high probable error. 



The nearer samples approach to 50 per cent attack, the more 

 accurate the results for the same number of bolls. Thus the middle 

 positions of the tables may be taken to be accurate to a higher degree 

 than the lower ends. 



Inaccuracy such as is due to defective sampling observable at the 

 low end of the tables returns at the higher end, but in much more 

 obvious form. Indeed it may be open to doubt whether results obtained 

 from one thousand bolls are usable for infestation records of over 

 80 per cent. At the low end of the table one worm more or less per 

 hundred bolls produces a change of one per cent. At the high end 

 of the table there must be on an average four, five, or more worms per 

 infected boll. Obviously the chances of distribution of worms non con- 



