VALIDITY OF ANALYSES 



35 



The mean is 1012, a difference of 1 per cent of the number actually known to be 

 there (1000). The mean variation is 22-4 per cent of the known number — or of 90 per 

 cent of the sub-samples it is 16-3 per cent of the known number. 



It is the possible range of variation which is important. We may discount 10 per cent 

 of the sub-samples as being ones which we can fairly assume would in practice have 

 been detected as being too high or too low when the pipette was used. Combining 

 these results with those of the other two series of sub-samples we may conclude that 

 in the majority of instances the error will not exceed 30 per cent, but occasionally the 

 error may reach 45 per cent. 



Calanus spp. 



The same sample made up of a known number of organisms, which we have just 

 referred to in the case of medium Calanoids, contained 1000 specimens of Calanus. 

 The sample was sub-sampled forty-four times by eight different methods. The numbers 

 estimated as being in the sample are as follows, according to the sub-samples taken: 



50 per cent of the sub-samples show a range of variation from the actual number in 

 the sample of from — 20 to + 18 per cent. 



75 per cent of the sub-samples show a range of variation from the actual number in 

 the sample of from 28 to + 30 per cent. 



90 per cent of the sub-samples show a range of variation from the actual number in 

 the sample of from — 52 to + 44 per cent. 



100 per cent of the sub-samples show a range of variation from the actual number in 

 the sample of from — 64 to 4- 125 per cent. 



Again we may say that exceptional cases (10 per cent of the sub-samples) would 

 actually in practice be detected when taking the sub-sample and the pipette emptied 

 and refilled. So that we may say that usually the error will not exceed 30 per cent, but 

 at times may reach 50 per cent. Where there are only a small number of Calanus in the 

 sample these would be picked out and counted separately. 



The mean of these estimated numbers is 919, being lower than the actual number 

 by 8 per cent. The mean variation from the known number is 28 per cent of the known 

 number, or if we eliminate the exceptionally high and low estimations (10 per cent of 

 the sub-samples) the mean variation is 17 per cent of the known number. 



