PoDul:2 



On the Continent, the float is usually searched in an enamel or porce- 

 lain bowl containing water, the cysts being collected from aroimd the 

 edge with a camfil hair brush by the unaided eye. With such a techniaue 

 there is considerable scope for personal errors. In Great Britain, 

 collection of cysts is usually accomplished from filter papers or from 

 the Fenwick tray with the aid of a binocular microscope. 



Control of Laboratory Errors 



Repeated covints of cysts from samples dra^^m from the same mixed bulk of 

 soil should fit the Poisson distribution, so that, whatever the extrac- 

 tion technique employed, its efficiency may be tested by a ■y.'- test 

 (Fisher 1938). In practice, the error variance generally exceeds the 

 expected Poisson variance because of imperfections in technique. Pro- 

 vided this difference is not too great, the technique may be regarded 

 as satisfactory. Because of the approximate fit to the Poisson distri- 

 bution, certain important consequences follow: 



(1) The accuracy of a count is determined solely by its 

 magnitude. Thus is x = the count, then Vx" = its 



standard error, and the degree of accuracy may be expressed 

 by the standard error as a percentage of the count. 



(2) For covmts of 10 or more, the Poisson distribution 

 approaches the normal distribution so that the ^S% range 



is given by x i 2SE. 



(3) Nothing is gained by estimating a series of small 

 samples and taking the mean. In fact, there are defi- 

 nite pitfalls to such a procedure. If a = the first count, 

 there may be a tendency to search until subsequent counts 

 reach the same magnitude and to cease searching thereafter. 

 This introduces bias and gives error variances significantly 

 less than expected. A case may be made, however, for the 

 estimation of two samples, the one as a check on the other 

 and intended to avoid gross mistakes. 



The implications of (1) and (2) above are illustrated in Tables 2 and 3. 



Table 2 shows the relative inaccuracy of small counts, the rapid in- 

 crease in accuracy up to covmts of 100, and the slower improvement 

 thereafter. Table 3 illustrates further the uncertainty connected x-dth 

 small counts. On a basis of 100 g. samples, certainty of detection is 

 not reached until the equivalent field population reaches a concentra- 

 tion of $0 to 60 millions per acre. This is of considerable importance 

 in connexion with the so-called certificates of freedom from infestation 

 already mentioned. The conclusion to be drawi from the infom\ation 

 shown in Tables 2 and 3 is that, for research, it is desirable to work 

 v:ith counts of the order of 100 or more and to vary the sample sise 

 accordingly, rather than to work vxith samples of fixed sise. 



