WiNSLOVV : A STATISTICAL CRITERION FOR BACTERIA' 37 



obtained and tabulated, the first important thing Is to determine 

 the major groups included in the series. This may be easily done 

 by a general study of the correlation of different characters, with- 

 out any elaborate mathematical methods. In our work on the 

 Coccaceae we constructed a correlation table for each pair of char- 

 acters and noted the general coincidence of one with the other. 

 It must be remembered, in estimating the importance of the 



F 



occurrence of correlated characters, that their common presence 

 may be due to one of two causes. The characters may be corre- 

 lated simply because ancestral forms developed them both, under 

 the action of independent causes. Or, on the other hand, the 

 properties in question may not be really independent but subtly 

 bound up in the physiological balance of the organism so that a 

 modification of one leads to a corresponding change in the othen 

 In either case profound modifications which have altered the whole 

 center of gravity of the organism may rightly be considered of 

 generic or family rank. 



Within the genera, specific types may be defined by variations 

 in single or subordinate characters. 



The final test as to whether two related organisms deserve 

 recognition as species, or are only variants from a single type, 

 must be made by an examination of the curve of frequency for the 

 character on which their identity is supposed to rest. . If there is 

 a single center of frequency of occurrence it is fair to consider that 

 the type is a simple one, however variable it may be. Thus 

 Goodman (Journal of Infectious Diseases 5 : 421) has recently 

 made a study of the B, DiphtJicriae group, which indicates that 

 w^idely differing strains of these organisms belong to a single type. 

 If on the other hand two distinct modal points indicate two separate 

 centers of distribution each may well receive a specific name. In- 

 termediate cases will of course be found, in which the curve of 

 frequency has two peaks separated only by a shallow trough. In 

 such cases Davenport and Blankinship's criterion, to which refer- 

 ence has been made above, may be helpful ; although any such 

 arbitrary standard is hkely to be modified by enlarged knowledge. 



• - 



It may be maintained with reason that genera and species of 

 bacteria defined by the statistical method will not be invalidated by 

 the discovery that the characters on which they are based may be 



