IMPORTANCE OF UNIFORM RESULTS. 223 



observation can be relied on as doing, renders it the 

 most certain resource in all physical enquiries, 

 where the discovery of a general law is desired. If, 

 for instance, we found, in ornithology, that twenty 

 out of twenty-three sub-families, particularly abun- 

 dant in species, could each be divided into seven 

 groups or genera, and that each of these subordi- 

 nate divisions was in itself circular, we should be 

 justified in believing the determinate number to be 

 seven ; because the preponderance of evidence sanc- 

 tions the conclusion, and leads us to believe that a 

 more extended analysis of other groups will produce 

 the same result. But if, in the remaining three, 

 equally abundant in materials, we can by no pos- 

 sibility make out more than five circular divisions, 

 we must either seek to equalise the results, or, if 

 that fails, abandon our first theory, and commence 

 anew. It will not be sufficient to argue that the 

 two missing types of these groups may be supplied 

 by future discoveries ; because such a singular co- 

 incidence, of two missing types in each of three 

 genera, carries on the face of it a high degree of 

 improbability. It will be remembered, also, we are 

 now supposing all the groups before us to be perfect ; 

 and, if perfect, then without any violent or pal- 

 pable interruption in the line of continuity ; in other 

 words, presenting no interval, wherein, if these 

 missing groups happened to be discovered, they 

 could be naturally inserted. Nothing, indeed, can 

 be easier than to start a theory on the universal 

 prevalence of a determinate number, assumed upon 

 the partial arrangement of one or two insignificant 

 groups, and without complying with the con- 



