THE EVOLUTION OF SCIENTIFIC THEORIES 259 



failure in the design or conduct of the experiment, such as improper 

 preparation of the experimental subject, inadequate control of ex- 

 perimental conditions, or instability of the observational devices? 

 Or may the error actually lie in some earlier, related determination 

 the experimenter has taken for granted? Despite the immense num- 

 ber of such possibilities, confirmatory investigations usually permit 

 reasonably prompt dismissal or substantiation of the challenging fact. 

 Suppose that fact is established as a fact: Conant's reminder then 

 comes aptly. 



. . . few if any hypotheses on a grand scale or conceptual schemes 

 can be directly tested. 



A whole chain of reasoning connects a conceptual scheme with the 

 experimental test. 



Defending the theory against the challenge of the fact, we may 

 begin by re-examining the linkage of the fact with the particular 

 colligative relation it calls in question. Have appropriate corrections 

 been made for all the complicating factors that ( though unmentioned 

 by the relation) may aflFect the observation, e.g., for the eflFect of 

 atmospheric refraction on the observed positions of some body be- 

 yond the atmosphere? When we make such corrections we "refine" 

 the denotations attached to our concepts, but we may also think to 

 overhaul these denotations more drastically. Will the supposed 

 equivalence of any alternate denotations involved bear re-examina- 

 tion? Can the denotations be so amended that the relation C fur- 

 nishes an amended prediction Cn concordant with what has been 

 observed, or even so that the relation can no longer be applied in 

 situation 11? In such amendments we have substantial freedom but 

 not complete license: we must not impair the previously confirmed 

 predictions drawn from C and other relations involving the same 

 concepts. We may indeed find no acceptable amendment. 



We will pass on then to reconsider the relation C itself. Is C less 

 generally applicable than we have supposed? Does it perhaps apply 

 only in certain "ideal" conditions of which 11 is not one? Is C perhaps 

 only the limiting form of a more complex law C which, within ex- 

 perimental error, yields elsewhere the same predictions as C, but in 

 situation 11 the prediction Cn? As long as predictions drawn from C 

 were well confirmed we will have been little disposed to question 

 its derivation, from theory (MF). But now we may well ask: have 



