ii SCIENTIFIC RECONSTRUCTION 263 



We are always in contact with processes which escape 

 observation. How can we be sure (i) that in any case, 

 even when we appear to control the whole of the conditions, 

 A is the sole change introduced, (2) that the operation of 

 the concomitants BC or DE does not consist in inner pro- 

 cesses having points of agreement which we cannot detect ? 

 There are two possible answers. One is an appeal to the 

 theory of chances. It is against all probabilities that, if we 

 go on varying the circumstances of an experiment, we 

 should always light on specially favourable conditions. 

 The other is an appeal to verification. The generalisations 

 which we make on this basis conform to fact, and calcula- 

 tions derived from them, deductions, constructions, yield 

 results which observation confirms. I will confine my 

 remarks here to the second argument, and see whither it 

 leads us. 



That results conform to observation is not in itself any 

 proof of the principles on which the calculation is based. 

 But results may under given conditions be logically used 

 in the corroboration of principles. Suppose that from two 

 causes in conjunction we infer an effect. Then suppose it 

 established by an independent generalisation that those 

 effects actually follow from those causes when conjoined. 

 Let the simple causes be a and b, and their effects c and d. 

 It is clear that if we take one of the causal relations, say, a-c 

 as certain, and the other one b-d as less certain, the proba- 

 bility of b-d is raised. For if b-d did not hold true, then 

 the generalisation ab-cd could not be true, and of ab-cd 

 we have independent probable evidence. But further, if 

 a-c is itself not quite certain but only probable, still its 

 probability increases that of b-d. Again, a-c may be simi- 

 larly corroborated by a fresh conjunction, as ae-cf, where 

 e-f has an independent probability, and this process may go 

 on indefinitely. In point of fact, any well-developed 

 science does exhibit a network of such interconnected 

 generalisations, and these are the tacit, though seldom ex- 

 pressed, ground of confidence in the general adequacy of 

 the observational methods on which it rests. The con- 

 fidence in rational method is thus legitimately strengthened 

 by actual success in establishing rational interconnection. 



