EMPIRICAL TOOLS AND EMPIRICISM 167 



the knowledge we seek from the superficially more obscure data 

 yielded by what might be described as controlled observation. 



The limitation of statistical methods. Undeniably powerful, the 

 statistical tools of thought provide us with notably important new 

 criteria for appraising the relevance or irrelevance of possible varia- 

 bles, and for distinguishing between "error" and the action of some 

 uncontrolled variable. But in neither case does statistical analysis 

 ever identify tliose \'ariables for us: always these can be suggested 

 only by our own hypotheses. Statistical analysis may reveal in our 

 data some subtle relation we might otherwise overlook. But such 

 analysis never supplants the insight of the investigator: always he 

 must supi^ly the concepts in terms of which alone the analysis can 

 be conducted and its results expressed. 



Like specific empirical tools, these explicit conceptual tools are 

 powerful aids to human thought and judgment. But never can they 

 stand in lieu of diought or eliminate the need for judgment. Simplify- 

 ing judgment at one level, they demand it more heavily at another. 

 Thus, for example, the whole opplicahilitij of statistical analysis de- 

 pends on a judgment that our experimental design suffices to pro- 

 vide results that will constitute a sample truly "random"— and 

 this is no light undertaking when, as is usual, we have access to 

 only a few results. Nothing frees the scientist from an omnipresent, 

 need for such judgments. And generally he makes these decisions 

 rather well, without even thinking of elaborate statistical analyses. 

 Quite clearly he relies then on other than statistical criteria— most 

 clearly when, as often he does, he overrides the statistical indications. 

 Thus he may attach great importance to data statistical analysis 

 would entirely discount: Meyerson, Planck, and many others have 

 commented on the complete inadequacy of tlie data from which 

 Mayer, Joule, and Colding boldly inferred the invariant equivalence 

 of heat and work. Contrariwise, the scientist may totally reject, as 

 meaningless, data which meet searchmg tests of statistical adequacy. 

 If Rhine's data on ESP met ( or meet ) all such tests, still most scien- 

 tists would reject them out of hand. A case perhaps parallel in point 

 is noted by Polanyi, who refers to 



... a table of figures published in Nature {146 (1940), p. 620) 

 purporting to show that the days of gestation of difi^erent rodents is 

 an integer multiple of the number tt, . . . no amount of such evi- 

 dence could convince us today that this relationship is real . . . 



