32 SCIENCE (and common sense) 



results. Ordinarily we can achieve such control in the laboratory, or- 

 dinarily our results are reproducible, and ordinarily the laboratory 

 worker demands reproducibility of anything he is prepared to treat 

 as a result. Yet the demand for reproducibility is not an absolute one. 

 Particularly as regards certain observational results, we may waive 

 reproducibility in practice and accept the much weaker criterion of 

 reproducibility in principle. We do not expect to be able to repro- 

 duce the observations reported for the fall of a particular meteorite. 

 Even in the laboratory, studying cosmic rays we do not expect to 

 be able to reproduce in our cloud chamber exactly the "nuclear 

 event" previously photographed by ourselves or others. In both cases 

 what we observe depends too heavily on variables beyond our con- 

 trol: on the nature, the relative velocity, and the angle of incidence 

 of the intruding "particle." The "event" and the observations made 

 of it are then irreproducible. Yet we accept those observations as 

 facts because we conceive such events as recurrent, or reproducible 

 in principle. Judging that we know the relevant conditions, we agree 

 that the circumstances involved are very "special." And we main- 

 tain that— however improbable in practice— the event, and tlie ob- 

 servations made of it, would be reproduced if the special circum- 

 stances were reproduced in some second meteorite or nucleon. 



Confronting a particular indentation in the sand— an indentation 

 unique and irreproducible in detail— Robinson Crusoe sees it as a 

 significant datum only because he at once affiliates it with the re- 

 current class "footprints in the sand." Much the same common-sense 

 criterion of acceptability of data is active in the sciences. We accept 

 data on irreproducible things and events when we judge that they 

 belong with "objects of this class," "situations of this type," when we 

 feel that other exemplars of such things and situations are known to 

 us. Even when guided by the subtle sense of analogy with which we 

 are so fortunately endowed, such feelings, such judgments, remain 

 humanly fallible. Recognizing that sometimes we go astray in our 

 judgments of recurrent type and class, we seek ever, by establishing 

 experimental control of conditions, to achieve reproducibility in 

 practice. We acknowledge the less secure criterion of recurrence 

 only when we must. Yet often, even in science, we must. The astron- 

 omer accepts data on the unique trajectory of some particular me- 

 teor because meteors following analogous trajectories are recurrent. 

 And he is likely to reject— as "observational error"— data indicating 



