THE QUARTERLY 



JOURNAL OF SCIENCE 



JANUARY, 1873. 



I. ON THE PROBABILITY OF ERROR IN 

 EXPERIMENTAL RESEARCH. 



By William Crookes, F.R.S., &c. 



SCIENTIFIC man engaged in any special pursuit 

 has much difficulty in making clear, to even the 

 scientific public, the result of his experiments. The 

 benefit of his research may be perfectly apparent ; but if his 

 experiments have been conducted with rigour there will be 

 a certain individual departure from a general standard in the 

 results, which, if he merely state his conclusions, will 

 confuse the attentive reader. Perhaps certain of the ex- 

 periments were performed under better test conditions, and 

 their numerical results are therefore more nearly correct 

 than the results of another series of experiments. Should 

 this be the case, to take an average of the results would 

 yield an empirical result, deviating considerably from the 

 truth. Yet many of our most eminent experimentalists are 

 satisfied with recording their experiments, and leave the 

 student of their labours in an uncomfortable uncertainty as 

 to the exact value of the entire system of experiment. 



In his " Budget of Paradoxes," Prof. De Morgan, in the 

 consideration of the relation of facts to theory, asks the 

 question — " What are large collections of facts for ? To 

 make theories from, says Bacon ; to try ready-made theories 

 by, says the history of discovery ; it's all the same, says the 

 idolater; nonsense, say we ! " Whether, however, we take 

 facts in subordination to theory, or the reverse, matters 

 little for our present purpose ; we have to regard the ob- 

 servation of facts, ascertained experimentally or otherwise, 

 as the test of theory. But a difficulty immediately occurs 

 to the experimentalist, and may be framed in the question — 

 " Which, and how much, of these experimental facts am 

 I to regard as correct, absolutely or approximately ? " 



Absolute correctness evidently may not be expected of 

 any of the human senses, since the absence of error would 



vol. in. (n.s.) b 



