428 THE SCIENTIFIC MONTHLY 



parts in reliable operation, or much time and accuracy will be lost 

 later on by breakdowns. Often times a slipshod arrangement is al- 

 lowed to stand with the hope that it will hold together during the test, 

 but nothing can be more discouraging than to find at the completion 

 of a long run that some little thing had gone wrong and rendered the 

 results useless. 



Perhaps it would not be out of place at this time to touch on the 

 subject of precision of measurements. It is believed that a lack of 

 understanding of this subject leads to a large amount of extra work 

 being done, and accuracy sacrificed, in many branches of research. 

 There is obviously no use in obtaining data, or of computing results, 

 to a much higher degree of accuracy than the least precise component, 

 and it is the neglect to find the precision measure of this component 

 that leads to much needless computation. On the other hand, it some- 

 times happens that if the least precise component were recognized it 

 could be obtained with greater precision, thus increasing the accuracy 

 of the whole experiment. It often happens, too, that certain factors 

 having only a slight influence on the results, are either recorded when 

 their effect is smaller than the errors in the other factors, or they are 

 neglected when that effect is larger than the errors introduced from 

 all other sources. For this reason, every condition that can in any 

 way effect the final results, should be carefully analyzed, not only to 

 determine whether it can safely be neglected, but to find out how closely 

 it need be measured if it can not be neglected. 



The successful carrying out of an experiment requires the constant 

 checking up of the data obtained, in order to detect an error before it 

 has invalidated a long series of runs. It is only by constant vigilance 

 that errors can be excluded from the work, and it is the ability to de- 

 tect irregularites that will cause future errors, or to detect the errors 

 themselves before they can cause trouble, that distinguishes the true 

 experimenter. For example, let us consider a certain test to determine 

 the effect of varying the aspect ratio of a model airplane wing in the 

 wind tunnel. The procedure consisted in making a test, then cutting 

 off a small length of the wing, and repeating the test, continuing the 

 process until the span of the wing was reduced to a small amount. It 

 would be extremely unwise in an investigation of this kind to collect 

 all of the values without working up the data and constantly com- 

 paring it with the preceeding results each time before cutting off the 

 wing. Otherwise, it might be found that one or more of the runs did 

 not agree with the rest, due to a lack of alignment or to some other 

 type of error that is apt to creep into any experimentation. After the 

 wing had been cut down, however, it would be too late to make a check 

 run and the whole test would be invalidated or at least made to appear 

 of doubtful value. 



