176 HENRY A. ROWLAND 



e = + -00003 , 

 / = + '00003 , 

 a + 'b + c + d + e +f= '00017. 



For A and B, log "=11-4566587 

 For A and C, log "=11-2882590 

 For B and C, log " = 11-1917176 



The distance of the mirror from the scale varied between 192-3 and 

 193-5 cm. 



Should we reject the quantity 34-831 in the third experiment so as 

 to make the mean result of that experiment 34-744 instead of 34-773, 

 we should obtain as a mean result of the whole 



34-7156 -0053, 



which has a less probable error than when the above observation is re- 

 tained. The number of plus and minus errors are also more nearly 

 equal and the greatest difference from the mean 1 part in 1100. 

 However the two results do not differ more than 1 part in 10,000. 

 We shall take 



R = 34-719 -007 earth - at 17' C . 

 second. 



as the final result. 



DISCUSSION 



On glancing over the table we see that the number of negative errors 

 greatly exceed the number of positive, but, if we take only the four 

 errors which are greater than 1 part in 5,000, we shall find two of them 

 negative and two positive. 



Combining the results with the different coils we have 



A and B .................... 34-696 -005 



A and C .................... 34-744 -Oil 



B and C .................... 34-716 -007 



Had we no other results to go by, we might suppose that the value of 

 M might not have been found as exactly for these coils as we have 

 supposed them to be. But if we include the preliminary results re- 

 jected on account of the imperfect circle used, we shall find 

 A and B .................... 34-704 -006 



A and C .................... 34-718 -017 



B and C .................... 34-758 -016 



which has the greatest error in an entirely different place. 



From the first series the probable error of each determination of M 

 is 1 in about 2,000. But as this includes the experimental errors which 



