328 Prof. Spencer XL Pickering on Experimental 



discard the first hundred as consisting chiefly of preliminary 

 and imperfectly conducted experiments ; the next century of 

 determinations was also found to give an experimental error 

 decidedly higher than the rest ; these also I have therefore 

 omitted,, and will confine myself to the remaining 400 expe- 

 riments. Out of these, however, it will only be the ones 

 which were performed in duplicate which will be available for 

 the present purposes. 



Now the average difference between two such duplicate 

 experiments, performed on the same day and with one thermo- 

 meter only in the calorimeter, was found to be 26*6 calories 

 (deduced from 76 separate experiments), giving a probable 

 error in each experiment of +13*3 cal. ; this corresponds 

 to an error in the thermometric column of considerably 

 less than +'05 miilim. ; and as the results involve not only 

 four or five readings of the thermometer, but also the correct 

 reproduction of every action performed in dissolving the salt 

 during the intervals allowed for the determination of the rate 

 of cooling, the correct determination of the salt temperature, 

 and many other points where serious experimental error may 

 occur, I consider that I am fully justified in concluding that 

 the thermometers can be read to *05 miilim., and that even a 

 certain amount of reliance can be placed on such estimation 

 figures. 



Where the duplicate experiments were not performed on 

 the same day, the error was found to be greater than in the 

 previous case ; on an average, deduced from such experiments, 

 it amounted to ±19*6 cal. This number is, however, some- 

 what bigger than it should be for strict comparison with the 

 previous one ; for the experiments which yielded it belonged 

 chiefly to the earlier ones amongst the 400 under considera- 

 tion, and an increase of accuracy appeared to be made with 

 increased practice throughout the whole series. 



In what follows I deal with those duplicate experiments 

 which were performed consecutively only. We thus have so 

 far:— 



I. Difference of two experiments where one thermometer 

 was employed, 26*6 cal. 



Now in a large number of cases two thermometers were 

 used in the calorimeter at the same time ; from these was 

 deduced, 



II. Difference of two experiments where two thermometers 

 were employed, taking the results obtained with the same 

 instrument in the two experiments, 41*9 cal. (from 72 expe- 

 riments). 



Subtracting I. from II., w^e get 



