136 PROCEEDINGS OP THE AMERICAN ACADEMY. 



creased with rise of temperature. This difference, if constaut at each 

 temperature, would be eliminated by combining two sets of observations 

 made at the same mean temperature, one set being made with spiral 

 No. 1 the warmer, and the other with spiral No. 2 the warmer. Sets of 

 observations were, it is true, combined in pairs, but in each such pair it 

 was the meau temperature of the disk that was kept nearly constant, 

 while the mean temperature of the spirals was seven or eight degrees 

 warmer in one set than in the complementary set. Accordingly, the 

 mean result of a pair of sets, taken without individual correction for the 

 inequality of the spirals, would have been subject to the error caused by 

 ignoring the variation of this inequality through a rise of, we will say, 

 8 degrees in mean temperature. This error would have been about 1.5 

 per cent of the final result at any temperature. To prevent such an 

 error, corrections, based on careful observations made for this specific 

 purpose, were applied in each individual set of observations for conduc- 

 tivity ; and, after these corrections had been applied, the complementary 

 sets were put together for a mean result at any given temperature of the 

 disk. By this means the error in question was probably reduced to very 

 small dimensions. 



It appeared likely that some part of the apparent difference of resist- 

 ance of the spirals at high temperatures was due to difference of resist- 

 ance of the copper connecting wires, which near the spirals and for some 

 distance away from the latter were considerably heated. As the experi- 

 ments went on, increasing care was taken to make the condition of the 

 wires leading to one spiral as nearly as might be the same as the condi- 

 tion of those leading to the other spiral, in order that the differential 

 changes of resistance should be confined to the spirals themselves. The 

 result was a progressive diminution of the correction for isothermal ine- 

 quality of resistance at high temperatures ; but this correction remained 

 large and somewhat uncertain to the end, so that the value of the con- 

 ductivity calculated from any one set of observations, unbalanced by its 

 complementary set, was liable to a considerable error, as the details 

 presently to be given will show. If I were to go through such work 

 again, I should try to reduce the importance of temperature changes in 

 tl>e copper connecting wires by increasing the resistance of the spirals or 

 the change of temperature of the stream between them. 



Heating and Flow of Water. 



The method of heating and controlling the flow of the streams of water 

 has been describe<l in previous papers. The only feature of importance 



