VALUE OF THE BRITISH ASSOCIATION UNIT OF RESISTANCE. 
321 
Date. 
Temperature. 
Frequency by 
observation. 
Frequency by 
calculation. 
1881. 
13° 
128T80 
128-182 
1881. 
14°-6 
128-161 
128-160 
October, 1882 . 
15°-98 
128-141 
128-140 
October, 1882 . 
17°-45 
128-122 
128-120 
October, 1882 . 
17°-6 
128-119 
128-118 
October, 1882 . 
17°-3 
128-117 
128-122 
Of the small discrepancies which the table exhibits it is probable that the larger 
part is due to imperfect knowledge of the actual temperatures of the standard fork. 
The use of screens to cut off radiation from the observers would probably have effected 
an improvement. For the highest accui'acy some sort of jacket, or chamber, would 
have to be contrived. 
Second Appendix. 
(Added July, 1883.) 
On the Effect of the Imperfect Insulation of Coils. 
In a former paper (Phil. Trans., 1882, Part II.) it was pointed out that the method 
of the revolving coil, employed by the first B.A. Committee, possesses the important 
advantage that it is possible to detect the existence of leakage from turn to turn, or 
from layer to layer, of the coil of wire. The general influence of such leakage, if 
undetected, upon the final number x expressing the ratio of the resistance of the coil 
when measured (R) in absolute units to its resistance r X 10 9 as referred to B.A. units, 
is easily seen by supposing that one turn of the coil is simply short-circuited. The 
formula in C.G.S. measure is 
R 7 dn 2 a co cot d> , . 
x= ,WToi>=-7VTo»-. (l) 
During the revolutions the short circuited turn produces its full effect in deflecting 
the magnet, and error arises only in the comparison with the standard of resistance. 
The quantity r will evidently be under-estimated by 1/n, and this will lead to an over¬ 
estimate of x, also by 1/n. This result, however, is modified, if as in practice we take 
only the difference of effects observed when the wire contact is open and closed. The 
short-circuited turn will produce its effect in both cases, and its influence will therefore 
disappear from the result. For all purposes it will be virtually non-existent, and the 
error produced is the same as if n had simply been miscounted. The final number x 
will thus be over-estimated by the fraction 2 /n. 
MDCCCLXXXIII. 2 T 
