( 86 ) 
Two methods have been applied to derive the temperature-coeffi- 
cients. In the first method I used the deviations of the monthly means 
from the yearly means, while in the second the deviations of these 
same monthly means from approximate values of the term a, i.e. the 
non-periodic part of the rate, were used. These values were derived 
from a curve which represents as nearly as possible the yearly means 
for years beginning with May, with August, with November and with 
February '). 
These two methods gave the following series of corrections, headed 
I and II respectively, which must be applied to the value — 0.50268 
of the temperature-coefficient. They are expressed in units of one 
tenthousandth part of a second. 
I II I II 
Be dG ee a a (889 A: at AD 
Co ee NE Speers ae 1) 1800, A a oF 
Ts uae ey Sipe Bad LEDE oe ee 
ABD - 1. 70, “2 68 1892 + B + 98 
re ae ae Ba 1893 + 52 1+ 46 
Ien Mags mc 1894 14+ 6 + 64 
885 VE Dea 1895 + 9 +96 
188615 32 ee 1896 +1402 + 96 
agen, Erg Pe is Sy aoe NE DE 
te MF 
The two methods thus give practically the same results. Although 
this agreement is of course not a measure of the real accuracy of 
the corrections found, it is nevertheless evident that the temperature- 
coefficient has not been constant during the whole period, but that 
the adopted value requires a positive correction as well in the first 
as in the last years, the latter being the most marked. 
If the whole period is divided into three parts, we get the following 
mean results, according to the second computation (those of the first 
method are nearly the same) : 
18791884 A c=—= + 38 
1885-1892 de 
1893-1897 S78 
}) In first approximation it was assumed that e.g. the mean for the year from 
"78 May to ’79 April gave the value of a for ’78 Nov. 1. Afterwards these 
values were in some cases slightly altered. 
