EEPOKT OF THE SECKETAEY. 



85 



measurement. Hence, -vve must expect that all values shall cluster 

 about a point on C D if the sun is constant, but shall cluster about 

 C T> as an axis if the sun is variable. The latter condition is evi- 

 dentlj^ the fact. Assuming the mean point of C D as a center, the 

 average deviation from it is proportional to 8. Assuming the line 

 C D as an axis, the average deviation from it is proportional to 3. 

 Thus the observations are represented 8/3 times better by assuming 

 that the sun's radiation is variable than bv assuming it constant. 



<s> 



Fig. 2. — Simultaneous solar constant determinations. Bassour and Mount Wilson. 



The average deviation of the values from the line C D is 0.021 

 calories. Hence avc may conclude that simultaneous solar-constant 

 measurements at Bassour and at Mount Wilson, while differing hy a 

 constant factor of 2 per cent, exhibit accidental errors of only 1.2 per 

 cent due to variability of the sky, errors of observing, and the like. 

 Dividing by the square root of 2, we find that the average accidental 

 error of a single solar-constant determination at one station is 0.9 per 

 cent. AVlien one considers the multiplicity of the sources of error in 

 this complex investigation, and that the result just announced de- 

 pends on the uniformity of the sky during several hours, as well as 



85360"— SM 1912 7 



