2 SMITHSONIAN MISCELLANEOUS COLLECTIONS VOL. ']'] 



of error curve which would give the best fit for these observations 

 w^as determined from a plot on an arithmetical probability diagram 

 in which the probability integral is expanded so that plotted values 

 of the integral follow a straight line. The curve thus derived is 

 drawn through the values in figure I. 



This curve indicates a mean variability in the solar radiation 

 values of ± .on calorie (about 0.6 per cent). That is, 50 per cent 

 of the observed values will not dififer more than this amount from 

 the mean value 1.945. while 50 per cent will show a larger deviation. 

 According to the curve, only two per cent of the observations deviate 

 as much as 2 per cent (0.039 calorie) from the mean value. That 



1.880 1.900 1,920 1.940 lo960 1.960 2.000. 2.020 



Fig. I. — Frequencies of occurrence of different intensities of solar radiation, 

 July, 1918, to September, 1922. 



is to say, very few observed values should fall below 1.906 and 

 very few rise above 1.984. The observed frequencies of very low 

 and very high values exceed the theoretical expectancy. 



In dividing the observations into high, low, and medium, it was 

 found most convenient to call all values above 1.960 high values, 

 and all below 1.931 low values. Separating the observations into 

 these two classes, they were compared with the 8 a. m. observations 

 of pressure at various stations in the United States and Canada, in 

 order to discover what relations might exist at the selected places 

 between solar heat variations and weather changes. The mean results 

 are given in tables 2 to 5, for the interval beginning two days before 

 the solar heat measurements and ending 12 days after. The results 

 for the winter half-year given in tables 2 and 3 are plotted for 



