Kew Yoek Weather Bukeau. 513 



In order to obtain a numerical expression of the reliability 

 of the normals as derived both from independent records and by 

 comparison with adjacent stations, a method developed and ex- 

 tensively used by Dr. J. Hann was employed, the results of the 

 oomputations appearing in tables 31 and 32. The mean varia- 

 bility of temperature, as shown in table 31, is the average differ- 

 ence between monthly or annual normals and the individual 

 values from which the normals are derived. In the same way,. 

 table 32 gives the average variability to w^hich the temperature 

 differences between several pairs of stations are subject. 



To determine the probable error of the means or normals ob- 

 tained from data subject to the given degrees of variation, the 



v 

 following modification of Peters' formula*is used: i'm=- 845 -. 



where r™ is the probable error of the mean of normal, V is the 

 average deviation from the mean value, or the variability, and 

 n is the number of years covered by the record. 



The maximum variability of meanis was found by a trial of 

 several records to be fairly represented by the values for January 

 and February, and the minimum by those of July and August. 

 The character of the variation of differences is also best indicated 

 by the midsummer and midwinter rates, and hence only the above 

 four months and the year are included in the tables. It will be 

 seen that the normal of Cooperstown, whose record is the longest 

 of the series, is liable to errors, amounting to 0.6 degrees in 

 winter and 0.4 degrees in summer, while at the remaining sta- 

 tions the uncertainty is oonisidlerably greater than at Coopers- 

 town during the winter months. In several cases, however, er- 

 rors of observation, or in the published records, undoubtedly 

 affect the results to some degree; hence the average variability 

 at the foot of the table is derived from three stations whose data 

 were known to be reliable, rather than from the entire number. 

 For ithe same reason, only the pairs of stations in table 32 num- 



♦ Reducible substantially to Fechner's formula used by Dr. Hann. A strict accuracy would 

 require the probable error of the variability itself to be taken into account, but this is not 

 necessary in the present case. 



33 



