514 Eighth Annual Report of the 



bered 3, 4, 5, 6, 7, 8, 11 and 12 can be fully relied upon as deter- 

 mining the variability of actual temperature differences. The 

 results of pair No. 13 are also undoubtedly reliable for the winter 

 months. 



A glaince at the tables suffices to show the great stability of 

 differences, betweeoa .regions as widely separated ias the Atlantic 

 Coast and the Great Lakes, or the Coast and the lower Champlain 

 valley; and also, in the case of Cooperstown and Central Park, 

 proves the effect of differenoes of altitude to be small in Central 

 New York. Thus, wbile a single normal derived from a record of 

 thirty-eight years is isubject to a probable error of 0.6 degrees, a 

 sysitem; of ^concurrent observations extending over' twenty yeara 

 would suffice tq reduce the uncertainty of differences between even 

 remote stations! to about half of that amount. 



The case may also be stated thus : Adopting 4.0 degrees as the 

 average variability for January, a record of fonty-five years is 

 foundf necessary to reduce the probable error of the normal to 

 0.5 degrees; whereas^ with a variability of differences las great as 

 1.6 degrees between Central Park and Burlington, only ten years 

 are required to reduce the relative error to the same amount.* 



In view of the advantages thus to be derived from the use of 

 a series of records embracing the same yeans &t a large number 

 of stations, an effort was made at the bcglMiing of this investiga- 

 tion to deduce the normals from the first series of Regents' ob- 

 servations at the academies of the State during the period 1826- 

 1850. All of the longer records were examined and compared, 

 month by month, «nd in this manner a large number of errors 

 (many of which are systematic) were detected! in the published 

 ta'bles. In a few cases the records proved to be quite satisfactory 

 (notably those of Albany, Kinderhook, and in a lesser degree, 

 Rochester). It is a peculiarity^ of these early obsiervation, not 

 easily explained, that the mean temperatures' for January are 



♦The above formula may be transformed to n=0.71 ; where V is the variability, R is a 



definite value of the probable error (n this case 0.5°), and n is the number of years required 

 ti re luce the jjrobable error to the given amouut. 



