472 Prof. Silas W. Holman on 



The lines AB and CD on the diagram (page 483), 

 constructed with percentage deviations as ordinates and 

 temperature as abscissas, show clearly that the deviations are 

 systematic. Upon inspection of this plot it appears that the 

 data may be separated into two groups, one including 

 0°-300°, the other 400°-1000°, which appear to have entirely 

 distinct forms of systematic error. This division corresponds 

 to two distinct groups of data, one extending from 0° to 300°, 

 the other including the second group and extending from 

 350° to 1075°. The latter were given in the Bulletin as the 

 final results of the high temperature comparisons of the irido- 

 platinum couple with porcelain bulb air thermometers. The 

 detailed statement of the 0°-300° comparison I have not seen. 

 Although the discrepancy between the two sets of systematic 

 deviations is not extremely large, yet it has seemed to me 

 that it is beyond the limits of concordance in the higher 

 temperature work, and that it would be better for the present 

 purpose to deal solely with the 350°- 1075° data. Two points 

 regarding Barus's work should be noticed: one the strikingly 

 high degree of concordance between individual observations 

 even with different thermometer bulbs and different thermo- 

 couples ; the other the remark in which Bar us notes a possi- 

 bility of being able still further to reduce the " stem error " 

 entering into the result, which so far as I am aware has not 

 yet been done. 



The high temperature air-thermometer comparisons (Bul- 

 letin, Series I., II., III., IV., and V.) of Barus are so 

 numerous (108) and so distributed that the labour of utilizing 

 them simply for deducing constants and testing an equation 

 would be excessive. Also they are too concordant to permit 

 interpolation on a -direct plot without a sacrifice of some of 

 their precision. For the purposes of discussion, therefore, 

 I averaged them in nine groups. The first group contained 

 all where the E.M.F.lay between 3000 and 4000 microvolts ? 

 the second group between 4000 and 5000 mv. ; and so on by 

 steps of 1000 microvolts, except that the seventh group 

 covered 2000 mv. from 9000 to 11000. These groups were 

 not exactly equal in number of observations, and therefore in 

 weight, nor is the arithmetical average a strictly legitimate 

 value where the function is not linear ; but, as easily seen by 

 inspection of the originals, the errors thus introduced are 

 negligible. In Table II. columns 1 and 2 give the direct 

 values of the averages. Column 3 reduces £20 e to So 

 by adding 150 microvolts, the value of X%e being elsewhere 

 given by Barus as —150 microvolts. 



