204 M AX K L E I K K R 



and thus reducing the random error, which is that part of the error 

 for which the causes are unknown. If, for example, the mean error 

 of a heat determination amounts to 0.3% of the result, one may re- 

 duce the error of a mean to 0.1% by running nine measurements. If, 

 instead, one finds that ^^ of the error originates from uncertainties of 

 water evaporation and discovers a method of avoiding this evapora- 

 tion error, a single measurement is as reliable— or has as small a 

 standard error — as the mean of nine measurements before the particu- 

 lar source of error was discovered and corrected. 



2. Sampling Error 



Two determinations of the heat of combustion of a sample of coal 

 or food or similar mixed substances should agree within about 0.5% 

 (Swiss Institute of Fuel Investigations, personal communication). 

 The major part of the error of this determination is usually caused by 

 the lack in uniformity of the samples and only a small part by the in- 

 accuracy of the calorimeter measurement per se. 



3. Biological Variation 



White wrote in 1928 as follows: 'Tn most cases today the desir- 

 able calorimetric precision is less than might be readily attained but 

 is all that is worthwhile in view of errors elsewhere in the experi- 

 ment." 



Microcalorimeters can now be made with no more than 0.2 to 

 0.5% error (17). In combustion calorimetry even a crude apparatus 

 consisting of an open can with water for the bomb, with a hand stirrer, 

 and thermometer will often give a precision of 1% (33). The Russel 

 Sage respiration calorimeter for man has an error for oxygen consump- 

 tion of 1.6% and for heat flow of 0.9% {34)- In alcohol tests the 

 respiration chamber for indirect calorimetry with cows at California 

 showed a standard error of 1.2%. The biological variation involved 

 in the application of these instruments is considerably larger, as a 

 rule, than the instrumental error. 



Following White (33), one may consider an instrumental error 

 negligible when it does not exceed one-tenth of the biological varia- 

 tion. A coefficient of variation of 4% was found by us for the four- 

 hour measurement of the rate of oxygen consumption of normal rats. 

 For half hour measurements this coefficient amounted to 9%. This is 

 the variability that results when the inherent variance between rats 



