VI. CALORIMETRIC MEASUREMENTS 203 



niiml)er of liters of oxygen consumed. This is in line with Thornton's 

 rule, according to which the heat of combustion of organic com- 

 pounds, A/7, is equal to 53 kilocalories times the number of gram 

 atoms of oxj^gen used in the combustion (32) . The calculation on the 

 basis of the nitrogen and carbon balance is illustrated by the example, 

 in Table VII, of a fasting Holstein cow, and compared with the ap- 

 proximate estimate of 4.7 kcal. per liter oxygen consumed. 



D. LIMITATIONS AND ERRORS 



1. Instrumental Error 



A critical discussion of the instrumental errors in calorimetry is 

 found in White's book (33). White writes: "It is more important 

 to be assured that large errors are absent than to have very small 

 values for the small ones." This is common sense but seems to be 

 forgotten rather often. To take great pains in reducing errors from 

 heat leakage by conduction, convection, and radiation to less than 

 0.01% of the result, when evaporation may cause errors of 1%, would 

 not be reasonable. 



In applying a generally good technique, relatively large errors 

 may occur by "blunders or unnoted accidents" (33). Such errors 

 are detected, as a rule, by repetition of the experiment. If two re- 

 sults of the supposedly same process are relatively far apart, one 

 makes a third measurement; if the result then is close to one of the 

 previous measurements, one regards the mean between the two similar 

 results as the correct result and discards the one that differs as the 

 outcome of a mistake. This procedure is open to criticism in cases 

 when the accident cannot be verified. In that case more measure- 

 ments are desirable to make sure that one is justified in discarding 

 the result with the greater deviation. 



Also in the absence of accidents or blunders, repetition leads to 

 averages that have a smaller error than a single measurement, but 

 this decrease in the error of the mean is proportional to the square 

 root of the number of measurements. To reduce the error of a mean 

 to one-tenth, one has to make 100 times as many measurements. 

 Repetition therefore, is not a very effective method for reducing er- 

 rors. 



Increase in accuracy is more effectively accomplished by analyz- 

 ing the errors, finding the causes of as much of the error as possible, 



