METHODS OF MEASURING RESPIRATORY EXCHANGE 17 



per minute decreased slowly from 3*40 c.c. to 2*86 c.c. and then 

 dropped suddenly to 1*6 c.c. in the last ten minutes before the animal 

 died. 



It follows from the above that when artificial respiration is per- 

 formed the respiratory quotient is very apt to become abnormal and 

 must vary with the amount of ventilation. In experiments with 

 spontaneous breathing through cannulas, mouthpieces, or masks these 

 instruments will often influence the ventilation, especially in subjects 

 who are not accustomed to them and abnormal quotients will result. 

 It has been found by Becker and Olsen [1914] that mental work of a 

 certain type produces increased ventilation and consequently washing 

 out of carbon dioxide. 



As the possible variations in the store of carbon dioxide in the 

 body are limited it is obvious that errors from this source must be- 

 come proportionately smaller in experiments of long duration, and in 

 24-hour determinations they can almost always be disregarded. In 

 many cases, however, it is impracticable to make long experiments, 

 and the best plan then is to compare series of quite short experiments. 

 When it is found that the CO 2 output either decreases or increases 

 independently of the oxygen intake the quotients found are not to be 

 relied upon as indices of the catabolic processes going on. It cannot 

 be concluded with certainty on the other hand that a quotient which 

 remains constant is the true catabolic quotient. 



When repeated determinations of the same quantity have been 

 made it is customary to average the results. An average ought never 

 to be given, however, unless the necessary data for judging its value 

 are published also. This fundamental rule has often been neglected 

 in respiratory exchange work. 



When only few determinations have been made the best plan is to 

 publish them all, but when they number more than five the elements of 

 the statistical theory of errors ought always to be applied, because it 

 will often allow definite conclusions to be arrived at in cases where 

 nothing can be seen from the untreated figures. Usually it will be 

 sufficient to figure out what is called the mean " error" or "standard 

 deviation " of a single determination (denoted //,) and the me^n error of 

 the average. To do this the average is formed and the deviation (d) 

 of each result from the average calculated. The algebraic sum of these 

 deviations must obviously be o. The deviations are squared and the 

 squares added together. Let the sum be Sd* and the number of deter- 



2 



