APPENDIX. 21 



rejected if the rule were followed strictly. The advantage of apply- 

 ing the rule in this way is that under it more than one result may be 

 rejected at a time. 



After the omission of the eight runs already rejected, 108 results 

 are left for further consideration. Their average is 0.320, and their 

 mean discordance 0.011; so that all results greater than 0.364 

 and less than 0.276 are properly rejected. This includes four runs, 

 Nos. 4, 16, 22, and 26, and leaves 104 results to which the same 

 criterion is again applied. The average is now raised to 0.324, and 

 the average discordance reduced to 0.006, so that results greater 

 than 0.348 and less than 0.300 are properly rejected. Again, the 

 results of four runs, Nos. 2, 81, 93, and 100, are marked for rejection 

 and an even hundred results remain. Their mean value is 0.32486, 

 their mean discordance 0.00518, and only results above 0.345 and 

 below 0.304 can be rejected. None falls outside these limits. It is 

 noteworthy that every one of the eight results rejected in the two 

 successive applications of this criterion fell below those retained. 

 The reason why all the faulty results were low and none were high 

 will be shown later in discussing the reliability. of the average in the 

 light of the frequency curve. 



The 100 results finally indicated for retention are now searched for 

 systematic variation demanding corrections. The most obvious cor- 

 rection to apply would be for the varying temperatures at which 

 the wood entered the calorimeter. Accordingly the results are tested 

 for variation with initial temperature. This is done graphically, as 

 shown in figure 2, which is made by plotting the individual results, 

 giving each an abscissa determined by its temperature and an ordi- 

 nate determined by its specific heat. The result is to scatter the 

 points over the paper with seemingly no well-marked trend. But 

 the trend is more clearly revealed when the results for successive 

 intervals of temperature, such as half degrees, are averaged, as shown 

 in the figure. These averages vary considerably, their range extend- 

 ing from 0.321 to 0.330, but the course of the line joining them is 

 generally horizontal. Should any correction be decided upon, it 

 would be an inverse function of temperature, which is the opposite of 

 what results at lower temperatures indicate and would be small. 

 Therefore a correction for temperature appears unwarranted. 



Possible variation with species, density, etc., have already been 

 disproven. 



FREQUENCY CURVE. 



The next step is the construction of the frequency curve shown in 

 figures 3 and 5; figure 3 shows the development of this curve. The 

 results of the separate runs were first arranged in order of magnitude, 

 beginning with the smallest, 0.308, and ending with the largest, 



