20 THE SPECIFIC HEAT OF WOOD. 



due to chance errors which are unavoidable ; in addition they may be 

 biased by the presence of systematic errors. Chance errors obey 

 certain mathematical laws which have been formulated by Laplace; 

 their measure is obtained by purely mathematical processes and con- 

 sists either of the mean error or the "probable" error. Systematic 

 errors are susceptible of no mathematical treatment, and, unless they 

 are very simple, their discussion leads to no numerical corrections. 

 Their presence is best revealed by the form of the frequency curve 

 and its relation to the arithmetic mean. 



REJECTED RESULTS. 



In Table 5 are displayed the material used in each of the 116 

 runs made from near 106 C., the number and date of each run, the 

 corrected observations, and the calculated results. The records of 

 each one of these runs were carefully scrutinized, and where there 

 was valid reason for doubting the fitness of any result it was rejected. 

 Rejections were made for two reasons : 



1 . In the preparation of some of the material the wood was exposed 

 to temperatures of 110 C. while stiU wet and dried in nearly satu- 

 rated air at this temperature. Such treatment darkened the color 

 of the wood, produced abnormal shrinkage, and, in some cases, 

 appeared to raise its specific heat. To avoid the possible effects of 

 such drastic treatment the results from the five runs, Nos. 47, 48, 

 49, 50, and 51, for which such material was used, were rejected 

 entirely and a notation to that effect entered in the column headed 

 "Remarks." 



2. While nearly all of the runs were made from an initial tempera- 

 ture between 102 and 112 C., three runs, Nos. 23, 24, and 25, were 

 made from temperatures above 120 C. The departures of these 

 runs from the average of the remaining results is not sufficiently 

 consistent to warrant their rejection; but inasmuch as there are 

 indications that the specific heat of wood varies greatly with tem- 

 perature, the results of these three runs are also rejected. 



The arithmetical mean of the remaining results was now calculated 

 as a first approximation to the best representative value, and the 

 criterion of permissible discordance was applied to the individual 

 results. One of the practical rules resulting from the laws governing 

 the behavior of chance errors is, that if one result of a series differs 

 from the average of the other results by an amount which is more 

 than four tunes the average deviation of the other results, the dis- 

 cordant result should be rejected, because to include it lowers the 

 reliability of the average. If, in the application of this rule, the 

 doubtful result is included in the average, the permissible discordance 

 is increased and the criterion made looser. Obviously any result 

 whose rejection is indicated under this application would certainly be 



