VALUES OF CURVES OF LEARNING 141 



of time, and this period apparently bears no definite relation to" 

 the learning process. In fact, it may persist after the animal's 

 behavior has become automatic. The experimenter must decide 

 whether to count time from the moment the rat is introduced 

 into the maze or wait until the time he starts off upon business 

 intent. Some rats may also stop and hesitate at the entrance 

 to the food box. Again, one must decide whether to omit the 

 time devoted to scratching and sniffing during the course of 

 the run. Any decision upon these points is, however, relatively 

 easv to follow in a consistent manner. Time is, on the whole, 

 a more practicable criterion than error from the standpoint of 

 giving comparable results at the hands of different experimenters. 



The total distance criterion presents so many difficulties as 

 to render it impracticable for ordinary work. One difficulty lies 

 in the matter of taking records accurately. The rats, after a few 

 trials, run so rapidly that it is extremely difficult for one person 

 to observe and record at the same time. To do this, it is neces- 

 sary to mark off the maze into small segments and commit to 

 memory some scheme of representation, so that records can be 

 jotted down in a purely automatic manner. The work of tran- 

 scribing this record into distance terms, and computing the same, 

 is very laborious. Eliminating these practical difficulties, the 

 distance criterion in some ways is an ideal one. There can be no 

 divergence of practice as to what shall be omitted or included, 

 and results obtained by different experimenters upon the same 

 maze will be strictly comparable. 



The distance and error criteria are alike in that both represent 

 the same thing, viz., the progressive elimination of unnecessary 

 or surplus distance, and this fundamental similarity must be 

 borne in mind in considering their relative value. If errors are 

 so defined as to include all returns over the true path, then the 

 distance curve forms the mathematical limit of the error curve 

 as a smaller and smaller segment of the maze is taken as the 

 unit of error. In this sense the distance curve is the ideal one 

 inasmuch as it attempts to portray accurately all of the details 

 of the process of the gradual elimination of surplus distance. 

 If errors are confined to cul de sacs and no attempt is made to 

 evaluate the varying degrees of error, and this, I take it, has 

 been the common practice, then an error ciirve will differ markedly 

 from a distance curve, and one is confronted with the problem 



