THE ELIMINATION OF ERRORS IN THE MAZE 133 



the food compartment and explore the inner cul de sacs twice 

 before the trial was ended, and in the second trial the animal 

 left the food and explored the inner alleys five times. If the 

 final data is based upon trials conducted with this technique 

 it can have no bearing upon the problem of the elimination of 

 errors, for we can not determine whether the smaller number of 

 trials required for the elimination of the errors near the food box 

 resulted from retroactive association or from the fact that addi- 

 tional trials in the inner part of the maze were ignored. 



Additional evidence dealing with the question of retroactive 

 association is now at hand and makes possible a more positive 

 conclusion concerning its role in habit-formation than could be 

 drawn before. The evidence consists of data upon the elimina- 

 tion of errors during the training of 56 rats in the Watson circular 

 maze. The method of analysis differs from that employed 

 earlier by Hubbert in that different types of errors are treated 

 separately. 



The ground-plan of the maze is given in figure 1. As will 

 be noted, two chief types of error are distinguishable. The 

 first of these (type I) is the passing of a doorway through which 

 the animal should go. Errors of this type are marked with 

 even numbers in the figure. The other (type II) is that of an 

 incorrect turn after passing through the doorway. Errors of 

 this type are given odd numbers in the figure. In addition to 

 these types are errors of turning back upon the correct pathway. 

 While common enough in the early part of training these errors 

 are soon eliminated. In the 616 possible cases in the data 

 examined only one was found in which an error of turning back 

 followed the last error of types I and II in any alley (with the 

 exception of /, which, being without a partition, permits only 

 errors of turning back), so that they may be disregarded in any 

 study which deals only with the last error made at each given 

 point in the maze. In compiling the data any turn which car- 

 ried the animal for its own length or more into a blind alley 

 was counted as an error. Thus figure 2, a tracing of the path 

 followed in a single trial, shows errors at 2, 5, 7, and 11. The 

 records of all the animals were examined and note made of the 

 number of trials preceding the last in which each of the 11 

 possible errors was made. The averages of these for the 56 rats 

 are given in table 1. 



