THE ELIMINATION OF ERRORS IN THE MAZE 137 



objects above them, would obviously remain nearly constant, 

 no matter in what alley the animal happened to be. A similar 

 orientation with respect to right and left turns could not be 

 maintained equally well because the direction of the doorways 

 alternates with successive alleys. We have not had oppor- 

 tunity to test the mechanism of orientation but there can be 

 little doubt that the elimination of errors of type I is affected 

 by it to a greater extent than is that of errors of type II. We 

 can find, on the other hand, no evidence for any influence which 

 might hide a retroactive association in the elimination of errors 

 of type II. Such an influence would have to be equally strong 

 and equally uniform in its action with retroactive association and 

 the behavior of the rats gives no evidence of anything which 

 might produce this result. 



Whatever the explanation of the serial elimination of errors 

 of type I, the process is evidently a complicated one and can 

 not be advanced as proof of retroactive association. No such 

 complication can be demonstrated in the elimination of errors 

 of type II. They show a pronounced uniformity in the amount 

 of practice necessary for their elimination; a greater uniformity 

 than should be expected if chance agents obscured the effects 

 of retroactive association, and it seems certain that their elimi- 

 nation presents a purer case of learning by the method of trial 

 and error than does that of type I. The number of animals 

 is large enough to make the averages dependable while still 

 allowing as great a chance variation as is found between the 

 averages in type II. Clearly the principle of retroactive asso- 

 ciation does not apply to errors of this type. 



Since, as we have pointed out, the elimination of this type 

 of error is probably not complicated by transfer of training 

 from one part of the maze to another, by orientation toward the 

 center of the maze, or by reactions to the curvature of the pas- 

 sages, etc., we may extend our conclusion and say that retro- 

 gressive association is not an effective mechanism in the selec- 

 tion and association of simple series of motor activities in habit- 

 formation. 4 



Evidence for the absence of retroactive association does not, 

 of course, throw much light upon the physiological mechanism 



4 Here and elsewhere in this paper the word " association " is used to express 

 merely the production of a new temporal relation of stimulus and reaction. 



