DEPARTMENT OF EXPERIMENTAL EVOLUTION. 



127 



'training proper' (learning to enter the first door to the right or left of a 

 variable series of opened doors) are based upon 100 trials per rat; 'test set-ups' 

 (a different variable series of opened doors requiring the same solution) are 

 based upon 20 trials per rat; the averages for 'memory' are based upon 40 

 trials per rat. Each series of trials may be judged in two ways — upon the 

 number of times a rat chooses the correct door first and the number of wrong 

 choices he makes on the average before choosing the correct door. Accord- 

 ingly, the table is divided into two parts, the two sides being different methods 

 of describing the same reactions of the same rats. In the columns headed 

 'Controls better' the difference between the test and the control averages is 

 shown with a plus sign when the controls have more correct first choices and 

 fewer wrong choices. 



"In table 2, the results of the training on the maze, the right and left halves 

 represent the two different experimental situations, and the first column 

 gives three different methods of judging the same reactions. The meanings of 

 these headings is fairly obvious. 



"An 'error' is a wrong turn in the maze; its definitions have been carefully 

 worked out and different types of error are recognized. These will form a 

 special study later on. A perfect trial is run when no error is made. The 

 same method of averaging has been used in the averages for all experiments 

 as used for the multiple-choice results. 



"In observing these figures it must be clearly realized that there are many 

 factors influencing the results that are entirely blotted out by this sort of 

 lumping together. No final conclusion should be drawn from such averages 

 until all the influences back of them have been thoroughly analyzed. At the 



