ance and exit doors. Since there was at this time
no mechanism for locking them when closed, it was difficult for the
experimenter to prevent him from escaping by way of the entrance door or
from raising the exit door sufficiently to obtain the food. Indeed, the
longer confinement worked so unsatisfactorily that on the following day
I substituted for it the punishment of forcing him to raise the entrance
door of the wrong box in order to escape for a new choice. He was
rewarded with food in the alleyway H, beside door 15 (figure 17), only
when he chose correctly on first attempt.

This method discouraged him extremely and proved wasteful of time.
Consequently, in a second series on the same date return was made to the
former method, and he was rewarded with food whenever he found the right
box. But on April 28, the two methods were again employed, the first in
the initial series and the second in a final series of trials. The
animal's persistent attempts to raise the doors gave the experimenter so
much trouble that on April 29 barbed wire was nailed over the windows of
the entrance doors with the hope that it might prevent him from working
at them. But he quickly learned to place his fingers between the barbs
and raise the doors as effectively as ever.

On April 30 the reward of food was given only when the first choice was
that of the right box and in that event it was placed in the alleyway H
as stated above.

As it seemed absolutely essential to break the unprofitable habit of
choosing the nearest door, on May 3 a new series of settings was
presented, in which only the doors to the left of the middle of the row
of nine boxes were used as right doors. That is, in this new series,
doors 1 to 4 occur as right doors; 5 to 9 do not. As punishment for
wrong choices on this date, Julius was confined in the wrong box from
one to five minutes. It was difficult to keep him in, but by means of
cords which had been attached to the doors, this was successfully
accomplished. Yet another and slightly different series of settings was
employed on May 4, and this, proving satisfactory, was continued in use
until the end of the experiment, with punishment ranging from sixty to
one hundred and twenty seconds for each mistake.

Naturally the modification of settings introduced May 3 greatly
increased the proportion of wrong first choices. Indeed, as appears in
table 8, the ratio of right to wrong immediately changed from 1:0.67 to
1: