Reprinted with permission
from
Psychological
Review.
1938.45,
1-41.
Chapter
18
The Determiners
of
Behavior
at
a
Choice Point
Edward
Chace
Tolman
University
of
California
The question
I
am going to dis-
cuss
is the very straightforward and specific one of “why rats turn the
way they do, at a given choice-point in a given maze at a given stage of
learning.”
The first item in the answer is fairly obvious. They turn the way
they do because they have on the preceding trials met this same choice-
point together with such and such further objects or situations, down the
one path and down the other, for such and such a number of preceding
trials. Let me, however, analyze this further, with the aid of a couple of
diagrams. First, consider a diagram of a single choice point (Figure
1).
In this figure the point of choice itself is designated as
0,;
the
complex of stimulus-objects met going down the left alley, as
O,,
that
met going down the right alley, as
OR;
the goal at the left, as
0,;
and
that at the right, as
OcR.
The behavior of turning to the left is represented
by the arrow
B,;
and that of turning to the right, by the arrow
BE.
And
Presidential address delivered before the American Psychological Association, Minne-
apolis, September
3,
1937.
337
338
EDWARD CHACE TOLMAN
I
Figure
1.
the point
I
am now making is that the relative strength of the tendency to
turn, say, left (rather than right) will be, first of all, a result not only of
the present presentation of
Oc
but also of all the previous presentations
of it together with the
OL,
OGL, OR,
and
OGR
consequences of having
behaved by
BL
and
BR
on all these preceding occasions. In short,
I
would
schematize this feature of the causal determination of the left-turning
tendency by the diagram shown in Figure
2.
The expression
BJBL
+
BR)
at the right-hand side of Figure
2
is the
"dependent variable" (we may call it the behavior-ratio).
It is the
percentage tendency at any given stage of learning for the group as a
whole to turn left. And the hieroglyphic at the left-hand side of this
figure is the "independent variable" which determines this behavior-
ratio. This hieroglyphic is to be read as meaning: the
sum
of all the
preceding occasions in which
Oc
has, by virtue of
BL,
been followed by
OL
and
OGL
and by virtue of
BR
been followed by
OR
and
OcR.
This
diagram is thus no more than a schematic way of representing the, shall
we say, (to use the term we theoretical psychologists have of late taken so
violently to our bosoms) "operational" facts. The expression at the left is
DEPENDENT
fi
VARIABLE
INDEPENDENT
VARIABLE
Figure
2.
THE DETERMINERS
OF
BEHAVIOR AT A CHOICE POINT
339
DEPENDENT
fi
VARIABLE
IN
DEPENDENT
VARIABLE
Figure
3.
an "operationally defined" independent variable and that at the right, an
"operationally defined" dependent variable.
For brevity's sake,
I
shall often substitute, however, an abbreviated
symbol for the left-hand term, viz.: simply
Z(OBO),
as shown in Figure
3.
One further point-the
fl
in each of these figures indicates merely
the fact of the functional dependence of the dependent variable upon the
independent variable.
To
indicate the "form" of this function we would
require a more analytical diagram, such as that shown in Figure
4.
Figure
4.
340
EDWARD CHACE TOLMAN
1
I
I
I
--
I
OR
f
oGR
OGL
hoL
I
But this, or course, is no more than our old friend, the learning
curve. It results when we plot the independent variable along an
X
axis
and the dependent variable along
a
Y
axis. Nothing very new
so
far. It
seems surprising, however, that in spite of the thousands, not to say
millions, of such learning curves which have been obtained in the last
four decades in American rat laboratories there are still a variety
of
quite
simple things about this function which we do not yet know or with
regard to which we are still in dispute.
For example, we are still in dispute, first
of
all, as to the relative
importance of the occurrences of the
two
alternative behaviors
BL
and
BR,
where
BL
is
“wrong” and
BR
is “correct.” (See Figure
5.)
Thorndike
(118,
119)
and Lorge
(69)
and their co-workers, as you all
know, working with human beings in analogous, though verbal, set-ups
have now concluded that the occurrence of the wrong behavior has no
such general causative effect. They find that learning appears only as a
result of the occurrences of the rewarded sequence
OC-BR4OR:
OGd.
On the other hand, still more recently, Muenzinger and Dove
(93,
working with set-ups similar to Thorndike’s have found that the occur-
rence of the wrong response
OC-BL-4OL:OG,)
does weaken its tend-
ency to re-occur. Also Cam, as a result of a series of experiments done by
his students
(54,
72, 132,
135,
137)
some time since in the Chicago
laboratory, was finally forced to conclude that
. . .
a
certain number of errors must be made and eliminated before the subject is
ever able to
run
the maze correctly. Comct modes of response are established in
part by learning
what
not
to
do
(26,
p. 98, italics mine).
A
second point about which we are still surprisingly ignorant is that
we do not yet know the importance of the rat’s being permitted, or not
Figure
5.
THE
DETERMINERS
OF
BEHAVIOR
AT
A
CHOICE
POINT
341
permitted, to return out of the wrong choice. In some experiments, when
the animal takes the wrong alley, he passes through a one-way gate and
is started over again. In others, he is allowed to treat it as a blind and
back out. But,
so
far as
I
know, there has been no carefully controlled
comparison between these two procedures.
Thirdly, the question of the relative effects of concentrated versus
distributed repetitions has not as yet received the thoroughgoing experi-
mental analysis that it deserves. But I understand that Professor Stone
and his co-workers are now directing their attention to it and are getting
some very significant findings.
Fourthly, we are ignorant concerning the difference between animals
which have an initial left-hand bias and those which have an initial
right-hand bias. We usually lump the results for both types together in a
single curve. But we might well separate them and study them inde-
pendently.
Fifthly, Brunswik
(14)
has recently brought to light a new point in
our ignorance. He has been trying the effect of rewarding on the right
and rewarding on the left different proportions of times. In other words,
it was no God-given rule but apparently some merely human predilec-
tion on our part which made us heretofore tend almost invariably to
make one of the alternative behaviors always rewarded and the other
always punished. But other frequencies of reward and punishment are
equally possible and equally deserving of study.
Sixthly, experiments by Krechevsky
(59,
60,
61),
seem to indicate
that there may be certain general features about the content of the
OBO's
such, for example, as their containing variable or non-variable paths,
which are very important in determining the resultant behavior-ratios
and about which we need more information.
Seventhly, a further point which needs more investigation is, as
Muenzinger and his co-workers
(87,
88, 89, 90, 92, 92, 93)
have beauti-
fully brought out, the fact that punishment or obstacles to be overcome,
even
on
the
correct
side,
may sometimes seem to aid rather than hinder
learning. (See also Tolman,
125,
and Tolman, Hall and Bretnall,
227.)
Eighthly, there is the question of what happens when
qOBO),
the
number of trials, has become very great. This seems to induce a special
sort of result for which the term fixation has been suggested.' And
further studies of such "fixations" are needed.
Ninthly, the problem as to the effect of temporal intervals between
Oc
and the resultant
OGL
and
OGR
are still by no means altogether
completely worked out in spite
of
all the beautiful work of Hunter and
1.
For one
of
the first experiments indicating that there are such biases, see Yoshioka
2.
See the original experiments on fixation
by
Gilhousen
(31,
32),
Krechevsky and
(249).
Honzik
(62)
and Hamilton and
Ellis
(28,
38,
39).