The Lessons of the Rubberbands

[from Gary Cziko 951111.1722 GMT]

In recent discussions involving Rick Marken, Bruce Abbott, and Chris
Cherpas, Rick has mentioned my "improved" rubberband demo a few times as a
way of demonstrating the weakness of a independent variabe (IV) ->
dependent variable (DV) model of behavior. Since I suspect that perhaps
Bruce and certainly Chris are not familiar with this awesome demonstration
of the behavioral illusion and the limitatin of the IV-DV approach to
understanding behavior, I am going to my best to explain it here (again?)
and request that all those who have not done it or seen it before give it a

Equipment needed: two rubberbands knotted together, plus a string about a
30 cm long tied to the end of the one of the two rubberbands (a third
rubberband can be used instead of the string, but the string works a bit
better and makes the procedure easier to describe; you can use a shoestring
if you have a shoestring budget for research).

Procedure: Ask a subject (S, female in this example) to hold one end of
one rubberband while you the experimenter (E) hold the end of the other
rubberband (the one with the string attached). Ask the subject to find a
spot on a wall or table over which the knot will be kept. Now, you (E)
pull and release your end of the rubberband and observe how S mirrors your
hand movements with hers. If you just disturb in a straight line toward
and away from the reference spot, you will have a very nice relationship
which gives S's behavior (as measured by her hand position) as a function
of E's behavior (as measured by his hand position). If both rubberbands
are or equal size and thickness and using the reference spot as zero on the
number line, the equation will look like:

S = -1*E

Now this can be a very useful equation. You can predict where S's hand is
if you know where E's is. And you can get S to put her hand where you want
by putting E's hand in the corresponding spot. So we have both prediction
and control of behavior--the two golden objectives of psychology!

But there are problems with this equation. The first is that it may
suggest to some that S perceives where E's hand is and responds to this
independent variable (IV). Instead, it is easy to show by blocking S's
view of E's hand that S is not perceiving E's hand at all, but is rather
perceiving the knot and it's relation to the reference spot. So the
equation fails as an actual (and not just an apparent) model of S's

The second problem is that the model S = -1*E also fails as a useful
predictor of S's behavior if any changes occur (a) to S's reference level,
or (b) to the relationship between E's hand and the knot. Concerning (a),
if S no longer wants to keep the knot over the reference spot, S = -1*E
will no longer work.

But (b) is more fun to demonstrate. This can be done a number of ways.
One is to have E keep his originally disturbing hand still while using his
OTHER hand to pull and release the string attached to his end of the
rubberband (the originally disturbing hand should relax it's grip of the
rubberband so it can slide through the hand). S = -1*E now says that the
subject should NOT move her hand (since E has not changed). But what
happens instead is that S DOES move her hand, even though the equation, so
useful up till now, says she should not. Or, E can hold the end of the
string taut and slide the originally disturbing hand up and down the
rubberband without actually pulling on it. S = -1*E now says that S should
move her hand, but in fact she does nothing.

Some other variations can be also used to show the very limited use of the
IV-DV model for explaining, controlling, and predicting behavior. As Rick
has recently mentioned, E's rubberband can be changed to one which is half
as thick as S's. Now we will see that a new equation pops up which is S =
-2*E. It appears that S is now twice as senstive to E's movements, but in
fact she hasn't changed at all, rather the rubberband has.

Finally, imagine that it is not just E's movement which disturbs the
position of the knot, but also things like the wind, variations in
gravitational field, rapid deterioration of the rubber, or other strings or
rubberbands also attached to the knot manipulated by other distrubing
agents. Now the coefficient -1 in the original equation S = -1*E will move
towad zero as the relationship between S and E becomes more noisy. But S
is just as concerned as before in controlling her perception of the knot
and may be controlling just a well, even with all these other disturbances

So the traditional IV-DV relationship does have some use in explaining and
predicting behavior, but its uses are quite limited and can actually be
misleading when perceptual control is involved. Knowing which perceptual
variable is under control allows us to easily explain (indeed, predict)
what will happen if (a) E pulls on the string while keeping the originally
disturbing hand still, (b) E moves the originally disturbing hand up and
down the rubberband while keeping the string taut with his other hand, (c)
E's rubberband is changed from thick to thin, or (d) other disturbances
begin to influence the position of the knot. The IV-DV model is absolutely
incapable of explaining (a) through (d) (and, boy, are in you in for a real
mess if you want to use the position of the KNOT as the IV!). Finding an
apparent IV-DV relationship does suggest that some variable is being
controlled, but the IV-DV relationship itself is rather useless in itself
when perceptual control is happening.

Hmmm. I juice the shock grid with 10ma and the rat jumps.
You're saying the shock doesn't control the rat's behavior?

Chris, how about an attempt to to apply the lessons of the rubberband to
your shocking example of the jumping rat? (Hint. Rubberband: E pulls, S
pulls. Shock: E shocks, S jumps.)

And don't let Bruce Abbott get near your rat unless it wants a rather
severe haircut!