Puzzling answers, Rubber band demos

[From Rick Marken (951112.1400 PST)]

Bruce Abbott (951111.1655 EST) --

Since I've frequently heard (from SOME quarters anyway) that I don't
really understand PCT, I thought I would just take the test in public and
either demonstrate that it's true or lay this issue to rest.

Understanding is not all or none (it took Bill several years to convince me
of this); Bill's test is a good way to find out how well we understand PCT;
what we know and what we need to learn.

Bill's question:

how do they do this [ make a square of rubber bands]?

Bruce's answer.

The simple answer is that everyone will begin to pull so as to move the
other end of their band to a given position while resisting the
disturbances produced by the resultant forces produced by the pulling of
the other three participants.

My answer:

I think it's important to include the fact that all participants must have an
internal reference for perceiving the bands in a "square" (in a way that
is not inconsistent with reference perceptions of the other participants).
So the answer to the question "how do they make the bands a square" is "by
controlling their perceptions of the shape of the bands relative to their
references for a "square" shape.

Bill's question:

what can one of the people do to make the square be some different size,
and still square?

Bruce's answer.

Just adjust her band tension and direction of pull.

My answer:

Yes. But I think it's important to note that this will change the shape
and/or orientation of the square only if none of the other three participants
is controlling for a square of a particular size and orientation. If some are,
then there is nothing one person can do to make the square a different size
or orientation (except to try to convince the others to change their

Bill's question:

How can the experimenter prove that one knot is most likely under
control and the other most likely isn't?

Bruce's answer:

Pull in directions that would move a selected the knot off-target without
doing so to the other knot; observe whether the controller-person acts to
bring the selected knot back to target.

My answer:

Agree. I presume, however, you mean that part of the process of observing
"whether the controller-person acts to bring the selected knot back to
target" involves watching the hypothetical controlled knot to see if it
moves as expected. You can't just look at the controller's actions to
determine if those actions are protecting a particular variable from
disturbance; you must also monitor the hypothetical controlled variable.

Bill's question:

can the experimenter find the answer by calculating the correlation of the
experimenter's hand movements with those of the controller?

Bruce's answer.


My answer:

An emphatic "No!".

The demo shows that you cannot possibly determine which knot is under
control by examining the correlation between the experimenter's hand
movements (IV) and those of the controller (DV). That relationship tells
you nothing at all about which knot is being controlled -- or even whether
a knot is being controlled.

Suppose you find that the correlation between the experiment's hand
movements and those of the controller is -.97. Which knot is under
control? Suppose the correlation is -.9. Which knot is now under control?
Is either knot under control? (Such correlations would exist even if the
subject were controlling some other part of the rubber bands besides the

Bill's question:

how could correlations be used to determine which knot is
under control?

Bruce's answer:

1. If the correlation between E's and C's hand movement is high, it's the
selected knot. If the correlation is low, it's the other one.

My answer:

How do you know if a correlation is "high"? If the subject is controlling
either knot (or some other location on the rubber bands) correlations
between E and C's hand movements will be in the -.9 plus range regardless
of which knot or location is being controlled. In order to know which
correlation qualifies as "high", you would have to first compute a
correlation when E's hand movements are a disturbance to one knot
and then compute another correlation when E's hand movements are
a disturbance to the other knot. The highest (negative) correlation
between E's and C's hand movements occur when E's hand movements
are a disturbance to the knot that is under control. In other words, to
determine which correlation indicates which knot is under control, you
have to do The Test.

Bruce's other answer:

2. If the correlation between E's hand movement and the position of the
selected knot is low or zero, it's the selected knot; if the correlation is
high, it's the other knot.

My other answer.

Correct. Now you're talking!

Bill's question:

What will C do if a third party grasps the third rubber band and pulls it
toward the direction of C's hand?

Bruce's answer:

C will move her hand in a direction opposite to that of the third party,
just fast enough to keep the knot over the dot.

My answer:


Bill's question:

C and E are interacting as usual, but C is making the knot move in a
repetive SLOW sine wave laterally across the dot (at right angles to the
line of the rubber bands). What can E do that will bring C's hand to a
standstill while the knot continues to move side-to-side in a sine-wave
as before?

Bruce's answer:

E can begin to move her hand laterally (side to side) in phase with C's.
As E's input and C's on the position of the dot will be additive, the
E's excursions, the smaller C's will have to be to keep the knot moving
along the same path. E only needs to adjust the excursions until C does
not need to move at all to keep the knot moving in the same pattern.

My answer:

This is incorrect. Try it (everyone on the list should; they will be VERY

You could have predicted the result if you had looked at the situation
from a control rather than an S-R perspective. If E were actually able to
mimic C's actions precisely, C's sine wave actions would decrease, leading
to a decrease in E's sine actions, causaing a reduction in the sine wave
movements of the knot (the variable C is controlling ) resulting in an
increase in C's sine wave actions. Your analysis is wrong because you forgot
about the controlled variable (sine wave movements of the knot). You also
forgot that, if you use C's hand as the "stimulus" for your movements
that produce the desired effect on the knot, then if C's movements actually
did stop, your movement would stop to (because you are mimicing C) and
C would have to start moving her hand again. S-R does not produce control.

E can bring C's hand to a standstill by moving his hand back and forth as
necessary until E perceives that C's hand no longer moves; E can't know
the details of the back and forth actions required to control his percepion
of C's hand movements; as in all control, the detailed actions that control
a perception (C's hand mvoements in this case) are largely determined by
disturbances to the controlled variable (such as slight variations in C's
effects on the controlled variable), disturbances that are typically
unpredicatble and undetectable.

How'd I do? (;->

Pretty darn good. Your problems come mainly (I think) from clinging to
the S-R point of view.

You seem to believe that S-R (IV-DV) relationships reveal (and account
for) the existance of controlled variables. They don't, as these rubber band
demos illustrate.

Gary Cziko (951111.2300 GMT) to Bruce Abbott --

I understand the "IV-DV" approach to be more than "push-observe." It
also tells you WHAT to observe, that is, some variable that changes along
with the "IV."... "IV->DV" does not say "vary the IV and observe what
doesn't change." But that's indeed how we find out what is being

Excellent, Gary. You're not just another pretty Net God.

Bruce Abbott says:

If your disturbance produces actions, you have disturbed a controlled
variable, no? So this method IS telling you something valuable.

Not necessarily. If a disturbance produces actions it may just be a
cause-effect relationship. You can't know if anything is controlled unless
you do The Test. If you type questions into a computer (disturbances) and
the computer prints out an appropriate answer (action) each time, have you
disturbed a controlled variable?

Gary --

We have no argument if you agree that the "IV-DV" method can at best be
only a first step in understanding behavior that is controlling a
perception. One then needs to follow it up with The Test of the controlled
variable to understand what is really going on. Do you agree?

Bruce --

Yes, that has been my position all along,

If so, then why not put a description of The Test in your methods book. If you
want to call it an IV-DV methodology that's fine with me. But The Test is a
pretty important variant of IV-DV methodology since it is the one used to
study living systems.

Gary Cziko (951112.1610 GMT) --

since Rick also agrees with me, then you must agree with Rick and Rick
with you. But I suspect that Rick won't find that agreeable.

I agree.

Bill Leach (951111.21:09 U.S. Eastern Time Zone) --

I also don't agree that Bruce asserts that THE TEST is not central to the
study of behaviour.

Bruce is saying that it is not important to describe The Test in his
textbook on how to study behavior. That's a perculiar thing to say about
a methodology that one believes is central to the study of behavior.