Maier, Read it again, PCT Research

[From Rick Marken (930409.0900)]

Greg Williams (930409) --

Great timing! I've been wanting to instigate a CSGnet discussion of the
classic research by Maier on "frustration."

Excellent. Sounds like a fine example.

Are you game? If so, look for a copy of the book, and I'll post fuller
references and summaries of some of the experiments.

Yes. I just hope the book has a lot of data in it. I'm much more
interested in getting as close to the actual numerical data
(and exactly how it was obtained) as possible -- rather than
reading pages and pages of the researcher's interpretation of it.
But it sounds like a great selection -- just post the references.

You misinterpreted what I meant by "WHAT'S IT TO ME?" as I suspected
some folks might when Gary posted it without elaboration, but you did
so even WITH my elaboration-after-the-fact! Please read me post of
yesterday to understand that I was not referring to "what's at stake,"
but rather questioning Gary's specifications of "information" and

I'm sorry. After all the posting on this topic I just thought we could
let people interpret those words as they wished. For the record,
I don't think there is information IN ANY SENSE about the disturbance
in perception. But when I say "information" in the context of this
debate I am usually thinking that the word refers to "something about
the time waveform of p that makes it possible for the system to
reconstruct another time waveform, o, that compensates for the
net effect of disturbances, d, on p". Allan's last post suggests that
even the IT people are willing to consider thinking about "information"
this way too. Given this definition of "information", there is
unquestionably no information about d in p (except in the special case
where p = o + d and O() is given). The word "useful" means (I think
to all parties involved) that the information can be used BY THE SYSTEM
to compute outputs, o, that produce control (p=r). There may be
information in p in the blog(R/r) sense (just like there is variance
or kurtosis in p) but it cannot be used by the system to compute
outputs that produce control.


PCT Research 1.0

As a first installment in this (possibly one way) discussion of
PCT research methodology, let me try to give a concrete demonstration
of what is meant by o = g (d) and why it is a hugh problem for the
IV-DV approach to doing behavioral research.

The demo is based on a familiar variant of the "parable of the
rubber bands". It is novel (to PCT regulars) only in the way it
is related to the conventional IV-DV research set up. The easiest
way to do this is to hook up two pairs of rubber bands (hook each
pair as shown in the "parable of the rubber bands'" section of BCP
-- which I am ashamed to admit I do not have here in the office
with me). One pair of rubber bands consists of a LARGE and a
MEDIUM sized band (L -- M). The other pair consists of a SMALL and
a MEDIUM band (S -- M). The MEDIUMs should be the same size in both
pairs. As usual, the subject's task is to keep the knot linking
the pairs of rubber bands on a target dot. The subject (with
finger in the L or S band) does this with the L -- M pair and the
S -- M pair; a within subject's design (don't forget to randomize
the order of presentation).

The independent variable in this experiment is "type of rubber band".
There are two levels of this variable, L (large) andS (small).
The dependent variable in this experiment is the distance the subject's
finger move's in response to a fixed distance rightward movement of the
M rubber band by the experimenter. In both conditions, start when the
subject has the knot over the dot and there is a little tension in the
rubber bands (the experimenter pulling slightly to the right on the M
band, the subject pulling slightly to the left on the S or L band).
Make a pencil mark under the subject's finger position. Now the
experimenter moves his/her end of the rubber band pair three
inches to the right. In order to keep the knot on the dot the subject
must pull his/her end of the rubber band pair to the left. Measure how
far to the left the subject's finger moves from the pencil mark. This
measure is the dependent variable.

If you collect several measurements from several subjects in the
two conditions (call them S and L) then you might get numbers like
this (but try it!). The numbers are measures of subject movements
in inches.

          S L
s1 5.5 .6
s2 5.7 .8
s3 5.2 .5

etc. (s1, s2 .. are subjects but they could be trials, for those
of you, like me, who have no friends).

So the subjects produce MUCH bigger responses to approximately
the same "stimulus" (the experimenter's pull) in the S condition
than in the L condition.

In this experiment, the controlled variable (the knot) and the
environmental connections to it (rubber bands) are visible but
in "real" IV-DV research they are MUCH more difficult to detect, if
they are detected at all (imagine what this experiment would be
like if you COULD NOT SEE the knot or your own connection to it!)
So the results of research like this tends to be taken at face value.
What it looks like is that subjects are far more "reactive" with
small rubber bands than with large ones. Theories would be developed,
trying to explain why people are so "reactive" or "hyper" when they are
pulling on small rubber bands instead of large ones. These theories
would be based on the idea that DV = f(IV) where f() is assumed to be
a characteristic of the subject. Big DV's happen because subjects are
more sensitive to S bands; or because there is more information in S
bands,or (for the Freudians) because they are trying to compensate for
having a small band (I can already see the next experiment to test that
one -- do females react less that men to having a small one?).

In fact, the relationship between IV and DV has nothing to do with
the subject and everything to do with the subject's effect on the
controlled variable. The size of the rubber band determines g(), the
effect of the subjects output (DV) on the controlled perception, p
(the knot). The large rubber band has more effect on p than does the
small one so the subject's output, DV, with the large rubber band is
small and with the the small rubber band it is large. Since the
disturbance (the expermenter's pull) is the same in both cases, the
output, DV, is proportional to the inverse of the feedback function,
g(), relating o to p, ie,
DV = g (d).

where, in this case, g() is the IV.

So you can see there is at least the possibility of coming to
the wrong conclusion about "how people work" if people are, indeed,
control systems and you use the IV-DV approach to studying their
behavior. Note that the problem of interpretation of this IV-DV
experiment has nothing to do with the use of statistics. It is a
result of ignoring the fact that people are controlling perceptual
input variables.