Research Methods

[From Bruce Abbott (941230.1700 EST)]

Rick Marken (941229.2045)]

Me:

The operant research you are planning with Bill will give us all an
opportunity to see whether rats are really controlled by food pellets or
whether they are controlling some perceptual characteristic of those
pellets.

Bruce Abbott (941229.1530 EST)--

What, do you think there's any doubt?

I think there is a huge amount of doubt.

Not on my part.

Me:

did you [Bruce] always know that organisms were perceptual control
systems? If so, is your research methods text all about how to study
perceptual control systems?

Since you haven't answered this for me yet, and since I'm on vacation,
I took a pleasant walk over to UCLA and looked over your research
methods text myself. I'm afraid I was unable to find much in it about how
to study living control systems. I looked for "controlled variable" in
the index but didn't find it. Maybe you called it something else?

I did find quite a bit about how to determine the causes of behavior
by manipulating independent (environmental) variables and determining
whether or not they had a statistically significant effect on a dependent
(behavioral) variable. I even learned that I should be careful to
_control_ extraneous variables that might also have an effect on behavior.

So? What's your point? (;->

Let's say that I want to learn how pushing the stick around on my RC
transmitter affects the angles of the rudder and elevator. Being a
quantitatively minded scientist, I carefully measure the starting positions of
the stick, rudder, and elevator. Then I systematically vary the stick
position over a large number of positions and record the rudder and elevator
positions at each stick position. I discover that every stick position yields
a unique combination of rudder and elevator positions, and that I can
decompose the "influence" of the stick into two orthogonal axes: up and down
moves the elevator without affecting the rudder, whereas left and right moves
the rudder without affecting the elevator. I also learn that the relationship
between stick position and rudder/elevator is one-way: forcing the rudder or
elevator to another position has no effect on the stick. The same test
reveals that the rudder and elevator resist my attempts to move them, and
return to their initial positions as soon as I stop applying force.

Isn't it interesting how much I was able to learn about the system using that
outdated methodology described in my text? I may have no idea that there is a
servomechanism involved in all this (in fact, a pair of them), yet (without
knowing how to correctly describe what is happening) I have learned the
correct relationship between stick position and changes in the reference
levels of the two servomechanisms involved. I have even learned that the
control surfaces seem to resist being moved from their reference positions by
applying a strong counterforce to my disturbances.

If strong variable disturbances were acting on the rudder and elevator while I
was conducting these tests, the "clean" relationships I observed would have
been obscured to some degree. Resorting to statistical techniques, I would
have been able to extract, to a fair approximation, the relationships I
observed under steady-state conditions. Of course, before resorting to
statistics, I would have attempted to "control" these disturbances, i.e.,
reduce them to a minimum, and thus might have been able to do away with the
statistical analysis altogether.

I think I'll just go right on teaching the scientific method, thank you. It
seems to work pretty well most of the time.

Actually, the book reminds me of one I wrote in my deep, dark conventional
psychological past. My book was called "Methods in experimental psychology";
it was completed just about a year after I started to really understand
PCT.

Late 70's then? I tried to look up your text, too. Apparently, there is not
a single copy of it in the entire Indiana library network. Who published it?

By the way, Bruce. What's your snail mail address; I've got a copy of
the "Blind men..." paper ready to go.

Check the header on one of those programs I wrote that you didn't read. (;->
You'll find my full address given there.

And Happy New Year!

Regards,

Bruce

[From Rick Marken (941230.2045)]

Bruce Abbott (941230.1700 EST)--

Isn't it interesting how much I was able to learn about the system
using that outdated methodology described in my text?

The IV-DV methodology is not "outdated"; it is the right methodology
to use when studying cause-effect systems. It is the wrong methodology
to use when studying living control systems.

The reasons why IV-DV methodology is the wrong way to go about
studying living control systems are described with elegant precision in
Powers (1978) _Quantitiative analysis of purposive systems_ which is
reprinted in _Living Control Systems_. I particularly recommend
Experiment 4 (p. 156-157 in the book) where Bill demonstrates the
"behavioral illusion". In that experiment, Bill manipulates an
independent variable (q.d) and determines (with extraneous variables
controlled;-)) its relationship to a dependent variable, q.o. The
relationship between q.d and q.o is nearly perfectly cubic. The
conclusion from IV-DV methodology would be that this function
characterizes how the subject responds to variations in the IV. In fact,
this function has nothing to do with how the subject responds to input;
it is actually the inverse of the feedback function relating the
dependent variable, q.o, to the (typically unnoticed) controlled variable.
If the feedback function becomes linear, the relationship between
independent, q.d, and dependent, q.o, variable becomes linear. This
occurs despite the fact that there is no change in the subject. The
"behavioral illusion" refers to the fact that, when you are dealing with
a control system, the IV-DV relationship that is supposed to tell you
something about the subject actually tells you nothing about the subject
at all.

I think I'll just go right on teaching the scientific method, thank you.

The scientific method you teach is excellent when it is applied to what
Bill refers to (in the "Quantitative analysis..." paper) as "Type Z"
systems that have "zero loop gain". Rocks and batteries are examples of
Z systems.

It seems to work pretty well most of the time.

This is correct, since most systems in the world are probably Z systems.
But when you suspect that you are dealing with an N system (one with
negative loop gain) you must try a variant of what you call "the
scientific method": "the test for controlled variables". It's just as
scientific but it's specially designed to study N systems. The only
way you can actually learn something about how N systems work is by
using "the test".

Me:

My book was called "Methods in experimental psychology"; it was
completed just about a year after I started to really understand PCT.

Bruce:

Late 70's then?

Close. 1981 is the copyright date. When it was published, I already knew
the ugly truth about the behavioral illusion (ugly if, like me, you were
uncomfortable about promoting hallucination as science;-)). It was
NOT fun when I finally understood what this illusion meant for conventional
psychology (it meant "The End") but I think it's a hurdle that a
conventional psychologist has to get over in order to see what PCT is
really all about. It's a hurdle that very few conventional psychologists
have been willing to jump (which is understandable)-- and that's why
PCT has not (and will not) taken conventional psychology by storm.

I tried to look up your text, too. Apparently, there is not a single copy
of it in the entire Indiana library network.

Thank god!

Who published it?

Brooks/Cole which is (was?) a division of Wadsworth.

The "Blind men..." paper is in the mail.

Best

Rick