HPCT as the Behaviorist Views It

[Bruce_Abbott]

[From Bruce Abbott (970223.2125 EST)]

Rick Marken (970223.1540) --

Your description of the Little Man is fascinating. Thank you.
It is certainly a correct description. What's interesting is
what is emphasized and what is played down. What is emphasized, it seems
to me, is what the model does (the objective results of its output);
what is played doen is what the model tries to perceive.
You have described the behavior of the model from the observer's rather
than from the model's perspective.

Not really. I wanted to describe what the model does (as seen by the
observer, the person playing with the simulation) as well as how the model
works, which includes a description of what the model controls and the
organization of its parts.

Actually, there is a system that controls the perception of the
retinal location of the target. This system controls it's perception by
varying the pointing angle of the eyes relative to the target.
An observer would see this as "a system to point the eyes at a
target" but, of course, it's really a system to bring a percpetion
(of retinal location) to a reference state.

Actually, there isn't, not in the model I was describing (Little Man I). As
I recall, the head turns so that the eyes face the target; what the model
computes is the angular difference between the position of the target and
the line of sight projected from the rotational axes of the head. However,
this would have essentially the same _effect_ as perception of retinal
location that you describe: if the simulation actually _had retinas, this
system would keep the image of the target approximately centered on them.
Nes pa, mon ami?

I think it's a good description of the behavior of an HPCT model --
as the behaviorist views it;-)

A behaviorist wouldn't have been talking about hierarchically organized
control systems with higher-level systems setting internally-specified
references for the lower-level ones. Sorry Charlie, your description won't
wash.

Regards,

Bruce

[Human_Factors_Lab]

[From Rick Marken (970223.2200 PST)]

Me:

Actually, there is a system that controls the perception of the
retinal location of the target.

Bruce Abbott (970223.2125 EST) --

Actually, there isn't, not in the model I was describing
(Little Man I). As I recall, the head turns so that the eyes
face the target;

How is this done if not by a control system controlling a
perception? Are you saying that Little Man I generated head
movements open loop?

A behaviorist wouldn't have been talking about hierarchically
organized control systems with higher-level systems setting
internally-specified references for the lower-level ones.

If your definition of a "behaviorist" is a person who wouldn't be
talking about "hierarchically organized control systems with
higher-level systems setting internally-specified references
for the lower-level ones" then I agree that it would be hard to
find a behaviorist talking about "hierarchically organized control
systems ...".

If, however, your definition of a "behaviorist" is (like mine) "a person
who believes that observable behavior is causad by external or internal
events" (as opposed to "a person who believes that behavior is the
control of perception") then it is easy to find behaviorists talking
about "hierarchically organized control systems ..." Albus, who wrote a
series of articles that appeared in the same 1979 Byte issues as Bill's
did, is one. Don Norman is another (he would prefer to call himself a
cognitive psychologist but all cognitive psychologists are behaviorists
from a PCT perspective). Of course, there is also the whole group of
behaviorists who apply MCT to behavior. I think EAB people like
Timberlake would be happy to talk about "hierarchically organized
control systems with higher-level systems setting internally-specified
references for the lower-level ones". Of course, Timberlake's references
(like those of any good behaviorist) are for behavior, not perception.

Sorry Charlie, your description won't wash.

Then why does it feel so clean and springtime fresh?

Best

Rick

[Bruce_Abbott]

[From Bruce Abbott (970223.0855 EST)]

Rick Marken (970223.2200 PST) --

Me:

Actually, there is a system that controls the perception of the
retinal location of the target.

Bruce Abbott (970223.2125 EST)

Actually, there isn't, not in the model I was describing
(Little Man I). As I recall, the head turns so that the eyes
face the target;

How is this done if not by a control system controlling a
perception? Are you saying that Little Man I generated head
movements open loop?

No, of course not. It is done closed loop, as I described in the part of
the above paragraph you didn't quote. Deviations of target angle from line
of sight (both horizontal and vertical) are brought to zero, the reference
value of both systems, via closed loop error correction.

If, however, your definition of a "behaviorist" is (like mine) "a person
who believes that observable behavior is causad by external or internal
events" (as opposed to "a person who believes that behavior is the
control of perception") . . .

Then your description of my analysis of Little Man I as "HPCT as a
behaviorist views it" is incorrect no matter which definition of behaviorist
is adopted -- yours or mine. I specifically described an organization of
heirarchically organized control systems, which of course produce their
actions as needed so as to control their perceptions.

Sorry Charlie, your description won't wash.

Then why does it feel so clean and springtime fresh?

(Sigh.) If feels dirty to me.

Bill P. agrees that my description was accurate, and so did you
(grudgingly). Why are you working so hard to make it _seem_ otherwise?
What is your hidden agenda?

Regards,

Bruce

[Richard_Marken1]

[From Rick Marken (970224.0910 PST)]

Bruce Abbott (970223.0855 EST) --

Bill P. agrees that my description was accurate, and so did you
(grudgingly). Why are you working so hard to make it _seem_
otherwise? What is your hidden agenda?

I'm not sure myself. It's hard to articulate, though I've tried in
various ways. My agenda is even hidden from me. Let me see if I can
figure out what it might be.

I think my agenda (relative to you) has to do with the fact that you
are such a fan of PCT, have such an excellent understanding of so
many of the details of the PCT model, and you have so many skills that
would be useful to the development of PCT as a science; yet I feel like
I am running up against an invisible wall when we get to issues that
seem to revolve around the basic concept of organisms as controllers
of their own perceptual experience.

One way this invisible wall appears is in your defensiveness about
conventional scientific psychology. I just don't see how it is possible
to grasp the basic idea of PCT (that organisms control their own
perceptual experience) and at the same time keep using (or advocating)
a conventional approach to psychological research. That's why I am
always nudging you about your psychological methods textbook. I just
can't believe that it is possible to understand the PCT model as well
as you do and and still think that the approach to research decribed in
conventional methods books -- an approach that completely ignores the
possibility that organisms are controlling perceptual variables -- is
valid.

I guess it's kind of a reality test. Every scientific psychologist I
know who has "gotten" PCT (and that's a precious small number) has
seen immediately the problems with the conventional approach to
psychological research. So I can't believe that you keep "pushing
back" when I suggest that _conventional_ IV-DV methodology (the kind
described in your textbook) is not appropriate for the study of
living control systems.

I think one of my agendas might be "reality testing"; repeating things
like "conventional_ IV-DV methodology is not appropriate for the study
of living control systems" in the hopes that I will hear what I would
expect to hear from someone who understands PCT as well as you do.
Something like: "Of course conventional_ IV-DV methodology is not
appropriate for the study of living control systems. Psychologists
have to learn how to Test for controlled variables. A few have
unintentionally done what is esentially The Test but we have to start
a systematic science based on the discovery of the perceptual variables
the organisms control".

I don't understand why I'm not hearing something like this and it's a
little nightmarish. I have never run into a situation quite like this
before. Usually people who don't want to get PCT (becuase it conflicts
with their own agendas) just don't bother understanding the model; they
reject what they THINK the model is about. But here you are, clearly
understanding the technical details of the model and yet, it seems, not
understanding what the model means (at least, not the way I do).

Maybe I'm wrong; maybe PCT fits seamlessly into conventional psychology.
Maybe the conventional methods textbook I wrote was just fine and I was
wrong to think that everything I said in it was invalidated by PCT.
Maybe PCT is just another model that explains the data of conventional
behavioral science. Maybe. But I've thought about this stuff for an
awfully long time now and it seems to me like this is not the case.

When I finally "got" PCT I _knew_ that nearly everything I had learned
from conventional psychology was wrong. I _saw_ that PCT cut away
the foundation of conventional scientific psychology (the causal
model). I guess I find it hard to believe that one can "get" the
PCT model as well as you do and still not "get" it the way I do. My
agenda, I guess, is to work through my befuddlement about this. If we
are both on the same side (the PCT side) then why do you always seem to
push back whenever I say things that seem to me so clearly compatible
with the most basic facts about control and PCT?

Best

Rick

[Bruce_Abbott]

[From Bruce Abbott (970225.1410 EST)]

Rick Marken (970224.0910 PST) --

Thanks for the candid answer to my question about your agenda. I can
understand the dilemma I raise for you, although I'm not sure there's much I
can do about it. We have different backgrounds, experiences, and
personalities, and for these reasons tend to see things very differently.

One way this invisible wall appears is in your defensiveness about
conventional scientific psychology. I just don't see how it is possible
to grasp the basic idea of PCT (that organisms control their own
perceptual experience) and at the same time keep using (or advocating)
a conventional approach to psychological research. That's why I am
always nudging you about your psychological methods textbook. I just
can't believe that it is possible to understand the PCT model as well
as you do and and still think that the approach to research decribed in
conventional methods books -- an approach that completely ignores the
possibility that organisms are controlling perceptual variables -- is
valid.

I'm afraid we're going to continue to disagree as to the value of scientific
research undertaken without an explicit recognition of behavior as the
control of perception; I've already attempted to communicate to you why I
think much of such research continues to be of value, without much success,
and have little hope that further argument will be any more convincing.
Research is conducted for many purposes, to answer different sorts of
questions (both practical and theoretical), and many methods have been
developed that will yield reasonable answers to those questions, given that
certain assumptions required by the methods hold at least approximately.
For example, there's a whole line of research examining how people (and
animals) make certain kinds of judgments, given certain information. This
research has turned up interesting facts; for example, it turns out that
people are not very good at dealing with probability (e.g., they often do
not behave rationally in placing bets, where "rationally" means behaving in
ways that will tend to maximize winnings and minimize losses, or some such,
according to the mathematical rules of probability. Further research has
attempted to disclose how people do make these judgements, which lead to
systematic kinds of errors such as the gambler's fallacy. So far as I can
see, these questions have more to do with the cognitive machinery we use to
reach certain conclusions than with issues of control, although how that
machinery works will certainly have implications for how well or poorly we
human beings will be able to control certain variables (e.g., risk). These
systems developed in order to facilitate control, but they can be studied in
their own right, and using standard methods of research as described in my book.

I guess it's kind of a reality test. Every scientific psychologist I
know who has "gotten" PCT (and that's a precious small number) has
seen immediately the problems with the conventional approach to
psychological research. So I can't believe that you keep "pushing
back" when I suggest that _conventional_ IV-DV methodology (the kind
described in your textbook) is not appropriate for the study of
living control systems.

Make no mistake about it, I do clearly see the problems with the
conventional approach to psychological research --- when applied to the
analysis of control systems. In such cases, open-loop analyses, or
closed-loop sequential analyses simply give the wrong answers. But when you
then assert that _all_ "conventional" research (conducted for whatever
purpose, to answer whatever question) becomes inappropriate the moment one
recognizes that living organisms control their perceptions, that's were we
part company. You seem to be constructing the following syllogism:

        All organisms are living control systems.
        Control systems require special methods of study.
        Therefore, organisms can only be studied using special methods
            designed for analyzing control systems.

To see the flaw in this logic, consider the following parallel structure:

        All automobiles are self-powered systems.
        Self-powered systems require special methods of study.
        Therefore, automobiles can only be studied using methods designed
            for analyzing self-powered systems.

But this is surely untrue. I wouldn't need a dynamometer and tachometer to
study how the radio works, for example, or the fuel-injection system (even
though the latter is a component of the powerplant). Methods developed to
understand how the engine develops its power would be of no use in
determining why there is excessive wind noise at 60 mph, or how well the car
corners.

I think you identify IV-DV with stimulus-response and, having correctly
rejected the latter as inappropriate in control-system analysis, therefore
also reject the former. Yet they are _not_ identical: even ordinary PCT
research can be characterized as IV-DV. For example, we might apply a
sine-wave disturbance to a given control system and observe the changes in
phase-lag, amplitude, and shape of the system's output, across several
sine-wave frequencies. That's IV-DV, although it certainly is not
stimulus-response or lineal cause-effect. It is not the methodology that is
wrong, but its inappropriate application in the open-loop or sequential
analysis of how a closed system of simultaneously changing variables will
behave.

The implications of PCT for the understanding of human and animal behavior
are profound; among them are the recognition that behavior exists in a
closed loop and that methods of study commonly used in the behavioral
sciences are inappropriate when applied to analysis of closed loops. But
this does not automatically invalidate their application to other research
problems and questions involving those living control systems, and that is
why I "push back" when you assert otherwise.

Maybe I'm wrong; maybe PCT fits seamlessly into conventional psychology.
Maybe the conventional methods textbook I wrote was just fine and I was
wrong to think that everything I said in it was invalidated by PCT.
Maybe PCT is just another model that explains the data of conventional
behavioral science. Maybe. But I've thought about this stuff for an
awfully long time now and it seems to me like this is not the case.

This paragraph captures what I view is wrong with your thinking on this
matter. You seem to think only in terms of either-or. Either PCT "fits
seamlessly into conventional psychology" or it overturns conventional
psychology completely. I think that the truth is somewhere between these
two extremes. Many of the findings of conventional psychology only show
that stable relationships exist among externally measured variables, without
offering any structure out of which such relationships would naturally
emerge. They are purely descriptive accounts which merely say, "under these
conditions, this is what is observed," or "in this situation, change this
variable, and that one will follow." Some attempts to account for these
relationship assume incorrect open-loop or equilibrium systems when what is
responsible for the observed relationships is the operation of closed-loop
control. The PCT analysis of these systems provides a physical basis for
these observed relations and replaces the invalid analytic approaches with
those appropriate for closed-loop systems. That's a quantum leap. Yet for
many research questions (even some relating directly to control system
analysis, as suggested above), established research methods will continue to
be of service. These methods have their appropriate application even in a
post-PCT world, so it is not a matter of either-or.

I may be wrong, but I get the impression that when I say that conventional
methods are appropriate for a variety of research questions in psychology,
you translate this to mean that I am claiming that conventional open-loop or
sequential analyses are O.K. for understanding the behavior of a closed loop
of simultaneously changing variables. If that is your understanding of my
position, I hereby state for the record that it is _not_ my position. Only
methods designed for analysis of such closed loops of variables will yield
correct answers.

Regards,

Bruce

[Bruce_Gregory9]

[From Bruce Gregory (970225.1550 EST)]

Bruce Abbott (970225.1410 EST)

to Rick Marken

Thanks for the candid answer to my question about your agenda. I can
understand the dilemma I raise for you, although I'm not sure there's much I
can do about it. We have different backgrounds, experiences, and
personalities, and for these reasons tend to see things very differently.

I do not mean to minimize the importance of what you say. I do
want to note, however, that Rick and I have very different
backgrounds, experiences, and personalities, yet we tend to see
things in very much the _same_ way.

        All organisms are living control systems.
        Control systems require special methods of study.
        Therefore, organisms can only be studied using special methods
            designed for analyzing control systems.

To see the flaw in this logic, consider the following parallel structure:

        All automobiles are self-powered systems.
        Self-powered systems require special methods of study.
        Therefore, automobiles can only be studied using methods designed
            for analyzing self-powered systems.

But this is surely untrue. I wouldn't need a dynamometer and tachometer to
study how the radio works, for example, or the fuel-injection system (even
though the latter is a component of the powerplant). Methods developed to
understand how the engine develops its power would be of no use in
determining why there is excessive wind noise at 60 mph, or how well the car
corners.

I see the situation somewhat differently. Imagine that
experimental science had developed before Galileo and Newton
showed us the fundamental importance of mass, velocity, and
acceleration in describing the motions of mechanical systems.
One can envision that large amounts of data might have been
collected that would later be seen to be useless because the
experimenters failed to include determinations of the values the
essential mechanical variables. Data gathered about living
things and their interactions seem likely to be equally useless
if those who gathered the data did not realize that what they
were studying was the often inadvertent side effects of
purposeful action.

Bruce Gregory

[Richard_Marken1]

[From Rick Marken (970225.1420)]

Bruce Abbott (970225.1410 EST) --

there's a whole line of research examining how people (and animals)
make certain kinds of judgments, given certain information. This
research has turned up interesting facts

But did it turn up the variables people are controlling when they make
these judgments?

Further research has attempted to disclose how people do make these
judgements

Yes. The research probably showed that people "process" the input
information in certain ways (a process that overestimates the
probability of events that have not occurred in some time -- the
gambler's fallacy) and produce judgements as outputs. Since the people
in these studies are in a closed loop relationship with respect to
their inputs, it seems pretty obvious to me that this research had
no chance of "disclosing" how, why or what these people were doing
when they made these judgements -- and it didn't.

So far as I can see, these questions have more to do with the
cognitive machinery we use to reach certain conclusions than with
issues of control

The questions have to do with "cognitive machinery" because the
researchers assume that people are cognitive machines who process
inputs and produce outputs. Cognitive researchers have had a
particularly difficult time seeing the error of their ways because
it's easy to find relationships between inputs and outputs using
conventional research techniques. In fact, people are not cognitive
machines; they are input controllers becuase they exist in a closed-
loop negative feedback relationship with respect to their inputs.

These [cognitive] systems developed in order to facilitate
control, but they can be studied in their own right, and using
standard methods of research as described in my book.

No they can't because these cognitive processes occur in a closed loop.
So you are using standard research methods, which are appropriate to the
study of open loop systems, to study closed loop systems.

Standard research methods assume that people are organized like this:

                input (IV) -->Cognitive Processes -->output (DV)

But people adon't suddenly become organized like this just because
you want to do a cognitive or perceptual experiment. People are
ALWAYS organized like this (except when they are dead):

  input (IV) -->controlled (CV) --> Cognitive Processes -->output (DV)
                                 variable |
                    ^ |

···

                                         >

                     -------------------------------------------

The DV that influences the CV may not be the one that is actually
measured in the experiment, of course.

If this is the way people are organized then I don't see how standard
IV-DV research can tell you much about cognitive process.

Make no mistake about it, I do clearly see the problems with the
conventional approach to psychological research --- when applied
to the analysis of control systems. In such cases, open-loop
analyses, or closed-loop sequential analyses simply give the
wrong answers.

Well, there you have it! If you understand this, why in the world would
you want to talk about conventional research at all in a course on
psychological methods. Are there any times (other than when they are
dead) that organisms are anything other than control systems? If so, how
do you tell when they are and when they are not control systems?

But when you then assert that _all_ "conventional" research
(conducted for whatever purpose, to answer whatever question)
becomes inappropriate the moment one recognizes that living
organisms control their perceptions, that's were we
part company.

I don't believe I've ever said this. I acknowledge that conventional
research methods are appropriate for many purposes. In particular,
they appropriate for the study of non-living systems. Thus, they are
the methods of choice in physics, chemistry and engineering.

I think you identify IV-DV with stimulus-response

Not at all. I identify IV-DV as an approach to studying living systems
that fails to take into account the fact that these systems are
controlling perceptual variables. The conventional IV-DV approach is
never aimed at determining the environmental correlates of the
perceptual variables that organisms control. The conventional IV-DV
approach is just as inappropriate (and misleading) when used by a
trendy cognitive psychologist as when used by a non-trendy behaviorist.
All the different warring schools of psychology are built on the SAME
basic model -- the causal model -- and they all (mis)use the same
methods (standard IV-DV) to test this model.

even ordinary PCT research can be characterized as IV-DV.

I now make this point in the paper I re-submitted to Psychological
Methods. Yes, The Test can be characterized as a type of IV-DV
research. But it is quite different from standard IV-DV research. In
the basic version of The Test, the IV is a disturbance to a
_hypothetical_ controlled variable; the DV is a measure of the
hypothetical controlled variable. The DV is measured WHILE the IV
is varied and you look for LACK of effect of the IV.

The results of The Test are equivalent to the results of a conventional
IV-DV experiment if the DV is NOT a controlled variable. In that case,
the IV will have an effect on the DV. But, unlike the conventional
researcher, the PCT researcher would not rush out to publish such an
effect, even if it were VERY strong. Instead, the PCT researcher would
try to find another definition of the controlled variable (DV); one that
is NOT influenced by variations in the IV.

For example, we might apply a sine-wave disturbance to a given
control system and observe the changes in phase-lag, amplitude,
and shape of the system's output, across several sine-wave
frequencies.

It would ONLY make sense to do this kind of research once you KNEW the
variable that was controlled by this control system. Your example here
pre-supposes that you have already done some version of The Test.

You seem to think only in terms of either-or. Either PCT "fits
seamlessly into conventional psychology" or it overturns conventional
psychology completely.

OK. I'm either-or. But you're not? I would think that a moderate
person would devote about half of a psychological research methods text
to standard IV-DV methods and the other half to PCT methods (The Test).
In fact, I could find nohing (zero) in your methods textbook about
controlled variables, perceptual variables, disturbances or The Test.
Not one peep. That suggest either-or to me -- and your "either" seems
to completely exclude my (which I thought was _our_) "or".

My either-or attitude did not result from some desire on my part to be
"either-or". I tried for some time (several years) to imagine that it
was possible to reconcile my understanding of PCT with my understanding
of how to do standard psychological research. Tom Bourbon went through
the same thing. We finally (and unwillingly) realized that there was no
compromise: if organisms are control systems (and they are) then
conventional psychological methods can't be used to determine what
they control; it's as simple as that.

I hope you allow yourself to realize this someday, too.

Best

Rick

[Bill_Powers1]

[From Bill Powers (970225.1730 MST)]

Rick Marken (970225.1420)--
Bruce Abbott (970225.1410 EST) --
Bruce Gregory (970225.1550 EST) --

I'd like to point out another difference between conventional psychological
research and the kind of research we try to do in PCT.

Bruce A. says

there's a whole line of research examining how people (and animals)
make certain kinds of judgments, given certain information. This
research has turned up interesting facts

The problem is that this sort of finding usually doesn't mean what it seems
to mean. When it is said that "people" make certain kinds of judgements,
this is not like saying that "people" doing a tracking task (or other
control task) perform certain kinds of actions in response to disturbances.
The difference is that in the tracking task, essentially 100 percent of the
"people" behave in a quantitatively predictable way: "people" means very
nearly "all people." But when you look at normal psychological facts, you
find that "people" means "a small majority", with no ability to predict
ahead of time which individuals will behave in accordance with the "fact."
If you say that "people" make a certain kind of judgement, the chances are
very large that the next person tested will not do so.

If someone were to go through all existing psychological findings to select
just those facts that are predictably true of, say, 95% of the people
tested, then there would be some basis for comparing PCT research with other
kinds of research. Bruce A., how much of the existing backlog of
psychological facts would survive this kind of screening?

Best,

Bill P.

···

But did it turn up the variables people are controlling when they make
these judgments?

Further research has attempted to disclose how people do make these
judgements

Yes. The research probably showed that people "process" the input
information in certain ways (a process that overestimates the
probability of events that have not occurred in some time -- the
gambler's fallacy) and produce judgements as outputs. Since the people
in these studies are in a closed loop relationship with respect to
their inputs, it seems pretty obvious to me that this research had
no chance of "disclosing" how, why or what these people were doing
when they made these judgements -- and it didn't.

So far as I can see, these questions have more to do with the
cognitive machinery we use to reach certain conclusions than with
issues of control

The questions have to do with "cognitive machinery" because the
researchers assume that people are cognitive machines who process
inputs and produce outputs. Cognitive researchers have had a
particularly difficult time seeing the error of their ways because
it's easy to find relationships between inputs and outputs using
conventional research techniques. In fact, people are not cognitive
machines; they are input controllers becuase they exist in a closed-
loop negative feedback relationship with respect to their inputs.

These [cognitive] systems developed in order to facilitate
control, but they can be studied in their own right, and using
standard methods of research as described in my book.

No they can't because these cognitive processes occur in a closed loop.
So you are using standard research methods, which are appropriate to the
study of open loop systems, to study closed loop systems.

Standard research methods assume that people are organized like this:

               input (IV) -->Cognitive Processes -->output (DV)

But people adon't suddenly become organized like this just because
you want to do a cognitive or perceptual experiment. People are
ALWAYS organized like this (except when they are dead):

input (IV) -->controlled (CV) --> Cognitive Processes -->output (DV)
                                variable

                   ^ |
                   > >
                    -------------------------------------------

The DV that influences the CV may not be the one that is actually
measured in the experiment, of course.

If this is the way people are organized then I don't see how standard
IV-DV research can tell you much about cognitive process.

Make no mistake about it, I do clearly see the problems with the
conventional approach to psychological research --- when applied
to the analysis of control systems. In such cases, open-loop
analyses, or closed-loop sequential analyses simply give the
wrong answers.

Well, there you have it! If you understand this, why in the world would
you want to talk about conventional research at all in a course on
psychological methods. Are there any times (other than when they are
dead) that organisms are anything other than control systems? If so, how
do you tell when they are and when they are not control systems?

But when you then assert that _all_ "conventional" research
(conducted for whatever purpose, to answer whatever question)
becomes inappropriate the moment one recognizes that living
organisms control their perceptions, that's were we
part company.

I don't believe I've ever said this. I acknowledge that conventional
research methods are appropriate for many purposes. In particular,
they appropriate for the study of non-living systems. Thus, they are
the methods of choice in physics, chemistry and engineering.

I think you identify IV-DV with stimulus-response

Not at all. I identify IV-DV as an approach to studying living systems
that fails to take into account the fact that these systems are
controlling perceptual variables. The conventional IV-DV approach is
never aimed at determining the environmental correlates of the
perceptual variables that organisms control. The conventional IV-DV
approach is just as inappropriate (and misleading) when used by a
trendy cognitive psychologist as when used by a non-trendy behaviorist.
All the different warring schools of psychology are built on the SAME
basic model -- the causal model -- and they all (mis)use the same
methods (standard IV-DV) to test this model.

even ordinary PCT research can be characterized as IV-DV.

I now make this point in the paper I re-submitted to Psychological
Methods. Yes, The Test can be characterized as a type of IV-DV
research. But it is quite different from standard IV-DV research. In
the basic version of The Test, the IV is a disturbance to a
_hypothetical_ controlled variable; the DV is a measure of the
hypothetical controlled variable. The DV is measured WHILE the IV
is varied and you look for LACK of effect of the IV.

The results of The Test are equivalent to the results of a conventional
IV-DV experiment if the DV is NOT a controlled variable. In that case,
the IV will have an effect on the DV. But, unlike the conventional
researcher, the PCT researcher would not rush out to publish such an
effect, even if it were VERY strong. Instead, the PCT researcher would
try to find another definition of the controlled variable (DV); one that
is NOT influenced by variations in the IV.

For example, we might apply a sine-wave disturbance to a given
control system and observe the changes in phase-lag, amplitude,
and shape of the system's output, across several sine-wave
frequencies.

It would ONLY make sense to do this kind of research once you KNEW the
variable that was controlled by this control system. Your example here
pre-supposes that you have already done some version of The Test.

You seem to think only in terms of either-or. Either PCT "fits
seamlessly into conventional psychology" or it overturns conventional
psychology completely.

OK. I'm either-or. But you're not? I would think that a moderate
person would devote about half of a psychological research methods text
to standard IV-DV methods and the other half to PCT methods (The Test).
In fact, I could find nohing (zero) in your methods textbook about
controlled variables, perceptual variables, disturbances or The Test.
Not one peep. That suggest either-or to me -- and your "either" seems
to completely exclude my (which I thought was _our_) "or".

My either-or attitude did not result from some desire on my part to be
"either-or". I tried for some time (several years) to imagine that it
was possible to reconcile my understanding of PCT with my understanding
of how to do standard psychological research. Tom Bourbon went through
the same thing. We finally (and unwillingly) realized that there was no
compromise: if organisms are control systems (and they are) then
conventional psychological methods can't be used to determine what
they control; it's as simple as that.

I hope you allow yourself to realize this someday, too.

Best

Rick

[Bruce_Abbott]

[From Bruce Abbott (970225.2245 EST)]

Rick Marken (970225.1420) --

Bruce Abbott (970225.1410 EST)

there's a whole line of research examining how people (and animals)
make certain kinds of judgments, given certain information. This
research has turned up interesting facts

But did it turn up the variables people are controlling when they make
these judgments?

That's the problem with your viewpoint in a nutshell, isn't it? Not only
must the human nervous system be a hierarchical control system, but so must
every sub-component of that system. The variable people were controlling
was probably something like conformance to the experimenter's request to
make the judgment using the information supplied to them. It would be no
different if they had been asked to solve Y = a + bX given certain values of
a, b, and X. The issue here is not what they are controlling, but how they
go about solving the problem set for them. For you, wielding your
control-system hammer (the Test), it is indeed true that every research
problem looks like a nail (a question of what is controlled).

Further research has attempted to disclose how people do make these
judgements

Yes. The research probably showed that people "process" the input
information in certain ways (a process that overestimates the
probability of events that have not occurred in some time -- the
gambler's fallacy) and produce judgements as outputs. Since the people
in these studies are in a closed loop relationship with respect to
their inputs, it seems pretty obvious to me that this research had
no chance of "disclosing" how, why or what these people were doing
when they made these judgements -- and it didn't.

Rick, here is a component of a simple proportional controller:

      --------->[Perceptual Input Function]--------->

Please identify the closed loop in this diagram.

I claim that the heuristic rules people use to make such judgments represent
the perceptual input function for risk.

So far as I can see, these questions have more to do with the
cognitive machinery we use to reach certain conclusions than with
issues of control

The questions have to do with "cognitive machinery" because the
researchers assume that people are cognitive machines who process
inputs and produce outputs. Cognitive researchers have had a
particularly difficult time seeing the error of their ways because
it's easy to find relationships between inputs and outputs using
conventional research techniques. In fact, people are not cognitive
machines; they are input controllers becuase they exist in a closed-
loop negative feedback relationship with respect to their inputs.

Every component of a closed-loop, negative feedback control system can be
viewed as an input-output device; i.e., one that transforms its inputs
according to certain rules into outputs. The input function does this. The
comparator does this. The output function does this. The environment
function does this. As such these components can be quite properly analyzed
using ordinary IV-DV methods. People are not cognitive machines, but
cognitive machines lie within them to serve the needs of perceptual control.

These [cognitive] systems developed in order to facilitate
control, but they can be studied in their own right, and using
standard methods of research as described in my book.

No they can't because these cognitive processes occur in a closed loop.
So you are using standard research methods, which are appropriate to the
study of open loop systems, to study closed loop systems.

It is one thing to assert that these cognitive processes form part of a
closed loop; it is quite another thing to assert that they are themselves
closed loop processes. Even in the case of hierarchical control, in which
the output "function" of the upper level is in fact the lower-level control
system, when the lower-level system is doing its job well it may be treated
as a simple input-output function from the point of view of analyzing the
higher-level system it serves.

Standard research methods assume that people are organized like this:

               input (IV) -->Cognitive Processes -->output (DV)

But people adon't suddenly become organized like this just because
you want to do a cognitive or perceptual experiment. People are
ALWAYS organized like this (except when they are dead):

input (IV) -->controlled (CV) --> Cognitive Processes -->output (DV)
                                variable

                   ^ |
                   > >
                    -------------------------------------------

People may be organized as in your diagram, but the cognitive process under
study can be analyzed separately. If what you say were true, there would be
no hope of identifying input functions, output functions, comparator
functions, in short, any of the components of a control system. It would
remain a veritable black box. In fact we can and do conduct such analyses.

Make no mistake about it, I do clearly see the problems with the
conventional approach to psychological research --- when applied
to the analysis of control systems. In such cases, open-loop
analyses, or closed-loop sequential analyses simply give the
wrong answers.

Well, there you have it! If you understand this, why in the world would
you want to talk about conventional research at all in a course on
psychological methods. Are there any times (other than when they are
dead) that organisms are anything other than control systems? If so, how
do you tell when they are and when they are not control systems?

Experiments can be arranged to examine component systems apart from their
operation within a closed loop system. When I look at a line and report
that it is straight, I am not controlling the straightness of the line but
only the accuracy of my report (I can choose to lie or tell the truth, to
examine carefully or with barely a glance) and the act of reporting itself
(and higher-level variables such as the perceived approval of the researcher
and so on). My ability to judge the line accurately may be needed in order
to control some variable (such as the shape of a line), and at that point it
would become a component of a control system and participate in a closed loop.

But when you then assert that _all_ "conventional" research
(conducted for whatever purpose, to answer whatever question)
becomes inappropriate the moment one recognizes that living
organisms control their perceptions, that's were we
part company.

I don't believe I've ever said this. I acknowledge that conventional
research methods are appropriate for many purposes. In particular,
they appropriate for the study of non-living systems. Thus, they are
the methods of choice in physics, chemistry and engineering.

I'm talking about conventional psychological research, about which you
_have_ said this, not physics, chemistry, and engineering.

even ordinary PCT research can be characterized as IV-DV.

I now make this point in the paper I re-submitted to Psychological
Methods. Yes, The Test can be characterized as a type of IV-DV
research. But it is quite different from standard IV-DV research. In
the basic version of The Test, the IV is a disturbance to a
_hypothetical_ controlled variable; the DV is a measure of the
hypothetical controlled variable. The DV is measured WHILE the IV
is varied and you look for LACK of effect of the IV.

Switch to a measure of system output (the output that counteracts the effect
of the disturbance to the putative CV) and you have an ordinary IV-DV study.
In the bad-old-days this relationship was misidentified as "stimulus-response."

For example, we might apply a sine-wave disturbance to a given
control system and observe the changes in phase-lag, amplitude,
and shape of the system's output, across several sine-wave
frequencies.

It would ONLY make sense to do this kind of research once you KNEW the
variable that was controlled by this control system. Your example here
pre-supposes that you have already done some version of The Test.

True but irrelevant to my point, which is that it is still good ol' IV-DV
research, which you assert does not apply to the study of control systems.
Obviously this assertion is not true.

OK. I'm either-or. But you're not? I would think that a moderate
person would devote about half of a psychological research methods text
to standard IV-DV methods and the other half to PCT methods (The Test).
In fact, I could find nohing (zero) in your methods textbook about
controlled variables, perceptual variables, disturbances or The Test.
Not one peep. That suggest either-or to me -- and your "either" seems
to completely exclude my (which I thought was _our_) "or".

The book was substantially complete when I joined CSG-net. You're barking
up the wrong time-frame.

My either-or attitude did not result from some desire on my part to be
"either-or".

How do you know, until you conduct the Test? (;->

I tried for some time (several years) to imagine that it
was possible to reconcile my understanding of PCT with my understanding
of how to do standard psychological research. Tom Bourbon went through
the same thing. We finally (and unwillingly) realized that there was no
compromise: if organisms are control systems (and they are) then
conventional psychological methods can't be used to determine what
they control; it's as simple as that.

You and Tom Bourbon were examining the problem from an overly restricted
viewpoint. _What_ organisms control is not the only problem that must be
addressed. Other problems require other methods. It's as simple as that.

Regards,

Bruce

[Bruce_Abbott]

[From Bruce Abbott (970225.2345 EST)]

Bruce Gregory (970225.1550 EST) --

I do not mean to minimize the importance of what you say. I do
want to note, however, that Rick and I have very different
backgrounds, experiences, and personalities, yet we tend to see
things in very much the _same_ way.

Yes I know. If this is a play for sympathy it won't work! (;->

I see the situation somewhat differently. Imagine that
experimental science had developed before Galileo and Newton
showed us the fundamental importance of mass, velocity, and
acceleration in describing the motions of mechanical systems.
One can envision that large amounts of data might have been
collected that would later be seen to be useless because the
experimenters failed to include determinations of the values the
essential mechanical variables. Data gathered about living
things and their interactions seem likely to be equally useless
if those who gathered the data did not realize that what they
were studying was the often inadvertent side effects of
purposeful action.

One can also envision that large amounts of data might have been collected
that would later be seen as quite useful. In the example to which you
refer, Tycho Brahe's accurate observations of planetary transits provide one
very real instance, which Kepler employed to develop and test his view that
the planets revolve about the sun in elliptical orbits.

In addition, your position is true only insofar as it is true that all
psychological research is directed toward determining the (open loop) causes
of behavior, aka stimulus-response. A Kuhnian paradigm shift in the form of
PCT would indeed make much of that research irrelevant. However,
psychological research is much more broadly distributed, so that many of its
discoveries remain valid even after such a revolution. Did relativity
theory destroy the relevance of all prior chemical investigation? Was
Helmholz' measurement of the speed of neural conduction thereby rendered
obsolute? Were all the studies of pendulums and balls rolling down inclined
planes consigned to the trash can? Did Mendel's experiments with peas have
to be rethought? Did the Roman aquaducts fail at the moment of publication
of Newton's mechanics? No, of course not. By the same token, and despite
other assertions to the contrary, PCT for all its power does not negate the
value all previous research and theory in experimental psychology.

Regards,

Bruce

[Human_Factors_Lab]

[From Rick Marken (970225.2250 PST)]

Me:

But did it [the conventional IV-DV research] turn up the variables
people are controlling when they make these judgments?

Bruce Abbott (970225.2245 EST)--

That's the problem with your viewpoint in a nutshell, isn't it?

Yes, I think it summerizes my viewpoint in a nutshell. The "problem",
of course, is yours.

The issue here is not what they are controlling, but how they
go about solving the problem set for them.

See. I just can't believe someone who understands PCT could say
something like this. What in the world do you think people are doing
when "they go about solving the problem set for them"?!?!?

For you, wielding your control-system hammer (the Test), it is
indeed true that every research problem looks like a nail (a
question of what is controlled).

Two points: 1) the central problem in understanding the behavior of
a control system is determining what perceptual variables it controls.
We use the "control-system hammer" (The Test) becuase we are looking for
the control system nail (controlled variables). If the system is not a
control system there is no penalty for using The Test; the causal nature
of the system will be quickly revealed in the IV-DV relationships found
while doing The Test. 2) conventional psychologists, wielding _only_
their causal sysem hammer (standard
IV-DV methodology) have never and will never discover the variables
organisms control; they will simply labor under the assumption that
they are dealing with a causal system, whether they are or not.

So my hammer (The Test) is a hell of a lot better than yours
(standard IV-DV methods)

Rick, here is a component of a simple proportional controller:

      --------->[Perceptual Input Function]--------->

Please identify the closed loop in this diagram.

There is no closed loop in this system so it would be perfectly
appropriate to study the perceptual input function using standard
IV-DV techniques -- as is done in physiological studies liek those
of Hubel-Weisel. I don't think the researchers in your "judgment" study
directly monitored the output of the perceptual input function, did
they? In fact, they used standard IV-DV methods to study
perceptual/cognitive processing in a closed loop system system.

You can't break intact organisms out of their closed loop
relationship with respect to their sensory inputs and study the
input-output relationships in the loop. Sorry.

Every component of a closed-loop, negative feedback control
system can be viewed as an input-output device...As such these
components can be quite properly analyzed using ordinary IV-DV
methods.

Yes. As long as you study these functions in an open loop situation. The
organisms in conventional psychological experiments are typically _not_
in an open loop situation; their sensory inputs influence what what they
do WHILE what they do influences their sensory inputs. All the
experimenter can do is introduce disturbances (the IV) into this loop.
Since the conventional experimenter has no clue that these IVs are
possible disturbances to controlled variables, he interpretes
covariation of IV and DV as a reflection of the open loop
characteristics of the system -- but it's not.

It is one thing to assert that these cognitive processes form
part of a closed loop; it is quite another thing to assert
that they are themselves closed loop processes.

I only assert the former. That's all that's needed to invalidate all of
cognitive psychology.

People may be organized as in your diagram, but the cognitive
process under study can be analyzed separately.

Not if people are organized as in my diagram. Sorry.

If what you say were true, there would be no hope of identifying
input functions, output functions, comparator functions, in short,
any of the components of a control system.

That's what modeling is about. But you can also study the input- output
characteristics of components of the loop directly by separating these
components from the loop. For example, you can study the open-loop
relationship between efferent spike rate and muscle contraction using
standard IV-DV techniques in an open-loop preparation (one in which the
muscle contraction can have no effect
on the efferent spike rate).

Experiments can be arranged to examine component systems apart
from their operation within a closed loop system.

Yes. But most _behavioral_ experiments -- such as those described in
your methods textbook -- cannot be so arranged. Standard IV-DV
methods are virtually _never_ appropriate in the study of the
behavior of (intact) living organisms.

Switch to a measure of system output (the output that counteracts
the effect of the disturbance to the putative CV) and you have an
ordinary IV-DV study.

Sounds good but this is BS; you would learn NOTHING about a putative CV
from such a study. Suppose my IV is the height from which I drop a
person and the DV is the time from release to landing. I find a nice
relationship between IV and DV. What's the CV?

The book was substantially complete when I joined CSG-net.
You're barking up the wrong time-frame.

So I take it that the next edition will include several chapters on
Testing for Controlled Variables? Do you include a substantial
section on The Test when you teach research methods now? If so,
I think it would be VERY informative if you could post your course notes
to CSGNet.

You and Tom Bourbon were examining the problem from an overly
restricted viewpoint. _What_ organisms control is not the only
problem that must be addressed. Other problems require other
methods. It's as simple as that.

Apparently, conventional psychology addresses ONLY those "other
problems" since there are precious few conventional research
studies that were _intentionally_ designed to test for controlled
variables. All these "other problems" must be a heck of a lot more
important than what seems to me to be the _central_ problem of
understanding behavior: what perceptions do organisms control?

What are all those other problems that psychologists are studying?
Are they things like figuring out how people "go about solving the
problems set for them"? Boy, that sure wouldn't have anything to do with
determining what perceptions an organism controls -- now,
would it?;-))

Best

Rick

[Bruce_Abbott]

[From Bruce Abbott (970226.1100 EST)]

Rick Marken (970225.2250 PST) --

Bruce Abbott (970225.2245 EST

The issue here is not what they are controlling, but how they
go about solving the problem set for them.

See. I just can't believe someone who understands PCT could say
something like this. What in the world do you think people are doing
when "they go about solving the problem set for them"?!?!?

I answered that question, to wit:

The variable people were controlling
was probably something like conformance to the experimenter's request to
make the judgment using the information supplied to them.

Obviously you didn't read my post carefully.

Two points: 1) the central problem in understanding the behavior of
a control system is determining what perceptual variables it controls.
We use the "control-system hammer" (The Test) becuase we are looking for
the control system nail (controlled variables). If the system is not a
control system there is no penalty for using The Test; the causal nature
of the system will be quickly revealed in the IV-DV relationships found
while doing The Test. 2) conventional psychologists, wielding _only_
their causal sysem hammer (standard
IV-DV methodology) have never and will never discover the variables
organisms control; they will simply labor under the assumption that
they are dealing with a causal system, whether they are or not.

So my hammer (The Test) is a hell of a lot better than yours
(standard IV-DV methods)

There you go again with either-or. I am not advocating exclusive use of
open-loop analytic methods. I have a hammer for nails and a screwdriver for
screws, and I know which tool to use on which. My position has nothing to
do with whether or not "conventional" psychologists have in the past used
IV-DV methods incorrectly to analyze close-loop systems (they have), but
rather, whether there are problems in psychology for which such methods are
appropriate. Once again you are confused about what I am asserting and,
having gotten it wrong, appeal to irrelevant facts.

You can't break intact organisms out of their closed loop
relationship with respect to their sensory inputs and study the
input-output relationships in the loop. Sorry.

Yes you can, and I provided an example which you chose to ignore, that of
judging the straightness of a line.

Yes. As long as you study these functions in an open loop situation. The
organisms in conventional psychological experiments are typically _not_
in an open loop situation; their sensory inputs influence what what they
do WHILE what they do influences their sensory inputs. All the
experimenter can do is introduce disturbances (the IV) into this loop.

You seem to believe that variables cannot be perceived without being
controlled. How strange! As I said, for you every system is a control
system, and can only be studied as such. Here I give a nice example of
studying a function in an open-loop situation, and point out that
psychologists have done a lot of work of this type and that for such work
conventional experimental techniques work perfectly well, and you once again
change the subject and rant about how conventional psychologists have often
used IV-DV methods inappropriately to study closed-loop systems intact. I
agree! So what? It has nothing to do with the position I have argued for.

It is one thing to assert that these cognitive processes form
part of a closed loop; it is quite another thing to assert
that they are themselves closed loop processes.

I only assert the former. That's all that's needed to invalidate all of
cognitive psychology.

No, you must also assert that these cognitive processes are _invariably_
part of a closed loop in which the output of the cognitive process is
affecting the state of the CV, which is simultaneously affecting the input
to the cognitive process. I agree that in that case you have a problem
using open-loop analytic techniques, but I also know that these cognitive
processes can be studied under open-loop conditions (i.e., the output of the
process does not affect its own input).

If what you say were true, there would be no hope of identifying
input functions, output functions, comparator functions, in short,
any of the components of a control system.

That's what modeling is about. But you can also study the input- output
characteristics of components of the loop directly by separating these
components from the loop. For example, you can study the open-loop
relationship between efferent spike rate and muscle contraction using
standard IV-DV techniques in an open-loop preparation (one in which the
muscle contraction can have no effect
on the efferent spike rate).

That's exactly what I am claiming you can do, although my argument goes much
farther than that.

Experiments can be arranged to examine component systems apart
from their operation within a closed loop system.

Yes. But most _behavioral_ experiments -- such as those described in
your methods textbook -- cannot be so arranged. Standard IV-DV
methods are virtually _never_ appropriate in the study of the
behavior of (intact) living organisms.

You're not claiming that the only way to examine a function open-loop is by
dissection, are you?

Switch to a measure of system output (the output that counteracts
the effect of the disturbance to the putative CV) and you have an
ordinary IV-DV study.

Sounds good but this is BS; you would learn NOTHING about a putative CV
from such a study. Suppose my IV is the height from which I drop a
person and the DV is the time from release to landing. I find a nice
relationship between IV and DV. What's the CV?

And there you go again. The only valid question to ask of a living organism
is, "what are you controlling?" I'm not denying the importance of that
question, but it clearly is not the only interesting one that can be -- and
has been -- asked.

Suppose I give you a list of 20 nonsense syllables (CVCs) to go over --
once, and tell you that I will be asking you to recall as many of them as
possible. I allow you to read the list and then, after a 5 minute retention
interval, I ask you to repeat as many of the items as you can. You recall
12 out of 20. Why only 12? Would more repetitions improve retention? Does
the position of the item within the list affect its chances of being
correctly recalled? What would happen if one of the item were made to stand
out, for example by printing it in red? Would a longer retention time
affect the number correctly recalled?

As a psychologist, you are of course aware that all these questions were
asked and answered a long time ago, but I bring them up to illustrate
research questions that do not focus on what is being controlled and yet are
interesting in their own right. There's nothing wrong with identifying
controlled variables, but you seem to think that that's the only research
worth doing. I strongly disagree with that assessment.

You and Tom Bourbon were examining the problem from an overly
restricted viewpoint. _What_ organisms control is not the only
problem that must be addressed. Other problems require other
methods. It's as simple as that.

Apparently, conventional psychology addresses ONLY those "other
problems" since there are precious few conventional research
studies that were _intentionally_ designed to test for controlled
variables. All these "other problems" must be a heck of a lot more
important than what seems to me to be the _central_ problem of
understanding behavior: what perceptions do organisms control?

Now you are pretending that my argument is that I think other types of
research are more important than research on controlled variables. I argued
for no such idea. Again, you drift off into irrelevant side-issues.

What are all those other problems that psychologists are studying?
Are they things like figuring out how people "go about solving the
problems set for them"? Boy, that sure wouldn't have anything to do with
determining what perceptions an organism controls -- now,
would it?;-))

The knowledge developed from such research might have everything to do with
how well people are able to control certain variables, even though the
research itself is not about control. For example, how people evaluate risk
may have everything to do with their ability to effectively control their
economic and personal well-being. How well people are able to judge the
straightness or angle of a line may have everything to do with how well they
will be able to control the drawing of a picture. How well people are able
to remember certain kinds of information may have everything to do with
whether they can control for success at a task requiring that information.

Apparently, conventional psychology addresses ONLY those "other
problems" since there are precious few conventional research
studies that were _intentionally_ designed to test for controlled
variables. All these "other problems" must be a heck of a lot more
important than what seems to me to be the _central_ problem of
understanding behavior: what perceptions do organisms control?

You are arguing that more psychologists should be conducting research to
determine what perceptions organisms control. I agree, and I've put my
money where my mouth is by switching into such research. But again, that's
another issue entirely from whether all research not directed toward this
objective is worthless; this latter is the position you have been defending
and against which I have been arguing for lo these many moons.

By the way, I see the question of _what_ specific perceptions are under
control as of only limited interest. What is controlled often changes
moment by moment. As I take a drink, I control a number of variables
related to my coffee cup, but I stop controlling them as soon as I set the
cup down. The central question for me is, how is the system organized so as
to make such flexible control possible? Some of those questions about
organization can be answered using ordinary methods of research. I prefer
to keep my toolbox well stocked. That means that it contains more than just
a hammer.

Regards,

Bruce

[Bruce_Abbott]

[From Bruce Abbott (970226.1440 EST)]

Bill Powers (970225.1730 MST) --

I'd like to point out another difference between conventional psychological
research and the kind of research we try to do in PCT.

Bruce A. (970225.1410 EST) says

there's a whole line of research examining how people (and animals)
make certain kinds of judgments, given certain information. This
research has turned up interesting facts

The problem is that this sort of finding usually doesn't mean what it seems
to mean. When it is said that "people" make certain kinds of judgements,
this is not like saying that "people" doing a tracking task (or other
control task) perform certain kinds of actions in response to disturbances.
The difference is that in the tracking task, essentially 100 percent of the
"people" behave in a quantitatively predictable way: "people" means very
nearly "all people." But when you look at normal psychological facts, you
find that "people" means "a small majority", with no ability to predict
ahead of time which individuals will behave in accordance with the "fact."
If you say that "people" make a certain kind of judgement, the chances are
very large that the next person tested will not do so.

Bill raises an important problem with many psychological studies, those
conducted based on group methodology. The results are cast in terms of
differences between the mean performances of groups of individuals exposed
to different experimental conditions (or of a single group observed under
different experimental conditions). The logic behind such designs is that
many extraneous variables cannot be held constant in the way that, e.g., the
variables affecting a pendulum's swing can. People (and animals) vary in
their physical characteristics, knowledge, attitudes, and so on, and these
differences can influence the observed relationship between the variables
being manipulated and observed performance. Statistical theory, however,
shows that the influence of these extraneous variables can be "controlled"
and evaluated statistically using appropriate group methodology (averaging
over participants) and statistical assessment. This approach _can_ work,
but it introduces the danger that the average effects uncovered by this
method may not apply to any single individual (Sidman, 1960). As Bill
knows, my own preference is strongly in favor of single-subject designs
(which as the name suggests do not use group averaging but focus on the
behavior of individuals), of which PCT-style investigations into control are
an excellent example. Along with some notable others in psychology, I am
concerned about the overreliance on statistical hypothesis testing in
psychological research to the near exclusion of estimation and in the text
on statistics I am currently writing I am saying so and arguing for
alternatives.

If someone were to go through all existing psychological findings to select
just those facts that are predictably true of, say, 95% of the people
tested, then there would be some basis for comparing PCT research with other
kinds of research. Bruce A., how much of the existing backlog of
psychological facts would survive this kind of screening?

I think that this question is misplaced, because it suggests that the
purpose of most investigation is to uncover facts that are "predictably
true" of some high percentage of people (or animals). Imagine we were able
to show that, say, 60% of those who are exposed to a given virus develop flu
symptoms within three days, whereas of those not exposed to the virus, only
1% developed flu symptoms in that period. To my mind this is strong
evidence that this particular virus is capable of producing flu symptoms in
human beings, even though 40% did not develop these symptoms on exposure to
it, and 1% developed such symptoms in the absence of such exposure. The
less than 100% penetration rate indicates that exposure to the virus is a
factor in the production of such symptoms, but not the only factor. That
seems an important fact to me, but presumably you would have us discard this
finding because it is not "predictably true" of at least 95% of the
population. That requirement may be quite proper in some cases and quite
improper in others, as in this example.

It would take more time and effort than I care to invest at the moment to
fully develop my views here on the reliability and usefulness of information
established through group-based statistical designs, and I don't really feel
like arguing about it at present. Suffice it to say that, as usual, I have
adopted a rather complex view full of "ifs, ands, and buts" that falls
between the extremes of "ain't nuttin' wrong with it" and "ain't nuttin'
right with it."

Regards,

Bruce

[Human_Factors_Lab]

[From Rick Marken (970226.1730)]

I just noticed this tidbit:

Bruce Abbott (970226.1440 EST)

I am concerned about the overreliance on statistical hypothesis
testing in psychological research to the near exclusion of
estimation and in the text on statistics I am currently writing
I am saying so and arguing for alternatives.

Well, _of course_ you don't want to go all the way with PCT.
You've got a statistics book to write. Gee, it never would
have occured to me to write a statistics book after I learned
PCT. Interesting.

Sorry to bother you with all that "Testing for the Controlled
Variable" stuff. Perhaps you would rather discuss the relative
merits of parametric and non-parametric significance tests?

Best

Rick

[Bruce_Gregory9]

[From Bruce Gregory (970226.1045 EST)]

Bruce Abbott (970225.2345 EST)]

In addition, your position is true only insofar as it is true that all
psychological research is directed toward determining the (open loop) causes
of behavior, aka stimulus-response. A Kuhnian paradigm shift in the form of
PCT would indeed make much of that research irrelevant. However,
psychological research is much more broadly distributed, so that many of its
discoveries remain valid even after such a revolution. Did relativity
theory destroy the relevance of all prior chemical investigation? Was
Helmholz' measurement of the speed of neural conduction thereby rendered
obsolute? Were all the studies of pendulums and balls rolling down inclined
planes consigned to the trash can? Did Mendel's experiments with peas have
to be rethought? Did the Roman aquaducts fail at the moment of publication
of Newton's mechanics? No, of course not.

What you say is quite true. But only scholars are able to trace
the contributions of alchemy to chemistry. The vast array of
alchemical studies (including those of the Sainted Newton,
alas, proved valueless. The alchemists were looking at the
wrong things... You maintain that psychological studies are
analogous to the studies of early science. I am afraid they are
more analogous to alchemy.

Bruce Gregory

[Bruce_Gregory9]

[From Bruce Gregory (970226.1030 EST)]

Bruce Abbott (970225.2245 EST)

To Rick Marken

That's the problem with your viewpoint in a nutshell, isn't it? Not only
must the human nervous system be a hierarchical control system, but so must
every sub-component of that system. The variable people were controlling
was probably something like conformance to the experimenter's request to
make the judgment using the information supplied to them. It would be no
different if they had been asked to solve Y = a + bX given certain values of
a, b, and X. The issue here is not what they are controlling, but how they
go about solving the problem set for them. For you, wielding your
control-system hammer (the Test), it is indeed true that every research
problem looks like a nail (a question of what is controlled).

I sympathize with your feelings because I held similar beliefs
for many years myself. Only as a result of complete frustration
with trying to make a cognitive approach work in a classroom did
I begin to look at what was happening in terms of control. The
power of this new perspective is so great that I am willing to
make the outrageous statement that the only thing worthy of
being called learning is the enhanced ability to exercise
control. Solving problems _is_ exercising control.

You and Tom Bourbon were examining the problem from an overly restricted
viewpoint. _What_ organisms control is not the only problem that must be
addressed. Other problems require other methods. It's as simple as that.

Control is the _only_ problem. Beat that for simplicity

Bruce Gregory

[Bruce_Gregory9]

[From Bruce Gregory (979226.1145 EST)]

Bruce Abbott (970226.1100 EST)]

Suppose I give you a list of 20 nonsense syllables (CVCs) to go over --
once, and tell you that I will be asking you to recall as many of them as
possible. I allow you to read the list and then, after a 5 minute retention
interval, I ask you to repeat as many of the items as you can. You recall
12 out of 20. Why only 12? Would more repetitions improve retention? Does
the position of the item within the list affect its chances of being
correctly recalled? What would happen if one of the item were made to stand
out, for example by printing it in red? Would a longer retention time
affect the number correctly recalled?

Ah, nonsense syllables. It's been a good many years since I
wrestled with similarity, meaningfulness, and percentage of
reinforcement. I even addressed the questions you cite as
examples. The truth is that we learned _nothing_ that enhanced
our understanding of what real people do when they really learn
something. We learned less than nothing, in fact, because the
lessons we drew from these studies were counter-productive.
Thanks for reminding me that I have come quite a way, even
though it isn't always obvious.

Bruce Gregory

[Martin_Taylor3]

[Martin Taylor 970226 11:00]

Rick Marken (970225.2250 PST)] to Bruce Abbott

Perhaps this message will be seen as wilfully malicious. If so, I'm sorry,
but Rick posts such a tempting target sometimes, rather like a sniper
who stands in an open field to shoot his victims, instead of hiding
up a tree. Brave, but then ...?

The problem is that this sniper is a ghost or a vampire, that doesn't
even know when ordinary bullets pass through him. So I guess he's not
being so brave when he stands in an open field with his AK-47.

Two points: 1) the central problem in understanding the behavior of
a control system is determining what perceptual variables it controls.

This may well be true, but you usually seem to define "the central problem"
as "the only problem." You have asserted that it is illegitimate to seek
how these perceptual variables are produced (computed, learned, structured),
that it is illegitimate to look at how output systems may be organized,
that it is illegitimate to consider how "the controlled variable" changes
as disturbances change, etc., etc. The _only_ thing that may legitimately
be studied is _what_ perceptual variable(s) is/are controlled--not even
when, why, or how "what" changes.

We use the "control-system hammer" (The Test) becuase we are looking for
the control system nail (controlled variables).

Yep, and your buildings are not screwed down, nor are they glued. They
are all nailed, regardless of whether you are dealing with framing,
brickwork, plumbing, electric supply, or wallpaper.

2) conventional psychologists, wielding _only_

their causal sysem hammer (standard
IV-DV methodology) have never and will never discover the variables
organisms control; they will simply labor under the assumption that
they are dealing with a causal system, whether they are or not.

Depends what they are using the methodology for. No, they won't discover
the variables the organisms control, and they won't build a house, because
they won't have the framing on which the brickwork, the plumbing, the
drywall and the electric supply are hung.

So my hammer (The Test) is a hell of a lot better than yours
(standard IV-DV methods)

There's no "better" about it. Without your hammer, the building can't
even be started. Using only your hammer, it can't be finished. There's
a place for every tool. "They" try to use screwdrivers where hammers are
needed. Useless. You hammer screws into the brickwork. Pointless (the
screw is, after you've finished, at least).

> Rick, here is a component of a simple proportional controller:
>
> --------->[Perceptual Input Function]--------->
>
> Please identify the closed loop in this diagram.

There is no closed loop in this system so it would be perfectly
appropriate to study the perceptual input function using standard
IV-DV techniques -- as is done in physiological studies liek those
of Hubel-Weisel. I don't think the researchers in your "judgment" study
directly monitored the output of the perceptual input function, did
they? In fact, they used standard IV-DV methods to study
perceptual/cognitive processing in a closed loop system system.

Nope. What Bruce said was right, that the variable the subjects control
is a perception of satisfying the experimenter (probably--and along with
many other things). What, in most perceptual experiments, they cannot
control, is the pattern the experimenter provides as a "stimulus" for
which they are asked to judge some aspect and produce a "response". Their
"response" has no effect then or later on the "stimulus" they are being
asked to judge. If their responses are "good" in the experimenter's view,
they will thereby reduce error in the perception they are acting to control.
But the sub-part, or component, of that loop between "stimulus" and "response"
is itself appropriately studied by IV-DV methods, because it _is_ an
open-loop element just as Bruce sketched (though he omitted the part
from PF to muscle output, which might be important).

We went through all this some years ago, and I'm surprised (no I'm not)
that you have mindlessly reverted to "Om mane percepti Om".

You can't break intact organisms out of their closed loop
relationship with respect to their sensory inputs and study the
input-output relationships in the loop. Sorry.

Actually, in electronic/electrical systems there is a technique called
"clamping" that effectively breaks loops by ensuring that the signals
in one part cannot affect themselves. The dual of clamping (fixing
some voltage) is cutting the circuit (fixing the current at zero).
Same in psychological studies. A component can be studied by IV-DV
methods if the effects at the output cannot affect the values at the input.

What I will grant you is that these studies are very often misapplied
when brought into the real world.

> Every component of a closed-loop, negative feedback control
> system can be viewed as an input-output device...As such these
> components can be quite properly analyzed using ordinary IV-DV
> methods.

Yes. As long as you study these functions in an open loop situation. The
organisms in conventional psychological experiments are typically _not_
in an open loop situation; their sensory inputs influence what what they
do WHILE what they do influences their sensory inputs.

I don't know what you mean by "conventional psychological experiments."
Certainly not psychophysics, which is ordinarily open-loop as described
above, though some psychophysics is done deliberately closed loop--I've
done a lot of it that way myself in my younger days.

For example, I studied the motion after-effect by giving people control
over the speed of, say, a rotating disk, and asking them to keep its
rotation speed at zero (or some other prespecified value), after they
had looked at one they couldn't control rotating for some period of time.
I thought that was conventional psychology, and the journals to which I
submitted the articles seemed to have no problem with this view.
And when I asked people to say whether a particular time-interval had
or had not contained a tone, that was open-loop so far as the connection
between the waveform and the yes-no button push is concerned. Conventional
psychology, _and_ compatible with PCT.

Martin

[Bill_Powers1]

[From Bill Powers (970226.1215 MST)]

Bruce Abbott (970226.1440 EST)--

If someone were to go through all existing psychological findings to
select just those facts that are predictably true of, say, 95% of the
people tested, then there would be some basis for comparing PCT research
with other kinds of research. Bruce A., how much of the existing backlog
of psychological facts would survive this kind of screening?

I think that this question is misplaced, because it suggests that the
purpose of most investigation is to uncover facts that are "predictably
true" of some high percentage of people (or animals). Imagine we were
able to show that, say, 60% of those who are exposed to a given virus
develop flu symptoms within three days, whereas of those not exposed to
the virus, only 1% developed flu symptoms in that period. To my mind this
is strong evidence that this particular virus is capable of producing flu
symptoms in human beings, even though 40% did not develop these symptoms
on exposure to it, and 1% developed such symptoms in the absence of such
exposure. The less than 100% penetration rate indicates that exposure to
the virus is a factor in the production of such symptoms, but not the only
factor. That seems an important fact to me, but presumably you would have
us discard this finding because it is not "predictably true" of at least
95% of the population. That requirement may be quite proper in some cases
and quite improper in others, as in this example.

I'm thinking primarily in terms of theories of human nature, not mass
measures that are of interest to, say, epidemiologists or public health
officials. The use of population statistics is quite appropriate for those
whose professions are concerned with populations.

There's a very clever commercial that I see on the Weather Channel (it's
probably running elsewhere, too). It starts out with a big scare headline
saying that something like 30 percent or 40 percent of people don't survive
their FIRST heart attack, so even if you have no symptoms, you should take
precautions by swallowing their pills. This gives them a potential market of
250,000,000 customers, most of whom are not going to have a heart attack at
all. This is a nice example of the conflict between people wno deal with
populations and individuals who care primarily about one case.

The big problem with statistical measures arises when one party is concerned
with population statistics and the other is concerned with a specific case.
It's NEVER in your interests to take a test when you apply for a job, for
instance. The employer can benefit from a test that predicts performance
accurately only a few percent more often than half the time, but the job
applicant taking that test would have close to a 50% chance of being
rejected because of the test results. If you can skip taking the test, by
all means do.

Anyway, since I want to know how people work, I'm interested only in
theories that work out practically all the time, the more the better. But
I've given my spiel on the kinds of facts a real science needs too many
times already.

Best,

Bill P.

[HANS_BLOM2]

[Hans Blom, 970227c]

(Bruce Abbott (970226.1100 EST)) and (Rick Marken (970225.2250 PST))

Bruce:

You seem to believe that variables cannot be perceived without
being controlled. How strange!

The discussion that Bruce and Rick are having suggests strongly to me
that humans have a built-in goal to have others see and pay attention
to the same things. You find that in small kids: "Mama, look!" With a
lot of disappointment if mama doesn't. We see the same mechanism at
all times and everywhere. We see it daily in the news reported by
newspapers and TV. With a lot of disappointment if someone else saw
something first, whether important or just a gossip. And with even
more disappointment -- or even distrust -- if someone else sees
something different: "No, there really is no green giant under the
bed".

This built-in urge to perceive the same world makes us "social
animals" and is the origin of culture, I guess. With all its
advantages and disadvantages. We all have that urge, so we ought not
complain when we find it in others as well.

Yet it bothers me, occasionally, to discover that some (sometimes
significant) others have such a strong urge to want to define the
world that I live in...

Greetings,

Hans