cyclic ratio data: BRUCE IS RIGHT

[From Bill Powers (950731.1200 MDT)]

Bruce Abbott (950730.1920 EST) --

This is getting tiresome. Bruce, you are right again. Boy, are you
right!

I started setting up cyclic3.pas. First I fitted a straight line to the
original behavioral data, to get a regression and an intercept. Since
that used up the data, I then computed p by using that straight line for
each value of m instead of the actual data. This gave me a smooth curve
for the values of p as a function of m. I then discovered that by
adjusting the collection time slightly away from your values, I could
get p to be exactly constant, as close as I could read the displays.

The values of c for rats 1 through 4 were 5.60, 5.66, 5.70, and 5.44 --
not much different from the values you got.

That sent me back to the drawing board. Knowing that it was possible for
the pressing rate p to be exactly constant, I went through the equations
again with that in mind.

First, the function describing the mean behavior rate b as a function of
the food delivery rate r:

(1) b = G*(r0 - r) where r0 is the "reference level." You'll see why I
put that in quotes shortly.

b = m/(m/p + c)

where p is pressing rate between collections, c is collection time.

r = 1/(m/p + c), so

m/(m/p + c) = G*r0 - G/(m/p + c)

Collecting terms with a common demoninator, we get

(m + G)/(m/p + c) = G*r0, or

(2) m+G = G*ro*(m/p+c)

If p is a constant the only variable is m and we can differentiate both
sides with respect to m:

d/dm:

   1 = G*r0/p, or

(3) p = G*r0

Plugging that back into (2) we get

m + G = m + G*r0*c or

G = G*ro*c, or

r0 = 1/c

The so-called reference level turns out to be just the maximum rate at
which food deliveries can be obtained.

From (3), this gives us

G = pc

We can now put these derived values into the original "control system"
equation (1) and get

b = pc*(1/c - r)

For Rat 1, p = 5500 and c = 5.60/3600, so the equation is

b = 8.4*(643 - r)

This looks like a control system with a gain of 8.4 and a reference
level of 643 reinforcements per hour. The only problem is that it's not
a control system: it's a crock.

ยทยทยท

---------------------------------
In this experiment each rat is either collecting food or pressing the
bar at a constant rate. This is true for all schedules from FR-2 to FR-
64. The schedule has absolutely no effect on the behavior. All of my
elaborate modeling, and all of Staddon's, I presume, is worth exactly
nothing. The rat was either pressing the bar at a constant rate or
collecting the food, and that is all that happened in these experiments.
We aren't seeing reinforcement and we aren't seeing control. What we're
seeing is a very hungry rat pressing the bar as fast as it can, given
other limitations, and collecting the food whenever it appears.
---------------------------------
Obviously the same consideration will be seen in the Motherall data, at
least for the right limb where there is essentially a negative straight-
line relationship. So all of my modeling of those experiments and
Staddon's models as well are bull***t.

What remains now is either to get hold of the original data and verify
that the pressing rate is a constant, or to do an experiment where we
can get the same data. If the data check out with these deductions, a
very large pile of experimental data (and all the fancy analyses of
them) is bound for the garbage can.
---------------------------------
Bruce, you are a genius. When you thought of checking out the time per
reinforcement against the pressing rate, you pulled the foundations out
from under everything. Your initial deductions were right: the behavior
is not varying, and the curves we see are strictly an artifact.

This should not affect OPCOND5, because the collection delay is built
into it and I believe we can reproduce these effects. Maybe when we see
how the model parameters have to be set up to imitate the Staddon data,
we will get some insight into the kind of conditions needed to show a
true control phenomenon.
----------------------------------------------------------------------
Best,

Bill P.

[From Rick Marken (950731.1310)]

Mary Powers (950731) --

The idea is for people to learn it [PCT] well enough so that if they do go
out and apply it and problems arise they can, by thinking about the model,
generate the answers themselves.

I feel exactly the same way about it; I think it is important for applied
people to learn the model so that, when unanticipated problems arise, they
can generate answers themselves. I don't think one can apply PCT effectively
by just memorizing the mantras ("control of perception", "hierarchy of
goals", etc). I think the most effective practitioners will be those who
understand the nature of the systems on which they are practicing --
perceptual control systems.

Bill Powers (950731.1200 MDT) --

In this experiment each rat is either collecting food or pressing the
bar at a constant rate. This is true for all schedules from FR-2 to FR-
64. The schedule has absolutely no effect on the behavior. All of my
elaborate modeling, and all of Staddon's, I presume, is worth exactly
nothing...We aren't seeing reinforcement and we aren't seeing control. What
we're seeing is a very hungry rat pressing the bar as fast as it can, given
other limitations, and collecting the food whenever it appears.

So after tons of work we have managed to discover that the organisms in these
experiments are not even controlling reinforcement. This seems to me to be
the ultimate proof of exactly what Tom and I have been arguing: conventional
operant research is of absolutely no use to PCT.

Bill and Bruce have been trying like crazy to sqeeze some information out of
the operant data posted to the net, under the assumption that the organism in
these experiments is in a position to control at least one perceptual
variable (reinforcement rate); the organism can perceive this variable and
influence it so there is a closed loop relationship between the organism and
reinforcement rate. But operant researchers never did the _sine qua non_ of
research on control -- they never did The Test to determine that a variable
IS under control. So we assumed that reinforcement rate was under control
because it _looked_ like it was.

But reinforcement rate is apparently NOT under control in these operant
studies; the operant situation seems to be set up so that, even when the
"work demands" are low (low ratio requirement) the organism is unable to
control reinforcement rate . All the organism can do is respond as quickly as
possible and get whatever reinforcement results.

If the data check out with these deductions, a very large pile of
experimental data (and all the fancy analyses of them) is bound for the
garbage can.

Yes, indeed. If there is no evidence of control, then these data do go into
the garbage can. But I think we could have saved a lot of time and effort
(and embarrassment) if we had thrown this data into the garbage can right off
the bat. We embarassed ourselves because we tried to make PCT sense out of
data that were collected without any understanding that control might be
going on; there was no Test for controlled variables so there was no evidence
that we were even looking at a control phenomeneon (and, as it turns out, we
were apparently NOT looking at a control phenomenon). We embarassed ourselves
by trying to show that PCT could handle conventional data that "seems" to
involve control -- when, in fact, it doesn't. Why can't we remember --
phenomena first!

Bruce, you are a genius...Your initial deductions were right: the behavior
is not varying, and the curves we see are strictly an artifact.

This was genius on Bruce's part and I hope we learn a lesson from it. The
lesson is: when people come to us with data and ask how PCT explains it, the
proper reply is to ask how the researchers determined which variable is under
control. If the answer is not a description of the Test for the controlled
variable, then the proper reply is to walk away.

Best

Rick

[From Bruce Abbott (950801.1220 EST)]

Bill Powers (950731.1200 MDT) --

This is getting tiresome. Bruce, you are right again. Boy, are you
right!

Wow, right twice in six months! I guess if if guy keeps swinging the bat
he'll hit one eventually. Actually, on this one I'd rather be wrong.
Staddon is one of the few people in EAB with a grasp of the basic control
process, although he tends to apply it rather narrowly. So here we have
what appears to be a rather nice application, and it fails. I was happier,
Bill, when you were saying that the apparent linearity in the data was a
coincidental result of opposing effects [Bill Powers (950731.0815 MDT)].

The so-called reference level turns out to be just the maximum rate at
which food deliveries can be obtained.

It may be more accurate to state that it's the rate at which food deliveries
_are being_ obtained, rather than the maximum rate at which they _can be_
obtained. The observed rate may depend on such factors as the level of
deprivation, size of food pellet, etc. Clearly there must a maximum rate at
which the food can be collected given the characteristics of the apparatus
and the rat's own physical limitations, but I suspect that these animals
were taking more time than that. However, the amount of time they do
require is independent of the ratio.

In this experiment each rat is either collecting food or pressing the
bar at a constant rate. This is true for all schedules from FR-2 to FR-
64. The schedule has absolutely no effect on the behavior. All of my
elaborate modeling, and all of Staddon's, I presume, is worth exactly
nothing. The rat was either pressing the bar at a constant rate or
collecting the food, and that is all that happened in these experiments.
We aren't seeing reinforcement and we aren't seeing control. What we're
seeing is a very hungry rat pressing the bar as fast as it can, given
other limitations, and collecting the food whenever it appears.

This interpretation says that the rat has reached some upper physical limit
in its ability to generate a high rate of pellet delivery. This would
explain the constancy in response rates and the lack of actual control over
rate of reinforcer delivery. However, this appears inconsistent with the
fact that the rates are also constant _but lower_ at a lower level of food
deprivation (95%). It is _as if_ a given level of error in the "stomach
loading" system is determining the observed response rates, rather than
determining the reference level for rate of pellet collection.

The question is, does this make sense? In nature the rat must vary its rate
of food discovery and injestion as required to maintain itself in a proper
nutritional state. If circumstances are such that the usual level of
activity fails to produce the necessary rate of food injestion, the error in
the nutrient-system increases, leading to higher rates of food-seeking activity.

In the operant chamber, however, the state of the nutrient system is
carefully maintained at a nearly constant level of error. If the rate of
food-gathering activity depends on this error, that rate will remain roughly
constant so long as the error remains roughly constant. Activity itself and
other metabolic processes will be slowly raising that error, whereas
injection of food pellet will be lowering that error (at a rate dependent on
the rate of delivery of the pellets, their size, and their nutrient
content). Over the course of a 1-hr session, this error may change little
or, more likely, will be reduced through food consumption.

Whatever happens to it, the change will be corrected between sessions
through the experimenter's intervention. However, if the rat had to earn
_all_ its food by responding on a given ratio schedule, response rates that
failed to produce enough food would lead to increasing deprivation level
(error) and thus to higher response output. In other words, under this
condition we should see control. In the typical operant study, by holding
error constant, at best we are able to measure only the gain of the output
function at the particular level of error we have induced in the nutrient
control system.

Furthermore, there is the problem of competing control systems. Given a
certain level of error in the nutrient control system and in, say, a system
controlling whatever perception is involved in excercising, and given that
the rat can't both lever-press and run in a running wheel at the same time,
the availability of the running wheel means that part of the time that might
be spent obtaining food pellets will now be spent reducing the error in the
latter system instead. Thus, at a given level of deprivation, placing a
running wheel in the operant chamber is likely to lead to reduced average
output on the lever. Questions: (1) How will the rat "allocate" its time?
(2) Does the answer to this question fall naturally from a PCT analysis of
the situation? (3) If so, how?

Regards,

Bruce

<[Bill Leach 950801.23:10 U.S. Eastern Time Zone]

[From Rick Marken (950731.1310)]

Bill and Bruce have been trying like crazy to sqeeze some information
out of the operant data posted to the net, under the assumption that the
organism in these experiments is in a position to control at least one
perceptual ...

But Rick, they have "squeezed" a great deal of information directly due
to the effort. It is very important to understand with precise detail
exactly what is wrong with the data and why. We already know that it is
completely useless to just say that one is conducting a test in such a
fashion as to ignore the most important characteristic of the test
subject.

The effort that Bruce and Bill are making, if continued in a methodical,
dispassionate and ruthlessly honest manner could well provide the sort of
explainations that will help to influence other researchers in EAB that
are already uncomfortable with both results and existing explainations.

Bruce's ability to view this work from the perspective of an active EAB
researcher is doubtless invaluable. As a person on the "sidelines"
(still), I can not adequately express my appreciation for efforts at
intellectual honesty that I see in the exchanges between Bill and Bruce.

I wish that everyone, myself included, could learn from watching these
two wrestle with mutual understanding. People with significant
differences in beliefs concerning matters of great personal concern would
see that when there is agreement at levels dealing with matters such as
how to search out one's own assumptions (or inferences), how to question
even that which one is "certain" about; then it IS possible to begin to
develop a body of understanding and agreement from which both learn.

Watching the evolution of Bill and Bruce's struggle has been and is, to
me, exciting and inspiring.

-bill

[From Rick Marken (950803.1615)]

Me:

Bill and Bruce have been trying like crazy to sqeeze some information
out of the operant data posted to the net, under the assumption that the
organism in these experiments is in a position to control at least one
perceptual variable...

Bill Leach (950801.23:10) --

But Rick, they [Bill and Bruce] have "squeezed" a great deal of information
directly due to the effort. It is very important to understand with precise
detail exactly what is wrong with the data and why.

I am not convinced that a "great deal" of information has been "squeezed".
But I am certainly not against looking at existing data from a PCT
perspective. I am confident that organisms are, indeed, perceptual control
systems so I have no doubt that it is possible, in principle, to develop a
PCT model that will explain the results of ANY behavioral experiment ("in
principle" because the data needed to develop a control model is typically
not available). So what's my beef?

I want to see a science of living control systems started as soon as
possible (before I die would be nice). I don't think the basic insights of
PCT (that behavior is control; that organisms control perceptual inputs) will
become part of mainstream thinking about behavior until such a science
begins. PCT will not survive long as a belief system and it will not become a
science until people start doing it.

PCT science will not be anything like conventional behavioral science. The
goal of PCT science will be to undersatnd control, not to exert it (as in
conventional operant conditining experiments, for example). PCT science will
be about what variables organisms control, how they control them and how they
come to be able to control them (among other things).

PCT science will not start until conventional behavioral scientists drop what
they are doing and start doing PCT. It is simply impossible to build a
science of living control systems on the database of conventional behavioral
science. This database was collected under the assumption that organisms are
cause-effect rather than control systems. Therefore, the database contains
only incidental evidence of the variables controlled by organisms. This
does not mean that there is no evidence of controlled variables; and it
doesn't mean that there is no data based on what is essentially a test for
controlled variables. It only means that there is not much data that clearly
and precisely reveals the nature of the controlling done by living systems.

I think we should not ignore the database of conventional behavioral science;
but I also think we should not expect to be able to make precise and valid
models of what we find in that database, which consists largely of
irrelevant side effects of control. The conventional database is like
shadows on the wall; PCT provides a likely explanation of what's making the
shadows but only PCT research can clearly reveal the hand before the light.

As the discussion of the operant data continues, I think it becomes more and
more apparent that the data in the conventional database doesn't tell us what
we need to know (from a PCT perspective); I think we have to take our eyes
off the shadows and look directly at the cause of those shadows -- the
living control system. As Bill Powers (950803.1355 MDT) just said:

I think the conclusion is inevitable. We have to do these experiments
with your rats. There are just too many unknowns behind the published
data.

I would imagine that this will be the conclusion we eventually come to after
reviewing most conventional behavioral data: there are just too many
unknowns. And the unknowns are unknown because no one was looking, carefully
and systematically, for evidence of control.

So look at the old data if you must; but let's start doing PCT science soon.

Best

Rick

<[Bill Leach 950804.18:30 U.S. Eastern Time Zone]

[From Rick Marken (950803.1615)]

Bill P. said it much better than I, but PCT IS addressing the specific
questions to find out about what is going on in EAB experiments.

I still agree with Bruce (as I understand him); It is first necessary to
understand what is NOT known about an experiment before on can begin to
shift experimental methods toward PCT methods in an existing methodology.

EAB still looks like a good choice. There is a great deal of "similar"
experimental results from sufficiently similar experiments to allow a
well conducted set of PCT based experiments to provide data that can be
viewed from both forms of analysis.

Just think about how difficult it is for most reasonably intellegent
people to make the "jump" from recognizing a VERY simple control example
to understanding even in a small way some of the MOST obvious
implications of living systems as control systems.

It IS asking an awful lot of any bunch of behavioural scientists, trained
for years in a "non-control" school of belief to recognize the
significance of control to their own work.

SO, yes you are right! PCT based research needs to begin as soon as
possible in as many ways as possible. But someone has to begin this
extension to what has already been done and it might be nice if that
(those) person(s) could communicate their results to the general
scientific community and be understood.

-bill