Early PCT Research?

[From Bruce Abbott (950602.1455 EST)]

It would seem we agree that the Badia, Harsh, Culbertson, & Abbott (1976)
study I described (950601.1635 EST) does indeed apply "the Test" for the
controlled variable, but there were some issues raised about the research
that I'd like to address here:

Rick Marken (950601.1900) --

What happened was that the rats pressed the lever quickly and reliably
enough to spend 85-95% of session time in the signaled shock condition.

This is a nice piece of data because it provides at least a rough measure of
control. The presumed controlled variable was in a particular state 85-95%
of the time; if the rat had done nothing the controlled variable would have
been in that state only, what, 50% of the time? So the control loop is
definitely keeping shock signalling schedule under control.

Actually, if the rat had done nothing it would have been in the unsignaled
condition 0% of the time during the testing sessions.

I do think you
could have spent more time nailing down the controlled variable, though.
It seems like there were some other very plausible possibilities, given your
description of the study.

This research was part of a larger program designed to identify the variable
or variables actually being controlled (although these researchers would
have SAID that they were attempting to identify the controlling variables,
thus puting the cart before the horse). There are many variables that
differ in value between the signaled and unsignaled conditions; any of these
could be a (or the) controlled variable and they are all confounded in the
simple test: state of the houselight, presence/absence of warning,
opportunity to prepare for impending shock, identifiable periods of safety
during which shock is guarenteed not to occur, and others.

For example, the subject might have been
controlling for having the light on, regardless of the shock signalling
schedule.

This had been tested in an earlier study. Rats prefer the signaled schedule
whether it is associated with light-on or light-off. Also, if you place the
rat on the signaled schedule and lever-pressing produces the unsignaled
schedule, the rats learn to avoid pressing the lever, thus remaining on the
signaled schedule full time.

It would also have been nice if you had tried a number of

different disturbances to determine that it was, indeed, the signalling
schedule that was under control.

I'm not sure what you mean by "different disturbances." We did manipulate a
number of variables in an effort to identify which were controlled. For
example, in the cited study the "dependability" of the signal as a predictor
of shock [i.e., p(shock|signal] was manipulated, as was the "dependability"
of signal absence as a predictor of safety [i.e., p(shock|no signal)]. The
rats continued to control for the signaled schedule over a wide range of
values for p(shock|signal) but not when p(shock|no signal) was degraded.
False alarms seem to be less of a problem for the rat than failures to warn.

Bill Leach 950602.00:56 U.S. Eastern Time Zone --

Did the rats ever press the lever during the signaled condition? If so
then what does that mean?

Good question. Yes, they sometimes did, but they will do this even before
they have learned the contingency between lever pressing and schedule
condition (i.e., during initial training, when lever pressing has no effect
on the schedules). The levers are placed in the chamber in a position such
that they will occasionally get pressed as a byproduct of exploratory
activity--otherwise they would never learn what effect lever-pressing has.
This same activity during the signaled schedule can generate lever presses;
these were recorded but had no programmed consequences. We assume that
these presses are just byproducts of exploration, as opposed to the
purposive lever pressing that is observed immediately after the schedule
switches to unsignalled. At any rate, they occured infrequently.

I personally have quite a bit of trouble with this sort of experiment in
that I believe that the whole setup is contrived. The rats were
essentially placed into a situation that would never exist in their
"normal" environment.

This is a familiar concern; you are worried about what experimental
psychologists call the "external validity" of the research: how well the
findings can be generalized from the laboratory to the "real" world. An
excellent discussion of this issue can be found in

Mook, D. G. (1983). In defense of external invalidity. _American
  Psychologist_, April, 379-387.

The rat is still the same system and will behave according to the rules
imposed by its structure and organization, whether the situation is
"natural" or "artificial." When you wish to analyze the performance of a
designed control system, you no doubt subject the system to artificial
conditions (e.g., step functions, open-loop conditions) that the controller
will never encounter in the real world, because the response of the system
under such conditions reveals aspects of its performance that are difficult
or impossible to see under normal operating conditions. A rat in an operant
chamber may not be facing the rich complexity of conditions it usually
encounters in the wild, but its behavior will still reflect the basic
organization of its system, and the test conditions may reveal aspects of
its performance that would difficult or impossible to observe under those
more natural conditions. Furthermore, the artificial conditions may be a
good enough analog to give results that would generalize to the wider world.
How rats respond to signals warning of impending shock may, for example,
help us to understand how real-world conditions may contribute to such human
conditions as chronic anxiety or the development of peptic ulcers. Once a
phenomenon has been identified in the rat studies, one can then do followup
studies to determine the generality of the findings empirically.

Artificial conditions may also be used to test specific hypotheses or
predictions of theory. The cited study was conducted to compare predictions
of the "preparatory response" and "safety" hypotheses.

Typical behaviour for an animal experiencing a shock is, as far as I have
observed, to leave the place where the shock occurred and if possible
never return.

Yes, but what of the wild rat confronted with the dangerous job of foraging
for food? Stay home, stay safe, die of starvation; forage, eat, but
possibly get attacked by cats and other predators. Given the choice, would
the rat visit locations in which the arrival of predators is always signaled
(e.g., by the sounds of rustling leaves or by visual input as the predator
crosses open ground) or those in which it could be easily surprised? The
artificial situation provided in the operant chamber may provide a
reasonable analog.

It seems to me that attempting to learn what the controlled variables are
for "typically normal" behaviour would be the initial goal of PCT
research. Next would likely come studies of frequently observed
"abnormal" behaviour and then maybe studies of behavioural situations
that the subject would not normally ever encounter.

I agree, but the research program one follows is dictated by the general
questions being addressed, which may lead one to focus on particular methods
that would seem to provide the best avenue to a clear answer. This
particular research was initially spurred by the realization that the rat's
choice of the signaled condition was not consistent with extant theory.

You did not mention if there was a way for the rat to avoid the shock if
the light signaled its impending arrival.

Ideally, no. In practice, some rats did stumble on a solution: they learned
to roll onto their backs, where the insulation provided by their fir
prevented them from receiving further shocks. Rats are excellent little
autonomous control systems: with a bit of luck, they will learn how to
control variables the experimenter intends to be uncontrollable. Of course,
experimenters are good controllers too. If necessary, we shaved the rat's
back to remove this source of shock control.

Bill Powers (950602.0600 MDT) --

How did this experiment determine that the rat didn't just prefer the
house light to be on? In order to show that the rat preferred the
signalled condition, you would have to do the same test with "house
light off" indicating the signaled shock condition, and preferably
repeat the experiment with other kinds of indicators, too. If the rat
always turned on the signalled condition regardless of the kind of
behavior required to do that, we might suspect that the rat was
perceiving and controlling something related to the signaled condition
that was different in the unsignaled condition.

I've answered the first question, but I'll add that rats, being nocturnal
animals, prefer the houselight off, not on. The suggested test was done.
Also, rats choose the signaled schedule whether they have to press a lever,
not press a lever, run to the opposite side of a shuttlebox, or stay where
they are in order to be in the signaled condition. If pressing the lever
does not switch conditions from unsignaled to signaled, the rats stop
pressing (i.e., lever pressing extinguishes) even if that action continues
to switch the houselight on.

Even after proving that the manipulandum was not itself the primary
controlled variable, you would still have a job ahead in proving that
the controlled variable was "signaled condition present." That is the
human way of perceiving the situation, but my prejudice is that it is a
rather abstract perception for a rat to have. I would ask myself, "From
the rat's point of view, what is different between what I see as the
signaled and the unsignaled conditions, in terms of the experiences the
rat is having?" One difference, I would guess, is that in the signaled
condition the rat might be able to reduce the experience of shock, or
prepare itself in some way to receive the shock, whereas in the
unsignaled condition the shock might arrive at any instant and catch the
rat unprepared. Whatever it is that the rat is controlling, it has to be
something that a rat can perceive. What the human observer can see about
the situation is irrelevant.

A long series of studies was conducted to try to get at that question. For
example, in one study the intensity of shock in the signaled condition was
systematically varied while holding the intensity of shock in the signaled
condition constant. The rats stopped changing to the signaled condition
only when the shocks were about three times intense as those in the
unsignaled condition. This did not rule out the possibility that the rats
were getting prepared for the shock (reducing the aversiveness of the shock
experience), but did indicate that if preparation is involved, it is
extremely effective. I was able to demonstrate in another study that such
preparation, if it occurs, could not involve actions (such as postural
adjustments) that minimized shock contact.

I think it would be fair to characterize the entire research program as a
search for the actual controlled variable(s) in this rather complex
situation, from the rat's point of view and not our own.

Probably the best way to gain insight into the rat's experience is to
undergo the experiment yourself. While you would be able to characterize
the situation as "signaled" and "unsignaled", you could also look for
lower-level perceptions that are different even without this
characterization. You could ask, "What is better about having a signal
indicating that a shock is about to occur?" You might, for example, find
that at least some of the time you would be less surprised by the shock
if a signal immediately preceded it, so preferring not to be surprised,
you would try to find ways to make that signal appear. The ultimate
controlled variable would be to keep the experience of the shock as
untraumatic as possible, and to do this, the immediate variable that
needs to be controlled is the presence or absence of the signal. That is
less abstract than a perception of "signaled condition," and I would
consider it a more likely prospect for a perceptual variable that the
rat could control.

Some work WAS done using volunteer college student participants, and they
were asked to explain why they thought they behaved as they did. Human
participants sometimes behaved differently from the rats owing to their
having a deeper control hierarchy than the rats. One student stayed in the
unsignaled condition and then explained that "I thought it was worse but I
wanted to prove to you that I could take it."

As to the ultimate controlled variable, I have no doubt that the rats are
attempting to minimize the aversiveness of the shock experience as you
suggest (rats do not appear to develop motives like those of the college
student just quoted). This, I think, was taken for granted. The question
of interest was _why_ the signaled condition seems less traumatic to the rat
than the unsignaled one. After all, the _shock_ schedule in the two
conditions is identical. How do the warning signals change the rat's
perceptions? How is it that for the mere presence of a signal the rat is
willing to take shocks up to three times more intense?

    I maintain that this experiment performed the Test for the
    controlled variable.

I agree that it did, but it didn't carry it very far. I believe the
experimenters could have got a lot closer to what the rat was actually
controlling.

I hope I've corrected that impression.

I doubt that what the rat perceived was "the schedule in effect." That
is what the _experimenters_ perceived. The experimenters stopped the
test when they found a variable that _they_ could perceive as being
under control, but that apparent control might merely have been a side-
effect of the rat's controlling another, and much simpler, variable.

You should appreciate now that this assessment was premature.

I think that experimenters doing the test should try to distinguish
between their own perceptions and those of the test subject, and realize
that there can be a considerable difference. Whenever possible they
should test perceptual variables of the lowest level they can, to
minimize overinterpretation.

This was done.

For example, in varying "dependability" of a signal, could this not also
be interpreted simply as varying the number of times the signal occurred
during a session? If the rats controlled for perceiving the signal,
would this not make it appear that they are controlling for an abstract
condition called "dependability" or "probability?"

Every manipulation introduces confounding with other variables; the only
choice one has is which variables to confound. In this case you can
manipulate signal dependability by deleting shocks or by adding signals.
The first way confounds dependability and shock frequency; the second
confounds dependability and signal frequency. Both ways were investigated,
and they yielded the same result.

On a somewhat different topic, my master's thesis examined whether rats
preferred to have physical control over shock (being able to prevent it or
to terminate it once it occurred) in the same way that preference for
signaled over unsignaled shock schedules was investigated. Yes, this is
control in the PCT sense of the term (shock is a perception). The answer
was that they were not interested in changing from a condition in which they
lacked control over shock to one in which they had such control, so long as
the shocks in the two conditions were identical (i.e., same frequency,
intensity, and duration). In fact they were indifferent to the two
conditions. The procedure equated the actual shock experience in the two
conditions; apparently just the perception of _having_ control is not itself
a controlled variable, at least for the rat.

Regards,

Bruce

<[Bill Leach 950602.18:33 U.S. Eastern Time Zone]

[From Bruce Abbott (950602.1455 EST)]

Actually, if the rat had done nothing it would have been in the
unsignaled condition 0% of the time during the testing sessions.

Huh?

I want to express some additional critical comments on this experiment
but I believe that the experiment was carried out with even greater care
than I can infer directly from your last posting (which is itself
considerable). The comments will obviously be my opinion and of course
may in part be due to additional aspects of the acutal project with which
I am not familiar.

... or variables actually being controlled (although these researchers
would have SAID that they were attempting to identify the controlling
variables, thus puting the cart before the horse).

I really do believe that this is a matter of a GREAT deal more than that
the researchers "would have SAID...".

The real subject view vs. experimenter view is the most critical aspect
of designing and analyzing experiments. The problem with "controlling
variables" is NOT that the wrong CEV is found it is deeper than that.
Indeed, the researcher may well HAVE identified the correct CEV (even if
he thinks it is a controlling variable).

The first problem is that the approach of a research that is looking for
something in the environment that controls observed (by the researcher)
behaviour of the subject is that what the researcher sees is incidental
to what the subject is doing.

The second problem is that the researcher "projects" onto the subject his
own perception as though the subject also perceives the same thing (as
indeed you did with the "control of schedule" statement).

The third problem is that the focus really IS on the environment an not
on the subject. Thus, it is easier to miss a basic CEV because the
observer's perception is possibly at a higher or even lower level than
the subject's.

As I see it, the recognition that one can not control the behaviour of a
control system and thus must alway focus on the subjects perceptions is
the single most important view point for a PCT researcher. And I mean by
"can not control" that while it is often possible to create a situation
where YOU will observe behaviour on the part of the subject that appears
to be "controlled" in the manner that you wish to perceive, the actual
reasons why what you observe happening may have little to do with what
you think "is causing" what you see.

Good question. Yes, they sometimes did, but they will do this ...

Was there any data collected then to distinguish between what appeared to
be intentional actuations vs. accidental?

Was their any noted change in that data as a function of different
experimental configurations?

The assumption may well be incorrect though if they occured infrequently
then it might well not matter.

The rat is still the same system and will behave according to the rules
imposed by its structure and organization, whether the situation is
"natural" or "artificial." When you wish to analyze the performance of
a designed control system, you no doubt subject the system to artificial
conditions (e.g., step functions, open-loop conditions) that the
controller will never encounter in the real world, because the response
of the system under such conditions reveals aspects of its performance
that are difficult or impossible to see under normal operating
conditions. A rat in an operant ...

I will give you that this is fundamentally correct as long as one
recognizes that the experimenter IS introducing disturbances to CEVs with
which the experimenter is completely unaware and that the errors in the
perceptual systems associated with these disturbances. And also
recognizes that "extreme" disturbance or disturbance that produces an
error in a controlled perception such that the error can not be
"controlled away" (or more accurately, reduced to acceptable (to the
subject of course, control limits) will result in reorganization or other
control actions (reference changes) that will result in changes that are
unique to the specific subject.

Additionally, many of the variables that are manipulated (intentionally
or otherwise) may be changing perceptions for several CEVs. I believe
that all of these sorts of things are at least marginally easier to
identify only if the researcher continually reminds himself that it is
not his perceptions that count.

Once a phenomenon has been identified in the rat studies, one can then
do followup studies to determine the generality of the findings
empirically.

I can imagine for various reasons that this step would literally be
forced without consulting the researchers (or ignoring their
recommendations if such represent a disturbance to whoever wants to "take
the next step")... Unfortunately linking studies together that were made
with as fundamental an error as thinking that a Stimulas "triggers"
behaviour has a very low probability of producing better results or more
useful knowledge.

Artificial conditions may also be used to test specific hypotheses or
predictions of theory. The cited study was conducted to compare
predictions of the "preparatory response" and "safety" hypotheses.

Naturally the problem here is that if the underlying philosophy for the
hypothesis is wrong then an experiment "designed" to allow only two
options is going to produce results subject to misinterpretation.
Possibly the MOST significant observation made was the one that you
mentioned about the rats that would roll over on their backs to
appearently avoid the shock altogether!

Yes, but what of the wild rat confronted with the dangerous job of
foraging for food? Stay home, stay safe, die of starvation; forage,
eat, but possibly get attacked by cats and other predators. Given the
choice, would the rat visit locations in which the arrival of predators
is always signaled (e.g., by the sounds of rustling leaves or by visual
input as the predator crosses open ground) or those in which it could be
easily surprised? The artificial situation provided in the operant
chamber may provide a reasonable analog.

I will provisionally accept this as long as you accept that the
comparison is probably stretching things a bit. I will also give you the
fact that attempting experiments with the experimenter having some
control over the limits to the options for the subject is necessary in
animal research as long as you will "give me" that when you begin
establishing such limits you really do not know what perceptions you are
disturbing for the subject, how you are disturbing them (from the
subject's view point of course) and that the answers to these questions
_could_ be crucial to the results obtained.

conditions is identical. How do the warning signals change the rat's
perceptions? How is it that for the mere presence of a signal the rat
is willing to take shocks up to three times more intense?

Rats don't like surprises and will control to reduce the probability for
having one?

Bill P.

under control, but that apparent control might merely have been a side-
effect of the rat's controlling another, and much simpler, variable.

You should appreciate now that this assessment was premature.

I don't think so. I do think that the experiment was well run to the
point of obtaining what is probably a correct evaluation of what the rats
were observed to be doing but I still don't think that a rat
intentionally control such an abstract concept as "the schedule in
effect". My earlier "off the cuff" comment about "control to reduce
probability for having one?" is not much better.

What might be better would be if I said "That the rats have a reference
level for surprises that is set rather low (subjective at this point) and
the signaled event represents a smaller disturbance (or no disturbance)
TO THIS perception than does the unsignalled event. The rats should
reorganize if unsignalled events are experienced that represent a
sufficient disturbance to the controlled perception to result in an
error. When a means, sequence of controlled perceptions or program is
discovered that brings this CEV under control then the rat will use the
new found method (unless other error(s) exist in higher priority CEV(s)
or reorganization "disassembles" the learned method.

Even still there is in addition always the question of CEVs of which the
experimenter has no knowledge and their possible effect(s) on the
results.

It is beginning to occur to me as to why Rick has always had (for the
year and a half that I have known him and presumably for quite a bit
longer than that) had such a strong adversion to attempting to review
past work using a PCT analysis.

Though it does sound to me as though in your experiments there seemed to
be a high level of recognition that the rats themselves seemed to "have
something to say" about what went on even to the point of thinking about
and testing for "what the rat might be doing", the initial design,
conducting and analysis for an experiment without a rigorous PCT
methodology will fall short at every point largely because many of the
"right questions" will not be asked before and during design.

I suspect that at some point, experiments of the sort that you described
would ultimately have to result in an understanding of control theory as
applied to living systems. The question would be how long?

-bill