Apologies if this gets through. A couple of recent postings of mine never showed up.
-- Richard Kennaway
We have been having server problems the last week, and am sending this
to see if it gets through. Thanks.
Jim Chriss
Kansas Newman College
From Rick Marken (970331.1100)]
Jim Chriss --
We have been having server problems the last week, and am
sending this to see if it gets through.
It worked!
Best
Rick
[From Bruce Abbott (980727.2040 EST)]
Rick Marken (980723.2050) --
Bruce Abbott (980723.0910 EST)
In my studies on preference for signaled over unsignaled shock,
the rats definitely were acting to oppose the disturbance created
by the apparatus when it moved the animals back into the unsignaled
condition at the end of each changeover period. They immediately
pressed the lever to restore the signaled condition. Simple as
this demonstration is, you refuse to recognize it as a Test
for the controlled variable. I think it pains you to admit it.
I will be happy to admit that your studies of preference involved
aspects of the test for the controlled variable. But it would pain
me to admit that they _were_ a test for the controlled variable
only because they were not. You can tell that they were not a
test for the controlled variable because we still have no idea
what variable (if _any_) was being controlled in those studies.
Sure we do: presence of a signal beginning 5 sec prior to shock and ending
with the shock. The presence or absence of such a warning signal was the
only difference between the signaled and unsignaled conditions. When the
apparatus disturbed the perception that the shocks would be preceded by
signals, the rat acted immediately, by pressing a lever, to restore the
perception that the shocks would be preceded by signals. When the
perception was that shock would be preceded by signals, the rats did not
press the lever.
Thus we observed the following simple result:
Disturbance ----> corrective action (disturbance effect counteracted)
No disturbance ----> no corrective action
Now, as I understand it, the Test for the controlled variable involves (a)
hypothesizing a controlled variable, (b) applying a disturbance to the
variable, and (c) observing whether the effect of the disturbance on the
variable is opposed via the organism's actions. This little experiment
involves all three elements, and therefore constitutes a genuine Test. Not
"aspects" of the Test, the Test, period.
Regards,
Bruce
[From Bruce Gregory (980728.1128 EDT)]
Rick Marken (980728.0800)]
> You can tell that they [shock preference studies] were not a
> test for the controlled variable because we still have no idea
> what variable (if _any_) was being controlled in those studies.Bruce Abbott (980727.2040 EST)--
> Sure we do: presence of a signal beginning 5 sec prior to shock
> and ending with the shock...If this is the reference state of the controlled variable (which
I presume is "signal prior to shock"), then the rat should also
act to restore the shock if the signal begins 5 sec prior to
the shock and ends with no shock. Did you test to see whether
the rats protect the "signaled shock" perception from the
disturbance of shock removal?
Perhaps the controlled variable is "either no shock or signal prior to
shock."
Bruce Gregory
[From Bill Powers (980728.0949 MDT)]
[From Bruce Gregory (980728.1128 EDT)]
Rick Marken (980728.0800)]
> You can tell that they [shock preference studies] were not a
> test for the controlled variable because we still have no idea
> what variable (if _any_) was being controlled in those studies.Bruce Abbott (980727.2040 EST)--
> Sure we do: presence of a signal beginning 5 sec prior to shock
> and ending with the shock...
Rick:
If this is the reference state of the controlled variable (which
I presume is "signal prior to shock"), then the rat should also
act to restore the shock if the signal begins 5 sec prior to
the shock and ends with no shock. Did you test to see whether
the rats protect the "signaled shock" perception from the
disturbance of shock removal?
Bruce G.:
Perhaps the controlled variable is "either no shock or signal prior to
shock."
This is more like the Test. You must not only show that your hypothesized
controlled variable fits the data, but that alternatives are ruled out. It
is very unlikely that your first guess, even though it seems to work, is
right. When you look into your initial proposal, you can usually find
alternatives to it that would also work. Then the process is one of
devising disturbances that would cause one hypothetical variable to change
without affecting the other. In this way, you use disturbances to eliminate
all the alternative controlled variables you can think of.
You also have to look carefully at what you're proposing. Your suggestion of
"either no shock or signal prior to shock" might reduce to a simpler
equivalent statement, and thus a simpler possible controlled perception.
For example, if you render your suggestion as a logical statement, you get
not shock or (signal and shock)
I believe this reduces logically to
signal and (shock or not shock) which is equivalent to
signal.
In other words, if the animal controls for appearance of the signal, the
result will be just the same as if it were perceiving the more complex
logical function.
Similarly, the signal appears only during the "signaled condition," so the
animal, simply by controlling for appearance of the signal, would also be
controlling for the whole signaled condition where the signal is followed
by a shock. Rick's suggestion would then be appropriate for distinguishing
whether occurrance of the shock was a necessary part of the definition.
Perhaps this should be made part of our formal definition of the Test --
finding as many alternative definitions as possible and varying the
disturbances in such a way as to eliminate as many of them as possible.
Best,
Bill P.
[From Rick Marken (980728.0800)]
You can tell that they [shock preference studies] were not a
test for the controlled variable because we still have no idea
what variable (if _any_) was being controlled in those studies.
Bruce Abbott (980727.2040 EST)--
Sure we do: presence of a signal beginning 5 sec prior to shock
and ending with the shock...
If this is the reference state of the controlled variable (which
I presume is "signal prior to shock"), then the rat should also
act to restore the shock if the signal begins 5 sec prior to
the shock and ends with no shock. Did you test to to see whether
the rats protect the "signalled shock" perception from the
disturbance of shock removal?
Best
Rick
···
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bruce Abbott (980728.1240 EST)]
Rick Marken (980728.0800)]
Sure we do: presence of a signal beginning 5 sec prior to shock
and ending with the shock...
If this is the reference state of the controlled variable (which
I presume is "signal prior to shock"), then the rat should also
act to restore the shock if the signal begins 5 sec prior to
the shock and ends with no shock. Did you test to to see whether
the rats protect the "signalled shock" perception from the
disturbance of shock removal?
The rats would certainly prefer no shock over signaled shock (yes, that has
been Tested), but that option was not allowed here. Given that the shocks
would occur (shock occurrence not under the rat's control), the rats
controlled for having them signaled.
Regards,
Bruce
[From Bruce Abbott (980728.1305 EST)]
Bill Powers (980728.0949 MDT) --
Bruce Gregory (980728.1128 EDT)
Perhaps the controlled variable is "either no shock or signal prior to
shock."
This is more like the Test. You must not only show that your hypothesized
controlled variable fits the data, but that alternatives are ruled out. It
is very unlikely that your first guess, even though it seems to work, is
right. When you look into your initial proposal, you can usually find
alternatives to it that would also work. Then the process is one of
devising disturbances that would cause one hypothetical variable to change
without affecting the other. In this way, you use disturbances to eliminate
all the alternative controlled variables you can think of.
This, of course, is precisely what was done in this _series_ of experiments.
For example, if you render your suggestion as a logical statement, you get
not shock or (signal and shock)
I believe this reduces logically to
signal and (shock or not shock) which is equivalent to
signal.
This absurd result shows that Bruce's proposal isn't quite as I imagine he
intended. As I noted to Bruce, the option is not "not shock or (signal and
shock)", because not shock is not available as an option. The rat would
control for not shock over signal and shock if it had the chance. But it
doesn't, and so must take second best: signal and shock over (not signal)
and shock.
Regards,
Bruce
[From Bill Powers (980728.1610 MDT)]
I believe this reduces logically to
signal and (shock or not shock) which is equivalent to
signal.
This absurd result shows that Bruce's proposal isn't quite as I imagine he
intended. As I noted to Bruce, the option is not "not shock or (signal and
shock)", because not shock is not available as an option. The rat would
control for not shock over signal and shock if it had the chance. But it
doesn't, and so must take second best: signal and shock over (not signal)
and shock.
You seem to use that word "absurd" a lot. The result above shows that the
more complex-looking proposal for the controlled variable is logically
equivalent to simply controlling for the presence of the signal. If the rat
controls for presence of the signal, and shocks occur whether the signal
occurs or not, it's "obvious" (to counteract your "absurd") that it is
sufficient for the rat to control for presence of the signal, ignoring the
occurrances of the shock (which it can't affect).
To rule out this possibility, you would have to present the signal in a way
that is unrelated to the shocks. If your proposed controlled variable is
the correct one, then you would find that the rats would not control for
presence of the signal if it occurred (for example) immediately after the
shock. By disturbing the timing of the signal, you can test for its
significance to the rat, and pin down the temporal relationship that is
significant.
Best,
Bill P.
[From Bruce Abbott (980728.2020 EST)]
Bill Powers (980728.1610 MDT) --
I believe this reduces logically to
signal and (shock or not shock) which is equivalent to
signal.
This absurd result shows that Bruce's proposal isn't quite as I imagine he
intended. As I noted to Bruce, the option is not "not shock or (signal and
shock)", because not shock is not available as an option. The rat would
control for not shock over signal and shock if it had the chance. But it
doesn't, and so must take second best: signal and shock over (not signal)
and shock.
You seem to use that word "absurd" a lot.
Not to be picky or anything, but that's a load of crap. I use it sparingly,
when warranted.
The result above shows that the
more complex-looking proposal for the controlled variable is logically
equivalent to simply controlling for the presence of the signal. If the rat
controls for presence of the signal, and shocks occur whether the signal
occurs or not, it's "obvious" (to counteract your "absurd") that it is
sufficient for the rat to control for presence of the signal, ignoring the
occurrances of the shock (which it can't affect).
What is absurd about the conclusion is the notion that wanting either (a) no
shock or, failing that (b) shocks preceded by signals should be considered
logically the same as wanting signals by themselves. Look Bill, perhaps it
is the case that you would rather have no steak than steak without salt.
Does this imply that what you really want is a pile of salt? I'll bet not.
It is the salt _with_ the steak you want, not the salt per se, and it is the
signal preceding the shock the rat wants, not the signal per se. The fact
that the analysis yields an absurd conclusion indicates that the logical
relations we started with do not model the actual logicic of the situation
correctly. A steak and a pile of salt are not the same as a salted steak.
A signal and a shock are not the same as a signaled shock. Your logical
construct (signal and shock) ignores that distinction.
To rule out this possibility, you would have to present the signal in a way
that is unrelated to the shocks. If your proposed controlled variable is
the correct one, then you would find that the rats would not control for
presence of the signal if it occurred (for example) immediately after the
shock. By disturbing the timing of the signal, you can test for its
significance to the rat, and pin down the temporal relationship that is
significant.
Ah, good. Been there, done that. The signal must precede the shock, by at
least 2 seconds. Five seconds appears to be about optimal. Are you now
ready to grant that I've been conducting the Test with these studies? (:->
Regards,
Bruce
[From Bill Powers (980729.0429 MDT)]
Bruce Abbott (980728.2020 EST)--
What is absurd about the conclusion is the notion that wanting either (a) no
shock or, failing that (b) shocks preceded by signals should be considered
logically the same as wanting signals by themselves.
The parallel to your case above is the choice between (no shock) or (shock
preceded by a signal)). That is not the choice that is offered. Actually, I
think I mis-stated the cases. The logical cases you used are (shock and
not-signal) or (shock and signal). This is (A and not-B) or (A and B). By
the distributive law, this is equivalent to [A and (not-B or B)], which is
equal to (A and 1), which is equal to A. If this result seems absurd to
you, that is because common sense sometimes conflicts with logic.
The only difference between the cases is whether the signal occurs or not;
the shock occurs in either case. You may be imagining that the rat would
prefer not to be shocked at all, but this experiment as you have described
it doesn't allow that case. A different test is required to show that the
rat's preference for being signaled depends on whether it is being
inescapably shocked. Perhaps you did that test, but if so, it should be
included as part of your proof that the rat is controlling for the specific
case (shock and signal). Your common sense is not a substitute for a proof.
However, the control may not be at the logic level. You may be proposing
that it is at the sequence level: the preference is for the signal to
precede the shock by some time interval. Using the time interval as an
independent variable, you can vary it over the entire available positive
and negative range (the maximum interval is half the interval between
shocks). If your hypothesis is consistent with the observations, you should
observe that the rat controls for the case in which there is a warning
signal some short time before the shock, but not the case in which there is
no signal or the signal occurs either after or too long before the shock.
In the latter case, you have to be concerned with the perceptual groupings
that define "before" and "after." Suppose the shocks occur 20 seconds
apart. Then a signal that occurs 15 seconds _after_ one shock also occurs 5
seconds _before_ the next shock.
Look Bill, perhaps it
is the case that you would rather have no steak than steak without salt.
That is not the case we are dealing with. The case is (steak) or (steak
with salt). If that is the only choice offered, you can conclude only that
the person prefers salt (or no salt). You can't say what the preference is
for the steak.
Does this imply that what you really want is a pile of salt? I'll bet not.
It is the salt _with_ the steak you want, not the salt per se, and it is the
signal preceding the shock the rat wants, not the signal per se.
If that's your hypothesis, you have to test for it. You have to offer salt
alone (or salt with something else) versus no salt as two more choices.
The fact
that the analysis yields an absurd conclusion indicates that the logical
relations we started with do not model the actual logicic of the situation
correctly. A steak and a pile of salt are not the same as a salted steak.
A signal and a shock are not the same as a signaled shock. Your logical
construct (signal and shock) ignores that distinction.
You have not made any distinction. You've merely asserted that several
things are not equivalent, without proof. The conclusion of the analysis is
not absurd at all; it is exactly correct, given the model (as you say). The
problem is that you haven't presented all the choices, or haven't yet
presented them correctly in terms of the rat's perceptions. Your instincts
about the logical conclusion are probably correct, but you're blaming the
wrong thing: the analysis rather than your way of conceiving or
communicating the situation.
To rule out this possibility, you would have to present the signal in a way
that is unrelated to the shocks. If your proposed controlled variable is
the correct one, then you would find that the rats would not control for
presence of the signal if it occurred (for example) immediately after the
shock. By disturbing the timing of the signal, you can test for its
significance to the rat, and pin down the temporal relationship that is
significant.Ah, good. Been there, done that. The signal must precede the shock, by at
least 2 seconds. Five seconds appears to be about optimal. Are you now
ready to grant that I've been conducting the Test with these studies? (:->
It would be good if you could present ALL the evidence concerning your
conclusions, rather than an inconclusive fragment of it. Now that you have
revealed more of your experimental manipulations, I can readily "admit"
that you were doing something equivalent to the Test. You speak as if you
are wringing an admission from me against my will, when in fact you simply
left out critical facts about the experiment.
I should point out that the rat was not (by your description so far) given
control over the interval by which the signal preceded the shock. Your
manipulations of that interval thus qualify as disturbances, not aspects of
the controlled variable itself. So the best you can say now (barring
further unrevealed facts) is that the rat is controlling for something that
is affected by the interval between the signal and the shock.
I have some doubts about the overall interpretation you'e suggesting. I
presume (you haven't yet said) that your measure of the effect on the rat's
behavior of your manipulations was the frequency of pressing, or the
probability per unit time of pressing, or the (inverse) latency of
pressing, of the key that switched to the signaled condition. You would
presumably have used the frequency, or probability, or inverse latency, as
a measure of the amount of behavior. When you say that an interval of 5
seconds was "optimum," I presume you mean that this value gave the highest
frequency or probability, or minimum latency, of the key-press: the highest
measure of behavior.
This, however, is not what we would expect a control system to do. If you
make the input match the reference condition more closely, you would expect
the measure of behavior or action to decrease, not increase. So while your
experiment appears at first to have followed the methodology of the Test,
the results are not what we would expect from the Test: minimum behavior
when the input is closest to the reference level, not maximum behavior.
Could you discuss that point?
Best,
Bill P.
[From Bruce Abbott (980729.0855 EST)]
Bill Powers (980729.0429 MDT) --
Bruce Abbott (980728.2020 EST)
What is absurd about the conclusion is the notion that wanting either (a) no
shock or, failing that (b) shocks preceded by signals should be considered
logically the same as wanting signals by themselves.
The parallel to your case above is the choice between (no shock) or (shock
preceded by a signal)). That is not the choice that is offered.
Yes, I am aware of that. That was Bruce G.'s proposal as to what the rats
were controlling for, not a statement of the alternatives offered.
Actually, I
think I mis-stated the cases. The logical cases you used are (shock and
not-signal) or (shock and signal). This is (A and not-B) or (A and B). By
the distributive law, this is equivalent to [A and (not-B or B)], which is
equal to (A and 1), which is equal to A. If this result seems absurd to
you, that is because common sense sometimes conflicts with logic.
See below.
The only difference between the cases is whether the signal occurs or not;
the shock occurs in either case. You may be imagining that the rat would
prefer not to be shocked at all, but this experiment as you have described
it doesn't allow that case. A different test is required to show that the
rat's preference for being signaled depends on whether it is being
inescapably shocked. Perhaps you did that test, but if so, it should be
included as part of your proof that the rat is controlling for the specific
case (shock and signal). Your common sense is not a substitute for a proof.
When I started down this road, I thought it was only necessary to
demonstrate that I was applying the Test for the controlled variable in
these studies. Now it seems that the ante has been upped, and I must now
demonstrate that all possible Tests have been conducted to rule out all
possible alternatives. I agree that this needed to be done in order to
identify the precise nature of the controlled variable. But I don't see
what this has to do with the question of whether a particular experiment
applied the logic of the Test or not. That logic requires that one (a)
apply a disturbance to an hypothesized controlled variable, and (b) observe
whether action is taken to oppose the effect of that disturbance. The
experiment I described does that.
However, the control may not be at the logic level. You may be proposing
that it is at the sequence level: the preference is for the signal to
precede the shock by some time interval. Using the time interval as an
independent variable, you can vary it over the entire available positive
and negative range (the maximum interval is half the interval between
shocks). If your hypothesis is consistent with the observations, you should
observe that the rat controls for the case in which there is a warning
signal some short time before the shock, but not the case in which there is
no signal or the signal occurs either after or too long before the shock.
In the latter case, you have to be concerned with the perceptual groupings
that define "before" and "after." Suppose the shocks occur 20 seconds
apart. Then a signal that occurs 15 seconds _after_ one shock also occurs 5
seconds _before_ the next shock.
The appropriate studies to test these alternatives have been done. Rats do
not prefer signals over no signals when shock is absent. The placement of
the signals is critical; they prefer signals just prior to shocks over
signals and shocks randomly related. Signals that occur at too brief an
interval prior to shock (less than 1 or 2 seconds) are not as preferred as
those that are longer, with 5-10 seconds being about optimal. Signals are
preferred over no signals prior to shock over a wide range of shock
frequencies. If you offer a conflict situation involving unsignaled weaker
shocks versus signaled stronger shocks, the rats choose the signaled
stronger shocks up to an intensity ratio of three to one. This series of
studies reveal clearly that rats receiving a schedule of unavoidable shocks
control for having the shocks preceded immediately by a warning signal of at
least 2 seconds (preferably longer) duration. That doesn't answer the
question of why, however.
Look Bill, perhaps it
is the case that you would rather have no steak than steak without salt.
That is not the case we are dealing with. The case is (steak) or (steak
with salt). If that is the only choice offered, you can conclude only that
the person prefers salt (or no salt). You can't say what the preference is
for the steak.
True, but that is irrelevant to the point I was making, which is that the
logical construction salt and steak is not necessarily equivalent to a
salted steak. If not, then the logical construct fails to represent the
actual situation adequately and deductions may lead to absurd conclusions,
not in terms of Aristotelian logic, but in terms of what may actually be
expected in the case being modeled.
Does this imply that what you really want is a pile of salt? I'll bet not.
It is the salt _with_ the steak you want, not the salt per se, and it is the
signal preceding the shock the rat wants, not the signal per se.
If that's your hypothesis, you have to test for it. You have to offer salt
alone (or salt with something else) versus no salt as two more choices.
The fact
that the analysis yields an absurd conclusion indicates that the logical
relations we started with do not model the actual logicic of the situation
correctly. A steak and a pile of salt are not the same as a salted steak.
A signal and a shock are not the same as a signaled shock. Your logical
construct (signal and shock) ignores that distinction.
You have not made any distinction. You've merely asserted that several
things are not equivalent, without proof. The conclusion of the analysis is
not absurd at all; it is exactly correct, given the model (as you say). The
problem is that you haven't presented all the choices, or haven't yet
presented them correctly in terms of the rat's perceptions. Your instincts
about the logical conclusion are probably correct, but you're blaming the
wrong thing: the analysis rather than your way of conceiving or
communicating the situation.
See above.
To rule out this possibility, you would have to present the signal in a way
that is unrelated to the shocks. If your proposed controlled variable is
the correct one, then you would find that the rats would not control for
presence of the signal if it occurred (for example) immediately after the
shock. By disturbing the timing of the signal, you can test for its
significance to the rat, and pin down the temporal relationship that is
significant.
Ah, good. Been there, done that. The signal must precede the shock, by at
least 2 seconds. Five seconds appears to be about optimal. Are you now
ready to grant that I've been conducting the Test with these studies? (:->
It would be good if you could present ALL the evidence concerning your
conclusions, rather than an inconclusive fragment of it. Now that you have
revealed more of your experimental manipulations, I can readily "admit"
that you were doing something equivalent to the Test. You speak as if you
are wringing an admission from me against my will, when in fact you simply
left out critical facts about the experiment.
Interesting that where I said "grant" you read "admit." But I must "admit"
that it does feel as though I am having to jump through hoops to win your
agreement that the experimental design under discussion does constitute a
proper Test for a controlled variable. We should be able to decide this
matter without hearing specifics about any real application of the method.
I offered the signaled versus unsignaled shock contrast as an example of the
method as it has been applied to a real problem, but now I'm being faulted
for not having laid out the whole experimental program. Come on, Bill,
what's the deal here?
I should point out that the rat was not (by your description so far) given
control over the interval by which the signal preceded the shock. Your
manipulations of that interval thus qualify as disturbances, not aspects of
the controlled variable itself. So the best you can say now (barring
further unrevealed facts) is that the rat is controlling for something that
is affected by the interval between the signal and the shock.
There are technical problems with giving the rat direct control over the
interval by which the signal preceded the shock, having to do with the lack
of immediate feedback. Any change the rat made in the interval would not
become apparent until the next signaled shock occurred, which on average
would have been about a minute after the change (on the typical VT 2-min
shock schedule). Given the delayed feedback, it is unlikely that the rat
would learn the relationship between its behavior and the signal-to-shock
interval. The alternative (which we pursued) was to observe whether the rat
would act to restore the signaled shock condition while Testing at a number
of different signal durations. The rats resisted did so act for signals
longer than 1 or 2 seconds but not for shorter ones.
If there's a lesson here, it's that as a practical matter, you sometimes
have to develop alternative methods to the straightforward procedure of
giving direct control over the aspect of the CV being investigated.
I have some doubts about the overall interpretation you'e suggesting. I
presume (you haven't yet said) that your measure of the effect on the rat's
behavior of your manipulations was the frequency of pressing, or the
probability per unit time of pressing, or the (inverse) latency of
pressing, of the key that switched to the signaled condition. You would
presumably have used the frequency, or probability, or inverse latency, as
a measure of the amount of behavior. When you say that an interval of 5
seconds was "optimum," I presume you mean that this value gave the highest
frequency or probability, or minimum latency, of the key-press: the highest
measure of behavior.
Actually the measure reported was the percentage of session time spent in
the signaled condition.
This, however, is not what we would expect a control system to do. If you
make the input match the reference condition more closely, you would expect
the measure of behavior or action to decrease, not increase. So while your
experiment appears at first to have followed the methodology of the Test,
the results are not what we would expect from the Test: minimum behavior
when the input is closest to the reference level, not maximum behavior.
Could you discuss that point?
Yes. The amount of time that would be spent in the signaled condition in
the absence of control action is essentially zero. The method used is like
logic-level control in that the rat is either in or not in the signaled
condition. A single response fully corrects a not-in error, and the
disturbance fully produces a not-in error. Consequently, we're not going to
see a graded decrease in behavior as the actual state of the CV approaches
the reference; we're either going to see behavior (when the CV is not at
reference) or no behavior (when the CV is at reference). And this is
precisely what we do see.
The proper way to assess the outcome is to compare the amount of time spent
in the changeover (signaled shock, in this case) condition to the amount of
time that would have been spent there in the absence of negative feedback
control action on the part of the rat (essentially zero). If the gain is
high, the animal will be spending nearly all its time in the changeover
condition. Lower gain translates into a more leisurely correction of error
and thus to a reduced time spent in the changeover conditon. With zero gain
(the animal is not controlling for being in the changeover condition) there
will be essentially no time spent in the changeover condition. (I say
essentially because very occasionally the response lever does get tripped as
a side-effect of other activity. The level at which this occurs is assessed
as part of the experimental procedure.)
Regards,
Bruce
[From Bruce Abbott (980729.1050 EST)]
Rick Marken (980729.0730) --
Bruce Abbott (980728.2020 EST)
Been there, done that. The signal must precede the shock, by at
least 2 seconds. Five seconds appears to be about optimal. Are
you now ready to grant that I've been conducting the Test with
these studies? (:->
Bill Powers (980729.0429 MDT) already gave an excellent reply to
this post.
Ah, appeal to Authority. If Bill said it, it must be true.
I'll just try to explain quickly why Bruce's
experiments are not a Test for the Controlled Variable.
This ought to be amusing. After that, I would be interested in hearing your
explanation for why the Earth is actually flat.
The reason
is that the subjects in his experiments clearly were not in
control of the purported controlled variable. If the variable
that is presumed to be under control is "signal 5 sec before shock"
then the subject would protect that variable from _all_ disturbances,
including (as I noted) removal of shock. In fact, the subject does
not protect this variable from shock removal, which shows that
"signal 5 sec before shock" is not under control.
I've already pointed out to you that the rats were not permitted to remove
the shock. Given that shocks would occur and could not be avoided, the rats
control for having them signaled. I never said that the rats would control
for having signaled shocks (as opposed to no shocks); I said that they
control for having signals before the shocks (as opposed to no signals
before the shocks). So much for Reason Number 1.
Another
indication that "signal 5 sec before shock" is not under control
is (as Bill Powers (980729.0429 MDT) notes)
Another appeal to authority? When did he receive a pipeline to Truth?
the fact that behavior
(press rate) is maximum (rather than minimum, as would be expected,
if error were zero) when the controlled variable is presumably at
its reference ("signal 5 sec before shock").
I've already explained why this is not a correct analysis in my reply to
Bill (980729.0855 EST). So much for Reason Number 2.
I think what we are probably seeing in Bruce's experiments is a
rat trying, unsuccessfully, to control the perceived effects of
shock. A signal prior to the shock apprently let's the rat do
_something_ that reduces the perceived effects of the shock
somewhat (assuming that the rat is controlling for zero perceived
effect of the shock).
You're speculating. I guess you think that's better than doing actual
research. And it's "lets," Rick, not "let's."
What Bruce is doing to the rats is equivalent
to infusing acid into your stomach at random times then offering
you the option of pressing to get a tiny fragment of Pepto
Bisbol before each infusion or not. I think you would quickly learn
to control for getting the tiny Pepto prior to acid infusion,
not because you are controlling for "Pepto 5 sec before acid" but
because the perception of stomach pain you get in this condition
is a little less horrible than the perception you get in the "no
Pepto" or "Pepto 1 secong after the acid" condition. The only
perception being controlled (unsuccessfully) is stomach pain
but (because the experimenter is unaware of this) it seems like
a sequence ("Pepto 5 sec before acid") is being controlled. I think
this is what is happening in Bruce's experiments. The rat is doing
what it can (in the signal time before the shock) to (unsuccessfully)
control the pain of shock. The experimenter makes it impossible for
the rat to control this perception; all the rat can do is move the
perception of shock _slightly_ in the desired direction (toward, but
not very far toward, zero).
Apart from the fact that your "analogy" is a bad one, it ignores the fact
that all these alternatives have already been Tested, and also ignores the
fact that even if Rick's speculation were true, it still would not establish
that the rats were not controlling for having signals prior to shocks. Even
if having signals prior to shocks were a means for reducing perceived shock
intensity, it would still be the case that the rats were controlling for
having signals prior to shocks as opposed to having the shocks unsignaled.
Also, although I haven't gone into it, the experimental evidence contradicts
the hypothesis that rats use the signals as opportunities to prepare for
shock and thereby reduce perceived shock intensity. We already Tested for
that possibility. So much for Reason Number 3.
A very important aspect of the Test is to allow the subject
to control the hypothetical controlled variable.
You mean like being able to control whether shocks will or will not be
preceded by signals? That was done.
That means, the
subject has to be able to bring that variable to _any_ value
he might want.
Incorrect. I've never been able to make the cursor in a tracking experiment
move off the computer screen and onto the wall, but that didn't seem to
prevent the experiment from Testing.
This is why it's important to apply disturbances
carefully when doing the Test; overwhelming disturbances can make
it look like a variable is not controlled when it is or make
it look like some variable is controlled when it is not.
True. What's it have to do with my experiments?
The
latter seems to be what happened in Bruce's experiments.
Oh. So you are asserting that the rats were not able to control whether or
not a signal preceded each shock after all. But Rick -- they could, and did!
In his
experiment the rat is not able to _control_ the relationship
between signal and shock; the rat cannot, for example, act to
restore the shock if shock is removed.
Red herring. See earlier comment on this. That was rejected Reason No. 1.
So it _looks like_ shock
is part of the variable the rat is _controlling_.
Apparently Rick has forgotten my statement that Tests were conducted to rule
out possibilies like the possibility that the rats simply wanted to hear
tones, whether shock was present or not, or that they simply wanted to hear
tones in a shock environment, whether prior to shock or not. So much for --
what are we up to? -- Reason Number 4.
The reason why Bruce was unable to conduct the Test properly
when he did his "preference" experiments was because he didn't
understand, at that time, the nature of control.
Ah, Rick can read my mind, too. As I noted in my reply to Bill, the reason
why the Test was conducted as it was is that it was not practical to do it
in the fashion Rick and Bill (and I) would have preferred. As conducted,
however, it yields the same information as to what the rat is controlling for.
Well, that exhausts Rick's argument. None of his reasons is valid. Nice
try, Rick, but no cigar. Why can't you see the truth?
Regards,
Bruce
i.kurtzer (980729.1200)
[From Bruce Abbott (980729.0855 EST)]
Bruce, you've been making a strong case that your experiment fit the bill of
the Test. I am not convinced, but i am listenening. Since you have written
so much already one is bound to make typological errors. But as you have made
this error before (your listing of the 10 biggies of PCT---roughly "in the
abscence of disturbances the organanisms does nothing" ) i will point it out.
< Consequently, we're not going to
< see a graded decrease in behavior as the actual state of the CV approaches
< the reference; we're either going to see ***behavior***** (when the CV is
not at
< reference) or ****no behavior**** (when the CV is at reference).
I suppose what you meant was "change in output" and not behavior. They are
,according to PCT, in no way identical. I trust this was a mistake.
On your proposal, it seems to me that the Test is meaningless without the
theoretical apparatus that gives credence to looking at invariation rather
than variation. Was this theoretical apparatus in place, or was this a good
guess by some rogue behaviorists, or maybe something else. Tell me.
i.
[From Hank Folson (980729)]
Bruce Abbott (980729.0855 EST)
The appropriate studies to test these alternatives have been done. Rats do
not prefer signals over no signals when shock is absent. The placement of
the signals is critical; they prefer signals just prior to shocks over
signals and shocks randomly related. Signals that occur at too brief an
interval prior to shock (less than 1 or 2 seconds) are not as preferred as
those that are longer, with 5-10 seconds being about optimal. Signals are
preferred over no signals prior to shock over a wide range of shock
frequencies. If you offer a conflict situation involving unsignaled weaker
shocks versus signaled stronger shocks, the rats choose the signaled
stronger shocks up to an intensity ratio of three to one. This series of
studies reveal clearly that rats receiving a schedule of unavoidable shocks
control for having the shocks preceded immediately by a warning signal of at
least 2 seconds (preferably longer) duration. That doesn't answer the
question of why, however.
Consider what a control system does. It controls perceptions. It resists
disturbances to controlled variables. Shocks are big disturbances. Given
the opportunity, the control system would reduce the error signal to
zero. But, in this (unnatural) experiment, that option is not available.
What is worse, for the control system, is that it learns that the shocks
will keep coming. So what can the poor control system do? If it does
nothing, doesn't a new higher level error appear? I think this would be
something like "painful disturbance may occur at any moment". This is a
variable that cannot be controlled. Won't the control system internally
amplify the meaning of this disturbance, due to the lack of control?
Wouldn't this cause huge error signals, even between shocks?
Now, consider the remaining option where the control system controls
only, or at least, the _timing_ of the uncontrollable disturbance. Now a
form of control can be maintained. The rat will call for the unavoidable
shock once it figures out (if it can figure this out) the maximum spacing
it can get away with and not lose control of when the %$&*@# shock will
occur. The rat will prefer at least that there is enough time between
pushing the lever & getting the shock to get ready mentally for it
(apparently >2 seconds). This way, there is no/less error signal between
shocks.
This logic would agree with the preference for predictable large shocks
over unpredictable small shocks.
Sincerely, Hank Folson
[From Bill Powers (980729.1014 MDT)]
Bruce Abbott (980729.0855 EST)--
I'm not as ready to dismiss your efforts to determine what state the rats
preferred as Rick seems to be, but I do agree with him that it seems
improbable that your objective, at the time the experiment was run, was to
determine what the rat was controlling. Insights are not retroactive.
You object to my suggestions that some more alternative hypotheses be
tested, but isn't doing just that the reason for which you tried all those
different variations, such as varying the timing of the signal prior to the
shock, and testing for preferring the signal with no shocks? This is simply
good scientific practice, checking to see whether alternative explanations
might be possible and thinking up variations that would rule them out. Of
course you believe that you tested every important alernative explanation
and stopped looking for alternatives at just the right time, but you can't
be surprised that someone else might have a different opinion.
... that is irrelevant to the point I was making, which is that the
logical construction salt and steak is not necessarily equivalent to a
salted steak. If not, then the logical construct fails to represent the
actual situation adequately and deductions may lead to absurd conclusions,
not in terms of Aristotelian logic, but in terms of what may actually be
expected in the case being modeled.
Don't blame the logical construction. There is nothing in your description
of the situation, as first given, that would have ruled out that
construction. As to what "may actually be expected," that is too subjective
a criterion to be taken very seriously. You have your own perception of a
"salted steak," but unless you can say how that differs from steak plus
salt you have no _scientific_ grounds for claiming that any difference
exists. If you want to claim that a difference does exist, it's your
respionsibility to think up an experimental test that bears out your claim.
Until you can do that, you're just trying to impose your opinions on others.
Interesting that where I said "grant" you read "admit." But I must "admit"
that it does feel as though I am having to jump through hoops to win your
agreement that the experimental design under discussion does constitute a
proper Test for a controlled variable.
Yes, and so would anybody else claiming to have conducted the Test. Part of
your "jumping through hoops" uses the same hoops you made yourself jump
through, but neglected to mention when you first described how you
conducted the Test. I am not asking anything unreasonable of you -- if I
were, you wouldn't be countering my objections by explaining that you
actually did the variations to test what I'm asking about.
We should be able to decide this
matter without hearing specifics about any real application of the method.
Oh, is that so? Then the "we" of whom you speak does not include me.
I offered the signaled versus unsignaled shock contrast as an example of the
method as it has been applied to a real problem, but now I'm being faulted
for not having laid out the whole experimental program. Come on, Bill,
what's the deal here?
The deal here is that the part of the program described did not satisfy me
as a good example of application of the Test. Obvious alternatives to the
proposed definition were left untested, and it seemed that you applied a
disturbance to the first variable that struck your fancy. When the
hypothesis was not contradicted, you assumed the job was finished. That is
not my idea of applying the Test. My idea of applying the Test is to try
all the alternatives you can think of, and accept a definition of the CV
only when you have diligently but unsuccessfully tried to rule it out and
it is the only one left. The Test is not a matter of going through a
certain ritual. It must be applied with some understanding of what you're
trying to do, which is to find out what, if anything, is being controlled.
There are technical problems with giving the rat direct control over the
interval by which the signal preceded the shock, having to do with the lack
of immediate feedback.
So what? Since when have technical difficulties made it all right to assume
things without proof? If your experiment created such technical
difficulties, it should have been revised to eliminate them -- and you
should have reserved judgement about the points in question that you could
not test, or stated that your conclusions were contingent on certain
untested assumptions.
If there's a lesson here, it's that as a practical matter, you sometimes
have to develop alternative methods to the straightforward procedure of
giving direct control over the aspect of the CV being investigated.
Agreed. But the alternative methods should actually test the point in
question.
Actually the measure reported was the percentage of session time spent in
the signaled condition.
Another delayed revelation. This implies that you did not keep a
press-by-press and shock-by-shock record of the experimental results, or
that if you did, your conclusions were based on whole-session averages. You
should know my opinion of that by now.
I agree that time spent in the signaled condition is definitely of
interest, but whether it is the sort of thing a rat might perceive and
control is another matter. I agree that the rat might _appear_ to control
that variable, but as has been pointed out numerous times on CSGnet, one of
the problems with the Test is that you have to be careful about misleading
appearances. The time spent in the signaled condition may be related to a
controlled variable, but to assume it is the _same_ as the controlled
variable is being too bold.
The proper way to assess the outcome is to compare the amount of time spent
in the changeover (signaled shock, in this case) condition to the amount of
time that would have been spent there in the absence of negative feedback
control action on the part of the rat (essentially zero). If the gain is
high, the animal will be spending nearly all its time in the changeover
condition. Lower gain translates into a more leisurely correction of error
and thus to a reduced time spent in the changeover conditon. With zero gain
(the animal is not controlling for being in the changeover condition) there
will be essentially no time spent in the changeover condition. (I say
essentially because very occasionally the response lever does get tripped as
a side-effect of other activity. The level at which this occurs is assessed
as part of the experimental procedure.)
All that is fine, except for the assumption that the first variable to pass
this test is the controlled variable. I would not be satisfied by this.
Certainly we can agree that the rats spend more time in the changeover
condition than they would if there were nothing under control, but this may
be merely an intermediate variable in the environmental feedback function.
I would ask what the effect of spending more time in the signaled condition
might be on something of more likely relevance to the rat's experiences.
For example, does having a signal to warn of an impending shock enable the
rat to do something to lessen the duration or intensity of the experienced
shock? With a detailed record of events, as in your current series of
experiments, we might be able to check up on measurable differences in the
experienced shock durations and intensities, and get closer to the right
definition.
Identifying the length of time spent in the signaled condition as a
possible controlled variable is a large step in the right direction, but
it's not the end of the trail. In my world, it would not be a publishable
result. I would go on to bet that with proper instrumentation, I could
actually measure a large difference in the experienced shock
intensity*duration product, greater than the 3:1 ratio you found when
comparing different shock intensities. In other words, I would bet against
the rat's ability to perceive something as abstract as the "signaled
condition."
I could lose that bet, of course, and be forced reluctantly to admit that
perhaps rats can perceive and control such an abstract concept. But that's
how it should be in science. We accept as truth only that which we
temporarily can't think of a way to disprove.
Best,
Bill P.
P. S. Reading your subsequent response to Rick, I realize that the proper
way to describe your proposed controlled variable is this: When shock is
inevitable, rats control for receiving signals that occur 2 to 5 seconds
before each shock. Rats appear to be controlling not just for the "signaled
condition," but for signals that occur in a particular temporal relation to
the shocks. And this is true only when shocks are qualitatively inevitable.
So far we have no indication of whether this control process is a means to
reduce the quantitative experience of the shocks (energy delivered).
[From Rick Marken (980729.1500)]
Bill Powers (980729.1014 MDT) to Bruce Abbott (980729.0855 EST)
re: Rick Marken (980729.0730)
I'm not as ready to dismiss your efforts to determine what state
the rats preferred as Rick seems to be
Gee. And I thought you would love that post. C'est la vie.
My aim was not to dismiss Bruce's efforts at testing rat preference.
The main point of my post was simply that we can't conclude that an
animal is controlling a particular variable if the Test is done in
a situation where the animal cannot possibly control the presumed
controlled variable. I was hoping that this might be one easy way
to determine whether the Test was being carried out in _any_
behavioral reseach project project, whether or not it had been
based on an understanding of PCT. I really don't think it's possible
to determine what variables a system is controlling without a good
grasp of PCT. But I'm trying to be open-minded about this. So I was
proposing a possible criterion for recognizing whether any type
of behavioral research constitutes a Test for the Controlled Variable.
If the hypothesis of a research project is that the animal is
controlling "sound 5 seconds before shock" then you can only test
this hypothesis if the animal is able to control that perception.
In Bruce's experiments, the animal was not able to control this
variable; it could not have restored shock if shock had been deleted
from the sequence and it could not have restored the temporal
relationship between sound and shock if this temporal relationship
had been changed.
All that the rat could actually control in Bruce's studies was
the condition they "preferred" to be in. As you note in your post
whether it [time spent in a particular condition] is the sort of
thing a rat might perceive and control is another matter. I agree
that the rat might _appear_ to control that variable, but as has
been pointed out numerous times on CSGnet, one of the problems
with the Test is that you have to be careful about misleading
appearances. The time spent in the signaled condition may be
related to a controlled variable, but to assume it is the _same_
as the controlled variable is being too bold.
This was the point of my Pepto example. I now realize that I failed
to make it clear that the person in that experiment (forced to drink
acid but allowed to press a bar to select one condition -- a Pepto
5 sec before the acid -- or another -- no Pepto before the acid)
could only perceive the click of the Pepto delivery device. The
example was meant to illustrate the point you make above; that the
selection of the "Pepto 5 sec before the acid" condition was _not_
the same as the controlled variable -- though it might appear to
be the controlled variable to an observer. The person had simply
learned to control for hearing the click because when the click was
happening the pain was less. The subject, like the rat in the
"signalled shock" study, would not even have to perceive the fact
that the click occurred 5 sec before acid infusion. All the
person would have to know is that when there is a click, the pain
is less.
I think it's unlikely that rats can control for a complex perception
like "signal starts 5 sec before shock". But if the rats could
control for spending the majority of their time in condition "signal
starts 5 sec before shock" over being in condition "signal starts
<>5 sec before shock" it would provide some evidence that they can.
I forget. Were there trials where the rats could press to be in
one signal/shock onset condition over another signal/shock onset
condition? My impression is that the rats always had the option
of a particular signal/shock onset condition vs the no signal
condition. If there were a signal in both conditions, then any
marked preference for one signal/shock onset condition over another
would certainly be evidence (though more tests would have to be
done on this) that rats can, indeed, perceive and control
signal/shock onset interval.
Anyway, the point of my post was that you can't tell whether or not
a variable is or is not under control unless the variable _could
be_ controlled in all the dimensions in which it could be disturbed.
Best
Rick
···
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bruce Abbott (980729.1800 EST)]
Bill Powers (980729.1014 MDT) --
Bruce Abbott (980729.0855 EST)
I'm not as ready to dismiss your efforts to determine what state the rats
preferred as Rick seems to be, but I do agree with him that it seems
improbable that your objective, at the time the experiment was run, was to
determine what the rat was controlling. Insights are not retroactive.
The objective at the time was to identify what was reinforcing the choice of
the signaled over the unsignaled condition, once it had been established
that the rats do indeed so choose. That does not prevent me from
reinterpreting the data in control-theoretic terms today, does it? Insights
_can_ be retroactive (applied to old data).
You object to my suggestions that some more alternative hypotheses be
tested, but isn't doing just that the reason for which you tried all those
different variations, such as varying the timing of the signal prior to the
shock, and testing for preferring the signal with no shocks? This is simply
good scientific practice, checking to see whether alternative explanations
might be possible and thinking up variations that would rule them out. Of
course you believe that you tested every important alernative explanation
and stopped looking for alternatives at just the right time, but you can't
be surprised that someone else might have a different opinion.
I'm sorry, but I don't recall making any such objections. What alternative
hypotheses did you suggest be tested, to which I objected? In my
recollection, my response was to inform you that the requested tests had
been performed. If that recollection is correct, then I have done exactly
as you suggest would follow good scientific practice.
... that is irrelevant to the point I was making, which is that the
logical construction salt and steak is not necessarily equivalent to a
salted steak. If not, then the logical construct fails to represent the
actual situation adequately and deductions may lead to absurd conclusions,
not in terms of Aristotelian logic, but in terms of what may actually be
expected in the case being modeled.
Don't blame the logical construction. There is nothing in your description
of the situation, as first given, that would have ruled out that
construction. As to what "may actually be expected," that is too subjective
a criterion to be taken very seriously. You have your own perception of a
"salted steak," but unless you can say how that differs from steak plus
salt you have no _scientific_ grounds for claiming that any difference
exists. If you want to claim that a difference does exist, it's your
respionsibility to think up an experimental test that bears out your claim.
Until you can do that, you're just trying to impose your opinions on others.
Perhaps it is because I know that these alternatives were Tested quite early
on and ruled out that your assertion of a logical equivalence between
(signal and shock) and signaled shock is incorrect, just as I know from
experience that (steak and salt) is not necessarily the same culinary
experience as salted steak. The distinction needs to be made in both cases,
not on my opinion, but on the evidence, which does argue for the distinction
made.
Interesting that where I said "grant" you read "admit." But I must "admit"
that it does feel as though I am having to jump through hoops to win your
agreement that the experimental design under discussion does constitute a
proper Test for a controlled variable.
Yes, and so would anybody else claiming to have conducted the Test. Part of
your "jumping through hoops" uses the same hoops you made yourself jump
through, but neglected to mention when you first described how you
conducted the Test. I am not asking anything unreasonable of you -- if I
were, you wouldn't be countering my objections by explaining that you
actually did the variations to test what I'm asking about.
I'm not so sure that you're being "reasonable." When you have in the past
illustrated Testing for the Controlled Variable in a tracking-task setting,
you have talked about how in a single experiment the Test (applying
disturbances to the putative CV) reveals whether the CV (or a close
correlate of the CV) is being controlled by the subject. What follows after
this are further Tests to refine that definition. You have never to my
knowledge asserted before that the tracking experiment by itself does not
constitute a Test for the Controlled Variable. Yet in the context of my
experiment you now wish to redefine the Test as having been performed _only_
if an exhaustive series of whatever-they-are-that-aren't-Tests has been
conducted to specifically identify exactly what aspect of the putative CV is
what the subject is "really" controlling for.
I'll say it again -- I AGREE with you that such a series is necessary to
refine one's definition of the CV; that is not the issue between us here.
(And in fact, as previously noted, just such an extensive series was
conducted in an attempt to identify what factor or factors are responsible
for the observed preference for signaled over unsignaled shock schedules.)
My assertion is that each experiment using the procedure I described does
constitute as much of a Test as a single tracking experiment does. Either
both perform the Test or neither does. Your choice.
We should be able to decide this
matter without hearing specifics about any real application of the method.
Oh, is that so? Then the "we" of whom you speak does not include me.
What a strange thing to say. Well, let's try this approach on a different
method. I ask a person to keep a certain variable at a certain value. I now
apply disturbances which tend to push the variable away from the stated
value, and observe that the person responds in such a way that the effects
of the disturbances are all but negated by these actions, so that the
variable remains close to the given value, much closer than would be the
case if the person had not so acted. Is this a Test for the Controlled
Variable? Can you determine that without having a specific application
(e.g., cursor tracking) of the method? If you can, why can't you do the
same when evaluating the changeover method in the abstract?
I offered the signaled versus unsignaled shock contrast as an example of the
method as it has been applied to a real problem, but now I'm being faulted
for not having laid out the whole experimental program. Come on, Bill,
what's the deal here?
The deal here is that the part of the program described did not satisfy me
as a good example of application of the Test. Obvious alternatives to the
proposed definition were left untested, and it seemed that you applied a
disturbance to the first variable that struck your fancy. When the
hypothesis was not contradicted, you assumed the job was finished.
What "obvious alternatives" were left untested? You suggested several
alternatives and if memory serves, my reply to you was that these had been
Tested. Your assertion that I "applied a disturbance to the first variable
that struck [my] fancy" is does not appear to arise from the facts as I
described them; I don't know where you're getting that.
That is
not my idea of applying the Test. My idea of applying the Test is to try
all the alternatives you can think of, and accept a definition of the CV
only when you have diligently but unsuccessfully tried to rule it out and
it is the only one left. The Test is not a matter of going through a
certain ritual. It must be applied with some understanding of what you're
trying to do, which is to find out what, if anything, is being controlled.
Now that I can agree with. My idea of applying [successive refinements of]
the Test is the same as yours.
There are technical problems with giving the rat direct control over the
interval by which the signal preceded the shock, having to do with the lack
of immediate feedback.
So what? Since when have technical difficulties made it all right to assume
things without proof?
Assume things without proof? What things? I'm sorry, but I don't recall
making any such assumptions.
If your experiment created such technical
difficulties, it should have been revised to eliminate them
Ah, but Bill, the experiment as conducted _was_ the revision that overcame
the technical problem noted.
-- and you
should have reserved judgement about the points in question that you could
not test, or stated that your conclusions were contingent on certain
untested assumptions.
But I could and did test them, nor were my conclusions "contingent on
certain untested assumptions." I don't see where you've shown that I've
made such assumptions.
If there's a lesson here, it's that as a practical matter, you sometimes
have to develop alternative methods to the straightforward procedure of
giving direct control over the aspect of the CV being investigated.
Agreed. But the alternative methods should actually test the point in
question.
Agreed. They did. On what basis do you assert that they didn't? Here's
the assertion, with nothing to back it up.
Actually the measure reported was the percentage of session time spent in
the signaled condition.
Another delayed revelation. This implies that you did not keep a
press-by-press and shock-by-shock record of the experimental results, or
that if you did, your conclusions were based on whole-session averages. You
should know my opinion of that by now.
Actually I did keep a press-by-press and shock-by-shock record of the
experimental data. The data presented were indeed whole-session averages,
but in this case those averages did not differ in any important way from
data collected at any point within a session. In tracking studies, you
report "whole session averages" like the various correlations, mean square
error, fitted gain value, and so on. The percentage of time spent in the
signaled shock condition is no different from these measures, Bill. It's an
index of overall performance.
I agree that time spent in the signaled condition is definitely of
interest, but whether it is the sort of thing a rat might perceive and
control is another matter. I agree that the rat might _appear_ to control
that variable, but as has been pointed out numerous times on CSGnet, one of
the problems with the Test is that you have to be careful about misleading
appearances. The time spent in the signaled condition may be related to a
controlled variable, but to assume it is the _same_ as the controlled
variable is being too bold.
The same goes for tracking studies, doesn't it? The percentage of time
spent on target is definitely of interest, but whether it is the sort of
thing that a person might perceive and control is another matter. I agree
that a person might _appear_ to control that variable, but as has been
pointed out numerous times on CSGnet, this could be misleading. [Does this
make as little sense to you, now that I have put it into a tracking context,
as your statement does to me? I hope so. I never asserted that the rat was
controlling the percentage of time in the signaled condition, only that it
was a measure of the rat's success in controlling for being in the signaled
condition.]
The proper way to assess the outcome is to compare the amount of time spent
in the changeover (signaled shock, in this case) condition to the amount of
time that would have been spent there in the absence of negative feedback
control action on the part of the rat (essentially zero). If the gain is
high, the animal will be spending nearly all its time in the changeover
condition. Lower gain translates into a more leisurely correction of error
and thus to a reduced time spent in the changeover conditon. With zero gain
(the animal is not controlling for being in the changeover condition) there
will be essentially no time spent in the changeover condition. (I say
essentially because very occasionally the response lever does get tripped as
a side-effect of other activity. The level at which this occurs is assessed
as part of the experimental procedure.)
All that is fine, except for the assumption that the first variable to pass
this test is the controlled variable. I would not be satisfied by this.
I have never asserted or implied that the first variable to pass this test
is the controlled variable. I would not be satisfied by this either, and I
think you know that.
Certainly we can agree that the rats spend more time in the changeover
condition than they would if there were nothing under control,
Hallelujah. That's a start, at least.
but this may
be merely an intermediate variable in the environmental feedback function.
I would ask what the effect of spending more time in the signaled condition
might be on something of more likely relevance to the rat's experiences.
For example, does having a signal to warn of an impending shock enable the
rat to do something to lessen the duration or intensity of the experienced
shock? With a detailed record of events, as in your current series of
experiments, we might be able to check up on measurable differences in the
experienced shock durations and intensities, and get closer to the right
definition.
As I have described previously on CSGnet, I did provide data relevant to
that question, by observing how the rats behaved during each shock and by
measuring the amount of time during each shock that the rats were in contact
with the grid. They run during the shock (in the same way whether signaled
or unsignaled), a behavior not conducive to the adjustments that would be
necessary to control the level of current, which was highly regulated by the
shocker. Another study conducted in Japan measured current flow directly,
with the same results I reported for shock contact duration. If anything,
the rats received slightly more contact and total current-flow in the
signaled condition than they did in the unsignaled condition.
Identifying the length of time spent in the signaled condition as a
possible controlled variable is a large step in the right direction, but
it's not the end of the trail. In my world, it would not be a publishable
result. I would go on to bet that with proper instrumentation, I could
actually measure a large difference in the experienced shock
intensity*duration product, greater than the 3:1 ratio you found when
comparing different shock intensities. In other words, I would bet against
the rat's ability to perceive something as abstract as the "signaled
condition."
As I just described above, it was not the end of the trail, and you would
lose your bet. Of course, I was not able to get inside the rat's head and
experience its perceptions, so it remains possible that the signals before
shock permit the rat to somehow internally reduce the perceived shock
intensity. If you find a way to do that, please let me know. As for
perceiving the abstract concept of "signaled condition," I am not asserting
that rats do perceive such a concept. I am asserting that they control for
having their shocks signaled.
P. S. Reading your subsequent response to Rick, I realize that the proper
way to describe your proposed controlled variable is this: When shock is
inevitable, rats control for receiving signals that occur 2 to 5 seconds
before each shock. Rats appear to be controlling not just for the "signaled
condition," but for signals that occur in a particular temporal relation to
the shocks. And this is true only when shocks are qualitatively inevitable.
It's also true when shocks are avoidable or escapable (Badia & Culberston,
1972).
So far we have no indication of whether this control process is a means to
reduce the quantitative experience of the shocks (energy delivered).
The energy delivered is slightly higher when shocks are signaled, because
the rats "freeze" during the signal and therefore are slower off the dime
when the shock begins. It is still possible that the signals somehow reduce
the perceptual experience. It's a possibility that is tough to test, but if
it happens, then it must be a potent effect, since rats are willing to
change to the signaled shock condition even when shocks are up to three
times as intense (in mA) as those delivered in the unsignaled condition.
(All this was discussed on CSGnet some time ago.)
Perhaps we need to find a somewhat simpler analogy from which to reason
about the changeover procedure. Bill, imagine that you are sitting in your
favorite easy chair, reading a book. The lamp on the stand next to your
chair is on. Mary hypothesizes that you are controlling for having the
light on, so she performs the Test (or whatever it is you now wish to call
it) by switching the lamp off. You immediately reach over and turn it back
on. She turns it off again. You turn it back on. She turns it off.
"Mary," you say with some irritation as you flip the light on again, will
you _please_ leave the light alone?" Mary concludes that you are indeed
controlling for the light being on.
Questions: Has Mary performed a Test for the Controlled Variable? If so,
is she correct that you are controlling for having the light on? What if
you wanted the light on so that you could read? Does the fact that you are
"really" controlling for having enough light on your book to read it mean
that you are NOT simultaneously controlling for having this particular light
on? Would it be fair to say that you are controlling for having this light
on SO THAT you can see the book well enough to read? That is, does the fact
that you are using this light as the MEANS by which you control for the
preception of reading the book negate the fact that you are controlling for
having the light on? (I ask this while fully aware of the fact that if this
means becomes a problem (e.g., Mary won't stop turning off the lamp, or the
bulb burns out), you may well find some other way to get sufficient light on
your book.)
I'll stop here and await your answers.
Regards,
Bruce
[From Bill Powers (980729.2315 MDT)]
Rick Marken (980729.1500)--
My aim was not to dismiss Bruce's efforts at testing rat preference.
The main point of my post was simply that we can't conclude that an
animal is controlling a particular variable if the Test is done in
a situation where the animal cannot possibly control the presumed
controlled variable.
As I understand it, the proposed controlled variable was the state of the
apparatus -- whether it was in the signalling or non-signalling mode. In
the non-signalling mode, shocks occurred on some schedule with no warning
being given. In the signalling mode (indicated by a light in some trials,
or absence of the light in others), shocks continued to be given on the
same schedule, but with a warning (I forget whether it was a light or a
tone) occurring 2 to 5 seconds prior to every shock. In the non-signalling
mode, the rat could switch to the signalling mode by a single press on a
lever. Again, I don't recall exactly how, but after some time in the
signalling mode, the apparatus would spontaneously revert to the
non-signalling mode. The rat would then have to press the lever to get it
back into the signalling mode.
The rats kept the apparatus in the signalling mode almost all of the time,
and the disturbance of the apparatus reverting to the non-signalling mode
was quickly countered by the rat's pressing the lever to restore it to the
signalling mode. So the rat was controlling the state of the apparatus,
which was interpreted as showing that the state in which it got a warning
of an impending shock was positively reinforcing. This interpretation can't
be directly verified, but it's a reasonable anthropomorphization to assume
that the rat preferred being "warned."
The rats also showed that they preferred the signalling mode with the shock
set at three times the strength of the shock in the non-signalling mode. So
the signalling was very important to them (assuming that the increased
shock current was actually more painful than the other level of current).
You are objecting that the rats had no control over either the timing of
the warning or the occurance of shocks, but those are not claimed to be
controlled variables. The only claimed controlled variable is the one that
the rat's actions could affect, the mode of the apparatus. This claim is
not contradicted by the results, although one can question whether the
results have isolated the perceived variables that were actually under
control.
I join you in questioning whether the Test can really be performed
accidentally, without understanding the principles of control. I believe
that Bruce's stand is that even though he was doing this experiment under
the impression that he was investigating reinforcers, the procedures were
identical to those involved in the Test. If I were to accept any part of
this argument, it would be on the basis that the conceptual framework
actually accepted the idea of purposiveness in the rat's behavior, and thus
control. Whatever words were used, it is clear that the rat was given
control over the state of the apparatus, and exerted it. There is no
evidence that anything in the experimental apparatus controlled the rat's
actions (unless, of course, you use some non-technical meaning of the word
control).
Best,
Bill P.