My research on controlled variables

[From Bruce Abbott (980313.1330 EST)]

Rick Marken has asked me to describe my research on identifying controlled
variables. Here is an early (mid-1970s) example (not the one I described on
nutrient control), conducted while I was in grad school and before I read B:CP.

Rats were exposed to two conditions, both of which included the delivery at
random of mild (1.0 mA), half-second footshocks through the grid floor of
the test chamber at an average rate of one per two minutes. When the
chamber houselight was on, each shock was immediately preceded by a 5-s
warning tone, which ended with the shock (signaled shock condition). When
the houselight was off, the shocks were not preceded by the tone (unsignaled
shock condition). Each rat was given several training sessions with each
condition to familiarize them with the conditions and to associate the
conditions with the state of the houselight.

Following this training, each rat was placed in the unsignaled condition.
By pressing a lever, the rat could "buy" on minute of time in the signaled
condition. When the lever was pressed, the houselight would immediately
turn on, and any shocks that happened to be programmed while the houselight
was on would be preceded by the signal. (Lever presses during this time had
no programmed effect.) At the end of the minute, the houselight
extinguished and the rat was returned to the unsignaled shock schedule. By
pressing the lever once immediately after being returned to the unsignaled
schedule, the rat could buy another minute in the signaled condition, and by
doing so on each occasion, could spend nearly the entire session in the
signaled condition. This phase was called changeover to signaled shock.

In another phase of the experiment, the rats were placed in the signaled
shock condition and could buy 1-min periods of unsignaled shock in the same
way. This phase was called changeover to unsignaled shock.

_Control-system analysis_

If, through their training with the signaled and unsignaled shock schedules,
the rats had developed a reference for being in the signaled (as opposed to
unsignaled) condition, then being placed in the unsignaled condition would
have constituted a disturbance. The rats could counteract that disturbance
by pressing the lever, thus switching themselves to the signaled condition.
After one minute the apparatus would disturb the CV again by automatically
switching the rat back to the unsignaled condition, and once again the rat
could counter the disturbance by pressing the lever and thereby returning to
the signaled conditions.

If the rats had developed a reference for being in the unsignaled (as
opposed to the signaled) condition, then being placed in the unsignaled
condition would not consitute a disturbance to the CV, and they should not
press the lever, which would in fact produce the disturbance. Conclusion:
If the rats were controlling for being in the signaled rather than the
unsignaled condition, then when in changeover to signaled shock, they should
immediately press the lever each time they are returned to the unsignaled
condition. If they were controlling for being in the unsignaled condition,
they should avoid pressing the lever.

In changeover to unsignaled shock, the reverse would be true: If the rats
had a reference for being in the signaled condition, they should avoid
pressing the lever, but if they had a reference for being in the unsignaled
condition, they should press the lever immediately upon being automatically
returned to the signaled schedule at the end of each 1-min changeover period.

Finally, there is the possibility that the rats would not care which
condition they received. In that case the amount of lever-produced changing
over from one condition to the other should be about the same in both
changeover phases and occur at essentially baseline (low) rates as a
side-effect of other activities (e.g., exploration).

_Results_

The rats strongly controlled for being in the signaled condition, spending
from 85% to 95% of session time in the signaled condition. When placed in
changeover to signaled shock, they changed over to signaled shock. when
placed in changeover to unsignaled shock, they stayed away from the lever
and did not change over. Subsequent research attempted to tease out what
specific elements, differing between the signaled and unsignaled conditions,
was responsible for this result, i.e., what were the rats really controlling
for when they acted so as to remain in the signaled rather than the
unsignalled schedule?

Regards,

Bruce

[From Bruce Gregory (980313.1245 EST)]

Bruce Abbott (980313.1330 EST)

_Results_

The rats strongly controlled for being in the signaled condition, spending
from 85% to 95% of session time in the signaled condition. When placed in
changeover to signaled shock, they changed over to signaled shock. when
placed in changeover to unsignaled shock, they stayed away from the lever
and did not change over. Subsequent research attempted to tease out what
specific elements, differing between the signaled and unsignaled conditions,
was responsible for this result, i.e., what were the rats really controlling
for when they acted so as to remain in the signaled rather than the
unsignalled schedule?

I can see that many of the rats developed a preference under
these conditions. This does not seem surprising. Further, I am
not sure what we learn by knowing this. Is this a fair question
to ask?

Bruce

p.s. I just noticed that I am prescient. I am responding to your
post 45 minutes before you wrote it...

[From Bruce Abbott (980313.1320 EST)]

Bruce Gregory (980313.1245 EST) --

I can see that many of the rats developed a preference under
these conditions. This does not seem surprising.

Actually, all of the rats developed a preference under these conditions.
Why does this not seem surprising to you?

Further, I am
not sure what we learn by knowing this. Is this a fair question
to ask?

That is always a fair question to ask, but it is not one that is always easy
for a scientist to answer. But I will make the attempt. First, the result
is unexpected under traditional (i.e., reinforcement) theory. That in
itself should have made the phenomenon worth investigating, as the answer
would shed light on -- and possibly alter -- accepted fundamentals in the
field. Second, understanding why this preference develops could help us to
understand how better to deal with life stresses, because whatever these
factors are that lead to a preference for predictable over unpredictable
aversive events, they appear also to be associated with reduction in
physiological activity of the sort associated with the weakening of the
immune system, adrenal hypertrophy, elevated blood pressure and cortisol,
ulceration of the stomach, and so on, i.e., stress-related changes as
identified and discussed by Hans Selye.

p.s. I just noticed that I am prescient. I am responding to your
post 45 minutes before you wrote it...

No doubt one of many fine talents . . .

Regards,

Bruce

[From Bruce Gregory (980310.1346 EST)]

Bruce Abbott (980313.1320 EST)

>Bruce Gregory (980313.1245 EST) --

>I can see that many of the rats developed a preference under
>these conditions. This does not seem surprising.

Actually, all of the rats developed a preference under these conditions.
Why does this not seem surprising to you?

Nobody seems to like unpleasant surprises. On the other hand,
there some evidence that suggests that some people would rather
not hear news that _might_ prove to be bad. Rats apparently do
not indulge in this sort of magical thinking.

>p.s. I just noticed that I am prescient. I am responding to your
>post 45 minutes before you wrote it...

No doubt one of many fine talents . . .

Aw gee, Your just saying that....

Bruce

[From Rick Marken (980313.1050)]

Bruce Abbott (980313.1330 EST)--

Rick Marken has asked me to describe my research on identifying
controlled variables. Here is an early (mid-1970s) example (not
the one I described on nutrient control), conducted while I was
in grad school and before I read B:CP.

Why do you think this study tells us anything more about controlled
variables than any other operant study? Remember how we thought
that operant "scheduling" experiments told us that rats are
controlling rate of reinforcement (because rate of reinforcment
remains _nearly_ constant across variations in schedule)? And
you proved that they don't show that because the rats are always
responding as fast as they can; the output is not compensating for
the change in feedback function.

The rats strongly controlled for being in the signaled condition,
spending from 85% to 95% of session time in the signaled condition.

I bet the rats would also appear to control for being in a "food
delivery" condition, pressing to spend 85% to 95% of session time
in that condition. This study tells us very little about what
the rat is controlling; only that it is controlling _some_ aspect
of the food delivery (or "signaled condition", as the case may be).

Try thinking about your experiment in terms of the pattern of x's
example that I gave in an earlier post. If a person is controlling
for _something_ about the x's then some movements of some x's will
lead to compensating behavior and some movements of these and
other x's will lead to no compensating behavior. Similarly,
some variations in signaling will lead to some variations in
actions if the rat is controlling _something_ about the shocks and
signals and other variations will not. But without a hypothesis
about the controlled variable and systematic Testing to determine
whether the hypothesized variable is under control there is no
way to know which of the many possible aspects of the x's (or
signals/shock) is under control.

Subsequent research attempted to tease out what specific
elements, differing between the signaled and unsignaled
conditions, was responsible for this result, i.e., what were
the rats really controlling for when they acted so as to remain
in the signaled rather than the unsignalled schedule?

Now _this_ is the research I would like to hear about! How
did they (you?) test to determine what aspect of the signaled
condition the rats were controlling? It may be that this whole
series of experiments (yours and the follow-ups done to nail down
the controlled aspect of the "signalled" condition) constitutes
a Test for the controlled variable.

···

--
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 (980313.1410 EST)]

Bruce Gregory (980310.1346 EST) --

Bruce Abbott (980313.1320 EST)

Actually, all of the rats developed a preference under these conditions.
Why does this not seem surprising to you?

Nobody seems to like unpleasant surprises.

Rats included. The question is, why?

On the other hand,
there some evidence that suggests that some people would rather
not hear news that _might_ prove to be bad. Rats apparently do
not indulge in this sort of magical thinking.

Yes, rats are very down-to-earth (literally).

>p.s. I just noticed that I am prescient. I am responding to your
>post 45 minutes before you wrote it...

No doubt one of many fine talents . . .

Aw gee, Your just saying that....

No, I mean it. Besides, anyone who likes to fly can't be _all_ bad . . .

Regards,

Bruce

[From Bill Powers (980313.1414 MST)]

Bruce Abbott (980313.1330 EST)--

Rick Marken has asked me to describe my research on identifying controlled
variables. Here is an early (mid-1970s) example (not the one I described on
nutrient control), conducted while I was in grad school and before I read

B:CP.

Rats were exposed to two conditions, both of which included the delivery at
random of mild (1.0 mA), half-second footshocks through the grid floor of
the test chamber at an average rate of one per two minutes. When the
chamber houselight was on, each shock was immediately preceded by a 5-s
warning tone, which ended with the shock (signaled shock condition). When
the houselight was off, the shocks were not preceded by the tone (unsignaled
shock condition). Each rat was given several training sessions with each
condition to familiarize them with the conditions and to associate the
conditions with the state of the houselight.

Perhaps I'm missing something here, but how did you rule out the
possibility that the rats simply preferred the house lights to be on
instead of off?

Best,

Bill P.

[From Bruce Abbott (980313.2215 EST)]

Bill Powers (980313.1414 MST) --

Perhaps I'm missing something here, but how did you rule out the
possibility that the rats simply preferred the house lights to be on
instead of off?

I didn't describe the control for that possibility: for half the rats,
houselight-on was associated with the signaled condition; for the remainder
it was associated with the unsignaled condition.

Also, bear in mind that rats are nocturnal and on this ground would be
expected a priori to prefer a dark chamber over an illuminated one.
Associating the signaled condition with houselight-on would bias the outcome
against preference for the signaled condition. However, the rats preferred
the signaled condition either way.

Regards,

Bruce

[From Bill Powers (980314.0515 MST)]

Bruce Abbott (980313.2215 EST)--

I didn't describe the control for that possibility: for half the rats,
houselight-on was associated with the signaled condition; for the remainder
it was associated with the unsignaled condition.

Ah, that does make a difference!

Also, bear in mind that rats are nocturnal and on this ground would be
expected a priori to prefer a dark chamber over an illuminated one.
Associating the signaled condition with houselight-on would bias the outcome
against preference for the signaled condition.

Was there any sign of this bias in the data?

However, the rats preferred the signaled condition either way.

It's difficult for me to accept that rats could understand a concept like
"signaled condition." I would keep looking for something that is different
between these two conditions, something that a rat incapable of verbal
generalizations might be able to perceive. For example, was there any
difference in the number of shocks received in the two conditions?

Best,

Bill P.

[From Bruce Nevin (980314.1607)]

Bill Powers (980314.0515 MST)--

(Reply to Bruce Abbott (980313.2215 EST))

It's difficult for me to accept that rats could understand a concept like
"signaled condition." I would keep looking for something that is different
between these two conditions, something that a rat incapable of verbal
generalizations might be able to perceive. For example, was there any
difference in the number of shocks received in the two conditions?

Here are two observed variables which could be perceptions:

1. Sometime about every 120 s, a shock.

2. Sometime about every 120 s, a sequence of 5-s tone ending with a shock.

Suppose the rats have some way to control the perception "no shock".
Suppose (for example) that the rats learn to jump off the conductive floor
a tad just as the tone ends and the shock happens. Or suppose they can
"steel" themselves to the shock, maybe make it less unpleasant by tensing
muscles in advance, but they have know when to do this. Or suppose they
don't know when to relax and attend to other things without the tone as the
first part of a sequence, and they prefer to be able to relax and attend to
other things at least some of the time.

Barring some such means to control a perception something like "no shock"
or "relax and enjoy life", the inference Bruce A seems to be making is that
the rats control for knowing when a mild unpleasantness will happen, as
opposed to having it catch them unaware. Even this requires no perception
called "the signalled condition." There are two perceptions. One is the
sequence, 5-s tone ending with shock. The other is shock.

Bruce Abbott (980313.1330 EST)--

After one minute the apparatus would disturb the CV again by automatically
switching the rat back to the unsignaled condition, and once again the rat
could counter the disturbance by pressing the lever and thereby returning to

I'm having trouble with the time specification here. After one minute? The
shocks are on average 2 minutes apart. How often would signalled shocks
ever occur within a 1-minute grace period? Is this a typo for 10 minutes,
or some other number greater than 2?

Thanks for describing this experiment. What explanation did you have at the
time? How has your understanding of PCT affected that explanation?

  Bruce Nevin

[From Rick Marken (980315.1150)]

Bruce Abbott (980313.2215 EST)--

However, the rats preferred the signaled condition either way.

Bill Powers (980314.0515 MST)

It's difficult for me to accept that rats could understand a
concept like "signaled condition." I would keep looking for
something that is different between these two conditions...
For example, was there any difference in the number of shocks
received in the two conditions?

I'm pretty sure that even researchers who believe that "signals"
and shock _cause_ behavior would be careful to make sure that there
was no difference in shock levels for the signaled and unsignaled
conditions. This is true, right Bruce A.?

Anyway, as Bruce noted, he was doing this research to test for
controlled variables. That's why I [Rick Marken (980313.1050)]
expressed interest in hearing about the "subsequent research"
Bruce mentioned [Bruce Abbott (980313.1330 EST)] that attempted
to tease out...what the rats [were] really controlling for when
they acted so as to remain in the signaled rather than the
unsignalled schedule?"

I imagine that this subsequent research involved systematically
disturbing other aspects of the "signaling" situation that the
rats could control. For example, they probably did studues where
the phase relationship between signals and shocks was disturbed.
The rats were presumably given a means to control this phase
relationship (for example, by pressing the bar faster or slower
they could advance or retard the phase). The rats must have
controlled for the signals occuring just before the shocks, since
Bruce A. told us that the rats were controlling for "signaling",
and Bruce A. wouldn't lie (like I do;-))

I'm sure there were a whole bunch of tests like this to determine
what the rats were controlling because, as Bruce A. said, he knew
how to test for controlled variables before he ever even heard of
PCT (and Bruce A. is an honorable man). This is great news because
I have been under the impression that "conventional" psychologists
have not been doing much testing for controlled variables. But,
apparently, some (like Bruce A.) have so now we will have many
examples of research aimed at testing for controlled variables.

I'm really interested in hearing about the research that nailed
down what the rats in these "signaled shock" experiments are
controlling for, Bruce. Could you describe it for us? It would
sure make a great substantive contribution to the discussions
on the net.

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 (980316.1245 EST)]

Bill Powers (980314.0515 MST) --

Bruce Abbott (980313.2215 EST)

Also, bear in mind that rats are nocturnal and on this ground would be
expected a priori to prefer a dark chamber over an illuminated one.
Associating the signaled condition with houselight-on would bias the outcome
against preference for the signaled condition.

Was there any sign of this bias in the data?

Very little: Preference for the signaled over the unsignaled condition is
strong, tending to swamp out weaker preferences like those for darkness over
illumination.

However, the rats preferred the signaled condition either way.

It's difficult for me to accept that rats could understand a concept like
"signaled condition." I would keep looking for something that is different
between these two conditions, something that a rat incapable of verbal
generalizations might be able to perceive. For example, was there any
difference in the number of shocks received in the two conditions?

The number, intensity, duration, and distribution of shocks was identical in
the two conditions. The apparatus programmed the shocks independently of
conditions.

Bruce Nevin (980314.1607) --

Suppose the rats have some way to control the perception "no shock".
Suppose (for example) that the rats learn to jump off the conductive floor
a tad just as the tone ends and the shock happens. Or suppose they can
"steel" themselves to the shock, maybe make it less unpleasant by tensing
muscles in advance, but they have know when to do this.

This, in various versions, is called the preparation hypothesis.

Or suppose they
don't know when to relax and attend to other things without the tone as the
first part of a sequence, and they prefer to be able to relax and attend to
other things at least some of the time.

This is known as the safety-signal (or safety) analysis.

Barring some such means to control a perception something like "no shock"
or "relax and enjoy life", the inference Bruce A seems to be making is that
the rats control for knowing when a mild unpleasantness will happen, as
opposed to having it catch them unaware. Even this requires no perception
called "the signalled condition." There are two perceptions. One is the
sequence, 5-s tone ending with shock. The other is shock.

There are others. One is the houselight being on or off. In the case where
houselight-on was associated with the signaled condition, houselight-on plus
tone is associated with imminent shock, houslight-on by itself with absence
of shock, and houselight-off with some low (but non-zero) rate of shock
occurrence (120-s/shock on average).

After one minute the apparatus would disturb the CV again by automatically
switching the rat back to the unsignaled condition, and once again the rat
could counter the disturbance by pressing the lever and thereby returning to

I'm having trouble with the time specification here. After one minute? The
shocks are on average 2 minutes apart. How often would signalled shocks
ever occur within a 1-minute grace period? Is this a typo for 10 minutes,
or some other number greater than 2?

A single lever-press while in the unsignaled condition bought the rat one
minute of time in the signaled condition, with its associated state of the
houselight. Shocks were programmed on a variable-time schedule which
approximated a constant shock probability over time. Intervals between
shocks varied at random between an 8-s minimum to a mamimum of around 10
minutes. Thus there would have been many occasions when no shocks (or
signals) would occur during the 1-minute changeover period.

Thanks for describing this experiment. What explanation did you have at the
time? How has your understanding of PCT affected that explanation?

At the time we were evaluating several hypotheses, including the preparation
and safety hypotheses. According to the safety analysis, the rats changed
over from the unsignaled to the signaled shock condition because most often
this would result in an immediate transition from a stimulus associated with
a constant probability of shock (the stimulus identifying the unsignaled
condition) to one associated with a zero probability of shock (the stimulus
identifying the signaled condition, presented without the tone) and much
less often to a stimulus associated with shock (stimulus identifying the
signaled condition, plus tone). According the the preparation analysis,
rats changed over to the signaled condition so that they could execute
well-timed preparatory behaviors that would serve to minimize the perceptual
aversiveness of the shocks. Proposed behaviors included adopting a
particular posture on the grid, getting ready to spring off the grid and
thus minimize shock contact time or current flow, tensing muscles, and
releasing endorphines in the brain and/or spinal cord to produce a temporary
state of relative analgesia. There was also an information hypothesis,
which stated that rats have evolved a preference for having information
about biologically significant events (like the occurrence of food or of
danger) and will choose to have that information even when it is "bad news."

My understanding of PCT has not affected how I would go about explaining the
preference, but it does suggest alternative ways to conduct the
investigation which I would not have thought of otherwise.

Rick Marken (980315.1150) --

Anyway, as Bruce noted, he was doing this research to test for
controlled variables. That's why I [Rick Marken (980313.1050)]
expressed interest in hearing about the "subsequent research"
Bruce mentioned [Bruce Abbott (980313.1330 EST)] that attempted
to tease out...what the rats [were] really controlling for when
they acted so as to remain in the signaled rather than the
unsignalled schedule?"

Glad to hear that you're interested.

I imagine that this subsequent research involved systematically
disturbing other aspects of the "signaling" situation that the
rats could control. For example, they probably did studues where
the phase relationship between signals and shocks was disturbed.
The rats were presumably given a means to control this phase
relationship (for example, by pressing the bar faster or slower
they could advance or retard the phase).

Had I known more about the Test at the time, I might have conducted a study
very much like this, but alas, I didn't. What I did do might be considered
an alternative form of the Test. First, I established conditions under
which rats would immediately act to correct the disturbance (being switched
back into the unsignaled condition) by pressing the lever that returned them
into the signaled condition). Next, I systematically varied the parameters
and conditions of the signaled condition, looking for changes what would
destroy the rat's interest in correcting the disturbance. In other words, I
looked for features of the signaled condition which, when removed,
eliminated the rat's interest in controlling for being in the signaled
condition.

The rats must have
controlled for the signals occuring just before the shocks, since
Bruce A. told us that the rats were controlling for "signaling",
and Bruce A. wouldn't lie (like I do;-))

You are right about one thing: I don't lie. But speaking of lies, did not
say that the rats were controlling for "signaling." I said that the rats
controlled for being in the signaled, rather than the unsignaled, shock
condition, which is exactly what they did. After this had been demonstrated,
the next step was to identify what specific factors were responsible for the
rats' preference for the signaled over the unsignaled shock condition.

I'm sure there were a whole bunch of tests like this to determine
what the rats were controlling because, as Bruce A. said, he knew
how to test for controlled variables before he ever even heard of
PCT (and Bruce A. is an honorable man). This is great news because
I have been under the impression that "conventional" psychologists
have not been doing much testing for controlled variables. But,
apparently, some (like Bruce A.) have so now we will have many
examples of research aimed at testing for controlled variables.

I didn't know at the time that I was testing for controlled variables (From
my point of view I was testing for "preference"), but in retrospect I can
see that this is exactly what I was doing.

I'm really interested in hearing about the research that nailed
down what the rats in these "signaled shock" experiments are
controlling for, Bruce. Could you describe it for us? It would
sure make a great substantive contribution to the discussions
on the net.

Great! I've got quite a number of them. For starters, how about the one in
which I evaluated the version of the preparation hypothesis which says that
the rats use the signal to get set to hop off the grid floor, thereby
minimizing their contact with shock? For this study I designed and
constructed a piece of electronics which allowed me to record the pattern
and total duration of contact with each shock delivered. Each rat was
exposed to both the signaled and unsignaled shock conditions while that
pattern and duration of each shock was recorded. I also put together a
solenoid mechanism which tripped the shutter of a 35mm camera just at the
moment prior to shock delivery, in order to capture on film the rat's
posture on the grid.

It did turn out that the rats got into a particular posture during the
signal (they "froze," which is a species-typical response to certain signs
of possible danger), but in consequence they were slower to start running
when the shock began and thus received longer total contact with the grid
during the shock than when the shocks were delivered unsignaled. The rats
were actually receiving more total contact with shock when it was signaled
than when it was unsignaled.

This experiment ruled out differential grid contact as the explanation. An
earlier study had demonstrated that rats continued to prefer the signaled
shock condition when shocks delivered there were up to three times more
intense than those delivered in the unsignaled condition. Whatever it is
that the rats prefer about the signaled condition, it is powerful enough to
win out over even fairly strong preferences for weaker over stronger shock,
when these two factors are pitted against one another. If some other type
of preparatory response is at work, it would have to be extremely effective
to make these physically much stronger shocks seem weaker than those
physically weaker shocks of the unsignaled condition.

Regards,

Bruce

[From bruce Gregory (980316.1355 EST)]

Bruce Abbott (980313.1410 EST)

>Bruce Gregory (980310.1346 EST) --

>>Bruce Abbott (980313.1320 EST)

>> Actually, all of the rats developed a preference under these conditions.
>> Why does this not seem surprising to you?

>Nobody seems to like unpleasant surprises.

Rats included. The question is, why?

Is that really the question? It seems to me that you have
demonstrated that rats control for not being surprised. One can
imagine some evolutionary "just so" stories about why this is
the case, but it might be more helpful to know the extent to
which other species control for not being surprised.

Bruce

[From Rick Marken (980316.1120)]

Bruce Abbott (980316.1245 EST)--

What I did do might be considered an alternative form of the Test.

Only by you.

First, I established conditions under which rats would immediately
act to correct the disturbance (being switched back into the
unsignaled condition) by pressing the lever that returned them
into the signaled condition). Next, I systematically varied the
parameters and conditions of the signaled condition, looking for
changes what would destroy the rat's interest in correcting the
disturbance. In other words, I looked for features of the signaled
condition which, when removed, eliminated the rat's interest in
controlling for being in the signaled condition.

This technique can tell you what the rat is _not_ controlling for
but it can't tell you what the rat _is_ controlling for. So you
atill have not described research involing a Test for the controlled
variable. Let's see what the next paragraph brings.

You are right about one thing: I don't lie.

Maybe not intentionally. But lying is often an uncontrolled side-
effect of ignorance. Calling the study you describe above "a version
of the Test" is a lie. Draw your own conclusions;-)

I said that the rats controlled for being in the signaled, rather
than the unsignaled, shock condition... After this had been
demonstrated, the next step was to identify what specific factors
were responsible for the rats' preference for the signaled over
the unsignaled shock condition.

Gee. I thought the next step was to identify the variable the
rats were controlling.

I didn't know at the time that I was testing for controlled
variables...but in retrospect I can see that this is exactly
what I was doing.

Really?

Each rat was exposed to both the signaled and unsignaled shock
conditions while that pattern and duration of each shock was
recorded.

These measurements tell you precisely _nothing_ about what the
rats were controlling for in the "signalled" condition. For all
you know, the rats were controlling for amount of shock and were
able to get that amount closer to their non-zero reference in
the "signalled" condition". Unlikely, I agree, but you can't test
for controlled variables by just measuring variables; you have
to monitor these variables while they are being disturbed.

This experiment ruled out differential grid contact as the
explanation.

What explanation? I thought we were looking for controlled variables.
You said you were testing for controlling variables. You lied
to me Bruce. Lied. Lied. I'm crushed.

An earlier study had demonstrated that rats continued to prefer
the signaled shock condition when shocks delivered there were
up to three times more intense than those delivered in the
unsignaled condition.

Gee, maybe they _do_ control for the shock. Maybe they are controlling
for the continuity of the shock. You might have looked into this
if you actually knew what controlled variables are and how to test
for them.

Come on, Bruce. We want the emmis. Tell us about the research you
did that _really_ involved the Test.

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 (980316.1955)]

Rick Marken (980316.1120) --

Bruce Abbott (980316.1245 EST)

What I did do might be considered an alternative form of the Test.

Only by you.

This technique can tell you what the rat is _not_ controlling for
but it can't tell you what the rat _is_ controlling for. So you
atill have not described research involing a Test for the controlled
variable. Let's see what the next paragraph brings.

First, we already know that the rat is controlling for being in the signaled
rather than the unsignaled condition, because the Test tells us that. So
the rat is controlling for _something_ related to being in the signaled as
opposed to the unsignaled conditon. That narrows the search down quite a bit.

Second, we systematically vary features of the signaled condition and
observe whether the changes affect the rat's willingness to control for
being in the signaled condition. When we find changes that eliminate the
rat's interest in controlling for being in the signaled condition, we have
identified factors that are necessary for establishing that reference. I am
fairly sure that Bill P. can see the logic in this approach, even if you can't.

Calling the study you describe above "a version
of the Test" is a lie. Draw your own conclusions;-)

I described it as an alternate way to Test for controlled variables, and
that is what it is. No lie. Try it!

I said that the rats controlled for being in the signaled, rather
than the unsignaled, shock condition... After this had been
demonstrated, the next step was to identify what specific factors
were responsible for the rats' preference for the signaled over
the unsignaled shock condition.

Gee. I thought the next step was to identify the variable the
rats were controlling.

Yes, that's what I was doing. When you know what, specifically, the rats
were controlling _for_, you know what they were controlling.

I didn't know at the time that I was testing for controlled
variables...but in retrospect I can see that this is exactly
what I was doing.

Really?

Really. If you have some principled argument to offer against that
assertion, let's hear it. So far all I have heard from you is assertions
and sarcasm. I hope not too many find such "arguments" persuasive. I know
that I don't.

Each rat was exposed to both the signaled and unsignaled shock
conditions while that pattern and duration of each shock was
recorded.

These measurements tell you precisely _nothing_ about what the
rats were controlling for in the "signalled" condition. For all
you know, the rats were controlling for amount of shock and were
able to get that amount closer to their non-zero reference in
the "signalled" condition". Unlikely, I agree, but you can't test
for controlled variables by just measuring variables; you have
to monitor these variables while they are being disturbed.

You can't take each experiment in isolation, Rick. We already know that
rats have a zero reference for shock intensity (see the Verhave study Bill
P. modeled). This experiment ruled out the possibility that they were
controlling for reduced shock contact. One often can learn as much from
ruling out as from ruling in.

Regards,

Bruce

[From Rick Marken (980316..2025)]

Bruce Abbott (980316.1955)

First, we already know that the rat is controlling for being in
the signaled rather than the unsignaled condition, because the
Test tells us that.

That's a pretty vague conclusion. Think about this in terms of
the coin game. What you have shown is equivalent to showing that
the subject in the coin game will repeatedly return the same
coins to their original position (the "signaled" condition) each
time they are moved to the same new position (the "unsignaled"
condition). . So you know that the subject is controlling
_something_ about the coins, just as you know that the rats are
controlling _something_ about the "unsignaled/signaled" condition.
This is better than knowing _nothing_ about what the subject is
controlling, but not much. This is where The Test really begins.
But rather than Testing (see "The Coin Game" discussion in B:CP to
see how to do the Test) you do the following:

Second, we systematically vary features of the signaled condition
and observe whether the changes affect the rat's willingness to
control for being in the signaled condition. When we find changes
that eliminate the rat's interest in controlling for being in the
signaled condition, we have identified factors that are necessary
for establishing that reference.

This is equivalent to trying to vary "features" of the coin
pattern (the one equivalent to the "signaled" condition), such as
it's shape, size, relative positions of coins, etc., so that
the subject makes no effort to change the coins back to this
pattern (the only one they can produce by "pressing") from the
pattern that is equivalent to the "unsignaled" condition.

For example, in you original experiment you found that rats
would press to get pattern 1 back to pattern 2.

1 (unsignaled) 2 (signaled)

  * * * *
       * *

Now you try to find a new pattern 2 that rats will _not_ work
to change pattern 1 into. Lets say you find it's this:

                     2' (altered signaled)
                  * * *

Have you learned what it is about pattern 2 that the subject
was controlling? That's how much your procedure tells you
about what the rat was controlling for in the signaled condition.

But I was wrong about one thing. I said that your psychedelic
version of The Test "can tell you what the rat is _not_ controlling
for". That was an (unintentional) lie on my part. As you can now
see, your version of the Test tells you precisely _nothing_ about
what the subject is controlling for and _nothing_ about what the
subject is _not_ controlling for. Very impressive.

I am fairly sure that Bill P. can see the logic in this approach,
even if you can't.

I'm sure Bill (like me) can see that the logic of this approach
is the logic of a conventional psychologist who is _desperate_
to see PCT as "just another theory of behavior". Sad.

Me:

Calling the study you describe above "a version of the Test" is a
lie. Draw your own conclusions;-)

Bruce:

I described it as an alternate way to Test for controlled
variables, and that is what it is. No lie. Try it!

I did (see above). It doesn't work. Try it!

If you have some principled argument to offer against that
assertion, let's hear it.

My reasons (diagrammed above) are based on an understanding of
the "principle" of control and control theory. I thought you
understood these principles too. My mistake.

I hope not too many find such [my] "arguments" persuasive.
I know that I don't.

And I know that you never will -- unless, by some miracle, you
are able to admit to yourself that all of the painstaking research
you have done outside the context of an understanding of control
was a lot of sound and fury, signifying (nearly) nothing.

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 Bill Powers (980317.0239 MST)]

Bruce Abbott (980316.1245 EST)--

The number, intensity, duration, and distribution of shocks was identical in
the two conditions. The apparatus programmed the shocks independently of
conditions.

Does that mean the number, intensity, etc. of _experienced_ shocks, or of
_programmed_ shocks? If the signal made no ultimate difference in the rat's
experience of the shocks, it's difficult to understand why where was a
preference for the signaled condition. The rat must have been able to do
something different in the signaled condition that reduced the unwanted
effects of the shocks, whatever they were -- at least if we assume that the
point of pressing the lever was to reduce some sort of error.

There's a bit of technical information missing here. Did you in fact verify
that in the signaled condition, the duration and magnitude of the current
passing through the rat was identical to that in the unsignaled condition?
Or did you record just the turning on of the apparatus as if a shock were
being administered? I suspect the latter, since your "preparation"
hypothesis implies that being prepared could have some effect on the
experienced shock, but that you couldn't directly verify this. If you could
directly verify that the rat's behavior reduced the effect of the shock on
the animal (as with the rat that flipped over on its back after the
signal), you wouldn't have to hypothesize about it. When you turned the
current-limiting up to three times its previous value, were you able to
determine that in fact three times the current flowed through the rat for
the same length of time under both conditions? Or did the rat's
"preparation" (or whatever) succeed in reducing the
intensity-times-duration of the current flow through the animal in the
signaled condition?

We could easily explain the preference for the signaled condition in the
case of the rat that flipped over on its back. In the signaled condition,
that rat would receive no shocks. That rat was able to control its
experience of the shocks, first by causing the signaled condition to occur,
and then by controlling the sequence (signal, shock) (at zero) through some
learned action. But after you shaved its back to prevent this mode of
control, I suppose its preference persisted, implying perhaps that it had
found some _other_ way to affect the duration and intensity of the
experienced shock in the signaled condition. That hypothesis could easily
be disproven by measuring the intensity and duration of the current flowing
through the rat.

The crux of the matter here, as far as the Test is concerned, is whether
the signal enabled the rat to alter what it experienced when the shock
occurred. If it did, we can hypothesis a two-level control system; one
level would reduce the shock experience when possible, and the other would
try to make it possible.

You say that you were trying to eliminate the preference by eliminating
supposed differences between the two conditions. That is a valid aspect of
the Test. However, I am asking whether you were really instrumented to
determine what the rat's experience (probably) was. The nature of the
guesses you were making suggests that this instrumentation was lacking.

Best,

Bill P.

[From Bill Powers (980317.04023 MST)]

Rick Marken (980316..2025) --

Now you try to find a new pattern 2 that rats will _not_ work
to change pattern 1 into. Lets say you find it's this:

                    2' (altered signaled)
                 * * *

Have you learned what it is about pattern 2 that the subject
was controlling? That's how much your procedure tells you
about what the rat was controlling for in the signaled condition.

I think it's more pertinent to start by admitting that Bruce A. was,
indeed, trying to find out what it was about the signaled condition that
was preferred. One way to do this is to try to eliminate the hypothesized
feature of the signaled condition to see if the attempts to re-establish it
will eventually disappear. Unfortunately, that approach considers only the
reorganizing aspects of the situation. Under the control hypothesis, if you
make it harder to achieve the desired condition, the organism should at
first produce _more_ behavior, not less. If the behavior is already at its
maximum, you should see no change.

The real criticism here is that Bruce's strategy assumes that behavior is
controlled by its consequences. Under that assumption, if you make it
impossible for the consequence to appear, the behavior should disappear. In
the long run, PCT would predict the same thing, because eventually
reorganization will kick in, but in the short run, PCT would say that
behavior should increase, not decrease. However, when the behavior involved
is binary, so it can only occur or not occur, and when its effects are also
binary (signaled condition or not), it's impossible to measure quantitative
changes in it: it's on or it's off.

So the strategy of tripling the programmed shock current in the signaled
condition to see if it affects the preference doesn't really test anything
relating to PCT. Doing so increases the error in the shocked condition, but
that would only serve to increase the behavior that is already in place to
counteract the error. Since the behaviors available are very limited --
turn on the signaled condition, flip over on your back, jump up in the air,
etc. -- there isn't any obvious way to increase the attempt to correct the
error.

What this variant on the experiment does show is that the consequence is
not controlling the behavior. Even though turning on the signaled condition
triples the programmed shock current, the animal continues to prefer the
signaled condition. Under PCT, this implies that in the signaled condition,
the animal does NOT experience three times the shock current; in fact, it
still probably experiences less shock than in the unsignaled conditionl
although not as much less as when the programmed shock currents are equal.

Remember that in EAB, when it is said that an animal "prefers" some
condition, this does not mean that the animal has a desire or intention
relating to that condition, and acts purposively so as to bring it about.
It means that the condition in question, when it happens to occur, alters
the animal's behavior to make that condition occur more frequently. Thus in
EAB, if we artifially prevent the "preferred" condition from occurring, the
result should be a decrease in the behavior being maintained by that
condition. In PCT, of course, we would predict an immediate _increase_ in
the behavior as the error increases.

To test that idea, we must of course set up the experiment so an increase
in the behavior could have an increased effect on the controlled variable,
and so we could measure it.

Best,

Bill P.

[From Rick Marken (980317.0710)]

Rick Marken (980316..2025) to Bruce A.

Now you try to find a new pattern 2 that rats will _not_ work
to change pattern 1 into. Lets say you find it's this:

                    2' (altered signaled)
                 * * *

Have you learned what it is about pattern 2 that the subject
was controlling? That's how much your procedure tells you
about what the rat was controlling for in the signaled condition.

Bill Powers (980317.04023 MST) --

I think it's more pertinent to start by admitting that Bruce A. was,
indeed, trying to find out what it was about the signaled condition
that was preferred.

I humbly (but firmly) disagree. I think my point was precisely
pertinent. Bruce Abbott was saying that he was doing a version
of The Test. My post (pertinently) showed that he was not.

Your [Bill Powers (980317.04023 MST)] descriptions of what Bruce A.
was actually doing (from a PCT and reinforcement theory perpective)
were spot on, though! Nice post.

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 Gregory (980317.1047 EST)]

Bill Powers (980317.04023 MST)

What this variant on the experiment does show is that the consequence is
not controlling the behavior. Even though turning on the signaled condition
triples the programmed shock current, the animal continues to prefer the
signaled condition. Under PCT, this implies that in the signaled condition,
the animal does NOT experience three times the shock current; in fact, it
still probably experiences less shock than in the unsignaled conditionl
although not as much less as when the programmed shock currents are equal.

I am surprised to learn that PCT has something to say about
the magnitude of shock experienced by the animal. I would have
thought that the PCT explanation would suggest that the animal
is controlling for something other than minimizing the perceived
shock.

Bruce