From Tom Bourbon [930817.1140]
Earlier [930813.1112], I posted a review of the following
article:
Todd A. Nelson (1993). The hierarchical organization of
behavior: A useful feedback model of self-regulation. Current
Directions in Psychological Research, 2(4): 121-126.
I stopped the review at a point where I said what followed in the
article were examples of weak research Carver and Scheier cite in
support of their version of control theory. I stopped too soon.
(If you do not agree, hit "zap.") More points in their work, as
reported in Don Nelson's review, deserve comment. During the past
few days I re-read several of Carver and Scheier's publications to
assure myself that I have accurate recollections of their work and
that Nelson gives an accurate summary. I do and he does.
The next section in Nelson's article is titled "Attention." He
reports that C&S say attention can be directed either toward the
environment or toward oneself, which they call "self-awareness"
or "self-focus."
"Attention is usually focused on one's environment and only
occasionally shifts momentarily toward oneself (when comparing
perceptions and standards). Thus, only periodically does an
individual note how well his or her behavior matches the relevant
standard for the particular behavior" (p. 124).
COMMENT: Control occurs continuously, not "only occasionally" or
"momentarily." And one's own behaviors, one's actions, are not
often the object of control. If a person does attempt to control
his or her actions, it is the person's own perceptions of the
actions, not the actions themselves, as perceived by an observer,
that are controlled. For people to whom PCT is a new subject,
that distinction usually seems capricious, but it *must* be
understood before the revolutionary nature of PCT can sink in.
To verify my impressions that Carver and Scheier do not
understand, or at least do not adequately address, that
distinction, I re-read an article of theirs that Nelson draws on
heavily:
Charles S. Carver and Michael F. Scheier (1990). Origins and
functions of positive and negative affect: A control-process
view, Psychological Review, 97: 19-35.
There, C&S say:
"This article addresses the nature of certain aspects of
emotion, viewed from a control-theory perspective on behavior.
This perspective focuses on the feedback-based processes through
which people self-regulate their actions to minimize
discrepancies between actual acts and desired or intended acts"
(p. 19).
and
"When people move (physically or psychologically) toward goals,
they manifest the functions of a negative (discrepancy reducing)
feedback loop. That is, people periodically note the qualities
they are expressing in their behavior (an input function). They
compare these perceptions with salient reference values --
whatever goals are temporarily being used to guide behavior (a
comparison process inherent in all feedback systems). If the
comparisons indicate discrepancies between reference value and
present state (i.e., between intended and actual qualities of
behavior), people adjust behavior (the output function) so that
it more closely approximates the reference value" (p. 19).
and
"In terms of human behavior, the exercise of feedback control
means that the person acts to minimize any discernable
discrepancy between current actions and the behavioral reference
value" (p. 19).
COMMENT: All three quotes are from the first page of the
article. C&S wanted to be *certain* that readers could identify
the object of control. They got their point across and they got
it dead wrong. Period. As a general rule, control systems do not
specify and control their actions. In one sentence, C&S say that to
"note" the "qualities" of one's own behavior is "an input function"
and in another that for one to "adjust" behavior is "the output
function." That seems to say pretty clearly that actions are
output and perceptions of actions are input -- and control
systems specify and control their actions. In fact, systems that
act to make their actions match a predetermined plan, in a
variable world, fail exactly like the plan-driven model Bill
Powers and I described in "Models and Their Worlds." Always.
Concerning the object of control, Carver and Scheier reveal a
large gap in their understanding of control systems and what they
do. The gap is large enough and serious enough that it raises
serious doubts about the value of anything they say about control
systems. As if to confirm those doubts, they said (in the second
passage quoted above) people compare present perceptions of their
own behavior against "whatever goals are temporarily being used
to guide behavior." In a control system, goals (reference
signals) do not guide behavior. This is a point Carver and
Scheier, and other well-published "control theorists," seem not to
understand, or if they do they do not make it clear in what
they write. It *does* seem contrary to all traditional "laws" of
behavior, and it *is* exactly that. Reference signals specify
perceptions, not behavior. Behavior, or actions, *must* be
unspecified and free to vary as needed to produce the requested
perceptions.
ยทยทยท
++++++++++++++++++++++++++++
Back to Nelson, but skipping some "minor" points and focusing on
more egregious ones. (Egregious points that originate with
Carver and Scheier and are reported by Nelson; I still hold out
the possibility that he is more messenger than perpetrator.)
Section titled: Affective Consequences of Goal Attainment
"Recently, Carver and Scheier have proposed that the rate at
which discrepancies are reduced is in itself a 'meta-monitoring'
negative-feedback loop and has direct affective implications for
the individual" (p. 124).
COMMENT: The reference here is to the paper on affect that I
cited in the previous section. Carver and Scheier limit their
"model" to three high levels in the PCT hierarchy. They do not
acknowledge or take advantage of the fact that HPCT theorists
include the perception of transitions as a level lower in the
assumed hierarchy. There, the rates of transitions are perceived
and are subject to control. Unaware of that fact, or unconcerned
about it, C&S introduce a description of a new loop. Its general
properties are described in the article (1990) on affect.
"We have characterized people's conscious self-regulation as
a process of monitoring their present actions and comparing the
qualities they perceive therein with the reference values that
presently are salient, making adjustments as necessary to render
discrepancies minimal. In what follows, we will use the term
*monitoring* to refer to this feedback process. As mentioned
earlier, we see this monitoring loop as fundamental to the
control of intentional behavior."
"We suggest, however, that there is a second feedback
process that (in a sense) builds on this one, in a fashion that
is orthogonal to the hierarchical organization discussed earlier.
This second function operates simultaneously with the monitoring
function and in parallel to it, whenever monitoring is going on.
The second feedback system we will term a *meta-monitoring*
function."
" ... we propose that the perceptual input for the meta-
monitoring loop is a representation of the *rate of discrepancy
reduction in the behavioral (monitoring) system over time" (C&S,
1990, p. 22).
COMMENT: There is another occurrence of the idea that a control
system monitors and controls its own behavior. Then there is an
orthogonal-parallel loop-function that monitors rate of reduction
of discrepancies. The "output" of this loop is affect, or felt
emotion. I do not want to imply that within the literature on PCT, or
HPCT, there is a thorough treatment of affect; there is not.
Carver and Scheier are attempting to address a difficult topic
and I certainly do not fault them for that. In some of my own
(primitive) modeling of adaptive control in interacting control
systems, I use loops that monitor the rate of change in error
signals. It would be nice if there were communication and
collaboration between people like Carver and Scheier, on the one
hand, and the community of PCT modelers, on the other. Perhaps
such a collaboration could emerge around the important task of
explaining and modeling emotion in a hierarchical control system.
Among the contributions modelers could make would be a clearer
understanding of the sources of reference perceptions for systems
that specify or monitor affect. There is a need for such
clarity. For example, when C&S discuss the source of reference
values for their new meta-monitoring loop, they say:
"Sometimes the reference value is imposed from outside (as in
tenure review decisions), sometimes it is self-imposed (as in
someone who has a personal timetable for career development), and
sometimes it derives from social comparison (as when people are
in competition with each other)" (C&S, 1990, p. 25).
There is a need for PCT to develop a more complete explanation of
the origins of reference values (signals). There have been many
conversations on this net on the topic of whether, and if so how,
the world, and other people in the world, can affect a person's
reference perceptions. I suspect there will be many more. I do
not think we will conclude that one person can impose a reference
perception on another. Participation in competition, timetables
for career development, and pursuit of tenure all are reference
perceptions people set for themselves. Adoption of certain
higher-level references, especially ones at the program level
that Carver and Scheier emphasize, both creates and limits the
range of options for lower level references. At any level, those
reference perceptions cannot be imposed, unilaterally, by the
environment. It follows that the reference values for affect
cannot be set directly from outside. If not directly from
outside, how? And how would an observer know when one or another
reference for affect had been set, or when it had been changed?
And why do Carver and Scheier often say that there are no
reference perceptions for affect -- that people do not attempt to
control affect? Those are all questions that could be addressed
in a collaboration between researchers with the resources of
Carver and Scheier, and modelers like the people represented on
this net.
+++++++++++++++++++++++++++++++++++++++++++++
C&S seem willing to postulate a new loop for every function or
property they decide *might* be needed in their model. For
example, a little later in their paper on affect they discuss the
"opponent-process model" of emotion, by Solomon. C&S decide
perhaps *that* feature should be added to their model. For them,
that is no problem. They say:
"The opponent process itself implies the existence of a feedback
system beyond the ones on which we are focusing, in that
whichever the direction of the initial emotional response
(positive or negative), the opponent process acts to return the
person to a neutral state" (C&S, 1990, p. 28).
COMMENT: Loops upon loops, but with no reasoned development;
every new loop is offered up ad hoc and untested. Someone's idea
looks interesting, therefore it seems to imply another feedback
system. What is the assumed role of this new (orthogonal-
parallel?) function? C&S say it would return the person to a
neutral state, whatever the initial "sign" of the person's
affect. But wouldn't that be the job of the original
meta-monitoring system? No. As C&S describe *that* system, it
is decidedly "unipolar." Nelson summarized its features.
++++++++++++++++++++++++++++++++++++++++++
"When the discrepancy between the present state and the standard
is reduced at a rate faster than the reference value for rate,
positive affect results. When the rate matches the standard,
neither positive nor negative affect is generated. When the rate
of discrepancy reduction is slower than the individual's
standard, negative affect results" (Nelson, p. 124).
COMMENT: When I am a passenger in a car and we approach the red
traffic light at an intersection at a rate that exceeds *my*
standard, but presumably matches *the driver's* standard, the
driver feels no affect, but I feel positive affect because the
rate of reduction of discrepancy is greater than my reference
value. Is that right? Or when the mortgage company called and
said that the next day they wanted to close the deal on the home
we were buying, instead of in a week and a half as we had
planned, my wife and I felt positive affect, because the rate of
progression toward the closing exceeded the rate we had set as a
standard. Right? And an astronaut or pilot whose rate of
descent close to the ground is greater than the reference rate
feels positive affect. Ideas like this run counter to human
experience.
Elsewhere in their paper, C&S say there are no reference
perceptions or reference values for affect. In other words,
people cannot set references for perceived emotions. That idea,
too, denies much of human experience.
At the level of modeling, when they advance the idea of different
affect for different signs of error, C&S neglect the fact that
discrepancy is discrepancy, no matter what the sign, and a
control system will act to eliminate it. "A rate greater than
the specified rate" is as much an error as "a rate lower than the
specified rate." In each of the examples I gave two paragraphs
above, perceived rates exceeded reference values for rate and
whenever possible the person(s) would, or did, adjust the
perceived rates of change *downward* to the reference rates.
++++++++++++++++++++++++++++++++++
I could write much more, but if you have read this far, you are
probably as tired of this as I am. Carver and Scheier enjoy
great circulation for their many well-placed publications.
However, they did not achieve fame through offering a rigorous
and accurate treatment of a working model for control; "their
model" does not meet that description. To advance their cause,
Carver an Scheier have introduced flaws in the model, willy
nilly. One more telling example. In the paper on affect, they
mention a person whose ideas they want to incorporate into the
control model. They believe those ideas suggest the need for a
new kind of feedback. Again, that is no problem for Carver and
Scheier. In addition to the three loops already in place --
ordinary negative feedback for "behavior," meta-monitoring
feedback, and opponent-process feedback -- they simply say:
"A negative feedback loop, to which we have limited ourselves
thus far, is a discrepancy *reducing*, or *negating*, loop (thus
the term *negative*). This system has a positively valenced
reference value, a desired goal. This sort of system would be
construed by some people as reward based. A positive feedback
loop, in contrast, is a discrepancy *amplifying* loop (see
DeAngelis, Post, & Travis, 1986, for detail). The reference
value of this system is an undesired quality. Discrepancy
amplifying loops attempt to move the currently perceived value as
far away as possible from the reference value. This sort of
system would be construed by some as punishment based. Deviation
amplifying loops are believed to be less common in naturally
occurring systems because they are unstable. [TB: A fact, not a
belief.] Nevertheless, whenever the motive behind an act is the
desire to prevent a condition from existing, the behavior would
seem to reflect a positive feedback process [TB: Things are not
always what they seem.] (see Carver & Scheier, 1981, pp. 157-165,
Ogilvie, 1987, for examples)."
COMMENT: I wonder if Carver and Scheier, or any of their more
ardent supporters, have seen, and if they have seen if they
understood, the CROWD and E. COLI demonstrations or publications.
If they have seen them, and if they did understand, would they
write such things as this? They are recognized as authorities on
control theory, but in this passage they reveal little
understanding of control systems and of the nature of feedback.
I say that with a feeling of great disappointment, not with
pleasure. When we look at the loop that includes a living system
and its immediate environment, the living system exhibits high-
gain negative-feedback control of certain environmental
variables. Actually, the system controls its own perceptual
signals; the control of the environment that we observers see, or
think we see, is not the systems experience. So far as the
observed system is concerned, what we see happening in the
environment is often an unintended, uncontrolled and perhaps
unperceived consequence -- a side effect.
A reference perception that calls for "see no bear in the wild,
or if you do, see it small" does not require a positive feedback
system; all I do is act to make the perception zero or small.
If my perception of "bear in the wild" is non-zero, I act to
shrink it. The fact that I move away from the bear does not mean
I am amplifying deviation; as a negative-feedback control system,
I am decreasing the perceived size of the bear, which I take to
mean I am increasing the distance.
A reference for being two seconds behind the car ahead of me on
the freeway does not mean I first use a negative feedback system
to bring myself from three seconds back to two seconds back
("reducing the discrepancy"), then a positive feedback system to
drop from one second back to two seconds back ("amplifying the
discrepancy"). Carver and Scheier seem to adopt the observer's
perception of increasing and decreasing distances between the
organism and some feature of its environment, then use that
observer-specific view as evidence that the organism must employ
different control systems to produce changes in either direction.
That is not necessary.
This misconception of the roles of positive and negative error is
similar to the one when Carver and Scheier attributed different
affective states to each direction of error. They are not the
first people to be confused on these points.
On the more general notion that positive feedback might be part
of the phenomenon of control, perhaps certain processes in an
organism incorporate linear processes, and some might even
exhibit positive feedback, although probably over a limited
spatial or temporal range. But if an organism controls a
variable, the sign of the entire loop is negative. For living
systems, positive feedback at the global level of the whole loop
is unstable and is rare to non-existent. If Carver and Scheier
are right in their alternative interpretation of positive
feedback, and PCT modelers are wrong, then for C&S to convince us
all they need to do is some elemental modeling; I will do it for
them, if they will tell us, precisely, *how* positive feedback
fits into a simple PCT model in the place of negative feedback.
In all of the examples I have tried on my own, the model's
behavior "blew up," and became chaotic, in the real sense -- it
looked random, but was structured. But I am willing to listen if
C&S explain to us why positive feedback is essential in places
where we have seen no need for it, or where we "believe it to be
less common" than negative feedback, or where we "believe" it
would make a system unstable.
+++++++++++++++++++++++++++++++++++++++++
Enough.
Until later,
Tom Bourbon
Department of Neurosurgry
University of Texas Medical School-Houston Phone: 713-792-5760
6431 Fannin, Suite 7.138 Fax: 713-794-5084
Houston, TX 77030 USA tbourbon@heart.med.uth.tmc.edu