Controlling for; PCT and cognitive psychology

[From Bill Powers (930428.0815)]

Bruce Nevin (930427.1232) --

A while back (it has been a while) as I was going over
something that I was preparing to send to CSG-L, I made the
following correction:

controlling for the perception of X ==> controlling the
perception of X

Since then, I have noticed this locution a number of times,
even in something that Bill wrote. The reason I thought of it
as an error and corrected it is because it seems to say
"controlling actions for the perception of X," when what we
want to say is that perception and not action is what is

This was something that my students in a 1972-1973 seminar
started saying, and I picked it up. It's used in explaining one's
actions to someone else (or oneself). If you object that I'm
driving too close behind another car, I explain that I'm
controlling for reading the bumper sticker (which says "If you
can read this, you're too close").

To say that you're controlling for some perception is to explain
your actions by describing what they're supposed to accomplish.
First, it means that the actions aren't random even if they
appear so; they're part of a control process. Second, it names
the goal-perception. Third, it explains that the control is not
necessarily being successful -- the actual perception might, at
the moment, be rather far from its intended goal-state. Parallel
constructions might be trying for, aiming for, or striving for.
Controlling for is not the same as controlling, because
controlling implies that the error is very small and control is
working well. "Controlling for" allows for the existence of
visible errors and incomplete counteraction of disturbances. I
was quite pleased with my students for coming up with such a
concise way of referring to a rather sophisticated understanding
of control behavior.


Ken Hacker (930422) --

I think you agree now with my characterization of the S-R
approach, and that it is widespread. Now we have to talk about
the cognitive approach, which also has problems. The cognitive
approach is claimed by those who follow it to be the newest and
most prevalant version of psychology, but that isn't true -- it
only seems true if you read nothing but the cognitivist journals
and assume that there's nothing else going on. In fact, most of
the work in the behavioral and social sciences is still done
under the S-R paradigm, as I have broadly defined it. A good deal
of this actually goes on under the name of cognitive science.
When Greg Williams starts reporting his findings, and as I
continue my reading of the journals and reporting on them, I
think that will become clear.

The cognitivist approach doesn't have all the problems of the S-R
approach, but it shares one major problem: defining behavior.

In your post, you say "Behavioral regularities occur in various
forms. First, there are personal regularities, what we call
habits and routines..."

If habits and routines are behavioral regularities, I assume you
mean that they are _visible_ behavioral regularities: they are
things you can see people doing over and over. The usual
cognitivist interpretation, I presume, is that these observed
regularities are simply reflections of cognitive processes -- in
effect, cognitive commands to produce just those behaviors that
are observed. They are habits and routines primarily because of
habitual and routine cognitive processes.

Underlying this interpretation is a fundamental assumption: if a
governing cognitive process were to repeat itself exactly, the
behavior it produces would repeat exactly. There is also a
functional (or is it structural?) assumption: a cognitive process
has its effects by sending commands to lower systems, ultimately
telling the motor systems what to do, and thus producing the
indirect effects in the environment that we recognize as the
outcome of the cognitive process. So the basic arrangement is

cognition --> motor actions --> behavioral consequences.

PCT denies both the fundamental assumption and the
functional/structural assumption. It says that there can be no
regular chain of processes as diagrammed above. This has nothing
to do with the difference between "emitted behavior" and
"interaction." It has to do with the mechanics of behavior,
whether involved in an interaction or not. This is not even a
PCT-specific observation; it's simply a fact that has not been
taken into account in other theoretical approaches to behavior.
PCT is the only theory that does not have to assume the above

The true state of affairs should be diagrammed like this:

cognition --> motor actions --> behavioral consequences
              / / / / / / / / /
              d i s t u r b a n c e s

Disturbances enter into this output chain at every point,
starting with the muscles themselves and extending throughout the
effects that muscles have on the limbs and environment. Putting a
constant cognitive process into the input of this chain will NOT
lead to consistent behavioral consequences. Conversely, if
consistent (habitual, routine) behavioral consequences are
observed, the correct explanation CANNOT be that they are the
result of consistent (habitual, routine) cognitive processes.

PCT explains how it is that "behavioral consequences" can indeed
appear consistent, habitual, or routine even though there is no
reliable chain of cause and effect extending from the brain
outward into the environment. So at one stroke, PCT points out a
fact of nature that is glaringly omitted from all cognitivist
accounts of behavior and provides a model that can explain the
regularities of behavior even with this basic unpredictability in
the output chain.

In order for a consistent behavioral outcome to be produced, it
is necessary that the commands entering the muscles vary with
every change in environmental disturbances, anywhere along the
causal chain between the brain and the final pattern designated
as a behavior. The changes in action join with the disturbances
in precisely the way needed to create a constant outcome. The
only way to explain this phenomenon is to say that the organism
is perceiving the outcome, and that it is controlling the
perception of the outcome. That is the only known feasible way
for outcomes to be repeatable in a variable environment.

If we observe a repeatable behavioral consequence, as in the

cognition --> motor actions --> behavioral consequences
              / / / / / / / / /
              d i s t u r b a n c e s,

we can immediately conclude that the cognitive output on the left
is NOT repeating. It must be varying in exactly the right way so
that when its effects join with the effects of independent and
unpredictable disturbances, the result is a constant observed
outcome. The logic is perfectly clear. If the cognitive output
did NOT change as disturbances change, the outcome would
necessarily change. The outcome is repeatable enough to be
recognized as a regular behavior; therefore the cognitive output
must SOMEHOW be changing in just the required way.

We must distinguish between cognitive outputs and cognitive
inputs: cognitive actions and cognitive perceptions. It is only
the cognitive perception that is preserved; the cognitive
interpretation of the outcome of action. A certain cognitive
outcome is desired (the reference signal). It is compared with
the cognitive outcome that follows from the current way of
interpreting sensory inputs. If there is a difference, the
cognitive output is varied so as to eliminate the difference. The
cognitive output specifies different goal-perceptions in lower-
level systems, which proceed to provide the new requested
perceptions by varying the goal-perceptions of still lower-level
systems, until we get to motor activities. These adjustments go
on continually and simultaneously at all the levels, including
the cognitive levels, until the cognitive perceptions match the
desired ones.

This leads to a picture of cognitive behavior that is very
different from the one usually assumed. The usual assumption is
that cognition follows laws of rational (or at least rule-driven)
thinking, so that given major and minor premises the cognitive
systems arrive at a prescribed course of action which is then
carried out. We like to think that reasoning drives behavior.

But the PCT view says that the desired cognitive outcome drives
reasoning. The reasoning process is adjusted until the world is
perceived as being in the state that is desired. If one desires
to perceive that one is winning an argument, yet perceives in
fact that one is losing it, the cognitive reasoning processes go
into a mode in which various arguments are put forth which are
perceived as making one's position stronger. An active search for
arguments goes on, with the arguments varying until the final
perception of rightness is achieved. This perception may be in
completely internal terms, or in terms of the other person's
inability to counter an argument.

Of course this process is fraught with possibilities for logical
and other blunders. The primary criterion is not that the
argument itself be free of logical or factual blunders, but that
it lead to the desired perception: ascendancy over the opposing
argument. I think that every honest person will admit that during
an argument, the exact truth of every premise pulled up in
defense of one's position is not necessarily known, and that the
logical structure of all the statements connected by ifs and ands
and ors and therefores has not actually been checked out with
Boolean algebra. The objective is not to be logically and
factually correct, but right in one's own perception or that of
someone else.

This explains many of the phenomena we PCTers have encountered in
dealing with people who adhere to other theoretical frameworks.
The reviews of rejected papers show that arguments are concocted
with a specific objective, not with the idea that the arguments
themselves would stand up under detailed scrutiny. One can only
speculate about the cognitive reference signals involved, or the
modes of perception being applied to the content of the articles.
But the overall goal is clear. And it is obvious how the
cognitive outputs are manipulated to arrive at a perception that
matches the goal.

If this is the normal way that cognitive systems operate, science
itself is obviously an anomalous undertaking. But I've been
talking about cognition as reasoning, and in the HPCT model there
is a higher level of control involved: principles. Principles are
not, to my knowledge, discussed much in cognitivist circles as
things existing in the subjects (although the theorist's own
principles are freely discussed). Science has adopted certain
principles that curb the natural tendency to find a logical way
of arriving at desired conclusions. The reasoning processes are
supposed to be formally laid out in public, so that one can't
arbitrarily adjust them to support a foregone conclusion. One can
argue against them using a particular line of reasoning, but the
reasoning behind the objection must also be laid out publicly.
Furthermore, the premises on which rational argument depends must
themeselves be public, and must be supported by evidence, not
just preference or expediency.

With these principles in control, the lower-level processes of
rule-governed thinking can no longer be adjusted so freely as a
way of proving anything one desires to be true. There are
additional constraints: what one desires to be true must fit with
publicly-accepted kinds of reasoning, and with premises that have
been proven true to everyone's satisfaction. If, then, a
difference still exists between what one wants to be true and
what accepted interpretations show as being true, there are only
a few recourses: change the desire, or associate only with other
people who desire the same thing to be true and do not demand
such strict standards of proof.

So PCT leads to seeing cognitive processes in new ways, many of
which are contrary to common beliefs. With this revised
understanding, I should think that many subjects commonly pursued
in cognitive sciences would take on a different look, even when
interactions are the object of study.

Bill P.