[From Bill Powers (940921.0400 MDT)]
Jeff Vancouver (940913) --
One of Lord's major errors is in placing a decision mechanism
(function) between the comparator and the output function. It is an
attempt on his part to include a DME within control theory. I give him
credit for trying. Choices are made, incorporating that into the
control theory model is paramount on my list of things to do. He took
a stab at it.
My problem with a DME is that I don't know what it is supposed to do.
You present me with a piece of candy and a dime and tell me I can take
one or the other but not both. After I pick up one of them, I am said to
have made a "decision" or a "choice." But that is only the outcome; what
is the process? How did I get from the state of looking at both items to
the state of having picked one of them up? Can you describe that process
without using "decide" or "choose" or other synonyms?
"Decision" and "choice" are words we learned when we were very young.
They are "natural language" terms with no formal meanings -- that is,
meanings derived from any systematic understanding, any model. Like most
such terms, we use them without giving much thought to them; they seem
too obvious to define further. We are just supposed to know what they
mean.
As a parallel problem, consider the term "comparison." We all know that
a control system needs a comparator, a way of comparing the perceived
state of affairs with a reference state. We all know what "compare"
means. But how do we construct a comparator for a computer model of a
control system? If you look at the instructions available in any low-
level programming language for any computer, you will not find an
operation called "compare." Of course we all have a natural-language
meaning for the term; if I show you a lemon and a grapefruit and ask you
which is larger, you compare one size with the other and pick the
larger. But how did you do that? What operation did you perform on the
two sizes that resulted in knowing which was the greater? There is only
one operation you can find in the computer that will create this result,
and that is _subtraction_. You form a perception of size for each one,
subtract one size-measure from the other, and assign the terms "larger"
and "smaller" according to the sign of the result. Are there any other
operations that will qualify as "comparison?" Perhaps there are some
logical operations, but as far as I can see they reduce in the end to
subtracting one measure from another. And I haven't found a case where
this is not true. That doesn't mean that no such case exists; I simply
haven't found it.
So wherever we see the term "compare," we can substitute "subtract"
after a suitable analysis of the situation. Now we have a meaning for
compare that we can put into a model that will run on a computer.
What I am asking with regard to "choose" or "decide" is what operation I
need to program into a computer in order to do these things. If there is
no such operation, then I need a definition of what the operation should
be, whether or not it can run on any existing computer. If there is such
an operation, then we can hope to incorporate choice and decision into a
working model because now we know what operations to perform that will
lead to having made a decision or choice.
The only such operations that I have ever seen proposed are arithmetical
or logical in nature. You consider alternatives, you weigh them against
each other, and you subject them to some rational process that results
in an outcome. But as soon as you lay out all the details in this way,
there is no longer any decision or choice to be made. You consider that
you could spend the dime on something you like better than the candy,
but also that the candy is right here and requires no trip to a store to
buy a dime's worth of something else. You compare the relative effort
involved and let the course of action that involves the least energy
expenditure in comparison with the expected satisfaction take place.
There is no decision to be made, once you have picked the grounds on
which the decision is based. That is, "decision" involves nothing beyond
perceiving the elements of the situation and applying an algorithm that
yields an answer based on the elements and logic or arithmetic.
If that is all that decision means, then we don't need the term decision
except as a convenient shorthand. And we don't need to add anything to
the PCT model, because all the processes and perceptions involved in
making a decision are already part of the model, with a specific level
already assigned to doing the operations that result in decisions. The
only thing that would make us add a separate box labelled "decision"
would be to find something in the process that can't be explained by
applying rules or algorithms to perceptions. Of course if "decision"
means something else, then we have to specify what that something else
is, and we have to add a level to the model capable of doing that sort
of thing.
PCT resulted from spending a lot of time, a very LARGE amount of time,
analyzing common-language terms that are used to describe behavior and
experience. It was during this process that I realized that there were
large numbers of words that I used every day without ever having asked
myself what they are supposed to mean. I used to think of myself as
quite the intellectual; I could argue on any subject with anybody and as
far as I was concerned, win. Words, words, words, and none of the
important ones had any meanings that I had thought through. When I think
now about some of my youthful pontifications, I want to avouid meeting
anyone who knew me then.
It seems to me that Locke, Bandura, Lord, and others in their field rely
almost exclusively on words with no meanings that have ever been thought
through. They throw common-language terms around as if merely saying
them is enough. To pick a central term, what is a "goal?" This word is
used as if everybody knows what it means, but none of the writers who
use this term ever stops to ask what a goal is, how it can have any
effects, why the actions we perform have the result of achieving a goal,
how we know when there is a "goal-discrepancy."
Bandura is very proud of the term "self-efficacy", but what does it
mean? It's a bastard construction to begin with, which leads to
confusion with the forms of "self-starting" and "self-explanatory" and
the like. Does it mean "efficacy of the self in creating actions?"
"Efficacy of actions in producing effects pertaining to the self?" And
what does "efficacy" mean? What does "self" mean? When you hear
sentences constructed with terms like these you get a sense of knowing
what the speaker means, but it is a very vague sense and subject to much
idiosyncratic interpretation. The discourse is composed exclusively of
unanalyzed natural-language terms. As long as you just let the words
flow by, you get a feeling of understanding what the speaker is going on
about, but the moment you pause to examine any part of it, and ask what
it means, the sense of meaning disappears. Self-efficacy is simply a
conjunction of two terms, one sort of referring to the aspect of
experience we call ourselves, and the other sort of meaning being able
to have effects or be effective or accomplish things. It's much like
saying "food nice." Say self. Say efficacy. Now say them together. Get
it?
PCT puts a layer of formal meanings underneath the natural-language
terms in which we talk about behavior and experience. Of course these
are _proposed_ meanings, and are always open to modification. But when
we talk in ways that have no such layer of formal meanings beneath them,
we are just talking; we aren't being theorists. We're conveying meanings
as best we can without having any formal understanding of what terms are
to mean. This is how people naturally converse with each other, how
children talk and how adults talk who have never thought deeply about
what their own words mean to them. It is somewhat miraculous that people
are no worse at grasping what others are talking about, and at conveying
to others what they mean, than they are.
Science can happen only when people get together and say "Now look, we
just have to reach some agreements about what certain terms are going to
mean." When they actually get down to doing this, they find that they
were far less in agreement than they thought they were. They find that
they have to work their way down and down through layers of language
until they can find common experiences so simple that misunderstanding
is next to impossible. "I say that meter reads 3.0." "Well, I say it
reads 3.1." "Move over so you're looking straight down at it." "OK,
it's 3.0." Once that's settled, they can start working out the question
of 3.0 _whats_.
Rick Marken's example of the three blind men and the elephant has
another side to it. The three blind men, if they just rely on words, can
think they are talking about the same thing when they're really talking
about different experiences. The one at the tail says "I feel a
cylindrical appendage -- let's call it a pachyoid." The one at the legs
says, "OK, right, I feel four pachyoids," and the one at the trunk says
"There's another one here." When they write their report, they say that
elephants are made of six pachyoids. And they go their separate ways
thinking they are talking about the same thing just because they are
using the same word.
···
----------------------------------
New subject:
I also thought of a set of psychological experiments that many would
not think useless or dangerous (although McPhail beat me to it with the
Sherif studies). Specifically, the studies that Thurgood Marshall used
to argue the Brown v. Topeka BoE case. They were the studies were the
black children identified white dolls as better than the black dolls.
He used those studies to argue that the black children's self-esteem
(self-concept in today's parlance) was lower than whites.
This illustrates another problem I have with much (not all)
psychological research. What the studies you cite showed was that black
children treated white dolls as being better than black dolls. From
this, it is informally inferred that therefore black children think that
white children are better than black children, and from that we conclude
that black children consider _themselves_ worse than white children, and
therefore that black children have lower self-esteem than white
children.
All that seems reasonable enough to base a decision on in the terms in
which most decisions are actually made, but it's not enough to call it
science. The problem is that _plausibility_ is substituted for _proof_.
If the proposed logical connections were actually valid, then the final
conclusion would be valid. But not otherwise.
Proof in this case would amount to finding some independent way of
determining the black children's degree of self-esteem. This is the only
way to validate this procedure for determining self-esteem. Before this
procedure is applied to a general population, we must establish that
when black children say white dolls are better, they are also
experiencing low self-esteem. We can see which dolls they choose, but
how can we see their self-esteem? What we're trying to establish is that
the choice of dolls, which is easy to measure, is a reliable indication
of self-esteem, which is not so easy to measure. If we could establish
that connection, we could then substitute the easy measure for the hard
measure in testing a general population. So the problem of showing that
the test means something boils down to measuring self-esteem in some
direct and independent way.
But here we are stymied. We can't see the self-esteem directly, unless
we happen to be the particular black child under study, and have the
sophistication to think in such terms. We can't just ask a three-year-
old child how much self-esteem he or she feels; that's a meaningless
term to a three-year-old.
The only recourse is to try to devise a completely different test for
self-esteem. We see how black children behave when getting into lines
with white children. We see how "assertive" they are in competing for
toys with white children. We see what roles they play in pretend games.
But in each case we run up against the same problem: there is no direct
measure of self-esteem available for validating any of the tests. In
each case, we simply assume that the behavior we can observe is
indicative of self-esteem.
So what do we end up with? A series of tests which are validated against
each other, and none of which is actually known to indicate self-esteem.
This thing called self-esteem remains hidden from observation; in fact
we don't know what any of these tests is indicating, if anything.
This problem became evident to me when I began taking psychology courses
in college. And it led me to wonder about something that psychologists
didn't seem to care to think about: did the things that these tests
supposedly measure exist at all? In the above example, I would have
asked, is there really such a thing as self-esteem? Is this something
that people have? Or is it just a way of talking?
Now I think I would say that yes, there is such a thing as self-esteem,
but that term is only a rough indicator of what is involved. To get any
closer to a useful understanding, we have to ask what "self" means, and
what "esteem" means, so we can see some sort of process going on.
Obviously, self-esteem is a perception that can be had only by the
person involved. It is a perception of one's own characteristics. And
this perception, to have any value, must be compared with a reference
level for the each state of those characteristics one has chosen as a
target. If you perceive yourself as less than you want to be in each
relevant respect, then you are probably experiencing what people call
low self-esteem.
But there is another side to this, which becomes visible only when you
use the PCT model. What if your perception of your self matches the
reference level with respect to each variable involved in characterizing
yourself, but you have set your reference levels very low for
characteristics that other people usually set to a high level? Your
perception of self-esteem would then be exactly what you have chosen in
each respect. It would not seem low to you, but others would still judge
you as having low self-esteem. You, as a black child, would go behind
the white children in the line, hand over a toy to a white child without
protest, and play the black victim rather than the white victor in
pretend games. Perceiving yourself doing these things would fit
perfectly with your reference levels for how you should be.
How could such reference levels come to be set so low? Obviously,
because these are the settings that work, all things considered. If you
are raised in a society that punishes you for assuming equality or
ascendance in the company of white people, you can avoid that punishment
by choosing ways of being that do not cause punishment. You remember how
you were when you were getting along as well as possible, and you pick
those memories as reference levels. This does not require any thought;
all it requires is reorganizing until the pain is minimized.
So now we are beginning to get a handle on self-esteem. We can
distinguish between a person who perceives a self that is less than what
is wanted in various dimensions, and a person who has come to want less
in these same dimensions. And while we're at it, we have arrived at some
definitions of the dimensions that pertain to self-perception. With this
understanding, we're ready to apply control theory to the problem of
determining a person's self-esteem.
The procedure goes this way. You pick a person, and start interacting
with that person. What you're looking for are controlled variables that
pertain to self-perception, as you define it. You see how this person
reacts to disturbances of the variables you have chosen; if there is no
reaction, you strike that variable from the list. Eventually, after a
lot of interaction, you arrive at definitions of variables relevant to
self-perception, as you see it, that you can demonstrate to be under
control by this person. And you also find the reference levels relative
to which each variable is controlled, so you can tell if low self-esteem
is due to a large error or to a low setting of the reference levels.
Then you pick a second person and go through this procedure again. And a
third, and a fourth, until you have done all the people in your test
group, one at a time. By the end of this time you will know 50 or 100 or
200 people very well indeed, with respect to what you see as their self-
concepts.
Then you can start looking for common factors, for variables that
indicate common problems, for deviations of individuals' reference
levels from those of most of the group. You look for indirect indicators
that show highly reliable correlations with either felt lack of esteem
or low levels of aspiration for esteem. Now you have a direct measure of
self-esteem, as you have carefully defined it, for a large enough group
of individuals to allow validating other kinds of tests that are easier
to apply. Of course these other tests will NEVER yield the kind of
reliability that the basic study of one person at a time will yield, and
no important decisions should be made about any one person on the basis
of such tests. But such tests can be used as the basis for modifying
policies that apply to many people or for assessing the overall
effectiveness of a program, a teacher, or an administrator.
Compare where we started with where we ended. We began with children
playing with dolls, and we ended with a study of individuals aimed at
discovering how their self-concepts compare with their desires, and what
those desires are. We started with a vague common-language term, self-
esteem, and ended by talking about specific variables associated with
self-perception. Which approach is going to yield solid knowledge of how
people perceive themselves, how children learn to perceive themselves? I
submit that the answer is obvious.
It's much, much easier to hand out a set of questionnaires to 30 people
than it is to sit down with each one of them for hours and hours,
learning how each person ticks. There are pressures of time and
competition and pride and funding that make psychologists look for quick
and easy ways to get information about important aspects of human
nature. But the results reflect the amount of effort and thought
expended. If anyone wants to conduct psychological investigations in a
way that will yield reliable knowledge about human nature, there is
really only one way to start: one person at a time. Without that kind of
basic understanding, all generalizations are fuzzy and empty and hardly
worth the trouble.
-----------------------------------------------------------------------
Best,
Bill P.
