[From Bill Powers (971122.1321 MST)]
Bruce Abbott (971122.1335 EST)--
I agree that the independent-dependent variable approach works, but that is
not why engineers get the 100% correlations and do not need statistics.
Engineers get the 100% correlations and do not need statistics because (a)
they work with simple systems involving only a few influential variables,
(b) they can measure those variables with high accuracy and precision, (c)
extraneous variables that could upset the measured relationships can be
rigidly controlled, or at least measured, (d) the independent variables can
be manipulated over a wide enough range that their effects on the dependent
measures can be readily distinguished from background fluctuations resulting
from extraneous variables that cannot be held rigidly constant, (e) the
machine being tested isn't itself undergoing continuous, unmeasured (and
perhaps unmeasurable) change in its internal organization while it is being
tested. That was my point.
I think you are quite wrong about why engineers get 100% correlations, or
close to them. They get them because they will accept nothing less, not
because the problems they study are easier. The data and analyses in
engineering are actually far more complex than those any psychologist tries
to handle.
The problem with your asessment is that one can draw only one conclusion
from it: there's no hope of a systematic approach that will yield close to
perfect correlations, so why try to get them? The human system is so huge,
so complex, so variable, so unpredictable, so subject to unknowable
external influences, that there is hardly any order there to detect in the
first place. Statistical analysis is our only hope.
I think that is just a giant rationalization, meant to account for the
failure of psychology to discover any solid facts about how human beings or
other animals work. The very last thing a psychologist wants to conclude is
that the _approach_ was wrong, that the _theories_ were wrong, that the
whole concept of what kind of system he or she thought was being studied
was wrong from the very start. Instead, the blame goes on the subject
matter. We're not using the wrong approach; the subject matter itself is
just very difficult.
Well, I say hogwash. Human behavior is highly precise, repeatable, and
understandable. There's just a lot of it to understand. If you don't see
order in it, you're looking at it the wrong way. What's the probability
that a carpenter building a house from a blueprint will end up with a house
that matches the blueprint within a percent or so? About 1.00. What's the
probability that a bridge built by an engineer will hold up for 30 years
under the traffic it's designed to carry? About 1.000000. What's the
probability that a driver going on a 20-mile trip to work will fail to get
there in good shape? Tom Bourbon looked into the figures, and came up with
something like one part in ten million. Human behavior is chock full of
regularities, extreme regularities. Anyone who misses them is obviously on
the wrong track.
One thing engineers know that psychologists don't seem to know is that you
can't handle complex problems until you're sure you've solved the simple
ones first. Most of the things that psychologists try to study are poorly
understood and poorly analyzed because there are no basics on which to
build. The popular, flashy, dramatic, and lucrative questions are asked of
nature long before anyone is in a position even to understand what the
questions really are. As a result, the data are messy and contradictory,
nobody really tries to replicate an experiment in toto, low correlations
are not only expected but published, and the average number of citations of
published articles is 1, with that citation most likely being by the
author. The average lifetime of a theory in psychology is about the same as
that of a butterfly.
The basic concept behind PCT is not really control theory. It's the idea
that if we approach human behavior in the right way, with the right models,
we will be able to understand it as well as we understand anything -- as
well, for example, as a physicist understands matter. If we adopt this
attitude toward the study of behavior, we will have no reason to hang on to
theories that explain what happens only 80% or the time, or 90%, or 99%.
What we're after are explanations that work all the time, under all
circumstances, with no exceptions or time-outs, as close as we can measure.
What that means is that we're not going to jump into the subject with the
idea of solving big important problems from the start. We're going to look
for problems simple enough to be handled with what we know or can work out.
If the results we get are puzzling and equivocal, we're at too high a level
of complexity; we have to get simpler yet. At some level, even if it's only
the level of watching people wiggle sticks, we're going to find a problem
we can solve with engineering precision. When we can do that, we can think
about tackling something slightly more complex, like two people wiggling
two sticks.
That's how engineers do it, and that's how psychologists need to do it if
they hope to find out anything about human nature worth a pig's -- foot.
The opinion is frequently expressed in this forum that psychologists either
remain ignorant of or refuse to accept control theory. Often it is said
that control theory is not accepted by psychologists because such acceptance
would entail scrapping everything learned to date in psychological research,
because it would threaten careers, and all sorts of other nonsense. What
this argument overlooks is that control theory has been repeatedly
introduced in various contexts within psychology since the early 1950s. The
perception unfortunately has been that where it has been tried it has
failed, except for certain limited cases. Bill Powers has had to make
nearly a career out of objecting to these conclusions, which have often been
based on misconceptions (such as the idea that open loop systems are faster
than closed loop systems). Deserved or not, the perception that control
theory had not turned out to be the panacea that it promised to be is
probably as strong a reason as any that it is not main stream psychology
today.
I claim the reason is that psychologists can't tell a theory that works
from one that doesn't work. This is not part of their training, because the
data they take and the analyses they are taught are all aimed at fishing a
little signal out of a sea of noise. They have no experience with theories
that predict correctly every time. They're taught that no such theories are
possible. They're taught that a theory that predicts correctly 6 times out
of 10 is worth paying attention to. In short, psychology students are given
standards of scientific achievement that put them below the poverty level.
Our PCT experiments pass right over their heads: they don't know what
they're looking at.
I suppose you're going to tell me -- again -- that not all theories have to
work all of the time; that lesser theories have their uses, especially when
we have nothing better. And I will reply as before: that's the attitude
that will keep psychology from ever becoming a science.
Best,
Bill P.