Fishy Test; Medicine & PCT; 23-level diagram

[From Bill Powers (920709.0700)]

Martin Taylor and Bill Silvert (920708) --

The spillover from this little squabble seems to indicate the idea of
looking for control processes in large natural phenomena, using The Test.

I remind one and all that there is more to the Test than looking for low or
high correlations of various sorts. The idea is to track down an actual
control system. Once correlations or other measures have suggested that a
control system might be reponsible for an unexpectedly low or unusually
high correlation or for unnatural stability of some observable variable, it
is incumbent on the investigator to go on to identify the system in
question. An active system must be found producing outputs that
specifically oppose disturbances, and it must be shown that the controlled
variable is in fact sensed by the system, loss of the sensory link
destroying control.

If you find that the parrot clings to the perch only because its feet are
nailed to the wooden bar, I would advise not buying it as a live working
control system. To call a system a control system should be, I think, a
last resort after all other possibilities have been ruled out as being
highly improbable, technically impracticable, or factually false. And, of
course, after all the necessary aspects of control have been demonstrated
to exist. There will always be cases in which the only defensible
conclusion is "I don't know."


Francisco Arocha (920708;15:22) --

I think you have a nice idea there: what are doctors controlling for? This
could turn out to be a rather larger project than you may now envision. It
could also turn out to be a study of extraordinary importance.

The approach as you outlined it makes the assumption that doctors are
controlling primarily for patients in a good state of health, and that they
will pick treatments that do in fact help. If this is true, you should find
that disturbances of the state of health that make it worse result in
increased treatment activity or more drastic treatments, and disturbances
that improve it result in decreased amounts of treatment or milder
treatments. A simple study of this kind would make a nice demonstration

You should also find that the doctor's beliefs about the efficicacy of
treatments corresponds to the treatment used, as appropriate to the
hypothetical error signal. A treatment that the doctor does not perceive as
effective would tend not to be used, regardless of the statistical data
concerning that treatment.

The perceptual side of the control process can be investigated in terms of
diagnostic procedures used by the doctor and tests that the doctor orders.
"How do you know that this patient is ill, the kind of illness, and the
severity of the illness?"

Other hypotheses about other possible controlled variables can also be
investigated. For example, many doctors are suspected of preferentiually
employing treatments or tests that they have special expertise in, or that
use medical facilities in which the doctor has a financial investment,
quite independently of what is wrong with the patient. The costs of various
treatments (in terms of how much money the doctor can make by using them)
should also come under scrutiny. Doctors may be controlling for perceptions
of specific causes of illness, or for income, or for a chance to exercise
special skills -- in addition, one presumes, to controlling for the
restoration of health.

There is also the question of the efficacy of treatment. Large parts of
modern medical practice consist of reading brochures put out by drug
companies and trying out the latest magic bullet. There are many drugs
whose positive effects are found in only a small minority of patients, yet
because of statistical analyses they are considered "effective." Therefore
when many drugs are used, the result actually to be expected is negative,
most such drugs having adverse side-effects and only some of them
benefiting a clear majority of patients.

An important question, therefore, is how doctors explain failure of a drug
(or other) treatment to have the expected effect. If the doctor perceives
the treatment as effective, and it fails to improve the patient's health,
does the doctor try to correct the error by increasing the amount of the
same treatment, or does he/she change the reference level for using the
treatment? Is the failure blamed on invisible disturbances that were larger
than anticipated, or on a treatment that is less effective than supposed?
In short, does the doctor control for a continued perception of medical
practice as being effective, or for improving the condition of the
patients, one by one, or for not doing harm (a principle that at least some
doctors profess to hold)?

Finally, the way in which data about many participants in this experiment
are combined is of utmost importance. A complete control analysis for each
participant has to be done BEFORE the data are combined. If you subtract
the average perception of drug efficacy from the average reference signal
for efficacy, you will get a average error signal that doesn't represent
the error experienced by ANY individual. What must be done first is to look
at the relationships among reference signal, perception, disturbance, and
action for each individual; what should be found is that for all
individuals, action opposes disturbance and perception is brought near to
the reference signal. The possibility of bimodal or multimodal measures
should be kept strongly in mind, because there will be large individual
differences in reference settings and perceptions. The relationship between
disturbance and action should be the same for all those in whom you have
identified a true controlled variable. But the reference settings can be
different in every individual.

There is a lot about medical practice that you can discover through PCT
analysis. There is a lot that NEEDS to be discovered.
Dag Forssell --

I agree that you, I, and Bruce Nevin see the "23-level" diagram in the same
way. But I also think that Martin Taylor does, too. Martin and I often go
around and around in apparent disagreement, only to find that we are really
talking about the same thing but in different words. As you say,
establishing agreement is actually harder than establishing disagreement.

I think that often the parties to a disagreement, in explaining their views
and arguing against others, experience a gradual shift in their perceptions
and end up agreeing to a consensus position that does not exactly match
what any of them began with. There may still be residual disagreement
between the internally-held positions, but it no longer shows up in words.
All we can do is keep iterating this process, in the hopes that all actual
positions are moving in some meaningful direction.


Bill (back on the air) P.