[SPAM] Re: What's wrong with schizophrenics? (was: Control system description)

[From David Goldstein (2009.05.08.06:02 EDT)]
[About Bill Powers (2009.05.07.0-631 MDT)]

BP: The descriptions of problems above are too general to allow relating
them to PCT concepts.

DG: Agreed. I was trying to start from a commonly known viewpoint, instead
of a test viewpoint.
You glossed over the study that Psychotherapy can be helpful for the more
practical problems
that people with this diagnosis are experiencing. This is an important
point.

BP: I see correlations of the kind shown in the second paper as
indicating failure to find anything of interest.

DG: I knew you were going to say this. This is where you lose a lot of
applied Psychologist types.
They are not that impressed with any demonstrations in a research setting.
The best test that
Psychologists have is the IQ test and this correlates .60 with reading skill
level at the
end of first grade. Correlations of .90 or higher are usually associated
with test-retest situations. That is,
a person takes the test twice at different times.

BP: I would be utterly astonished to find that the brain of a person who
shows the symptoms associated with a diagnosis of schizophrenia
worked no differently from the brain of a normal person.However, I would
be uncertain of the cause of the difference: is it a lesion or
a brain chemistry defect, or is it that the programs running in the brains
are different? Is it a hardware problem, or a software
problem?

DG: Your summary is an overgeneralization. The differences were found in the
frontal lobes. Is one study required to answer all the
questions one could ask? Also, while most people would agree that the brains
of people with this diagnosis must be different, this
intuition does not answer the question of how are they different. I am
attaching another study which goes a little further in
saying how they are different. Interestingly, the study uses a tracking
study.

BP: How many formal tests do you know of that assess conflict?

DG: There are a few, but the definition of conflict is different than the
one you put forth.

315.pdf (182 KB)

[From Bill Powers (2009.05.08.1346 MDT)]

David Goldstein (2009.05.08.06:02 EDT) --

You glossed over the study that Psychotherapy can be helpful for the more practical problems that people with this diagnosis are experiencing. This is an important point.

Not very. If you had said that studies of this kind are highly likely to predict the results of psychotherapy accurately for every person needing help, say with a 90% probability of correct prediction for each case, I would be impressed. But you haven't said that, nor does anything in the papers you attached promise anything like that kind of result.

Richard Kennaway did some analyses that showed what kinds of correlations are needed to enable us to make predictions with a reasonable percentage of correctness. His results were not supportive of the idea that many published correlations are useful. And as far as I can see, that failure to support the way things are done now is the ONLY reason that his findings continue to be totally ignored, everyone going right on as if he had never said a thing. You are arguing as if the correlations cited in the papers give us useful information, when it has been incontrovertibly proven that they do not. There is a severe epidemic of denial going on here.

This would be a good time to ask Richard once again to review his findings. I say that anyone who doesn't accept them has the responsibility for showing what is wrong with them, and failing that, must accept them and their consequences. As I do.

BP: I see correlations of the kind shown in the second paper as
indicating failure to find anything of interest.

DG: I knew you were going to say this. This is where you lose a lot of applied Psychologist types.

BP: You don't understand, David. That is where THEY lose ME. Is majority rule all that matters in science?

DG: They are not that impressed with any demonstrations in a research setting. The best test that Psychologists have is the IQ test and this correlates .60 with reading skill level at the end of first grade.

BP: I keep asking some form of this question and you keep not answering it. Given that there is a 0.60 correlation between IQ and reading skill level, what are the chances that the IQ of the next child you test will predict correctly the quintile of reading skill level of that child after first grade? There should be plenty of data available. I'm only asking for a prediction within 20%. I think you don't ever answer this question because you know how low the probability of correctness would turn out to be, and you don't want to say it out loud. There, I've made it into a direct challenge. If you dodge it again, I can claim that I've made my point.

BP earlier: I would be utterly astonished to find that the brain of a person who shows the symptoms associated with a diagnosis of schizophrenia
worked no differently from the brain of a normal person.However, I would be uncertain of the cause of the difference: is it a lesion or
a brain chemistry defect, or is it that the programs running in the brains are different? Is it a hardware problem, or a software
problem?

DG: Your summary is an overgeneralization. The differences were found in the frontal lobes. Is one study required to answer all the questions one could ask?

BP:Come on. You can ask whether it's a hardware or a software problem about the frontal lobe as well as any other part of the brain. And that is a very basic question, which determines the kind of treatment that will be preferred. You can't cure software problems with drugs or surgery.

Also, while most people would agree that the brains of people with this diagnosis must be different, this intuition does not answer the question of how are they different. I am attaching another study which goes a little further in saying how they are different. Interestingly, the study uses a tracking study.

Pursuit tracking by the eyes, which is not exactly relevant to the frontal lobe. Most people can do visual pursuit tracking (the eyes smoothly following a target rather than jumping from one position to the next in saccades). Apparently schizophrenics do it more than control groups do, which I've heard before. How much more, I couldn't say since they used group differences rather than individual, implying that the differences were very small (otherwise 14 subjects would not have been needed to detect differences described as "subtle").

Sometimes I feel as if I'm trying to push a door open that's not just locked, but isn't even a door: it's a picture of a door painted onto a brick wall.

Best,

Bill P.

[From: David Goldstein (2009.05.08.06:26 EDT)]
[About Bill Powers (2009.05.08.1346 MDT)]

BP:Not very. If you had said that studies of this kind are highly likely
to predict the results of psychotherapy accurately for every person
needing help, say with a 90% probability of correct prediction for
each case, I would be impressed.

DG: You don't understand. Medication has been viewed as the only way to
help a person with the diagnosis of Schizophrenia. However, medication
does not help a person acquire the practical skills of everyday living which
the person has failed to acquire. Adding some other interventions which specifically
teach these skills adds some benefit, at least in some cases.

In your case, you seem to be willing to undergo the rehabilitation therapy for
improving your breathing functioning. Have you been given a 90% probablity that
it will be successful? If not, why are you doing it? If yes, have you seen the
data?

DG: Richard Kennaway's paper is very interesting and persuasive, as far as it goes. However, it does not address the
possibility of predicting a variable from a set of variables, as in multiple, step-wise
regression. This is what most psychological studies of the correlational type do. Each
subsequent variable entered into the regression equation adds something that the previous
variable did not. This applies to the case of predicting end of first grade reading achievement from
IQ and other variables (for example measures of phonemic awareness).

You seem to be stuck on the idea that one variable must do everything.

What say you Richard Kenneway? Have you gotten this comment from people more expert in statistical matters than I am?

DG: I don't think that you have not given the study "Neurobiology of Smooth Pursuit Eye Movement Deficits in Schizophrenia: An fMRI Study" your
usual thoughtful consideration. I will come back to it in another post, maybe after you have had a second look.

DG: Takes a 'painted door on a wall' to know another one. (smile, smile)

···

----- Original Message ----- From: "Bill Powers" <powers_w@FRONTIER.NET>
To: <CSGNET@LISTSERV.ILLINOIS.EDU>
Sent: Friday, May 08, 2009 4:37 PM
Subject: Re: [SPAM] Re: What's wrong with schizophrenics? (was: Control system description)

[From Bill Powers (2009.05.08.1346 MDT)]

David Goldstein (2009.05.08.06:02 EDT) --

You glossed over the study that Psychotherapy can be helpful for the more practical problems that people with this diagnosis are experiencing. This is an important point.

Not very. If you had said that studies of this kind are highly likely to predict the results of psychotherapy accurately for every person needing help, say with a 90% probability of correct prediction for each case, I would be impressed. But you haven't said that, nor does anything in the papers you attached promise anything like that kind of result.

Richard Kennaway did some analyses that showed what kinds of correlations are needed to enable us to make predictions with a reasonable percentage of correctness. His results were not supportive of the idea that many published correlations are useful. And as far as I can see, that failure to support the way things are done now is the ONLY reason that his findings continue to be totally ignored, everyone going right on as if he had never said a thing. You are arguing as if the correlations cited in the papers give us useful information, when it has been incontrovertibly proven that they do not. There is a severe epidemic of denial going on here.

This would be a good time to ask Richard once again to review his findings. I say that anyone who doesn't accept them has the responsibility for showing what is wrong with them, and failing that, must accept them and their consequences. As I do.

BP: I see correlations of the kind shown in the second paper as
indicating failure to find anything of interest.

DG: I knew you were going to say this. This is where you lose a lot of applied Psychologist types.

BP: You don't understand, David. That is where THEY lose ME. Is majority rule all that matters in science?

DG: They are not that impressed with any demonstrations in a research setting. The best test that Psychologists have is the IQ test and this correlates .60 with reading skill level at the end of first grade.

BP: I keep asking some form of this question and you keep not answering it. Given that there is a 0.60 correlation between IQ and reading skill level, what are the chances that the IQ of the next child you test will predict correctly the quintile of reading skill level of that child after first grade? There should be plenty of data available. I'm only asking for a prediction within 20%. I think you don't ever answer this question because you know how low the probability of correctness would turn out to be, and you don't want to say it out loud. There, I've made it into a direct challenge. If you dodge it again, I can claim that I've made my point.

BP earlier: I would be utterly astonished to find that the brain of a person who shows the symptoms associated with a diagnosis of schizophrenia
worked no differently from the brain of a normal person.However, I would be uncertain of the cause of the difference: is it a lesion or
a brain chemistry defect, or is it that the programs running in the brains are different? Is it a hardware problem, or a software
problem?

DG: Your summary is an overgeneralization. The differences were found in the frontal lobes. Is one study required to answer all the questions one could ask?

BP:Come on. You can ask whether it's a hardware or a software problem about the frontal lobe as well as any other part of the brain. And that is a very basic question, which determines the kind of treatment that will be preferred. You can't cure software problems with drugs or surgery.

Also, while most people would agree that the brains of people with this diagnosis must be different, this intuition does not answer the question of how are they different. I am attaching another study which goes a little further in saying how they are different. Interestingly, the study uses a tracking study.

Pursuit tracking by the eyes, which is not exactly relevant to the frontal lobe. Most people can do visual pursuit tracking (the eyes smoothly following a target rather than jumping from one position to the next in saccades). Apparently schizophrenics do it more than control groups do, which I've heard before. How much more, I couldn't say since they used group differences rather than individual, implying that the differences were very small (otherwise 14 subjects would not have been needed to detect differences described as "subtle").

Sometimes I feel as if I'm trying to push a door open that's not just locked, but isn't even a door: it's a picture of a door painted onto a brick wall.

Best,

Bill P.

[From Bill Powers (2009.05.09.0445 MDT)]

David Goldstein (2009.05.08.06:26 EDT) –

BP earlier:…If you had said
that studies of this kind are highly likely to predict the results of
psychotherapy accurately for every person

needing help, say with a 90% probability of correct prediction for

each case, I would be impressed.

DG: You don’t understand. Medication has been viewed as the only way
to

help a person with the diagnosis of Schizophrenia. However,
medication

does not help a person acquire the practical skills of everyday living
which

the person has failed to acquire. Adding some other interventions which
specifically teach these skills adds some benefit, at least in some
cases.

BP: What did your paragraph have to do with my paragraph that you
cited? Mine was about predicting results of psychotherapy, yours
was about medication and teaching skills.

DG:In your case, you seem to be
willing to undergo the rehabilitation therapy for improving your
breathing functioning. Have you been given a 90% probablity that it will
be successful? If not, why are you doing it? If yes, have you seen the
data?

BP: I had enquired of two doctors whether exercise would help. Both of
them said no, COPD is a progressive disease and will not get better.
Finally I was referred to a specialist, who recommended the
rehabilitation approach, which involves exercise. I think the exercise
will help me – I did better than I thought I would. There was even a
little biofeedback: the woman overseeing my efforts on the machine
propped up a pulse oximeter so I could see it, and told me not to let the
reading get below 90. That was a big help.

I never said I had to have 90% reliability of a method before I would try
it. When nothing else is available you use what’s there, if it doesn’t
send you in the wrong direction. I’d say the same about being treated by
a witch doctor. In fact I have a distinct feeling of being treated by
witch doctors. So I don’t expect too much.

DG: Richard Kennaway’s paper is
very interesting and persuasive, as far as it goes. However, it does not
address the possibility of predicting a variable from a set of
variables, as in multiple, step-wise regression. This is what most
psychological studies of the correlational type do. Each subsequent
variable entered into the regression equation adds something that the
previous variable did not. This applies to the case of predicting end of
first grade reading achievement from IQ and other variables (for example
measures of phonemic awareness).

BP: If that is the rationale behind most psychological studies, I think
you may have put your finger on what is wrong with them. While what you
say sounds plausible, it looks to me like a total fallacy.

You sent me a list of 16 items which are used to judge the presence of
schizophrenia. Each item was a general term followed by a set of other
terms that elaborated on the category. These items struck me as a random
collection of informal terms referring to bad things people could
feel and do. Each one could arise for any number of different reasons.
Some sets of conditions seemed to go naturally with other sets (anxiety
and tension) but others seemed unlikely to go together (grandiosity and
blunted affect). As a basis for identifying a person’s underlying
problems, they looked useless to me.

Yet they can be used, and that is the problem. They can be used to create
totally artificial categories that have no existence except in the
context of a particular way of analyzing data. A good example of this is
found in the pdf file you sent me titled "Neurobiology of Smooth
Pursuit Eye Movement Deficits

in Schizophrenia: An fMRI Study" (to which Gary Cziko also referred,
and is in Science for May 1). In this paper 28 subjects performed
a pursuit eye movement task, with 14 being classed as schizophrenic and
14 being “healthy.” The problem can be seen in this plot of the
results:

755877.jpg

In the left panel, 5 schizophrenic subjects (red) showed
performance in the range of the normal subjects’ performance, 8 did not.
In the right figure, 6 schizophrenics showed normal behavior, 7 did not,
2 just barely (there seems to be one missing schizophrenic).

Yet read the caption.“Subjects with schizophrenia had significantly
lower gain,” it says. It does not mention that almost half of the
schizophrenic subjects performed just like the normals. In fact,
schizophrenics who who showed higher scores than almost all the normals
in the left panel are referred to as having lower scores than the
normals, because the group mean for the schizophrenics is lower than the
group mean for the normals.

Something mysterious has gone wrong with the thinking here. Somehow a
schizophrenic who does better than most of the normals is affected by
other schizophrenics who do worse, so that his superiority is erased by
the fact that he has been classed as schizophrenic. He has been
“clustered” with his fellow-schizophrenics, and from now on
will be judged not by his actual score, but by the average score of the
group with whom he is now irrevocably associated. “Schizophrenics
have lower gain” has become a fact, in exactly the same way it once
became a fact that “Jews are pushy.” This is the sort of fact I
simply reject; it is a pseudo-fact. Psychology, it often seems to me,
consists almost entirely of pseudo-facts like this.

What the data for schizophrenics leads me to speculate is that we have
here a group of 13 people with different problems, selected because to
some psychologist they all looked sort of similar in their behavior. But
each person had something different wrong with him, and these various
problems had different indirect effects on the gain or other measures
that would be observed in a visual pursuit task. Other than that, the
problems had nothing else in common. However, it is possible to calculate
the mean gain of the pursuit model over all subjects, and while
this mean has no relevance to any of them, it characterises the
group.

The basic error here is to assume that because the mean is calculated by
using all members of the group, it is a measure of each of them. Because
a functional cluster is defined by all the people who lie within its
boundaries, its characteristics pertain to each individual in the
cluster. And so on for all the cases in which multiple measures are used
to refine some group characteristic, a characteristic which is then
turned around to be used as if it applies to each individual in the
group.

DG: I don’t think that you have
not given the study “Neurobiology of Smooth Pursuit Eye Movement
Deficits in Schizophrenia: An fMRI Study” your

usual thoughtful consideration. I will come back to it in another post,
maybe after you have had a second look.

BP: Was this time around more like what you wanted? (he asked
slyly).

I think the moral is clear. We have to treat individual cases
individually. Creating groups, analyzing the groups, and then applying
the result back to the individuals is not only cumbersome and
time-consuming, it is an offense against reason. The only time it is
justified is when the groups have been selected so wisely that
within-group measures are closely similar, and differ radically from
between-group measures – in other words, when the differences are
obvious to the naked eye and statistical analysis is not needed to see
them.

I think it is possible that the best approach to schizophrenia would be
to abolish the category and simply take each patient as a new person to
be investigated as if no other person had ever been observed. That is the
basic theme of the method of levels: every session stands alone, every
person stands alone. We do not see people through the filter of what we
supposedly know about “people.” We do not look for similarities
between cases.

We do not diagnose. We follow the trail where it leads and hope to be
surprised.

I’m reminded of the days, at least a decade ago, when I was pleading for
people to try MOL all by itself, rather than tacking it onto some
existing approach. Tim Carey committed to that first; you, David, did
that later with some patients. Now I would like to see what happens with
a straight MOL approach to schizophrenics, preferable without the
knowledge of the therapist that they have been so classified. How else
can we find out if anything else is required?

Best,

Bill P.

[From Dick Robertson, 2009.05.09.1658CDT]

David,

About Table one in the Granholm et al. paper: I noticed that the randomization process resulted in almost 50% more Ss with halucinations in the Usual Treatment group than in the “enhanced” Treatment group, and exaactly 50% more delusions in the Usual Treatment Group.

Given that these conditions often pose handicaps in sustaining contacts with other people and performing some of the other activities that were used as measures in this study–might we not wonder whether the same groups with no treatment at all might still have shown that difference if the outcome measures were taken on them after the same period of time?

Maybe a quibble, but doesn’t it raise even moe problems for interpreting the very small statistical differences in the outcome measures?

Best,

Dick R

···

----- Original Message -----
From: davidmg davidmg@VERIZON.NET
Date: Saturday, May 9, 2009 6:15 am
Subject: Re: [SPAM] Re: What’s wrong with schizophrenics? (was: Control system description)
To: CSGNET@LISTSERV.ILLINOIS.EDU

[From David Goldstein (2009.05.09.06:57 EDT)]
[About Bill Powers (2009.05.08.1346 MDT)]

BP: Given that there is a 0.60 correlation between IQ and
reading skill
level, what are the chances that the IQ of the next child you
test will
predict correctly the quintile of reading skill level of that
child after
first grade?

DG: Attached is the best answer I can come up with for now. It
is based on a
statistics program called ZumaStat. See ‘For Bill.pdf’

BP: Not very. If you had said that studies of this kind are
highly likely
to predict the results of psychotherapy accurately for every person
needing help, say with a 90% probability of correct
prediction for
each case, I would be impressed. But you haven’t said
that, nor does
anything in the papers you attached promise anything like
that kind of
result.

DG: Here is the actual study that had been summarized. I can
provide the
effect size using ZumaStat if you want.
It was helpful for some of the patients in some of the measures
used. Some
help is better than no help. These
are very difficult people to work with successfully. See ‘520.pdf’.

[From David Goldstein (2009.05.09.23:45 EDT)]

[About Dick Robertson, 2009.05.09.1658CDT]

Your right. The presence of more people with these symptoms in the Usual Treatment Group might mean that the randomization process did not

result in two groups that were equivalent.

They probably should have rank ordered the subjects according to these symptoms and then randomly assigned

each pair to one or the other group.

David

···

----- Original Message -----

From:
Robertson Richard

To: CSGNET@LISTSERV.ILLINOIS.EDU

Sent: Saturday, May 09, 2009 6:10 PM

Subject: Re: [SPAM] Re: What’s wrong with schizophrenics? (was: Control system description)

[From Dick Robertson, 2009.05.09.1658CDT]

David,

About Table one in the Granholm et al. paper: I noticed that the randomization process resulted in almost 50% more Ss with halucinations in the Usual Treatment group than in the “enhanced” Treatment group, and exaactly 50% more delusions in the Usual Treatment Group.

Given that these conditions often pose handicaps in sustaining contacts with other people and performing some of the other activities that were used as measures in this study–might we not wonder whether the same groups with no treatment at all might still have shown that difference if the outcome measures were taken on them after the same period of time?

Maybe a quibble, but doesn’t it raise even moe problems for interpreting the very small statistical differences in the outcome measures?

Best,

Dick R

----- Original Message -----
From: davidmg davidmg@VERIZON.NET
Date: Saturday, May 9, 2009 6:15 am
Subject: Re: [SPAM] Re: What’s wrong with schizophrenics? (was: Control system description)
To: CSGNET@LISTSERV.ILLINOIS.EDU

[From David Goldstein (2009.05.09.06:57 EDT)]
[About Bill Powers (2009.05.08.1346 MDT)]

BP: Given that there is a 0.60 correlation between IQ and
reading skill
level, what are the chances that the IQ of the next child you
test will
predict correctly the quintile of reading skill level of that
child after
first grade?

DG: Attached is the best answer I can come up with for now. It
is based on a
statistics program called ZumaStat. See ‘For Bill.pdf’

BP: Not very. If you had said that studies of this kind are
highly likely
to predict the results of psychotherapy accurately for every person
needing help, say with a 90% probability of correct
prediction for
each case, I would be impressed. But you haven’t said
that, nor does
anything in the papers you attached promise anything like
that kind of
result.

DG: Here is the actual study that had been summarized. I can
provide the
effect size using ZumaStat if you want.
It was helpful for some of the patients in some of the measures
used. Some
help is better than no help. These
are very difficult people to work with successfully. See ‘520.pdf’.

[From Bill Powers (2009.05.10.1013 MDT)]

David Goldstein (2009.05.08.06:26 EDT) --

DG: Richard Kennaway's paper is very interesting and persuasive, as far as it goes. However, it does not address the
possibility of predicting a variable from a set of variables, as in multiple, step-wise
regression. This is what most psychological studies of the correlational type do. Each
subsequent variable entered into the regression equation adds something that the previous
variable did not. This applies to the case of predicting end of first grade reading achievement from IQ and other variables (for example measures of phonemic awareness).

BP:I wasn't aware of this method, so thanks for bringing it up. I Googled it, and it seems to be a more powerful way of determining causal factors operating on a behavioral variable. It amounts to proposing that the measure of behavior b is a function of a collection of variables v1 .. vn, of the form

b = a0 + a1*v1 + a2*v2 + ... an*vn

Start with the proposal that only one variable, v1, is responsible for the behavior. The variable v1 will correlate with b over multiple observations through time and the regression equation can be found to give the regression coeffient a1 and a constant a0. There will be residuals representing the variance unaccounted for. Those residuals can then be correlated with other variables and the regression coefficient v2 found, with the constant term being added to a0. This can be continued indefinitely, with each residual being reduced by use of still another possible causal variable. If the value of the entire right side of the above equation is then used as the independent variable, that composite variable will, or might, have a much higher correlation with b than will any one term alone.

This would work very nicely to determine all the disturbances acting on a controlled variable by examining the correlation of many environmental variables with the action produced by a control system. It's not necessary to do a physical analysis of the environment to reason out the variables that could physically affect the controlled variable. One simply measures every variable in the local environment and does the correlations, computing the regression coefficients to adjust the coefficients in the overall equation. In principle, it would seem, the multiple regression equation could be extended to get as high a correlation as one pleases, but for random noise in the system.

Now, what's wrong with this approach? In the first place, it can't tell the difference between multiple stimuli that produce a response and multiple disturbances that tend to alter a controlled variable. That is, it can't tell you what kind of system you're analyzing.

Second, there is that noise which makes all the variable measurements uncertain. The random noise indicated by each correlation will add in quadrature to produce a noise in the whole system that can't be reduced because it doesn't repeat on successive measurements of the same variable. Each term in the multiple regression function contributes noise to the value of the function and reduces the overall correlation. This is in addition to the failure of correlation due to nonlinearities (the multiple regression equation above is a linear function in each variable).

Third, the multiple regression equation omits one variable that can alter b without having any effect on the input to the system: the reference signal. If this is a purely stimulus-response system without reference signals, there is no problem, but if it's a hierarchy of control systems, every control system that affects behavior receives a reference signal that can vary unpredictably, having very large effects on the behavior. So the ideal of simply calculating correlations with every possible variable falls to pieces; that strategy will work on a control hierarchy only if all the reference signals are somehow held constant. If the reference signals can change, there is a vitally important influence on b that is unaccounted for and will remain so. The effect will be just as if there is a variable noise level in the system with a large magnitude.

Fourth, the multiple regression equation assumes a purely algebraic relationship between each variable and the behavior. However, we know that in many cases there are dynamic relationships, as in the tracking experiment where an error signal produces not a magnitude of behavior but a rate of change of magnitude. The mouse moves at a speed proportional to the error signal. This means that not only the magnitudes of disturbances must be correlated with the behavior, but the first few derivatives as well, and because other kinds of temporal relationships can hold, the integrals also and probably time delays. This can be done, but we are beginning to look at an extremely complex undertaking here.

Fifth and last for now, the multiple regression equation can handle only one level of organization. In order to be used with a hierarchy of control systems of the kind we study in PCT, we would have to investigate multiple controlled variables at each level, and then multiple higher-level controlled variables at the next level, and so on. Perhaps a very clever mathematician (hint, hint) might see a way to extend the multiple regression method to multiple levels, but I don't see any way.

Of all these problems the most important one in my opinion is the variable reference signal. It is not directly observable, its variations can be large, and its distribution is far from Gaussian. The reference signal can be constant for a while, then start varying. It can vary a lot, then only a little, then a lot again. Its pattern of variation can change from moment to moment. And there is nothing directly observable to indicate its state.

···

==========================================================================================
Aside from all the above problems, I have one more objection. This method does not require or provide any understanding of how the system works. It is "purely empirical," meaning that there is no model behind it. Since I aspire to an understanding of how the human system works, I don't use purely empirical methods. I think that the history of science shows us that pure empiricism is not a very effective way of understanding anything; it doesn't reveal organizing principles that can substitute for vast catalogues of observed relationships. It immerses us in seas of meaningless detail. Modeling seems to be far more effective and efficient.

DG: You seem to be stuck on the idea that one variable must do everything.

BP: I think I'm unstuck now. I still haven't seen any examples in the literature, but will keep an eye out for them.

Best,

Bill P.

[From David Goldstein (2009.05.10.16:07)]
[David Bill Powers (2009.05.10.1013 MDT)]

Many of your objections led some people to develop structual equation modeling.
See the following website:
http://davidakenny.net/cm/causalm.htm
David

[From Bill Powers (2009.05.10.1555 MDT)]

David Goldstein (2009.05.10.16:07) --

Many of your objections led some people to develop structual equation modeling. See the following website:
Structural Equation Modeling (David A. Kenny)

Well at least someone else agrees with my objections! How do you come up with these things?

I found that web site rather weird. It's all about causation, but it doesn't give me any picture at all of the kind of system they're talking about -- just a lot of equations and mathematical manipulations. They even talk about feedback without, apparently, realizing that this is a completely different sort of causal system. It really looks like old-fashioned S-R theory dressed up in quantitative clothes and made very complicated.

As far as I can tell, which isn't very far, they are simply trying to explain one variable as a function of a set of other variables. Maybe you got more out of that than I did. What sort of organized system are they talking about?

Best,

Bill P.

[From David Goldstein (2009.05.09.06:57 EDT)]
[About Bill Powers (2009.05.08.1346 MDT)]

BP: Given that there is a 0.60 correlation between IQ and reading skill
level, what are the chances that the IQ of the next child you test will
predict correctly the quintile of reading skill level of that child after
first grade?

DG: Attached is the best answer I can come up with for now. It is based on a
statistics program called ZumaStat. See 'For Bill.pdf'

BP: Not very. If you had said that studies of this kind are highly likely
to predict the results of psychotherapy accurately for every person
needing help, say with a 90% probability of correct prediction for
each case, I would be impressed. But you haven't said that, nor does
anything in the papers you attached promise anything like that kind of
result.

DG: Here is the actual study that had been summarized. I can provide the
effect size using ZumaStat if you want.
It was helpful for some of the patients in some of the measures used. Some
help is better than no help. These
are very difficult people to work with successfully. See '520.pdf'.

For Bill.pdf (10.7 KB)

520.pdf (186 KB)