Do you support educational vouchers in schools?

bizby40 wrote:
"toto" wrote in message
...
On 14 May 2005 19:12:16 -0700, wrote:

There are a lot of GIGO situations. (Anyone remember GIGO?)

Puts up hand....... I do, I do...

I found myself wanting to use the terms LIFO and FIFO in my
post on bus routes.

:-)

Bizby

I used it in my fridge, for eggs. FIFO, of course, not LIFO.

Rupa

In article ,
Bob LeChevalier wrote:
(Herman Rubin) wrote:
In article ,
toto wrote:
On 12 May 2005 10:50:14 -0500, (Herman
Rubin) wrote:

In addition, time spent on a bus is almost entirely
LOST time. In some cases, busing added more than
an hour to the "school day". Before integration,
black children frequently went to schools farther
than the white schools they were asking to get in.
But busing for racial balance provided for a lot
less time for learning, homework, and other activities.

Why is it that busing for integration is questioned, but busing
for distance is not. The same amount of time is spent on
the bus. It shouldn't matter what the purpose of the busing is.
Of course, if you are against busing kids then you have to
build schools closer to their homes.

Suppose you have two areas, with an hour ride between
them. Each has schools. Busing students from one
area to the other just to get racial balance is what
is stupid.

Only if you think "racial balance" is stupid. Replace "racial
balance" by "an academically challenging school" and you probably
would have no problem with it.

Racial balance has NEVER been legislated; what has
been legislated is no racial discrimination.

It is one thing to travel to a school which gives a
substantially stronger education, but is spending
two hours a day doing nothing worth it? Busing two
ways to get racial balance certainly does not get
justified by this type of argument.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

THIS IS A WOODWORKING NEWS GROUP!!!!!!!!!!!!!


"Herman Rubin" wrote in message
...
In article ,
Bob LeChevalier wrote:
(Herman Rubin) wrote:
In article ,
toto wrote:
On 12 May 2005 10:50:14 -0500, (Herman
Rubin) wrote:

In addition, time spent on a bus is almost entirely
LOST time. In some cases, busing added more than
an hour to the "school day". Before integration,
black children frequently went to schools farther
than the white schools they were asking to get in.
But busing for racial balance provided for a lot
less time for learning, homework, and other activities.

Why is it that busing for integration is questioned, but busing
for distance is not. The same amount of time is spent on
the bus. It shouldn't matter what the purpose of the busing is.
Of course, if you are against busing kids then you have to
build schools closer to their homes.

Suppose you have two areas, with an hour ride between
them. Each has schools. Busing students from one
area to the other just to get racial balance is what
is stupid.

Only if you think "racial balance" is stupid. Replace "racial
balance" by "an academically challenging school" and you probably
would have no problem with it.

Racial balance has NEVER been legislated; what has
been legislated is no racial discrimination.

It is one thing to travel to a school which gives a
substantially stronger education, but is spending
two hours a day doing nothing worth it? Busing two
ways to get racial balance certainly does not get
justified by this type of argument.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

In article ,
Bob LeChevalier wrote:
wrote:
On 12 May 2005 11:34:49 -0500, (Herman
Rubin) wrote:

....................

On the contrary; what I consider to be the concepts are
what is needed to know what it means well enough to
formulate the problems, and to know what the assumptions
mean. It is not unusual for someone to come in with a
stack of observations, and ask to be told the state of
the universe.

Well yeah, I can see where THAT request would crop up in every day
conversation. It's pointless to even try to tell you that what you're
using as an example to support your position is an extreme application
and not one that is commonplace, isn't it?

At the risk of defending Herman, I think it is commonplace, but is
seldom put in the highly abstract terms he uses.

One example would be the assessment of risk in a medical procedure.
People don't have the information on the chances of success of a given
medical option (the data is seldom properly gathered so as to give
those chances) vs the level of risk.

This is where the medical profession should come in,
to provide the assessment, not of the risk, but the
probabilities of outcomes. The risk also involves the
importance of the outcomes, and this MUST be left to
the individual to assess.

Immediately coming to mind are the newsmaking arthritis medications.
If I have arthritis, is it worth it for me to take Vioxx or one of the
other suspect drugs? There is risk of side effects, but what is the
risk for me personally? It would probably take a complex (Bayesian?)
statistical analysis to come up with any idea at all, and I know I
don't have the math to do that.

You are right. But YOU have to be able to evaluate the
importance of the various consequences, and the medical
profession needs to evaluate the probabilities, taking
into account you, as an individual, and the computer can
then provide the procedure. Even better, the computer
can evaluate the risks of the various procedures, weighting
them with your weights, in the different situations, and
you might want to choose one which is slightly worse in
the accepted situation, but less subject to change if the
accepted situation turns out to be somewhat in error.

But I don't see any practical solution either, because most people
aren't going to be willing to learn what they would need to learn to
be able to apply math intelligently.

Read the above; it can be done, and it should be done,
but not by government fiat.

They DO assume that all that one has to
do to analyze observations is to look at them, with no
idea of assumptions about the structure.

So they mostly do NOT really need to be able to the math themselves,
as I said.

He claims that they need to, if they want to get "correct" answers.

No, they need to formulate the math, not to be able to
carry out the calculations, or even to decide how the
calculations are done. Only the individual can produce
the relative importance of the various results, and
some of the biological information is individual; the
medical people need to make their formulation of the
probabilistic assumptions in medicine so the posterior
probabilities can be calculated, and then the machine
can come up with the information.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

Herman Rubin wrote:
Bob LeChevalier wrote:

One example would be the assessment of risk in a medical procedure.
People don't have the information on the chances of success of a
given
medical option (the data is seldom properly gathered so as to give
those chances) vs the level of risk.

This is where the medical profession should come in,
to provide the assessment, not of the risk, but the
probabilities of outcomes.

I think there's a data issue here. How can the medical profession
provide anything but a subjective assessment unless the data are
collected to provide those probabilities? And who funds that data
collection?



Even better, the computer
can evaluate the risks of the various procedures, weighting
them with your weights, in the different situations, and
you might want to choose one which is slightly worse in
the accepted situation, but less subject to change if the
accepted situation turns out to be somewhat in error.

IMO, the mathematical model in this case would be very easy to build.
It's not even that sophisticated, just a bunch of probabilities. What
is really difficult is getting the data to feed into the model.

Take the Vioxx example. You would want independent probabilities of
cardiac risk for various factors: Gender; age; existing cardiac
problems; ethnicity; weight. (That's all I can think of for now, but
I'm sure there would be other factors.) You would want joint
probabilities for every combination, assuming that they're not
independent. (And I don't think you can assume that.)

So you'd need a *huge* study of Vioxx against various control groups.

Who's doing that?

So what it will boil down to is something like, for a sample of people
taking Vioxx as compared to people taking, oh, say, ibuprofen, there
was an increased mortality of x%. Of this y% was due to cardiac
problems. So you extrapolate from that, having no reason not to, that
it will raise *your* risk of mortality by x%.

Then you figure, did using Vioxx make me feel so much better, compared
to using ibuprofen, that it's worth the increased risk?

(I'd probably want to know the increased mortality from all causes in
the sample, but that's a separate issue...)


No, they need to formulate the math, not to be able to
carry out the calculations, or even to decide how the
calculations are done. Only the individual can produce
the relative importance of the various results, and
some of the biological information is individual; the
medical people need to make their formulation of the
probabilistic assumptions in medicine so the posterior
probabilities can be calculated, and then the machine
can come up with the information.

The people who need to understand the math are the people who are
gathering data.

Unless they know, they won't be able to set up decent data collection,
and after that, aggressive analysis of the data won't necessarily give
you a good answer.

Rupa

In article .com,
wrote:

Herman Rubin wrote:
Bob LeChevalier wrote:

One example would be the assessment of risk in a medical procedure.
People don't have the information on the chances of success of a
given
medical option (the data is seldom properly gathered so as to give
those chances) vs the level of risk.

This is where the medical profession should come in,
to provide the assessment, not of the risk, but the
probabilities of outcomes.

I think there's a data issue here. How can the medical profession
provide anything but a subjective assessment unless the data are
collected to provide those probabilities? And who funds that data
collection?

What do you think is being done now? With the data
available, no matter how much is collected, there will
necessarily be considerable subjectivity. This is not
the worst of the problems, nor even that much of a
problem. Subjectivity is unavoidable.

Even better, the computer
can evaluate the risks of the various procedures, weighting
them with your weights, in the different situations, and
you might want to choose one which is slightly worse in
the accepted situation, but less subject to change if the
accepted situation turns out to be somewhat in error.

IMO, the mathematical model in this case would be very easy to build.
It's not even that sophisticated, just a bunch of probabilities. What
is really difficult is getting the data to feed into the model.

Take the Vioxx example. You would want independent probabilities of
cardiac risk for various factors: Gender; age; existing cardiac
problems; ethnicity; weight. (That's all I can think of for now, but
I'm sure there would be other factors.) You would want joint
probabilities for every combination, assuming that they're not
independent. (And I don't think you can assume that.)

One can always do the best with the data available.

As a statistician, I find the great bulk of the medical
uses of statistics to be misuses. Objectivity is
impossible unless the relations are simple, and they
are not, and there are few "nuisance factors".

It is possible to carry out such studies; they are not
possible using what can be taught in methods courses.

So you'd need a *huge* study of Vioxx against various control groups.

No; one can work with what is available.

Who's doing that?

So what it will boil down to is something like, for a sample of people
taking Vioxx as compared to people taking, oh, say, ibuprofen, there
was an increased mortality of x%. Of this y% was due to cardiac
problems. So you extrapolate from that, having no reason not to, that
it will raise *your* risk of mortality by x%.

This is not the important problem.

Then you figure, did using Vioxx make me feel so much better, compared
to using ibuprofen, that it's worth the increased risk?

This one is MUCH more important. BTW, ibuprofen decreases
the heart protective value of aspirin; I do not know if
Vioxx has a similar effect.

(I'd probably want to know the increased mortality from all causes in
the sample, but that's a separate issue...)

That is another problem. The additions to the usual
NSAIDS to decrease stomach problems increase the risks
from infectious diseases; the mechanism is clear. One
needs to balance the situation.

No, they need to formulate the math, not to be able to
carry out the calculations, or even to decide how the
calculations are done. Only the individual can produce
the relative importance of the various results, and
some of the biological information is individual; the
medical people need to make their formulation of the
probabilistic assumptions in medicine so the posterior
probabilities can be calculated, and then the machine
can come up with the information.

The people who need to understand the math are the people who are
gathering data.

The ones now designing the experiments do not; they only
use simplistic approaches.

Unless they know, they won't be able to set up decent data collection,
and after that, aggressive analysis of the data won't necessarily give
you a good answer.

They do not even know how to analyze the data. The bigger
problem is they do not even know how to report it, and do
it in a very simplistic manner.



--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

Herman Rubin wrote:
In article .com,
wrote:

Herman Rubin wrote:

This is where the medical profession should come in,
to provide the assessment, not of the risk, but the
probabilities of outcomes.

I think there's a data issue here. How can the medical profession
provide anything but a subjective assessment unless the data are
collected to provide those probabilities? And who funds that data
collection?

What do you think is being done now? With the data
available, no matter how much is collected, there will
necessarily be considerable subjectivity. This is not
the worst of the problems, nor even that much of a
problem. Subjectivity is unavoidable.

Of course, but of what kind?

The Vioxx study showed that of 1300 participants using Vioxx for over
18 months, 15 had cardiac events (heart attack or stroke). Of the 1300
in the control group, apparently 5 or fewer had such events.

What does that tell a doctor?

It triples your risk?
Ah. Right. What is my risk?

If we assume that the study's participants are representative of all
potnetial Vioxx users, my risk is somewhere around 0.4% without Vioxx
and 1.1% with Vioxx. Is that a lot or a little? How is it impacted by
other risk factors? If I'm at a higher risk because, say, there is
heart disease in my family, does it just add on a little bit, or does
it triple my already higher risk?


One can always do the best with the data available.

The best with the data available...is that it increases your risk from
0.4% to 1.1%.

Since the mechanism involved isn't clear, we don't know if it works
independently of other risk factors, or if it multiplies them.

Depending on how much relief Vioxx gave you as compared to the next
best thing, you might want to go with it.

You can't, of course, since it's been withdrawn. Merck calculated the
probability of getting sued...


It is possible to carry out such studies; they are not
possible using what can be taught in methods courses.

I rather think it's a matter of lab methods meets real world. You have
to find people willing to sign up for double-blind studies for whatever
you can offer them to participate.

If you wanted to stratify your sample by age and gender and so forth,
you would need a pretty big sample, AFAIK.

So you'd need a *huge* study of Vioxx against various control
groups.

No; one can work with what is available.

How?


So what it will boil down to is something like, for a sample of
people
taking Vioxx as compared to people taking, oh, say, ibuprofen, there
was an increased mortality of x%. Of this y% was due to cardiac
problems. So you extrapolate from that, having no reason not to,
that
it will raise *your* risk of mortality by x%.

This is not the important problem.

No? But then there isn't one, is there? You take it if it makes you
feel better, and you stop if it doesn't. Unless you understand the
extent of the risk, how can you weigh a benefit against it?

Then you figure, did using Vioxx make me feel so much better,
compared
to using ibuprofen, that it's worth the increased risk?

This one is MUCH more important. BTW, ibuprofen decreases
the heart protective value of aspirin; I do not know if
Vioxx has a similar effect.

(I'd probably want to know the increased mortality from all causes
in
the sample, but that's a separate issue...)

That is another problem. The additions to the usual
NSAIDS to decrease stomach problems increase the risks
from infectious diseases; the mechanism is clear. One
needs to balance the situation.

I think you might be referring to omeprazole and similar treatments for
acid reflux disease. They decrease the level of acid in the stomach,
but thus also reduce its ability to burn out the nasty bugs.

I don't believe the coatings of NSAIDs have a similar effect.


The ones now designing the experiments do not; they only
use simplistic approaches.

They do not even know how to analyze the data. The bigger
problem is they do not even know how to report it, and do
it in a very simplistic manner.

Actually, I would imagine that most big pharma companies employ people
trained in stats and math to design their trials and interpret the
results. The reporting of the results, though, may be determined by
lawyers and journalists. The Merck site had no details of how much
Vioxx increased cardiac risk. The site set up by a liability attorney
did.

Rupa

In article . com,
wrote:

Herman Rubin wrote:
In article .com,
wrote:

Herman Rubin wrote:

This is where the medical profession should come in,
to provide the assessment, not of the risk, but the
probabilities of outcomes.

I think there's a data issue here. How can the medical profession
provide anything but a subjective assessment unless the data are
collected to provide those probabilities? And who funds that data
collection?

What do you think is being done now? With the data
available, no matter how much is collected, there will
necessarily be considerable subjectivity. This is not
the worst of the problems, nor even that much of a
problem. Subjectivity is unavoidable.

Of course, but of what kind?

Are you familiar with the likelihood function, risk, and
the problems of self-consistent decision making? To put
the matter as simply as possible, the decision to be made
should be decided by using a weighted combination of the
risks in the various states of nature, and the weights to
be used can only be purely subjective. If the data is
overwhelming, it matters little. If it is not, it can
matter much.

The Vioxx study showed that of 1300 participants using Vioxx for over
18 months, 15 had cardiac events (heart attack or stroke). Of the 1300
in the control group, apparently 5 or fewer had such events.

What does that tell a doctor?

Not as much as you think it does. In cases like this, what
one might call the "effective sample size" would be 20, not
a particularly large sample.

It triples your risk?

Roughly.

Ah. Right. What is my risk?

If we assume that the study's participants are representative of all
potnetial Vioxx users, my risk is somewhere around 0.4% without Vioxx
and 1.1% with Vioxx. Is that a lot or a little?

Part of it depends on your age and how much Vioxx helps you.

How is it impacted by
other risk factors? If I'm at a higher risk because, say, there is
heart disease in my family, does it just add on a little bit, or does
it triple my already higher risk?

Here is where I would have had those conducting the study to
get the additional data, even though the elementary statistical
rituals they know would be unable to use it. I do not know how
much I could get out of it in this situation, but even if it
only led to the places to look in subsequent studies, it should
be obtained.

One needs to understand that statistics is not a collection of
recipes which tells one what to do, but that everything must
be taken into account. Of course, this means that the nice
procedures they have learned do not have the connotations which
they already improperly assign to them.

One can always do the best with the data available.

The best with the data available...is that it increases your risk from
0.4% to 1.1%.

Since the mechanism involved isn't clear, we don't know if it works
independently of other risk factors, or if it multiplies them.

See my paragraph above. The statistical idiots did not collect
the information needed to even make a guess about this.

Depending on how much relief Vioxx gave you as compared to the next
best thing, you might want to go with it.

You can't, of course, since it's been withdrawn. Merck calculated the
probability of getting sued...

If we had decent laws on this, this would not be a problem.
If a company presents all the known information, including
the bad information, and keeps updating it, this should be
an absolute defense from lawsuits. This should apply to the
FDA and other government agencies as well; their prohibitions
should make the people responsible subject to lawsuits. It
should not be a go/nogo situation, but here is what we know,
and here is what we do not know; YOU decide.

It is possible to carry out such studies; they are not
possible using what can be taught in methods courses.

I rather think it's a matter of lab methods meets real world. You have
to find people willing to sign up for double-blind studies for whatever
you can offer them to participate.

If you wanted to stratify your sample by age and gender and so forth,
you would need a pretty big sample, AFAIK.

You can use multivariate methods. People in other fields have
been using multivariate methods for 200 years.

So you'd need a *huge* study of Vioxx against various control
groups.

No; one can work with what is available.

How?

See the above. I have no idea what the additional data would
yield, if it was obtained. There is a limit to what can be
done. At this time, about all that one can do is assume that
it is an increased risk.

...................

The ones now designing the experiments do not; they only
use simplistic approaches.

They do not even know how to analyze the data. The bigger
problem is they do not even know how to report it, and do
it in a very simplistic manner.

Actually, I would imagine that most big pharma companies employ people
trained in stats and math to design their trials and interpret the
results. The reporting of the results, though, may be determined by
lawyers and journalists. The Merck site had no details of how much
Vioxx increased cardiac risk. The site set up by a liability attorney
did.

They have some, but not the ones with this type of knowledge.

Most of what they do has to meet the rather rigid statistical
ritual set out by the FDA, and those are determined by people
who do not understand good statistics, nor how to consider
handling complexities such as the ones you have raised.
And the liability laws are as I said; honesty is no defense.
That you had no idea that a product would cause reproductive
problems 20 years later in daughters of women who take it,
and how could they possibly have known this, did not protect
the company from liability. However, it protected the doctors
who prescribed it. The drug is still being used, on cattle,
to prevent miscarriages, and is does not present problems.






--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

In article ,
toto wrote:
On 12 May 2005 21:13:43 -0500, (Herman
Rubin) wrote:

How many times do I have to say that one does not learn
a concept by merely being given the words? It may even
be likely for someone who understands mathematical logic
to "see the light", but I do not expect a second grader
to be able to do this. Concepts have to be illustrated,
and it could be reasonable to point this out explicitly,
and have them do enough examples that they see that the
quartets of "number facts" are really only one.

My ds as a second grader saw many examples and could
indeed produce new examples of the commutative propery
and the law of inverses. He could and did figure out
negative numbers by *seeing* the number line. I don't think
that most kids will do that however.

But can he see the order properties of integers, and
the quantitative properties, and see that they are
different properties.

The positive integers are a subset of the integers, and
they have properties as positive integers which one cannot
get from properties of integers, let alone properties of
rational numbers or real numbers. So there are lots of
concepts, and position on a line does not convey all.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
Phone: (765)494-6054 FAX: (765)494-0558

Herman Rubin wrote:

wrote:

Herman Rubin wrote:

What do you think is being done now? With the data
available, no matter how much is collected, there will
necessarily be considerable subjectivity. This is not
the worst of the problems, nor even that much of a
problem. Subjectivity is unavoidable.

Of course, but of what kind?

Are you familiar with the likelihood function, risk, and
the problems of self-consistent decision making? To put
the matter as simply as possible, the decision to be made
should be decided by using a weighted combination of the
risks in the various states of nature, and the weights to
be used can only be purely subjective. If the data is
overwhelming, it matters little. If it is not, it can
matter much.

The point I was trying to make was that obviously the patient makes a
subjective decision, but the doctor *should* be able to provide
reasonably objective data. He can't because it hasn't been collected.
He can only give you something like the summary I've given below.
Which, as you point out, tells you very little.


The Vioxx study showed that of 1300 participants using Vioxx for
over
18 months, 15 had cardiac events (heart attack or stroke). Of the
1300
in the control group, apparently 5 or fewer had such events.

What does that tell a doctor?

Not as much as you think it does. In cases like this, what
one might call the "effective sample size" would be 20, not
a particularly large sample.

Quite so.


It triples your risk?

Roughly.

Ah. Right. What is my risk?

If we assume that the study's participants are representative of all
potnetial Vioxx users, my risk is somewhere around 0.4% without
Vioxx
and 1.1% with Vioxx. Is that a lot or a little?

Part of it depends on your age and how much Vioxx helps you.

Yes. And part of it depends on what my overall mortality risk is
anyway.


How is it impacted by
other risk factors? If I'm at a higher risk because, say, there is
heart disease in my family, does it just add on a little bit, or
does
it triple my already higher risk?

Here is where I would have had those conducting the study to
get the additional data...

One can always do the best with the data available.

The best with the data available...is that it increases your risk
from
0.4% to 1.1%.

Since the mechanism involved isn't clear, we don't know if it works
independently of other risk factors, or if it multiplies them.

See my paragraph above. The statistical idiots did not collect
the information needed to even make a guess about this.

I rather think it's a matter of lab methods meets real world. You
have
to find people willing to sign up for double-blind studies for
whatever
you can offer them to participate.

If you wanted to stratify your sample by age and gender and so
forth,
you would need a pretty big sample, AFAIK.

You can use multivariate methods. People in other fields have
been using multivariate methods for 200 years.

As you pointed out, our 2600 person study yielded an "effective sample
size" of 20 cases of cardiac events. How do you do a multivariate
analysis of that? Assuming you had data on age, gender, existing
medical conditions.

So you'd need a *huge* study of Vioxx against various control
groups.

No; one can work with what is available.

How?

See the above. I have no idea what the additional data would
yield, if it was obtained. There is a limit to what can be
done. At this time, about all that one can do is assume that
it is an increased risk.

Let us assume that it was obtained (and it well may have been - AFAIK,
clinical studies do keep a reasonable amount of data on every
participant).
How can you make it yield any information on interactions of risk
factors?

My guess is you'd need a *much* larger study.

Rupa