The Many Faces of Vitamin C (was Re: Half-life of vitamins in the body - was Re: Green Tea)
Steve Harris
news:od2c8.18694$P21.1...@newsread1.prod.itd.earthlink.net...
> Another thought from the peanut gallery. I have no nutritional knowledge,
I
> am a computer programmer. But why are you guys fixating on the input
levels
> of vitamin C vs. the output levels in the urine? What if maintaining
> vitamin C at some threshold in the body for some period of time has a
*side
> effect* ? Sets off some process that ordinarily doesn't occur or
something.
> Maybe a healthy process. The vitamin gets flushed out pronto, it may not
be
> relevant, but the process it sets in motion may be.
COMMENT:
There's no way to tell, but with a prospective randomized study.
Retrospective epidemiology is good for ruling out causal relations, but it's
very poor evidence for ruling them in, due to the possiblity of confounders.
And that's even when it's used properly-- when in many cases it isn't. For
instance vitamin C tissue levels don't correlate at all with cardiovascular
disease in the NHANES II except in people who drink quite a lot of alcohol.
If you control for alcohol you get no significant effect, and that's quite
_strong_ evidence that if vitamin C has any big pro-vascular-health effect
in non-drinkers, it can't be _large_, or else it would have to show up here.
That should make people think about no drinking so much (and not drinking
anything but red wine when they DO drink) before it makes them think of
popping vitamin C pills. As for vitamin C plasma levels, if you ignore
drinking, they weakly correlate with cardiovascular disease risk across very
broad ranges from lowest to highest, but that doesn't mean there's any
epidemiologic difference between cardiovascular disease at (say) the levels
you equilibrate at, when you take a pill once a day, and your average higher
levels when you take MORE than one pill a day. We do have a confused person
here who thinks that a curve drawn through many points can be taken as
gospel across levels at the high end of the curve-- but he is (well) a
confused person. Ignore this stuff.
Now let me note some hair-raisers, while we're at it. Vitamin C may
epidemiologically decrease vascular risk in male drinkers a little, but the
exact same evidence from the exact same study suggests that vitamin C intake
increases risk of CANCER in women a LOT. You don't get to pick which one
you're going to believe, if you insist on believing every positive
correlation you get from post-hoc retrospective epidemiology. Taking this
stuff at face value, you might conclude that vitamin C pills might (say))
protect a male Nobel Prize Winner from vascular disease and let him die in
old age, but (at the same time) might kill his wife with some nasty tumor
<here I look at sky innocently and whistle>. But read the following and see
what you think. And don't listen to everyone who says they're a health
expert. There's no substitute for reading the studies yourself, and knowning
a bit about statistics. And for being wise about that fact that there are
few panaceas in life.
SBH
=========================
J Am Coll Nutr 2001 Jun;20(3):255-63 Related Articles, Books, LinkOut
Relation of serum ascorbic acid to mortality among US adults.
Simon JA, Hudes ES, Tice JA.
General Internal Medicine Section, Veterans Affairs Medical Center, San
Francisco, California 94121, USA. jas...@itsa.ucsf.edu
PURPOSE: To examine the relation between serum ascorbic acid (SAA), a marker
of dietary intake (including supplements), and cause-specific mortality.
SUBJECTS AND METHODS: We analyzed data from a probability sample of 8,453
Americans age > or = 30 years at baseline enrolled in the Second National
Health and Nutrition Examination Survey (NHANES II), who were followed for
mortality endpoints. We calculated relative hazard ratios as measures of
disease association comparing the mortality rates in three biologically
relevant SAA categories. RESULTS: Participants with normal to high SAA
levels had a marginally significant 21% to 25% decreased risk of fatal
cardiovascular disease (CVD) (p for trend = 0.09) and a 25% to 29% decreased
risk of all-cause mortality (p for trend <0.001) compared to participants
with low levels. Because we determined that gender modified the association
between SAA levels and cancer death, we analyzed these associations
stratified by gender. Among men, normal to high SAA levels were associated
with an approximately 30% decreased risk of cancer deaths, whereas such SAA
levels were associated with an approximately two-fold increased risk of
cancer deaths among women. This association among women persisted even after
adjustment for baseline prevalent cancer and exclusion for early cancer
death or exclusion for prevalent cancer. CONCLUSIONS: Low SAA levels were
marginally associated with an increased risk of fatal CVD and significantly
associated with an increased risk for all-cause mortality. Low SAA levels
were also a risk factor for cancer death in men, but unexpectedly were
associated with a decreased risk of cancer death in women. If the
association between low SAA levels and all-cause mortality is causal,
increasing the consumption of ascorbic acid, and thereby SAA levels, could
decrease the risk of death among Americans with low ascorbic acid intakes.
PMID: 11444422 [PubMed - indexed for MEDLINE]
--
I welcome Email from strangers with the minimal cleverness to fix my address
(it's an open-book test). I strongly recommend recipients of unsolicited
bulk Email ad spam use "http://combat.uxn.com" to get the true corporate
name of the last ISP address on the viewsource header, then forward message
& headers to "abuse@[offendingISP]."
Brandon J. Van Every
"Steve Harris" <sbha...@ix.RETICULATEDOBJECTcom.com> wrote in message
news:a4rubk$3v2$1...@nntp9.atl.mindspring.net...
> SUBJECTS AND METHODS: We analyzed data from a probability sample of 8,453
> Americans age > or = 30 years at baseline enrolled in the Second National
> Health and Nutrition Examination Survey (NHANES II), who were followed for
> mortality endpoints. [...] This association among women persisted even
after
> adjustment for baseline prevalent cancer and exclusion for early cancer
> death or exclusion for prevalent cancer.
It is a pity not to be able to sample much larger numbers of people. No
matter what statisticians say, life is complex and there are many variables.
I can only think that "adjustment" means throwing out various data samples.
The more adjustments you make, the less data you're really working with. In
3D computer graphics, my field, 8000 is nowadays a trivial number for all
sorts of things. A study of millions of people would impress me, but of
course it would be correspondingly expensive to undertake. And so there you
have it, dollars and economy of scale are a factor in science and
engineering.
--
Cheers, www.3DProgrammer.com
Brandon Van Every Seattle, WA
20% of the world is real.
80% is gobbledygook we make up inside our own heads.
Steve Harris
> It is a pity not to be able to sample much larger numbers of people. No
> matter what statisticians say, life is complex and there are many
variables.
> I can only think that "adjustment" means throwing out various data
samples.
> The more adjustments you make, the less data you're really working with.
In
> 3D computer graphics, my field, 8000 is nowadays a trivial number for all
> sorts of things. A study of millions of people would impress me, but of
> course it would be correspondingly expensive to undertake. And so there
you
> have it, dollars and economy of scale are a factor in science and
> engineering.
A study of a million people isn't necessary and is a waste of resources, in
much the same way you don't need to ask a million people a question to get a
good Gallup Poll result. If you want your sampling results within 1/2 % of
true, all you need is a couple of thousand samples. Anything more is just
making up for inhomogeneity.
--
Steve Harris
You can email me at sbhar...@ix.netcom.com
But remove the numerals in the address first.
==============================
Our nada who art in Nada
Nada be thy nada..
-- Dada Hemingway
==========================
Martin Banschbach
> It is a pity not to be able to sample much larger numbers of people. No
> matter what statisticians say, life is complex and there are many variables.
> I can only think that "adjustment" means throwing out various data samples.
> The more adjustments you make, the less data you're really working with. In
> 3D computer graphics, my field, 8000 is nowadays a trivial number for all
> sorts of things. A study of millions of people would impress me, but of
> course it would be correspondingly expensive to undertake. And so there you
> have it, dollars and economy of scale are a factor in science and
> engineering.
Brandon,
Having a different view of the issues facing nutrition research is
very
refreshing. I often wonder if the problem with getting enough
human subjects to study forces investigators to try to make math
(statistics) do more than it can.
Thanks to this newsgroup, I'm now aware of two areas where women may
be different than men. How they handle fructose and vitamin C appears
to be quite different than the way men handle these two dietary
components. But I know that from working with mice and rats that
the female reproductive system really alters metabolism to the point
of either using female animals all in the same reproductive stage or
just removing the ovaries and then adding back known ammounts of
female hormones.
One thing that the general public does not understand is that humans
have such a wide variation in genetic makeup that even if the
environment is faily constant, the gene drive is going to be much
greater than anything that you would ever see in lab animals.
Cindy Fuller pointed out that there is a known genetic defect that
sets some women up for being folate deficienct at the time of
conception and therefore greatly increases their risk of having a baby
with a poorly formed spinal column (spina bifida). I have pointed out
that
there is a genetic defect where the enzyme that takes homocysteine to
cysteine has been altered so that it's affinity for B6 is much lower
than
it should be so more B6 is needed to keep the enzyme saturated with
B6.
One poster pointed out that for humans there is probably a 10 fold
variation
in the requirement for specific essential nutrients. Those of us
involved
in collecting and analyzing data often loose sight of this problem
with
human data. It takes someone from outside to step in and say what you
just
said to bring us to our senses (at least it brought me to my
senses).
I will often say that there is no reason for a poster to do what they
are
doing based on what is known about human nutritional requirements.
The data
lulls me into a false sense of security. One person may do quite well
with a
selenium intake of 150 micrograms per day but someone else may need
500 micrograms per day to be at the same level metabolically as the
person getting
150 micrograms per day.
I just finished going over the data for 30 different corn hybrids
where only
the fatty acid profile and vitamin E profile was examined. The
genetics of
corn is much more stable than the genetics of humans.
The only data that I have been able to amass in all of my many years
of covering
human nutrition where this same kind of variability in humans was
accurately
studied was nitrogen balance data. I know that the RDA set for each
essential
amino acid in humans is way off for some of them (some people are
going to need
much more than the RDA). This data was collected using normal healthy
males.
NHANES III has data for over 30,000 Americans and this is considered
to be very
good data because of the sample size. But I suspect based on your
observations
that even this is not enough. If 1 in every 1,000 humans has a
genetic profile
that makes them need much more folic acid than the RDA, even this hugh
sample
size may not pick them up.
What we need for humans is exactly what we did for corn, accurately
identify the range of fatty acid profile and vitamin E profile. To my
knowledge we
have never done this for anything other than nitrogen balance.
I've read most of the RDA reports where the data is reviewed to try to
establish human variability in terms of essential nutrient
requirements and
it's a guess. Even when the hard data is there like it is for
individual
essential amino acid requirements, the range is not considered, only
the mean is used.
It's tempting to say that if some humans really had such a high
requirement for
specific essential nutrients, they would never have been part of the
human gene pool because supplements were not available as the human
species evolved. It sounds nice but it's probably not true, all you
have to do is survive long enough to reproduce and even if you die
young, others in the clan will probably raise your offspring.
The human genome project holds tremendous potential for finally
getting a handle on human genetic variability but I'm never going to
live long enough to see this converted into essential nutrient
variability. I've got to try to remember that when someone says that
they need to do a certain thing to make them feel better that they may
be the 1 in 1,000 that really does need to do it because their genes
set them up for a much higher essential nutrient requirement. I am
going to get my cheat sheet into the college website.
This is what I handed out last Friday to try to keep my students from
getting themselves in trouble if they wanted to try alternative
medicine. But even here the upper limit may not apply to every human.
Max Watt
> A study of a million people isn't necessary and is a waste of resources, in
> much the same way you don't need to ask a million people a question to get a
> good Gallup Poll result. If you want your sampling results within 1/2 % of
> true, all you need is a couple of thousand samples. Anything more is just
> making up for inhomogeneity.
>
>
It can be even worse than that. Too large a sample can find
differences that don't really exist.
If you sampled the entire population, you might find a 1% occurence of
some phenonomena, that is actually due to random fluctuation, not
because of the actual existence of the phonemona at that level of
certainty.
Herman Rubin
You are more likely to find random variations in a small
study. A larger source of random variations in demographic
studies is that the various groups are quite different
genetically or environmentally.
Where a large sample is needed is to find something rare.
Liver problems with Rezulin were only found after the drug
was in wide use; the incidence of serious problems seems to
be about one in 50,000, and one needs more than one case to
justify condemning the drug. So how many years do we have
to wait to get some drug? In fact, this particular drug
would still be on the market if alternatives had not come
out; they have other side effects.
--
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, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
hru...@stat.purdue.edu Phone: (765)494-6054 FAX: (765)494-0558
Brandon Van Every
"Steve Harris" <SBHar...@ix.netcom.com> wrote in message
news:a4t5er$8e6$1...@slb7.atl.mindspring.net...
>
in
> much the same way you don't need to ask a million people a question to get
a
> good Gallup Poll result. If you want your sampling results within 1/2 % of
> true, all you need is a couple of thousand samples. Anything more is just
> making up for inhomogeneity.
Why do you accept that as fact as opposed to a statistical leap of faith?
Jay Tanzman
Brandon Van Every wrote:
>
> "Steve Harris" <SBHar...@ix.netcom.com> wrote in message
> news:a4t5er$8e6$1...@slb7.atl.mindspring.net...
> >
> > A study of a million people isn't necessary and is a waste of resources,
> in
> > much the same way you don't need to ask a million people a question to get
> a
> > good Gallup Poll result. If you want your sampling results within 1/2 % of
> > true, all you need is a couple of thousand samples. Anything more is just
> > making up for inhomogeneity.
>
> Why do you accept that as fact as opposed to a statistical leap of faith?
Because it _is_ a fact that has ben proven mathematically, or if you prefer, can
be demonstrated by computer simulation. The fact that you have not studied
statistics might mean that _you_ have to take it on faith, but that is only
because you haven't studied the subject.
-Jay
Brandon Van Every
"Max Watt" <maxwa...@yahoo.com> wrote in message
news:870a5d01.0202...@posting.google.com...
>
> It can be even worse than that. Too large a sample can find
> differences that don't really exist.
>
> If you sampled the entire population, you might find a 1% occurence of
> some phenonomena, that is actually due to random fluctuation, not
> because of the actual existence of the phonemona at that level of
> certainty.
In 3D graphics benchmarking I don't get excited about anything unless I see
at least a 20% performance improvement. Are you saying that nutritional
studies are going to town with "provables" in the 1% range? I'd say that's
scientifically silly, any signal theorist will tell you it's noise.
Ergo, if you're not trying to fabricate silly details that don't exist, I
don't see why larger and larger samples are "worse." They will give you
more certainty about whatever shows up as a strong signal. They may prove
that other studies were guilty of undersampling.
Brandon Van Every
"Jay Tanzman" <jtan...@sph.llu.edu> wrote in message
news:3C72A591...@sph.llu.edu...
> Brandon Van Every wrote:
> > "Steve Harris" <SBHar...@ix.netcom.com> wrote in message
> > news:a4t5er$8e6$1...@slb7.atl.mindspring.net...
> > >
> > > A study of a million people isn't necessary and is a waste of
resources,
> > in
> > > much the same way you don't need to ask a million people a question to
get
> > a
> > > good Gallup Poll result. If you want your sampling results within 1/2
% of
> > > true, all you need is a couple of thousand samples. Anything more is
just
> > > making up for inhomogeneity.
> >
> > Why do you accept that as fact as opposed to a statistical leap of
faith?
>
> Because it _is_ a fact that has ben proven mathematically, or if you
prefer, can
> be demonstrated by computer simulation.
Pfffffffffshhh bring it *on*, baby! Show me your software, now you're
talking my language.
> The fact that you have not studied
> statistics might mean that _you_ have to take it on faith, but that is
only
> because you haven't studied the subject.
This is because you've studied telepathy?
Jay Tanzman
Brandon Van Every wrote:
>
> "Jay Tanzman" <jtan...@sph.llu.edu> wrote in message
> news:3C72A591...@sph.llu.edu...
> > Brandon Van Every wrote:
> > > "Steve Harris" <SBHar...@ix.netcom.com> wrote in message
> > > news:a4t5er$8e6$1...@slb7.atl.mindspring.net...
> > > >
> > > > A study of a million people isn't necessary and is a waste of
> resources,
> > > in
> > > > much the same way you don't need to ask a million people a question to
> get
> > > a
> > > > good Gallup Poll result. If you want your sampling results within 1/2
> % of
> > > > true, all you need is a couple of thousand samples. Anything more is
> just
> > > > making up for inhomogeneity.
> > >
> > > Why do you accept that as fact as opposed to a statistical leap of
> faith?
> >
> > Because it _is_ a fact that has ben proven mathematically, or if you
> prefer, can
> > be demonstrated by computer simulation.
>
> Pfffffffffshhh bring it *on*, baby! Show me your software, now you're
> talking my language.
It's hardly _my_ software. Go buy SAS and I'll gladly write you a program to do
the simulation.
> > The fact that you have not studied
> > statistics might mean that _you_ have to take it on faith, but that is
> only
> > because you haven't studied the subject.
>
> This is because you've studied telepathy?
How to measure the precision of a statistical estimate is taught within the
first few weeks of any introductory statistics course. Your earlier comment
showed that you didn't understand this most basic of statistical concepts,
implying that you never studied statistics or at least never grasped the
concepts.
-Jay
Max Watt
Experimental errors of the first type (missing a difference because it
doesn't reach significance due to small sample size) are more common
than errors of the second type, where due to too large a sample size,
a random fluctuation reaches statistical significance. Bigger sample
size is not necessarily better, though it will be more expensive. In
a statistical psychology class, they beat this into your head, because
most people intuitively assume bigger sample size is better. It is
not.
Steve Harris
news:PDxc8.685$ZC3....@newsread2.prod.itd.earthlink.net...
>
> "Steve Harris" <SBHar...@ix.netcom.com> wrote in message
> news:a4t5er$8e6$1...@slb7.atl.mindspring.net...
> >
> > A study of a million people isn't necessary and is a waste of resources,
> in
> > much the same way you don't need to ask a million people a question to
get
> a
> > good Gallup Poll result. If you want your sampling results within 1/2 %
of
> > true, all you need is a couple of thousand samples. Anything more is
just
> > making up for inhomogeneity.
>
> Why do you accept that as fact as opposed to a statistical leap of faith?
Analysis of variance is a topic in any beginning book on statistics. Pull
yours off the shelf and have a look. If you don't have one, you know the
basic problem in this conversation.
If you want to write a program to prove it to yourself nonmathematically,
make yourself up a dataset which contains a known and exactly fixed
proportion of X vs Y (like 61%/39%), and then see how many single random
samples (i.e, random picks) from it, that it takes in order to get a
population of picks which reflects the correct makeup of the sampled
population, to within 1/2% of true. It's in the low thousands, not
millions.
SBH
liaM
Oops.
Jay Tanzman
You are so confused I don't even know where to begin to correct what you've
written above.
> Bigger sample
> size is not necessarily better, though it will be more expensive.
Bigger sample size is always more precise (all else equal). There is no
statistical disadvantage to large sample sizes.
> In
> a statistical psychology class...
OK, that explains the source of the confusion.
> ...they beat this into your head, because
> most people intuitively assume bigger sample size is better. It is
> not.
OK, I see one thing you're confused about: the difference between statistical
and clinical significance. You seem to think that a big sample size is bad
because an unimportantly small effect might be statistically significant. So
what if it is? Statistical and clinical significance are entirely different
concepts. Just because an effect is statistically significant does not mean
that it is clinically important, and vice versa.
-Jay
Brandon Van Every
>
> How to measure the precision of a statistical estimate is taught within
the
> first few weeks of any introductory statistics course. Your earlier
comment
> showed that you didn't understand this most basic of statistical concepts,
> implying that you never studied statistics or at least never grasped the
> concepts.
Did you skip all those liberal arts courses on belief systems? Notice I'm
not telepathically assuming, I'm asking.
Brandon Van Every
>
> Experimental errors of the first type (missing a difference because it
> doesn't reach significance due to small sample size) are more common
> than errors of the second type, where due to too large a sample size,
> a random fluctuation reaches statistical significance. Bigger sample
> size is not necessarily better, though it will be more expensive. In
> a statistical psychology class, they beat this into your head, because
> most people intuitively assume bigger sample size is better. It is
> not.
Actually, it is. But statistical psychologists would be frightened to
realize that they'd need millions of samples per person, per day, for many
interacting people. You see, without all the lies and collective illusions
about what statistics do and don't represent, you'd break a lot of careers.
People would have to admit that their studies don't contain information.
You have valid statistical information only when the sample sizes and
methodologies happen to correspond to a sufficiently stable and appropriate
phenomenon. There is no *mathematical* way to prove such appropriateness,
because the appropriateness depends enitrely upon the real world physical
phenomenon, not your models for it. You simply test and test and test over
and over again until you're satisfied in pragmatic terms that you don't need
to explore anymore. For instance, Newtonian physics works great until you
put it under extreme conditions.
Brandon Van Every
>
> Analysis of variance is a topic in any beginning book on statistics. Pull
> yours off the shelf and have a look. If you don't have one, you know the
> basic problem in this conversation.
Well, the problem with variance is that you can only know if you're
sufficiently sampling something, i.e. you can reconstruct it flawlessly from
samples, is if you actually know exactly what's varying in the first place.
In the absence of exact knowledge of the domain, all you're saying is "gee
we have this common language we like to talk about called 'statistics', here
is some data described in the language of statistics." It is a language,
not a fact.
> If you want to write a program to prove it to yourself nonmathematically,
> make yourself up a dataset which contains a known and exactly fixed
> proportion of X vs Y (like 61%/39%),
In experimental science you are working with the unknown. I might as well
make models out of clay and say "these shapes I have sculpted sorta resemble
the paths these electrons seem to be tracing out. Maybe the shape of my
sculpture is a theory!" Now how do you see your sculpture? Elegant?
Lopsided? Are you tempted to see it as more elegant than the data says?
Are you tempted to ignore bits of the sculpture you don't like?
> and then see how many single random
> samples (i.e, random picks) from it, that it takes in order to get a
> population of picks which reflects the correct makeup of the sampled
> population, to within 1/2% of true. It's in the low thousands, not
> millions.
BTW your example doesn't work. You have completely ignored the size of the
points and the size of the sample filters. I can pick any filter sizes I
want, pick any grid resolutions I want, and thereby generate any number I
want to yield that "1/2% of true."
You say statistics always picks a specific point and filter size, by
convention? Well that's great, but has nothing to do with the reality of
unknowns.
Steve Harris
> Well, the problem with variance is that you can only know if you're
> sufficiently sampling something, i.e. you can reconstruct it flawlessly
from
> samples, is if you actually know exactly what's varying in the first
place.
> In the absence of exact knowledge of the domain, all you're saying is "gee
> we have this common language we like to talk about called 'statistics',
here
> is some data described in the language of statistics." It is a language,
> not a fact.
It's neither objective fact nor merely subjective agreement-- rather it's a
probabilistic estimate. Like tomorrow's weather forcast, it's neither an
analytic nor synthetic fact. But that's true of all inductive conclusions.
And so what? Such conclusions hold you up when you take a jet airliner. If
you're willing to bet your life on them, as you do whenever you drive, why
are you disrespecting them?
> In experimental science you are working with the unknown. I might as well
> make models out of clay and say "these shapes I have sculpted sorta
resemble
> the paths these electrons seem to be tracing out. Maybe the shape of my
> sculpture is a theory!" Now how do you see your sculpture? Elegant?
> Lopsided? Are you tempted to see it as more elegant than the data says?
> Are you tempted to ignore bits of the sculpture you don't like?
Elegance and simplicity are only the tiebreaker criteria when competing
theories are equally good at predicting the future. Don't get your
scientific scoring system mixed up.
> BTW your example doesn't work. You have completely ignored the size of
the
> points and the size of the sample filters. I can pick any filter sizes I
> want, pick any grid resolutions I want, and thereby generate any number I
> want to yield that "1/2% of true."
I used a computer model since you said you programmed. You can do it equally
well with a fishbowl containing 61 white marbles and 39 black ones. Shake,
pick one blindfold, record it, and then put it back (or if you like, do it
with a swimming pool containing 39 million black marble and 61 million white
ones, and then you don't have to put any back). How many picks do you need
to do this before the ratio of blacks to totals comes withing 1/2% of 39?
95% of the time, less than 2000. 99% of the time, less than 10,000. And so
on. Filter size has nothing to do with it because you're looking at
something which is inherently of a particular grain. If you voted for
president, you either voted for Bush or you didn't. Some thing are naturally
quantized at a given level.
Steve Harris
news:FhVc8.3417
> Did you skip all those liberal arts courses on belief systems? Notice I'm
> not telepathically assuming, I'm asking.
Whereas I assume that you took them all. Probably from profs who said they
don't hold to the idea of an objective reality, but yet still continue to
hold drivers licenses, and live. Strange to say. One imagines that they may
not be acting according to their stated beliefs.
Brandon Van Every
> "Brandon Van Every" <vane...@3DProgrammer.com> wrote in message
> news:FhVc8.3417
>
> > Did you skip all those liberal arts courses on belief systems? Notice
I'm
> > not telepathically assuming, I'm asking.
>
> don't hold to the idea of an objective reality,
Geez by 32 years of age aren't I allowed to think for myself by now? I'd
hate to think that your image of education is one of collegiate pedigree
owing totally to what one studied as an undergrad. Come to think of it, I
thought for myself back then. And furthermore I'd like to say that The
Stumbling Monk in Capitol Hill is the best darned pub in Seattle. The Big
Time Brewery being a pretty good second beer-wise, if not atmospherically.
Brandon Van Every
>
> It's neither objective fact nor merely subjective agreement-- rather it's
a
> probabilistic estimate.
Ok we agree. Great!
> And so what? Such conclusions hold you up when you take a jet airliner.
Understanding the real world error bounds, the limits of current knowledge,
and a pragmatic lack of faith will help you make good decisions about what
will hold you up and what won't.
> If
> you're willing to bet your life on them, as you do whenever you drive, why
> are you disrespecting them?
I don't bet my life on driving. I bet my life on my *skill* at driving. If
I believed that statistics were the main thing, I'd believe that the average
asshole would kill me. Whereas I know that the average asshole is
incompetent and not me. I know that there are risks I can defend against.
Habits that will increase my chances. A freak accident can still take you
out, but there's no particular reason that any of us have to live as
statistics.
> > BTW your example doesn't work. You have completely ignored the size of
> the
> > points and the size of the sample filters. I can pick any filter sizes
I
> > want, pick any grid resolutions I want, and thereby generate any number
I
> > want to yield that "1/2% of true."
>
> I used a computer model since you said you programmed. You can do it
equally
> well with a fishbowl containing 61 white marbles and 39 black ones.
Shake,
> pick one blindfold, record it, and then put it back (or if you like, do it
> with a swimming pool containing 39 million black marble and 61 million
white
> ones, and then you don't have to put any back). How many picks do you need
> to do this before the ratio of blacks to totals comes withing 1/2% of 39?
You want to talk about how much time you have to make those picks? Herein
lies the basis of cryptography. Don't go there.
You want to talk about whether any given reality is a bowl of 39 million
marbles, or 61 million marbles, or 1e10^17 marbles? We already discussed
that. You were wrong. If you have a lot of real world experiments to prove
that you were right, that you really do know the problem domain and can
reconstruct it faithfully with X number of samples, then you're right. In
the absence of that, you're wrong.
> 95% of the time, less than 2000.
Gee even 1 in 20 is a bitch if it means death.
> Filter size has nothing to do with it because you're looking at
> something which is inherently of a particular grain. If you voted for
> president, you either voted for Bush or you didn't. Some thing are
naturally
> quantized at a given level.
Prove that you know what this natural filter size is a priori. You cannot.
Max Watt
> "Max Watt" <maxwa...@yahoo.com> wrote in message
> news:870a5d01.02021...@posting.google.com...
> >
> > Experimental errors of the first type (missing a difference because it
> > doesn't reach significance due to small sample size) are more common
> > than errors of the second type, where due to too large a sample size,
> > a random fluctuation reaches statistical significance. Bigger sample
> > size is not necessarily better, though it will be more expensive. In
> > a statistical psychology class, they beat this into your head, because
> > most people intuitively assume bigger sample size is better. It is
> > not.
>
> Actually, it is.
Not always. It depends on what you are trying to measure, and with
what level of confidence.
> But statistical psychologists would be frightened to
> realize that they'd need millions of samples per person, per day, for many
> interacting people. You see, without all the lies and collective illusions
> about what statistics do and don't represent, you'd break a lot of careers.
> People would have to admit that their studies don't contain information.
>
> You have valid statistical information only when the sample sizes and
> methodologies happen to correspond to a sufficiently stable and appropriate
> phenomenon. There is no *mathematical* way to prove such appropriateness,
> because the appropriateness depends enitrely upon the real world physical
> phenomenon, not your models for it.
That's what a chi squared test is for, comparing distributions in two
groups.
>You simply test and test and test over
> and over again until you're satisfied in pragmatic terms that you don't need
> to explore anymore.
This methodology reminds me of the engineer's proof that all odd
numbers are prime:
Obviously, one is prime. Three is prime. Five is prime. Seven is
prime. Nine, um, we'll come back to nine. Eleven and thirteen are
prime. Obviously, all odd numbers are prime, nine was just
experimental error.
(Not that he didn't just use too small a sample, he chose a biased
sample. Satisfaction should have nothing to do with it.)
Brandon Van Every
> >
> > You have valid statistical information only when the sample sizes and
> > methodologies happen to correspond to a sufficiently stable and
appropriate
> > phenomenon. There is no *mathematical* way to prove such
appropriateness,
> > because the appropriateness depends enitrely upon the real world
physical
> > phenomenon, not your models for it.
>
> That's what a chi squared test is for, comparing distributions in two
> groups.
All that does is say you've tested 2 groups and that they differ according
to some linguistic convention defined by statistics. Says nothing about
whether only 2 groups are needed to converge to an accurate model of a real
phenomenon.
> >You simply test and test and test over
> > and over again until you're satisfied in pragmatic terms that you don't
need
> > to explore anymore.
>
> This methodology reminds me of the engineer's proof that all odd
> numbers are prime:
>
> Obviously, one is prime. Three is prime. Five is prime. Seven is
> prime. Nine, um, we'll come back to nine.
Engineers can divide by 3. Since you can't test whether a number is prime
except by doing division, and this is a trivial division at the beginning of
the dataset, your example is a strawman. The truth is, Engineers know
whether each of the first few hundred numbers are prime or not, because it's
easy enough to test them directly. You should argue about what Engineers do
when the computation gets more and more expensive. At some point they're
going to say "Dude, why do you give a rip? Is this going to help us build a
building or something? We don't have time for this, we've got contracts to
finish and kids to feed."
> Eleven and thirteen are
> prime. Obviously, all odd numbers are prime, nine was just
> experimental error.
Some engineers would call for more tests. In fact, that is my position in
the argument above. You're the one saying we get to stop at 13, because the
chi statistical crapola said so.
> (Not that he didn't just use too small a sample, he chose a biased
> sample. Satisfaction should have nothing to do with it.)
Satisfaction has everything to do with it, because it is only a process of
either peer review or the individual bravado of ego that will bring about an
end to the testing. We do not contain a priori knowledge of domain
knowability! Yet you act like statistics contains this thing. It does not,
statistics only makes comparisons.
Max Watt
> "Max Watt" <maxwa...@yahoo.com> wrote in message
> news:870a5d01.02022...@posting.google.com....
> > >
> > > You have valid statistical information only when the sample sizes and
> > > methodologies happen to correspond to a sufficiently stable and
> appropriate
> > > phenomenon. There is no *mathematical* way to prove such
> appropriateness,
> > > because the appropriateness depends enitrely upon the real world
> physical
> > > phenomenon, not your models for it.
> >
> > That's what a chi squared test is for, comparing distributions in two
> > groups.
>
> All that does is say you've tested 2 groups and that they differ according
> to some linguistic convention defined by statistics. Says nothing about
> whether only 2 groups are needed to converge to an accurate model of a real
> phenomenon.
Which is all one ever can do. You have focused on the basic paradox
of perception that is the basis of the philosophical school of
phenomenology. One can never perceive the thing itself, only one's
perceptions (which are a kind of model.)
>
> > >You simply test and test and test over
> > > and over again until you're satisfied in pragmatic terms that you don't
> need
> > > to explore anymore.
> >
> > This methodology reminds me of the engineer's proof that all odd
> > numbers are prime:
> >
> > Obviously, one is prime. Three is prime. Five is prime. Seven is
> > prime. Nine, um, we'll come back to nine.
>
You missed the point of my joke, below, though I admit it has other
interpretations. My point is that "being satisfied" is not enough.
Really, for those of us who aer not divine beings, the best we can do
is stumble along in the dark until we bump into something not in our
"model", then adjust the model. Nor is repetition (analagous to
increasing the sample size) a guarantee of accuracy of the model, and
it can lead to a false sense of confidence.
But we've strayed from science into philosphy, which is where such
discussions should continue.