Highest Posterior Density
AB
how can I find an HPD using a statistical software (e.g. Mathematica)?
Thanks in advance
Reef Fish
By HPD do you mean VALUE of the variable that corresponds to
the highest point of the posterior density in a problem using a
Bayesian method of inference?
You need to be able to supply the Bayesian ingredients of a prior
distribution for your parameter, be able to perform the integration to
obtain your posterior distribution of your PARAMETER (which is
the random variable in your posterior density). Most statistical
software should have some method for finding the maximum of a
function -- that part should be easy. You use it to find the value
of theta (the parameter) that maximized the posterior density.
-- Reef Fish Bob.
illywhacker
> AB wrote:
> > how can I find an HPD using a statistical software (e.g. Mathematica)?
>
> the highest point of the posterior density in a problem using a
> Bayesian method of inference?
>
> You need to be able to supply the Bayesian ingredients of a prior
> distribution for your parameter, be able to perform the integration to
> obtain your posterior distribution of your PARAMETER
Actually, you do not need to perform the integration as the
normalizing constant does not depend on the parameter you
wish to estimate, and hence does not affect the
maximization.
> Most statistical
> software should have some method for finding the maximum of a
> function -- that part should be easy.
Whether the optimization is easy very much depends on the
dimensionality of the parameter space on which the density
is defined, and on the complexity of the density itself. It
can be an almost impossible task to find a global (as
opposed to local) maximum in a high-dimensional space,
hence the large amount of research devoted to this topic.
There is also a conceptual difficulty with MAP (HPD)
estimation. While Bayes' theorem supplies you with a
measure on the parameter space, in order to perform the
maximization you need a function, i.e. a density. You
therefore need an underlying measure with respect to which
you can define the density of the posterior. The problem is
that different underlying measures give rise to different
estimates. (A similar thing happens with MMSE estimates
too.) Since the most common procedure is simply to drop the
'd\theta' symbol in the density (\theta is the parameter),
i.e. to use 'Lebesgue measure' in the current coordinate
system on the parameter space as the underlying density,
different choices of coordinates, e.g. using \theta^{2}
instead of \theta for a positive parameter, lead to
different estimates. In order to get around this, it is
necessary to choose an underlying density that is invariant
to the choice of coordinates on the parameter space.
The most sensible choice is to use a Riemannian metric to
define the underlying measure, since a metric generates
coherent notions of volume (MAP estimates) and distance
(MMSE estimates). The question is then, how to choose the
metric? If we are just given a density on some space, there
seems to be no good way to do this. However, in the context
of Bayes' theorem, there are good arguments for choosing
the Fisher information metric constructed from the
likelihood, as this is the only choice that does not
introduce extra information. In the case that the prior is
Jeffreys' prior (the volume element of the Fisher
information metric), this leads to maximum likelihood as
the estimation method.
To return to the OP's question: the first thing to know is
that there is no algorithm that will guarantee to find the
global maximum for an arbitrary function. The difficulty of
your problem largely depends first, on the dimensionality
of your space, and second, on the complexity of your
density.
Matlab has an fminsearch function that looks for an
unconstrained minimum of a multivariable function, but may
only return a local minimum. (Obviously finding a maximum
is the same as finding a minimum of the negative of the
function.) The Matlab Optimization toolbox provides
fminunc, which addresses the same problem, as well as other
optimization tools.
Mathematica has the functions NMinimize and FindMinimum,
with similar goals.
There is no real substitute, though, for analysing your
function and trying to find out as much as you can about
it, and then writing your own code. Many optimization
algorithms are not hard to implement (gradient descent,
simulated annealing), and can often provide good results
with all your assumptions known, because you wrote the
code.
illywhacker;
illywhacker
See my reply to Reefer.
illywhacker;
Anon.
> Reef Fish wrote:
>> AB wrote:
>>> how can I find an HPD using a statistical software (e.g. Mathematica)?
>> By HPD do you mean VALUE of the variable that corresponds to
>> the highest point of the posterior density in a problem using a
>> Bayesian method of inference?
>>
>> You need to be able to supply the Bayesian ingredients of a prior
>> distribution for your parameter, be able to perform the integration to
>> obtain your posterior distribution of your PARAMETER
>
> Actually, you do not need to perform the integration as the
> normalizing constant does not depend on the parameter you
> wish to estimate, and hence does not affect the
> maximization.
>
parameters, in which case some integration is needed (albeit not over
the normalising constant).
I think we really need the OP to tell us more about the problem to give
good suggestions.
Bob
--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland
Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB: www.jnr-eeb.org
illywhacker
> It's also possible that the OP wants to marginalise over nuisance
> parameters
Perhaps, but you would have thought (s)he would have mentioned it were
it the case. Anyway, my reply was to Bob, who at least in my reading
was not talking about marginalization.
> I think we really need the OP to tell us more about the problem to give
> good suggestions.
I agree.
illywhacker;
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
> > AB wrote:
> > > how can I find an HPD using a statistical software (e.g. Mathematica)?
> >
> > By HPD do you mean VALUE of the variable that corresponds to
> > the highest point of the posterior density in a problem using a
> > Bayesian method of inference?
> >
> > You need to be able to supply the Bayesian ingredients of a prior
> > distribution for your parameter, be able to perform the integration to
> > obtain your posterior distribution of your PARAMETER
>
> Actually, you do not need to perform the integration as the
> normalizing constant does not depend on the parameter you
> wish to estimate, and hence does not affect the
> maximization.
Your comments in this post clearly indicated that: (1) You are NOT
a Bayesian; (2) You don't know what Bayesian Statistics is about;
(3) you've NEVER assessed a proior distribution that reflects your
own opinionor belief about a parameter; and (4) you have never
carried out a Bayssian analysis in an applied and realistic way.
Your comments are of the type that might impress someone like
yourself who dosn't know anything about Bayesian statistics, and
all they know are some buzz words in the discipline. I'll be explcit
about the various ERRORS of yours in this post, about your use
of Bayesian buzz words.
Your first MAJOR error is in the paragraph above, about "you do
not need to perform the integration". Except for the use of
conjugate priors (which are seldom realistic for the personal prior
of anyone) or diffuse priors in stable estimation when the likelihood
function is sharp, you always need to integrate the product of the
prior and the likelihood function to get the posterior distribution.
While it is true that the posterior distribution needs to be known
only up to a proportionality constant (which is why the proportionality
constants in a joint likelihood function can be discarded before the
integration), but you always have to integrate! That's why Robert
Schlaifer wrote an entire BOOK providing numerical methods for
assessing (AND INTEGRATING) mostly univariate prior distributions
that are NOT of the conjugate type but are needed to reflect one's
true prior beliefs.
http://links.jstor.org/sici?sici=0162-1459(197312)68%3A344%3C1023%3ACPFEDA%3E2.0.CO%3B2-%23
Computer Programs for Elementary Decision Analysis
He is one of a very few Bayesians who has even done that, and I
applauded him heartily in my review of his book for JASA.
Beyond ONE dimention, I have not seen any Bayesian carefully
construct or argue why his prior distribution is represented by the
mathematical form for the parameter, in a bivariate case, or one
in a higher dimension. Those who do that kind of "Bayesian
Statistical Analysis" are mostly non-Bayesians, pseudo-Bayesians,
or just "mathematical statisticians" who can sling a few matrices
around without having the slightest idea of what they are doing, in
terms of representing his own PRIOR beliefs!
Thus, my comments to the OP were general (for ALL dimensions)
but sufficiently non-technical to cover all problems, but thinlking
mostly in a posterior distribution of ONE (perhaps two) dimensions.
>
> > Most statistical
> > software should have some method for finding the maximum of a
> > function -- that part should be easy.
>
> Whether the optimization is easy very much depends on the
> dimensionality of the parameter space on which the density
> is defined, and on the complexity of the density itself. It
> can be an almost impossible task to find a global (as
> opposed to local) maximum in a high-dimensional space,
> hence the large amount of research devoted to this topic.
Give me an example of a THREE dimensional PRIOR
distribution used by any Bayesian, that is of the kind that is
expressed in a mathematical form that requires a careful
elicitation to arrive that prior, and invariably would require
numerical integration because the product of the prior and
likelihood function is not in a form integrable by usual
mathematical integration methods. You can't even find
a TWO dimension PRIOR of that kind without the user
resorting to a conjugate prior because that's the ONLY
form that can be easily integrated, but unfortunately NOT
because they argued the form realistically reflected their
own prior.
THIS are the basic elements few Bayesian ever struggled
past to make any useful or realistically person Bayesian
inference. That's the inherent DEFECT of Bayesian
statistics -- the difficult in the EXECUTION, though perfect
in theory and foundation of how statistics should be done.
>
> There is also a conceptual difficulty with MAP (HPD)
> estimation. While Bayes' theorem supplies you with a
> measure on the parameter space, in order to perform the
> maximization you need a function, i.e. a density.
WHY would you need a density function? I thought you had
understood that a posterior distribution has to be known only
up to a proportionality constant. The posterior distribution,
whether it is a genuine density function (integrable to 1) or
just proportional to a density function, the MAXIMUM of the
function (or the parameter) is generally easy to find. For
one or two dimensions a many discrete search methods in
numerical analysis would apply. The function does not even
need to be continuous or integrable.
What follows below is complete utter nonsense of
"mathematistry". The finding of a maximum will be in the
Euclidean one- or two-dimensional space of the REAL WORLD,
where measure and integration does not come into play --
not even calculus is necessary.
Priors based on Fisher's information matrix, or Jeffrey's and
other kinds of uninformative prior, are precisely the kind of
EXCUSES used by non-Bayesians and pseudo-Bayesians
to justify certain aspects of NON-Bayesian inference, such
as that of MLE, by arguing that they don't know ANYTHING
about the parameters on which they are making statistical
inference.
That is the opposite extreme of a TRUE Bayesian who
incorporates whatever he knows about the parameters (in
his prior distribution) so that his inference depends not only
on the likelihood function from the DATA, but how it blends
in with his own prior information or beliefs.
That is what Bayesian Statistics is all about.
It is NOT an exercise in "Mathematistry" or pretending that
one knows NOTHING about the parameters.
>
> To return to the OP's question: the first thing to know is
> that there is no algorithm that will guarantee to find the
> global maximum for an arbitrary function. The difficulty of
> your problem largely depends first, on the dimensionality
> of your space, and second, on the complexity of your
> density.
You continue to bark up the wrong tree.
>
> Matlab has an fminsearch function that looks for an
> unconstrained minimum of a multivariable function, but may
> only return a local minimum. (Obviously finding a maximum
> is the same as finding a minimum of the negative of the
> function.) The Matlab Optimization toolbox provides
> fminunc, which addresses the same problem, as well as other
> optimization tools.
>
> Mathematica has the functions NMinimize and FindMinimum,
> with similar goals.
These are all standard mathematical methods in optimization
in NUMERICAL ANALYSIS that is the least of the problems in
performing any Bayesian analysis and inference.
>
> There is no real substitute, though, for analysing your
> function and trying to find out as much as you can about
> it, and then writing your own code. Many optimization
> algorithms are not hard to implement (gradient descent,
> simulated annealing), and can often provide good results
> with all your assumptions known, because you wrote the
> code.
>
> illywhacker;
More buzz words in numerical analysis and numerical
optimization, while MISSING the entire BOAT of Bayesian
Statistical inference and analysis.
-- Reef Fish Bob.
illywhacker
You are so scared that you may be forced to make a mental
effort and disturb years of rigidified thinking, that you
do everything you can to avoid dealing with Bayesian
methods. Your technique for doing this is admirably subtle,
however. You pick a particular Bayesian school (one for
which you can, if pressed, use worthless but impressive
arguments based on prestige, since you knew one or more of
the people involved in its founding), and then attack
everyone else for failing to conform to it, or at least
your vision of it. Now of course, those who use Bayesian
methods in practice cannot conform perfectly to such
religious dogma because they deal with the 'real world', as
you call it. In this way you can brand everyone who uses
Bayesian methods as a fake, and thus effectively dismiss
them and safeguard your indolence. Should you use this
technique if you choose to reply to this post, everyone
will notice.
Now to your specific points. Your post is so long and
rambling that it is pointless to answer it line by line in
the time-honoured Usenet fashion. Of course, my reply
assumes intellectual honesty about what you intended to
say, something that has been noticeably lacking in your
previous discussions. Still, we can but hope.
Yours points are the following:
-------------------------------
1) Computation of the posterior probability distribution,
or its density, requires calculating the normalizing
constant.
2) Robert Schlaifer is one of a very few Bayesians who has
been concerned with numerical integration for posteriors.
3a) No one uses priors that you consider worthwhile in more
than one dimension.
3b) All `real world' problems are low-dimensional.
4) Ed Jaynes (or anyone with a similar point of view) was
not a Bayesian.
5) That my points about invariance are wrong, and indeed
mere 'mathematistry'. (Actually, you simply misunderstood
what I was saying. I explain in more detail below.)
6) That the OP's question concerned trivialities because
"optimization <...> is the least of the problems in
performing any Bayesian analysis and inference".
-------------------------------
Now here are my replies.
-------------------------------
A1) Correct. But you do not need to calculate the
normalizing constant if all you want to do is to maximize
the posterior density, which was what the OP wanted to do.
Since performing the integration is often very hard, it
seems pointless to recommend it as a first step when it is
not required. It is rather like saying to someone who
wishes to change the tire on a car, 'first take out the
engine'.
A2) This is incorrect, even absurd. A great deal of
Bayesian research is devoted to numerical methods for
integration.
A3a) Since we do not need to accept your particular version
of `Bayesian', we do not need to pay attention to this
point.
A3b) Again, this is incorrect, even absurd. There are many
problems in the 'real world' as you call it, where the
space of parameters has a very large number of dimensions.
Signal processing provides a large class of examples, but
there are many others.
A4) This is merely a matter of definition and hence in
itself is not worth considering. However, what is worth
considering is that although the incorporation of prior
knowledge is characteristic of Bayesian approaches, the
most characteristic point, the one that makes the most
difference, is rather the acknowledgement that
probabilities have nothing to do with randomness, and that
therefore probabilities for hypotheses can be considered,
even though the hypothesis is not what an orthodox
statistician (such as yourself) would call a 'random
variable'.
A5) I am as against mathematistry as you are, Bob, but the
problem of which I speak remains. You ask why one needs a
density function (note I did not say probability density
function - normalization is indeed irrelevant). The problem
is that Bayes' theorem does not give you a function; it
gives you a measure, e.g.
d\theta f(\theta) .
To maximize you need a function, not a measure, so you have
to divide by another measure. Most of the time, people
simply drop the d\theta and maximize f (i.e. divide by the
measure d\theta). But suppose a second person was using a
different coordinate to described the parameter, e.g.
\alpha = \theta^{2}. They would construct the same
posterior measure in the different coordinates:
d\alpha g(\alpha) ,
where f(\theta) = 2\theta g(\theta^{2}). Now this person
follows the same procedure, drops d\alpha, and maximizes g.
They will therefore not find the same estimate, in the
sense that \alpha* = \theta*^{2}. Clearly, however they
should find the same estimates in this sense, as all their
information is the same.
A6) Had you ever addressed a high-dimensional inference
problem, you would never suggest that optimization is the
least of the problems involved. However, I do agree that
built-in functions in Matlab or Mathematica are not the
best way to address these problems, as I said in my post.
illywhacker;
Reef Fish
illywhacker wrote:
> Dear Bob,
>
> You are so scared that you may be forced to make a mental
> effort and disturb years of rigidified thinking, that you
> do everything you can to avoid dealing with Bayesian
> methods.
That is not a very good way to start your excuse of my errors
about Bayesian methods. You had once claimed that the only
real Bayesians are the physicists and you learned it from them.
That says it all.
>Your technique for doing this is admirably subtle,
> however. You pick a particular Bayesian school (one for
> which you can, if pressed, use worthless but impressive
> arguments based on prestige, since you knew one or more of
> the people involved in its founding), and then attack
> everyone else for failing to conform to it, or at least
> your vision of it. Now of course, those who use Bayesian
> methods in practice cannot conform perfectly to such
> religious dogma because they deal with the 'real world', as
> you call it. In this way you can brand everyone who uses
> Bayesian methods as a fake, and thus effectively dismiss
> them and safeguard your indolence. Should you use this
> technique if you choose to reply to this post, everyone
> will notice.
I did not dismiss Robert Schlaiffer, but applauded his effort
to provide tools to make the application of Bayesian methods
more realistic and useful. I did not dismiss MOST of the
Bayesians I know personally. The only exception is Arnold
Zellner who is a "matrix slinger". He takes a multivariate
prior that is mathematically integrable, matched it with a
sample mean, and called it his own prior. That's the mind
of mindless Bayesians I speak out against.
In your case, you are deficient even in the BASICs fo
Bayesian statistics, as evidenced by the explicit exhibits
I showed in my post which you snipped, only to misquote
some in your own words in this post.
>
> Now to your specific points. Your post is so long and
> rambling that it is pointless to answer it line by line in
> the time-honoured Usenet fashion. Of course, my reply
> assumes intellectual honesty about what you intended to
> say, something that has been noticeably lacking in your
> previous discussions. Still, we can but hope.
>
> Yours points are the following:
>
> -------------------------------
> 1) Computation of the posterior probability distribution,
> or its density, requires calculating the normalizing
> constant.
That is patently false. I even specifically said you can
DISCARD the multiplcative factor in the likelihood function
because the posterior distribution needs to be determined
only to a multiplicative factor.
That was against YOUR comment that no integration is
needed because the posterior distribution can be up to a
multiplicative constnat -- for entirely the WRONG REASON.
Next time, you QUOTE me because your misrepresentation
to try to hide your own error.
>
> 2) Robert Schlaifer is one of a very few Bayesians who has
> been concerned with numerical integration for posteriors.
NOT so much for numerical integration of posteriors but for
the assessent of REALISTIC PRIORS, which necessitated
the numerical integration of the posterior. He wrote an
entire book about his programming work.
>
> 3a) No one uses priors that you consider worthwhile in more
> than one dimension.
Misquote for the 3rd consecutive time.
>
> 3b) All `real world' problems are low-dimensional.
Misquote for the 4th consecutive time. I said it's a DEFECT
of Bayesian Statistics in application that one can only deal
with ones OWN personal beliefs as a prior distribution in one
or two dimensions, and challenge you and anyone else to
show one elicited through Bayesian methods of prior elictation
by self-interrogation (publications by Savage and others) in
three or more dimensions. There is NONE.
>
> 4) Ed Jaynes (or anyone with a similar point of view) was
> not a Bayesian.
5th consecutive misquote.
He is a theoretical Bayesian whose main contribution is in
his uninformative prior, which I consider a mere evasion of
the solution of how to elicit a REAL Bayesian prior. The
indictment was the use of UNINFORMATIVE prior as an
EXCUSE to pretend one has no information about the
prior when one has plenty of prior information, just don't
know HOW to put it down in a mathematic form.
> 5) That my points about invariance are wrong, and indeed
> mere 'mathematistry'. (Actually, you simply misunderstood
> what I was saying. I explain in more detail below.)
6th consecutive misquote of what I wrote. I said nothing
about invariance. I ridiculed at your siily mention of
Riemann and Lebesque in an APPLIED Bayesian
question asked by the OP.
>
> 6) That the OP's question concerned trivialities because
> "optimization <...> is the least of the problems in
> performing any Bayesian analysis and inference".
I clarified that to mean the low dimension that are feasible
for any realistic application. You should NOT have
snipped everything I wrote, and MISREPRESENTED
every point I made.
You made 6 itemized MISQUOTES of 7 items of what I
said!
The rest of your post is just your rationalization of your
own ignorance about Beyesian theory and methods
by replying to YOUR OWN ERRONEOUS positions
about what I said.
Your replies are just as POINTLESS as those in your
original post.
You have NOT correctly represented a SINGLE POINT
I made in my post, let alone refute it. You merely
MISREPRESENTED (by NOT quoting me) my points
for your own convenience of obfuscation and excuse
about your own ignorance about Bayesian Statistics.
-- Reef Fish Bob.
Reef Fish
The 8th misquote is illywhacker misquoting illywhacker. Ed Jayes
was not mentioned in his post. It was Jeffrey and Jeffrey's
uninformative prior. My response (before going back to check
illywhacker's post was exactly correct as it stands.
I just want to cut and paste the EXACT quotes and my response
which illywhacker completely distorted to the tune of even
misquoting Janes as Jeffrey.
ill> introduce extra information. In the case that the prior is
ill> Jeffreys' prior (the volume element of the Fisher
ill> information metric), this leads to maximum likelihood as
ill> the estimation method.
RF> Priors based on Fisher's information matrix, or Jeffrey's and
RF> other kinds of uninformative prior, are precisely the kind of
EF> EXCUSES used by non-Bayesians and pseudo-Bayesians
RF> to justify certain aspects of NON-Bayesian inference, such
RF> as that of MLE, by arguing that they don't know ANYTHING
RF> about the parameters on which they are making statistical
RF> inference.
RF> That is the opposite extreme of a TRUE Bayesian who
RF> incorporates whatever he knows about the parameters (in
RF> his prior distribution) so that his inference depends not only
RF> on the likelihood function from the DATA, but how it blends
RF> in with his own prior information or beliefs.
RF> That is what Bayesian Statistics is all about.
illywhacker, you had already proven beyond a shadow of doubt
that you're NOT a Bayesian and you are deficient in your
knowledge about Bayesian Statistics in general, and applied
Bayesian statistics in particular.
Distorting MY posted view, in EIGHT consecutive times within
6 items is not going to salvage any of your complete lack of
credibility on the subject of Bayesian Statistics!
It merely added to your LACK of credibility in general, in your
post. You had been discredited by Herman Rubin and myself
in a previous thread about Bayesian statistics, in which you
were not only unaware of the most important contributors to
Bayesian statistics, but made the ridiculous claim that (note
that I am QUOTING you in your argument with Herman Rubin):
>>2) Because physicists, not statisticians, were responsible for the
>>renaissance of Bayesian thinking, after decades of useless thinking on
>>behalf of statisticians, and very much against their will.
Rubin> The renaissance of Bayesian thinking came from the idea of
Rubin> rational behavior under uncertainty and the attempts to
Rubin> apply game theory, which became a subject by itself in the
Rubin> mid 40s.
>All of the attempts to come up with priors depending on the
> distribution of the data are not Bayesian, but a very rash
> form of pseudo-Bayesian.
The above quote was by Herman Rubin, in the same post,
lecturing illywhacker, the non-Bayesian who learned his version
of Bayesian statistics from misguided physicists!
-- Reef Fish Bob.
illywhacker
confrontational and enter into tedious and unenlightening
bickering about who said what, I will take a different
and more agreeable tack, and try to find a few
(unfortunately trivial) things on which we agree, and a
couple of (non-trivial) things on which we do not agree.
1) We agree that to find a normalized posterior density,
we need to perform an integration. If we did not agree on
this, there would be nothing more to talk about.
2) We agree that in order to maximize the posterior
density, we do not need this normalization constant.
Calculating it is often difficult, and if the OP is only
interested in maximization, it is counter-productive to
suggest that the OP calculate the normalizing constant.
This is what it seemed to me you were doing, based on
your statement 'You need to be able to supply the
Bayesian ingredients of a prior distribution for your
parameter, be able to perform the integration to obtain
your posterior distribution of your PARAMETER'. Perhaps I
misinterpreted the words 'You need to be able' here.
3) We agree that many interesting problems exist in high
dimensions, and that in high dimensions, the maximization
problem is far from trivial. Indeed, it is insoluble at
the present time, except in very specific cases.
4) I do not think we agree about what priors represent. I
think that the construction of priors in high dimensions
is a worthwhile exercise. However, it is difficult, which
perhaps explains your dissatisfaction with the results.
The problems are twofold. First, it is frequently unclear
how to construct a distribution that captures certain
types of prior information. This is why model building is
a subject of research. Second, there often has to be a
compromise between accuracy and tractability. Again, this
is a question of intelligent model building. In any case,
the OP presumably has a prior in mind, so this question
need not really detain us.
5) We seem not to agree on the relevance of post high
school mathematics to probability and statistics. I
repeat that I agree with your opinion of mathematistry,
but a Riemmanian metric and Lebesgue measure are scarcely
mathematistry. Regardless of the names you use for
various quantities, the fact remains that to create a
posterior density, you have to divide the posterior
probability distribution by another measure. What people
usually do is simply to drop the symbol 'd\theta'
(assuming \theta is the argument of the posterior), which
amounts to dividing by Lebesgue measure in the \theta
system of coordinates. My point is that this procedure,
that is, to drop the d'something' symbol from a
probability distribution, is not an invariant procedure,
a point that has been emphasized since Fisher.
Cheerfully yours,
illywhacker;
Reef Fish
illywhacker wrote:
> OK Reefer. I am in a good mood, so rather than be
> confrontational and enter into tedious and unenlightening
> bickering about who said what, I will take a different
> and more agreeable tack, and try to find a few
> (unfortunately trivial) things on which we agree, and a
> couple of (non-trivial) things on which we do not agree.
>
> 1) We agree that to find a normalized posterior density,
> we need to perform an integration. If we did not agree on
> this, there would be nothing more to talk about.
I am NOT going to waste any more time on you. This was
your opening sentence before you started the rest of your
nonsense:
ill> Actually, you do not need to perform the integration as the
ill> normalizing constant does not depend on the parameter you
ill> wish to estimate,
That really was more than sufficient to indicate that you did
NOT have any basic knowledge about Bayesian methods.
You misquoted EIGHT consecutive times in your last reply.
You still haven't quoted anything.
You are just mouth dancing in front of the mirror, to YOURSELF.
You had already proven (from your MIS-stated 6 points) of your
ignorance about Bayesian statistics and Bayesian inference.
This is all too TYPICAL of your irrelevance, which is a strong
indication of your ignorance:
> but a Riemmanian metric and Lebesgue measure are scarcely
> mathematistry.
Those are not themselves mathematistry. They are YOUR use
of "buzz words" when those words are completely irrelavant to the
question the OP asked about finding the max of the posterior
distribution!
> Cheerfully yours,
>
> illywhacker;
If you want to find out what I agree and don't agree about Bayesian
Statistics, there is a lengthy discussion between Herman Rubin and
myself, in April 2005 -- each of us had forgotten more about
Bayesian statistics than the totality of your knowledge about the
subject.
"A Debate on What Prior to Use as a TRUE Bayesian, Anyone?"
http://groups.google.com/group/sci.stat.edu/msg/7a8f69f5d1151bb0?hl=en&
with 6 posts each by Herman and myself, which was a subthread
after a 15 post discussion by us, about the two subtopics:
"Advanced Probability for Dummies?"
"Measure Theory" needed for Applied Statistics? You Jest!
In these posts you'll find a fairly definitive statements, by Herman
Rubin and by myself, on where we stand on Applied Statistics,
Bayesian Statistics, and Mathematical Statistics and Theory.
I don't think our points of agreements and disagreements have
changed the slightest, until Herman have withdrawn some of his
positions a year and a half ago.
Illywhacker, there is room for YOU to discuss anything serious
about Statistics OR Mathematics, as evidenced by what I had
read that you wrote in sci.stat.math,
-- Reef Fish Bob.
illywhacker
Reef Fish wrote:
> illywhacker wrote:
> This was
> your opening sentence before you started the rest of your
> nonsense:
>
> ill> Actually, you do not need to perform the integration as the
> ill> normalizing constant does not depend on the parameter you
> ill> wish to estimate,
>
> That really was more than sufficient to indicate that you did
> NOT have any basic knowledge about Bayesian methods.
My statement in context says that to perform the maximization, you do
not need to perform the integration, that is all. You obviously agree,
so why the angst? Calculating the normalization constant is a waste of
time if all you want to do is to maximize a posterior density. Of
course I agree that to normalize the posterior distribution you need to
calculate the normalizing constant; I never said otherwise. But it is
irrelevant to the question of maximizing the posterior density.
> This is all too TYPICAL of your irrelevance, which is a strong
> indication of your ignorance:
>
> > but a Riemmanian metric and Lebesgue measure are scarcely
> > mathematistry.
>
> Those are not themselves mathematistry. They are YOUR use
> of "buzz words" when those words are completely irrelavant to the
> question the OP asked about finding the max of the posterior
> distribution!
They are not irrelevant. I do not think you have understood the point
about invariance. Bayes' theorem does not give you a function that you
can maximize; it gives you a measure. Whether you use an ignorance
prior is neither here nor there. The problem remains because MAP
estimation uses a delta function loss. Delta functions in two different
coordinates systems are not necessarily the same thing, hence you have
a consistency problem.
illywhacker;
illywhacker
Do you have R? Then look here:
http://www.public-health.uiowa.edu/boa/
and, more specifically, here:
http://www.public-health.uiowa.edu/boa/help/boa.hpd.html
If you do not have R, then look here first:
and then at the previous links.
illywhacker;
Reef Fish
illywhacker wrote:
> >
> > > but a Riemmanian metric and Lebesgue measure are scarcely
> > > mathematistry.
> >
> > Those are not themselves mathematistry. They are YOUR use
> > of "buzz words" when those words are completely irrelavant to the
> > question the OP asked about finding the max of the posterior
> > distribution!
>
> They are not irrelevant.
They are 100% irrelevant, in the problem of finding the maximum
of a function of theta (the posterior distribution).
> Bayes' theorem does not give you a function that you
> can maximize; it gives you a measure.
Really? Can you imagine all these years I've missed this important
principle illywhacker has just discovered.
> Whether you use an ignorance
> prior is neither here nor there. The problem remains because MAP
> estimation uses a delta function loss. Delta functions in two different
> coordinates systems are not necessarily the same thing, hence you have
> a consistency problem.
That must be what you learned from your physicists who called
themselves
Bayesian. There is NO DELTA FUNCTION nor "delta loss function" in
the posterior distribution of any Bayesian inference problem.
>
> illywhacker;
That was why I said you didn't know anything about the basic theory
and methods of Bayesian inference!
If one uses a diffuse prior, then the posterior is, up to a
multiplcative
factor, the classical LIKELIHOOD function of theta, which does not
require and delta function to find the MLE.
If one uses a conjugate proior, then the posterior is in the same
family as the prior, such as a BETA density for the problem of the
Bernoulli/Binomial p. The mode of the posterior distribution is
immediate.
If one uses any of the prior distributions in Robert Schlaiffer's
book that required numerical integration to get the posterior, the
maximum of the posterior distribution is just the ordinary maximum,
without the use of any "delta loss".
That covers ALL possible situations governing any univariate
posterior inference in general, the inference about the parameter
p of a Binomial in particular, in finding the MAXIMUM of a
posterior distribution.
In short, the more you post, the more PURE NOISE, from ignorance
about Bayesian inference comes out.
> > > but a Riemmanian metric and Lebesgue measure
is something you think is NECESSARY in finding the maximum of a
posterior distribution puts my FREE lesson to you to a COMPLETE
HALT.
-- Reef Fish Bob.
illywhacker
Reef Fish wrote:
> illywhacker wrote:
> > >
> > > > but a Riemmanian metric and Lebesgue measure are scarcely
> > > > mathematistry.
> > >
> > > They are YOUR use
> > > of "buzz words" when those words are completely irrelavant to the
> > > question the OP asked about finding the max of the posterior
> > > distribution!
Actually, this is not the question the OP asked. The OP asked about HPD
intervals I believe. You posted about MAP estimation, but you made some
misleading comments that required correction. I have corrected them,
but you do not seem to understand. Let me reiterate.
> They are 100% irrelevant, in the problem of finding the maximum
> of a function of theta (the posterior distribution).
The posterior *distribution* is not a function of theta. It is a
measure. You give it a set, and it gives you back a number. If you
compute its value on a set consisting of just one point, it will, in
most cases of practical interest, give you zero. There is thus no sense
in maximizing it over points. There is sense in maximizing it over a
restricted set of sets, and this is what HPD intervals are, a point you
seem to have missed.
In contrast, a posterior *density* is a function, and it makes sense to
maximize it over points. However, the construction of a density from a
measure requires dividing by another measure. The density will depend
on this choice of measure. Different choices therefore lead to
different MAP estimates. In particular, dividing by the d\theta thingy
(I do not want to use mathematistry terms like Lebesgue measure so I
decided on something more user friendly) in a particular coordinate
system produces answers that depend on the coordinate system.
> There is NO DELTA FUNCTION nor "delta loss function" in
> the posterior distribution of any Bayesian inference problem.
Indeed not. But there is in MAP estimation, which is what you are
discussing. And there is not in HPD interval computation, the subject
of the OP's question.
> If one uses a diffuse prior, then the posterior is, up to a
> multiplicative
> factor, the classical LIKELIHOOD function of theta, which does not
> require and delta function to find the MLE.
No. The posterior, up to a factor, is the likelihood function
multiplied by d\theta.
I did not invent this idea Reefer: go and read Fisher and the many
others who comment on it.
illywhacker;
Anon.
> Dear Bob,
>
> Reef Fish wrote:
>> illywhacker wrote:
>>>>> but a Riemmanian metric and Lebesgue measure are scarcely
>>>>> mathematistry.
>>>> They are YOUR use
>>>> of "buzz words" when those words are completely irrelavant to the
>>>> question the OP asked about finding the max of the posterior
>>>> distribution!
>
> Actually, this is not the question the OP asked. The OP asked about HPD
> intervals I believe.
That wasn't terribly clear, though: I initially thought that, but then
the MAP interpretation would also make sense. Unfortunately, I think
we've scared the OP off, so we may never know.
Reef Fish
illywhacker wrote:
>
> The posterior *distribution* is not a function of theta.
Unmistakable BLUNDER. I even used the BETA distribution as an
example.
>
> > There is NO DELTA FUNCTION nor "delta loss function" in
> > the posterior distribution of any Bayesian inference problem.
> I did not invent this idea Reefer: go and read Fisher and the many
> others who comment on it.
>
> illywhacker;
This is the FIRST ever that anyone calls Fisher (I presume you mean
Sir Ronald A. Fisher, and not Debbie Reynold's former husband
Eddie Fisher, or the chess playing Bobby Fisher) a Bayesian.
Thanks for the laugh.
-- Reef Fish Bob.
illywhacker
Reef Fish wrote:
> illywhacker wrote:
> >
> > The posterior *distribution* is not a function of theta.
>
> Unmistakable BLUNDER. I even used the BETA distribution as an
> example.
I am sorry to see that someone with your reputation and experience does
not understand the difference between a measure and a function.
> > I did not invent this idea Reefer: go and read Fisher and the many
> > others who comment on it.
>
> This is the FIRST ever that anyone calls Fisher (I presume you mean
> Sir Ronald A. Fisher, and not Debbie Reynold's former husband
> Eddie Fisher, or the chess playing Bobby Fisher) a Bayesian.
I did not call Fisher a Bayesian. I said that Fisher, unlike Reef Fish
Bob, was aware of the difference between a measure and a function, and
was therefore also aware of the need for an underlying measure to
calculate MAP estimates.
illywhacker;
illywhacker
> illywhacker wrote:
> > Actually, this is not the question the OP asked. The OP asked about HPD
> > intervals I believe.
>
> That wasn't terribly clear, though: I initially thought that, but then
> the MAP interpretation would also make sense. Unfortunately, I think
> we've scared the OP off, so we may never know.
It is kind of you to say 'we', when in fact it is the Reefer and I who
may have scared the OP off. I agree that it was not completely clear,
but I do think the OP meant HPD intervals; were (s)he merely talking
about optimization, the answer would be obvious.
illywhacker;
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
> > illywhacker wrote:
> > >
> > > The posterior *distribution* is not a function of theta.
> >
> > Unmistakable BLUNDER. I even used the BETA distribution as an
> > example.
>
> I am sorry to see that someone with your reputation and experience does
> not understand the difference between a measure and a function.
I understand the difference between those two perfectly.
>
> > > I did not invent this idea Reefer: go and read Fisher and the many
> > > others who comment on it.
> >
> > This is the FIRST ever that anyone calls Fisher (I presume you mean
> > Sir Ronald A. Fisher, and not Debbie Reynold's former husband
> > Eddie Fisher, or the chess playing Bobby Fisher) a Bayesian.
>
> I did not call Fisher a Bayesian. I said that Fisher, unlike Reef Fish
> Bob, was aware of the difference between a measure and a function, and
> was therefore also aware of the need for an underlying measure to
> calculate MAP estimates.
>
> illywhacker;
Perhaps you're drawing Fisher, incorrectly, into your physicists camp
of butchers of Bayesian statistics. Fisher can be said to be a NON-
Bayesian, but he is NOT a butcher of Bayesian statistics like
illywhacker.
illywhacker, give it up.
You don't understanding ANYTHING about Bayesian Statistics and
Bayesian inference. All you throw around are some buzz words
completely unrelated to Bayesian inference as evidence by your
plenty of misuse and abuse of those words already.
-- Reef Fish Bob.
illywhacker
you have lost the argument but are unable to admit it, as usual.
illywhacker;
Reef Fish
What invective? I am citing EVERY WORD I posted which you snipped:
RF> I understand the difference between those two perfectly.
That was a statement of FACT.
RF> Perhaps you're drawing Fisher, incorrectly, into your physicists
camp
RF>of butchers of Bayesian statistics. Fisher can be said to be a
NON-
RF> Bayesian, but he is NOT a butcher of Bayesian statistics like
RF> illywhacker.
That was a statement of FACT.
RF> illywhacker, give it up.
That was a statement of opinion, based on illywhacter's complete
lack of understanding about Bayesian statistics.
RF> You don't understanding ANYTHING about Bayesian Statistics and
RF> Bayesian inference. All you throw around are some buzz words
RF> completely unrelated to Bayesian inference as evidence by your
RF> plenty of misuse and abuse of those words already.
That was a statement of FACT.
Where were the invectives?
Saying you don't understand ANYTHING about Bayesian Statistics
and Bayesian inference?
Saying all you did was to throw around buzz words unrelated to
Bayesian inference?
Saying that those were EVIDENCED by your misuse and abuse
of those buzz words?
You're were just running out of excuses for your own blunders.
-- Reef Fish Bob.
illywhacker
was rather, or in the alternative, also, invective that had nothing to
do with the substantial issues. All this has arisen because you refuse
to admit that you 1) misinterpreted the OP's question, and 2) given
your misinterpretation, gave poor advice.
> RF> I understand the difference between those two perfectly.
Evidently, unlike the rest of the world, you define a probabilty
distribution to be a density function with respect to some underlying
measure. So be it. This still does not change the invariance issue, or
the need to specify the underlying measure.
> RF> illywhacker, give it up.
>
> That was a statement of opinion
Actually, no, it is not an opinion. It is a suggestion or a command.
Which of course I will ignore.
illywhacker;
Reef Fish
illywhacker wrote:
> You can *call* anything a fact, FishBoy, but that changes nothing.
Not in the mind of someone as ignorant as illywhacker.
It makes a difference to those who are much better educated in the
subject matter than illywhacker. They can CHECK the facts
themselves.
> It
> was rather, or in the alternative, also, invective that had nothing to
> do with the substantial issues. All this has arisen because you refuse
> to admit that you 1) misinterpreted the OP's question, and 2) given
> your misinterpretation, gave poor advice.
The OP was asking a question about finding the maximum of a
posterior distribution in a Bayesian analysis.
Not exactly a deep question.
But apparently one sufficient to derail those who are completely
ignorant about Bayesian inference, such as illywhacker who learned
his mathematics (not even statistics) from physicists, and who
does not even know what a posterior distribution IS and thought
R.A. Fisher was an authority to appeal to, on matters relating to
Riemannian metric and Lebesque measure that are entirely
irrelevant to the Bayesian question, especially at the low level
at which the question was addressed.
> > RF> I understand the difference between those two perfectly.
>
> Evidently, unlike the rest of the world, you define a probabilty
> distribution to be a density function with respect to some underlying
> measure. So be it.
I took graduate level courses in Real Analysis and Functional
Analysis which contained plenty of measure theory; in addition
to a graduate course in Measure Theory from the book by
Halmos, all from decent Departments of MATHMATICS, before
I even began studying Statistics.
I did not learn your kind of pseudo mathematics and pseudo
statistics from physicists and computer packages and learned
nothing but some BUZZ WORDS which you don't understand
and thought it had something to do with Bayesian inference.
> This still does not change the invariance issue, or
> the need to specify the underlying measure.
Those issue never even came up. You were just bluffing on
something you don't understand yourself but heard of those
words somewhere in your physics lab.
> > RF> illywhacker, give it up.
> >
> > That was a statement of opinion
>
> Actually, no, it is not an opinion. It is a suggestion or a command.
If was a suggestion based on sound opinion. If it were a
command, I would at least put an exclamation mark after it,
or express it unambiguously as a command.
> Which of course I will ignore.
>
> illywhacker;
Which of course will only deepen your state of ignorance about
Bayesian statistics, and statistics in general.
Pity.
-- Reef Fish Bob.
illywhacker
> The OP was asking a question about finding the maximum of a
> posterior distribution in a Bayesian analysis.
First, the OP was asking whether certain *software* could compute
something, not how it could be computed in general:
AB wrote:
> how can I find an HPD using a statistical software (e.g. Mathematica)?
Second, you assumed that the OP wanted a MAP estimate. I followed
you for a couple of posts, before I realized that there was an
ambiguity in the OP, of which you are as yet unaware. That is: it is
not clear whether the OP wanted an HPD
interval or a MAP estimate. This possibility you will find confirmed by
Anon's post earlier.
> > Evidently, unlike the rest of the world, you define a probabilty
> > distribution to be a density function with respect to some underlying
> > measure. So be it.
>
> I took graduate level courses in Real Analysis and Functional
> Analysis which contained plenty of measure theory; in addition
> to a graduate course in Measure Theory from the book by
> Halmos, all from decent Departments of MATHMATICS, before
> I even began studying Statistics.
So what. You probably got D's. In any case, you should have gotten D's,
because you
evidently do not understand what the difference between a function and
measure implies, even if you could manage to parrot off the
definitions for your exams.
> I did not learn your kind of pseudo mathematics and pseudo
> statistics from physicists and computer packages
If you had, you might be of some use to the OP (see OP quote above).
> > This still does not change the invariance issue, or
> > the need to specify the underlying measure.
>
> Those issue never even came up. You were just bluffing on
> something you don't understand yourself but heard of those
> words somewhere in your physics lab.
That's right Fishy. Salve your ego. You still do not understand this
invariance issue: it always comes up with MAP estimation, as Fisher
pointed out in an effort to support MLE, which is invariant, by
attacking MAP.
illywhacker;
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
> > The OP was asking a question about finding the maximum of a
> > posterior distribution in a Bayesian analysis.
>
> > > distribution to be a density function with respect to some underlying
> > > measure. So be it.
> >
> > I took graduate level courses in Real Analysis and Functional
> > Analysis which contained plenty of measure theory; in addition
> > to a graduate course in Measure Theory from the book by
> > Halmos, all from decent Departments of MATHMATICS, before
> > I even began studying Statistics.
>
> So what. You probably got D's. In any case, you should have gotten D's,
LOL. You are really raking into the bottom of your barrel in trying
to salvage something out of your own ignorance, isn't it?
> > > This still does not change the invariance issue, or
> > > the need to specify the underlying measure.
You have already lost ALL credibility in the subject of Statistics and
Bayesian statistics. No amount of Mouth Dancing can salvage it.
So there's your old Bayesian friend Ronald A. Fisher again?
> as Fisher pointed out in an effort to support MLE, which is invariant,
>
> illywhacker;
Hey you WHACKO. Why would anyone need Fisher to support
the invariance of MLE?
Every UNDERGRAD in methematical statistics knows THAT!
That's why you thought the invariance of the MLE is a big deal,
since you learned it from the labs of a physicist's pseudo
school of statistics.
The more you fumble around with that ever-so-little statistical
facts you've heard, the clearer it becomes of the level of your
total knowledge about statistics -- somewhat below most good
High School level.
Keep whacking away, whacky, whacky, whacky.
That's why you chose that posting name, to pre-warn your
whacky self.
-- Reef Fish Bob.
Lou Thraki
illywhacker wrote:
> Reef Fish wrote:
> >
> > something you don't understand yourself but heard of those
> > words somewhere in your physics lab.
>
> That's right Fishy. Salve your ego. You still do not understand this
> invariance issue: it always comes up with MAP estimation, as Fisher
> pointed out in an effort to support MLE, which is invariant, by
> attacking MAP.
I guess that Reef Fish considers the choice of parametrization
part of putting all prior knowledge into the prior. The invariance
is not so much of an issue then.
Reef Fish
That COULD be one way to put it. But invariance is seldom an
issue in any application of Statistics whether it's in the choice of
parametrization or in the choice of estimation methods.
When one uses an unbiased estimate, INVARIANCE is already
thrown out of the window.
When one chooses the MSE criterion of estimation of sigma^2,
INVARIANCE is out of the window already. In fact, in the
estimation of sigma^2, INVARIANCE is preserved ONLY by
the use of MLE.
Furthermore, when there are transformations within a pre-chosen
statistical procedure, one has to be more concerned with the
underlying assumption of the error DISTRIBUTION rather than
the invariance of the parameter.
In short, and especially in a Bayesian estimation problem,
there is NEVER any issue of invariance for a true APPLIED
Bayesian, because said Bayesian has to ellicit his own
PRIOR about the chosen parameter theta.
His prior about theta is not going be invariant under any other
nonlinear function of theta -- by definition of CHOICE and
by REASON of probability assessment in the prior distribution.
So, if you consider ALL aspects of any statistical application,
the invariance property is seldom a problem, and in the case
of Bayesian inference, your characterization is ALMOST
right on the money:
> I guess that Reef Fish considers the choice of parametrization
> part of putting all prior knowledge into the prior. The invariance
> is not so much of an issue then.
Invariance is a NON-issue because whatever theta you may
wish to assess or use in your prior, it is NOT invariant.
Invariance comes into play only in the pesudo-Bayesian's
use of UNINFORMATIVE prior, which is a silly notion to be
a Bayesian in the first place. It is a COP OUT for ones
inability to represent one's prior knowledge and experience
in mathematical form, as a Bayesian.
So, Lou Thraki's comment is a succinct way of killing illywhacker's
red-herrings in any Bayesian problem.
-- Reef Fish Bob.
illywhacker
> I guess that Reef Fish considers the choice of parametrization
> part of putting all prior knowledge into the prior. The invariance
> is not so much of an issue then
Bob can consider what he likes, but he would be wrong. It is not an
issue of the prior. The prior, being a measure, does not depend on the
coordinate system because the coordinate system is just a labelling
mechanism. It is thus hard to see what knowledge can be put into the
prior about the coordinate system. The posterior does not depend on the
coordinate system either. What, in common practice, depends on the
coordinate system is the MAP estimate, as has been noted by many
people. Since a coordinate system is just an arbitrary ' name' we give
to points in a space, this is not acceptable.
illywhacker;
illywhacker
Reef Fish wrote:
> When one chooses the MSE criterion of estimation of sigma^2,
> INVARIANCE is out of the window already. In fact, in the
> estimation of sigma^2, INVARIANCE is preserved ONLY by
> the use of MLE.
Wrong Bob. You need to keep abreast of the literature.
> In short, and especially in a Bayesian estimation problem,
> there is NEVER any issue of invariance for a true APPLIED
> Bayesian, because said Bayesian has to ellicit his own
> PRIOR about the chosen parameter theta.
> His prior about theta is not going be invariant under any other
> nonlinear function of theta -- by definition of CHOICE and
> by REASON of probability assessment in the prior distribution.
Wrong again. The prior has nothing to do with it. For example, the
prior
ds f(s)
for some positive real parameter s, is the same measure as
dt g(t)
where g(t) = 2t f(t^{2}) and t is restricted to be positive. Since this
is the same measure, this change does not affect the posterior, which
as I have already explained is independent of the coordinate system.
What is not independent of the coordinate system is the MAP estimate as
it is commonly implemented.
Now are you going to continue to spout nonsense or are you going to
write some equations to prove your point?
illywhacker;
illywhacker
little of any great reknown, and who now feels the need to boost his
ego by trying to diminish people on this board. When you are in one of
your stable moods, you are quite a reasonable person to talk to, but
when bipolar disorder kicks in, you are insufferable.
You are wrong on the points we have been discussing, and you have
failed to acknowledge your misinterpretation of the OP. Even your
suggestion to the OP based on your misinterpretation would have had the
OP spending a lot of time trying to compute something that was
completely irrelevant to the problem. So far, I am the only person who
has actually answered the OP' s question.
There is only one point in your post that requires answering;
> Why would anyone need Fisher to support
> the invariance of MLE?
>
> Every UNDERGRAD in methematical statistics knows THAT!
I did not say that Fisher tried to support the invariance of MLE, even
you understand that. I said that Fisher used the invariance of MLE as
an argument for its superiority over MAP, which, as he understood it,
was not invariant.
illywhacker;
Reef Fish
illywhacker wrote:
> Lou Thraki wrote:
> > I guess that Reef Fish considers the choice of parametrization
> > part of putting all prior knowledge into the prior. The invariance
> > is not so much of an issue then
>
> Bob can consider what he likes, but he would be wrong.
>
I have a Ph.D. at Yale, in Statistics. I learned my Bayesian
Statistics
from L.J. (Jimmie) Savage himself, one sometimes called the
Founder of Bayesian Statistics in the USA (because of his book
on the Foundations of Statistics about Bayesian statistics).
I have taught Bayesian statistics, both theoretical and applied
at the Graduate level at the University of Chicago.
I was elected Fellow of the American Statistical Association in 1984.
I was listed the same year (and in several volumes later) in Marquis's
Who's Who in the World as a "Statistical Educator", with a citation
longer than that of Ronald Reagan who was President of the USA at
the time, and that of Bill Clinton who was Governor of Arkansas at
the time.
All those are easily verifiable FACTS about my statistical training
and credentials.
What can you say for yourself, Illywhacker, other than you
learned your statistics from some physicists?
You are not only ignorant about Statistics (which would be a
forgivable defect) if you only learn to keep your mouth SHUT
on subjects you know nothing about.
As soon as you open your mouth and spew NONSENSE about
statistics in general and Bayesian statistics in particular, it is my
obligation (as a statistician and a moderator of this group) to
speak out against your malpractice, in the same manner i speak
out against others of your ilk.
This is a self-moderated group on the accuracy of the Statistical
substance, as well as on the behavior of the posters.
Illywhacker is a serious offender in both, IMHSHO.
-- Reef Fish Bob.
-- Reef Fish Bob.
illywhacker
Reef Fish wrote:
<snip many claims to authority, but no substantial points whatsoever>
Is it not Reefer who is always calling for 'statistical substance'? But
here he provides nothing but a list of his educational experiences,
which may or may not have resulted in him being educated.
> I was listed the same year (and in several volumes later) in Marquis's
> Who's Who in the World as a "Statistical Educator", with a citation
> longer than that of Ronald Reagan who was President of the USA at
> the time, and that of Bill Clinton who was Governor of Arkansas at
> the time.
This is most amusing. I too was asked by this defunct book whether I
wished to have an entry, which shows how much it proves. I declined.
Bob's ego would not allow such a thing of course. What is more, you
write your own entry. I am thus sure that Bob's entry was suitably
self-congratulatory and enormously long, listing his many
'achievements'.
When you have something to say of statistical substance, Bob, I will
pay attention. You still understand nothing about our topic of
discussion, witness your conviction that ignorance priors are somehow
connected to the invariance issue for MAP estimates. But the invariance
issue for MAP estimates exists for any prior whatsoever.
illywhacker;
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
>
> <snip many claims to authority, but no substantial points whatsoever>
>
>
> > Who's Who in the World as a "Statistical Educator", with a citation
> > longer than that of Ronald Reagan who was President of the USA at
> > the time, and that of Bill Clinton who was Governor of Arkansas at
> > the time.
>
> This is most amusing. I too was asked by this defunct book whether I
> wished to have an entry, which shows how much it proves. I declined.
>What is more, you write your own entry.
That proved that you LIED. Marquis has many Who's Who books but
you DO NOT get to write your own entry in Who's Who in the WORLD.
The Editors of the book wrote the entry and asked me to verify for
correctness.
You not only can't show any EDUCATIONAL background in Statistics,
you even LIED about your Who's Who entry. I have received DOZENS
of other Who's Whos like the one You did, which I promptly trashed
without even looking.
Your Who's Who was probably something like
"Who's Who on the Internet"
*> Feel free to nominate anyone, no matter how small their claim
*> to fame (i.e., first person to access the Net from the Galapagos
*> island, maintainer of a cool list/archive, first administrator of
a
*> Net node in Antarctica, developer of <favorite Net tool/protocol
*> here>, etc.). And yes, you may nominate yourself.
And Illywhacker nominated himself because he knew how to post
in newsgroups and he claims to be the first person who lives in the
red light district of Amsterdam who is a physicist Bayesian who
learned about the invariance of MLE from Bobby Fisher.
> I am thus sure that Bob's entry was suitably
> self-congratulatory and enormously long, listing his many
> 'achievements'.
That' the kind of IGNORAMUS you are, Illywhacker!
>
> When you have something to say of statistical substance, Bob, I will
> pay attention.
I had plenty to say, but because of your poor education on the subject,
you couldn't understand any of it.
It's pure and simple.
>
> illywhacker;
If you want to do something USEFUL, go whack some Bushes. <;-)
-- Reef Fish Bob.
illywhacker
Anything of any statistical consequence, again.
- No reply to my concrete arguments about invariance, which he
incorrectly interprets as being connected to ignorance priors.
- No acknowledgement that he completely misinterpreted the OP.
- No acknowledgement that even in this misinterpretation, he made a
stupid and counterproductive suggestion, to whit, that the OP waste
time computing the normalization constant of the posterior, when in
fact it is irrelevant to MAP estimation, the goal that Reefer set for
the OP (although not of course the goal the OP set for himself).
No. Reef Fish wrote:
The usual blather he comes up with, very inventively one has to say,
whenever he realizes that he is wrong.
You are out of your depth, Bobby boy. Leave the deep waters of
invariance and other mathematics subtleties that are beyond your ken.
Scuttle back to your safe little reef, where you can continue to
pontificate about undergraduate statistics, which you may or may not
understand, judging by the reaction your self-important and pompous
'lectures' are producing. I am not wasting my time reading them. (Yes -
I know you think I should read them in order to shine a little light
into the dark statistical abyss that I inhabit, etc. etc.)
illywhacker;
Reef Fish
illywhacker wrote:
>
> No. Reef Fish wrote:
>
> The usual blather he comes up with, very inventively one has to say,
> whenever he realizes that he is wrong.
>
> You are out of your depth, Bobby boy. Leave the deep waters of
> invariance and other mathematics subtleties that are beyond your ken.
> Scuttle back to your safe little reef, where you can continue to
> pontificate about undergraduate statistics, which you may or may not
> understand, judging by the reaction your self-important and pompous
> 'lectures' are producing. I am not wasting my time reading them. (Yes -
> I know you think I should read them in order to shine a little light
> into the dark statistical abyss that I inhabit, etc. etc.)
>
> illywhacker;
As I said, go whack the Bushes!
Your credibility in Statistics and Bayesian Statistics has reached its
nadir.
-- Reef Fish Bob.
Reef Fish
illywhacker wrote:
>
>
> You are out of your depth, Bobby boy. Leave the deep waters of
> invariance and other mathematics subtleties that are beyond your ken.
> Scuttle back to your safe little reef, where you can continue to
> pontificate about undergraduate statistics, which you may or may not
> understand,
I actually read THAT part and had a little chuckle at the desparate
GASPING for air in even the shallow waters of Bayesian Statistics
that are already way over his head, having learned his BUZZ WORDS
from physicists who called themselves Bayesians.
I overlooked this part:
> I am not wasting my time reading them. (Yes -
> I know you think I should read them in order to shine a little light
> into the dark statistical abyss that I inhabit, etc. etc.)
>
> illywhacker;
Then why do you keep coming back with your impertinence?
You said DOZENS of posts ago that you weren't going to read
my "lectures" on Bayesian inference and statistics, then why
do you keep coming back?
Just to practice your Whacking, Whacking, Whacking, ...
Whack, Whack, Whack, Whack. Whack, Whack, Whack, ...
Whack, Whack, Whack, Whack. Whack, Whack, Whack, ...
Whack, Whack, Whack, Whack. Whack, Whack, Whack, ...
Whack, Whack, Whack, Whack. Whack, Whack, Whack, ...
You should do that in the privacy of your own bedroom and
not in a public forum, don't you know?
-- Reef Fish Bob.
illywhacker
Reef Fish wrote:
> Then why do you keep coming back with your impertinence?
> You said DOZENS of posts ago that you weren't going to read
> my "lectures" on Bayesian inference and statistics, then why
> do you keep coming back?
Did I? I am more consistent than I thought. I come to the board,
FishStick, not to read you, but I am forced to share the space with
you, just as I am forced to share the pavement with dog faeces. Whack
whack.
illywhacker;
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
> > Then why do you keep coming back with your impertinence?
> > You said DOZENS of posts ago that you weren't going to read
> > my "lectures" on Bayesian inference and statistics, then why
> > do you keep coming back?
>
> Did I? I am more consistent than I thought. I come to the board,
What board? LOL. This is a USENET newsgroup! It's an
unmistakable discriminant function for a Clueless Newbie to call
sci.stat.math "the board".
The only board you find here are those wooden chunks inside
your head!
> FishStick, not to read you, but I am forced to share the space with
> you, just as I am forced to share the pavement with dog faeces. Whack
> whack.
>
> illywhacker;
The fact that you share dog feces (I assume that's what you meant)
with your wacky physcists is not surprising.
Whack. Whack, Whack, Whack, Whack, ...
Whack. Whack, Whack, Whack, Whack, ...
You are getting down to your SPECIALTY now, aren't you?
-- Reef Fish Bob.
Reef Fish
Reef Fish wrote:
> illywhacker wrote:
> > Reef Fish wrote:
> >
> > <snip many claims to authority, but no substantial points whatsoever>
> >
> >
> > > I was listed the same year (and in several volumes later) in Marquis's
> > > Who's Who in the World as a "Statistical Educator", with a citation
> > > longer than that of Ronald Reagan who was President of the USA at
> > > the time, and that of Bill Clinton who was Governor of Arkansas at
> > > the time.
> >
> > This is most amusing. I too was asked by this defunct book
LIE LIE LIE LIE Shameless LIE by Illywhacker
TWO LIES in one line: "defunct book"? "asked"?
http://www.marquiswhoswho.com/products/WOprodinfo.asp
The webpage was about Marquis's Who's Who in the World
(24th Edition, 2007).
> > whether I
> > wished to have an entry, which shows how much it proves. I declined.
>
> >What is more, you write your own entry.
>
RF> That proved that you LIED. Marquis has many Who's Who books but
RF> you DO NOT get to write your own entry in Who's Who in the WORLD.
>
RF> The Editors of the book wrote the entry and asked me to verify for
RF> correctness.
One CANNOT request to be included in that Who's Who in the World
either. The Editors of the volume seeks out the listees, and those
in the volume are NOT charged one penny NOR are they required to
purchase any volume. There are enough libraries in the World to
pay over $500 for each of those volumes.
>
> You not only can't show any EDUCATIONAL background in Statistics,
> you even LIED about your Who's Who entry. I have received DOZENS
> of other Who's Whos like the one You did, which I promptly trashed
> without even looking.
>
> Your Who's Who was probably something like
> "Who's Who on the Internet"
>
> *> Feel free to nominate anyone, no matter how small their claim
> *> to fame (i.e., first person to access the Net from the Galapagos
> *> island, maintainer of a cool list/archive, first administrator of a
> *> Net node in Antarctica, developer of <favorite Net tool/protocol
> *> here>, etc.). And yes, you may nominate yourself.
>
> And Illywhacker nominated himself because he knew how to post
> in newsgroups and he claims to be the first person who lives in the
> red light district of Amsterdam who is a physicist Bayesian who
> learned about the invariance of MLE from Bobby Fisher.
I actually came CLOSE. There is actually a Marquis's
"Who's Who on the Web".
http://www.marquiswhoswho.com/products/ontheweb.asp
It's sold to suckers like Illywhacker on SUBSCRIPTION basis.
There are about 23 DIFFERENT Marquis's Who's Whos. The ONE
single Who's Who in the World is the most exclusive of them all,
citing ONLY leaders in various areas in 215 countries in the World
in one single volume.
Reef Fish
Reef Fish wrote:
> illywhacker wrote:
> > Reef Fish wrote:
> >
> > <snip many claims to authority, but no substantial points whatsoever>
> >
> >
> > > I was listed the same year (and in several volumes later) in Marquis's
> > > Who's Who in the World as a "Statistical Educator", with a citation
> > > longer than that of Ronald Reagan who was President of the USA at
> > > the time, and that of Bill Clinton who was Governor of Arkansas at
> > > the time.
> >
> > This is most amusing. I too was asked by this defunct book
LIE LIE LIE LIE Shameless LIE by Illywhacker
TWO LIES in one line: "defunct book"? "asked"?
http://www.marquiswhoswho.com/products/WOprodinfo.asp
The webpage was about Marquis's Who's Who in the World
(24th Edition, 2007).
> > whether I
> > wished to have an entry, which shows how much it proves. I declined.
>
> >What is more, you write your own entry.
>
RF> That proved that you LIED. Marquis has many Who's Who books but
RF> you DO NOT get to write your own entry in Who's Who in the WORLD.
>
RF> The Editors of the book wrote the entry and asked me to verify for
RF> correctness.
One CANNOT request to be included in that Who's Who in the World
either. The Editors of the volume seeks out the listees, and those
in the volume are NOT charged one penny NOR are they required to
purchase any volume. There are enough libraries in the World to
pay over $500 for each of those volumes.
>
> You not only can't show any EDUCATIONAL background in Statistics,
> you even LIED about your Who's Who entry. I have received DOZENS
> of other Who's Whos like the one You did, which I promptly trashed
> without even looking.
>
> Your Who's Who was probably something like
> "Who's Who on the Internet"
>
> *> Feel free to nominate anyone, no matter how small their claim
> *> to fame (i.e., first person to access the Net from the Galapagos
> *> island, maintainer of a cool list/archive, first administrator of a
> *> Net node in Antarctica, developer of <favorite Net tool/protocol
> *> here>, etc.). And yes, you may nominate yourself.
>
> And Illywhacker nominated himself because he knew how to post
> in newsgroups and he claims to be the first person who lives in the
> red light district of Amsterdam who is a physicist Bayesian who
> learned about the invariance of MLE from Bobby Fisher.
I actually came CLOSE. There is actually a Marquis's
"Who's Who on the Web".
http://www.marquiswhoswho.com/products/ontheweb.asp
It's sold to suckers like Illywhacker on SUBSCRIPTION basis.
There are about 23 DIFFERENT Marquis's Who's Whos. The ONE
single Who's Who in the World is the most exclusive of them all,
citing ONLY leaders in various areas in 215 countries in the World
in one single volume.
> That' the kind of IGNORAMUS you are, Illywhacker!
illywhacker
Reef Fish wrote:
> RF> That proved that you LIED. Marquis has many Who's Who books but
> RF> you DO NOT get to write your own entry in Who's Who in the WORLD.
I quote from the web page:
"Who's Who in the World provides instant access to over 50,000
personal profiles-reviewed, revised, and proofread by the listees
themselves-providing guaranteed accuracy and immediacy."
Providing guaranteed hagiography. Presumably one cannot actually write
fiction, but adding a degree of spin would certainly be possible.
> One CANNOT request to be included in that Who's Who in the World
> either. The Editors of the volume seeks out the listees...
Indeed. One must be invited, as I was, as you were. I declined. You
accepted.
> > You not only can't show any EDUCATIONAL background in Statistics,...
Neither can you. No one can: this is the Internet. All you can do is
make unprovable claims. This is why it is not worth bothering, unless
one is trying "proof by authority", a favourite tactic of yours.
illywhacker;
illywhacker
Reef Fish wrote:
> The fact that you share dog feces (I assume that's what you meant)
I see you use the simplified spelling of "faeces" designed for use by
US citizens.
illywhacker;
illywhacker
double post. Apology please, as I am sure you would demand from anyone
else.
illywhacker;
Robert F Ling
> Reef Fish wrote:
> > RF> That proved that you LIED. Marquis has many Who's Who books but
> > RF> you DO NOT get to write your own entry in Who's Who in the WORLD.
>
> I quote from the web page:
>
> "Who's Who in the World provides instant access to over 50,000
> personal profiles-reviewed, revised, and proofread by the listees
> themselves-providing guaranteed accuracy and immediacy."
>
> Providing guaranteed hagiography. Presumably one cannot actually write
> fiction, but adding a degree of spin would certainly be possible.
It is not only possible, but it has been proven:
The Hall of Lame
Tucker Carlson, 03.08.1999
MICHAEL BOLANOS MADE IT INTO Who's Who in America last year. To
celebrate, Bolanos, who runs a celebrity-oriented website in New York,
fired off a full-page press release to "entertainment and business
editors" heralding his triumph. "To be chosen for inclusion," read the
release, "candidates must have held a position of responsibility or
have attained a significant achievement in their field." Who's Who,
Bolanos reminded the editors, is a "guide to today's most influential
people."
Congratulations, Mr. Bolanos. Now meet Anita Dawn Sawyer, a fellow
influential person of significant achievement who also recently earned
a place in Who's Who in America. Sawyer, a 1986 graduate of the
University of Central Arkansas, teaches junior high school special
education classes in Little Rock. Since 1991 she's coached bowling and
floor hockey in the Alpena (Arkansas) Special Olympics. According to
Who's Who, her hobbies include cooking, reading, crafts, playing piano
and singing.
Or say hello to Stephen Geiman, who teaches gym at Wilson Memorial High
School in Fishersville, Virginia. From 1970 to 1972, Geiman, a graduate
of the physical education program at Appalachian State University, was
the school's driver's ed instructor. Or David Dolsen, an undertaker in
Denver. Or Amy Fung, an accountant from Staten Island. And let's not
forget Mary Morgan, a 55-year-old social worker in Elizabethtown,
Kentucky. Or Lila Licens, an administrative assistant from Tacoma,
Washington, who has been president of the Mt. Rainier chapter of
Professional Secretaries International since 1994. Or Courtland Paul, a
landscape architect in San Juan Capistrano, who implores Who's Who
readers to "Be on time, produce more than is expected and always,
ALWAYS be fair!!!" And of course there's Marguerite Gearhart, a school
nurse in Jupiter, Florida, who lists among her myriad accomplishments a
1968-69 stint as "co-leader" of a Campfire Girls troupe. Never heard of
these people? Then you haven't read Who's Who lately.
Not that anybody has read Who's Who lately, or ever, at least not very
closely. The point of Who's Who is not to read it, but to be in it. One
hundred years after it was first published by Chicago newspaper
publisher Albert Nelson Marquis (who despite his ostensible commitment
to accuracy pronounced his name "Markwis"), Who's Who has been a fairly
reliable guide to who has made it and who has not. That's been the
marketing strategy, anyway. Flip through the latest volume, however,
and it's hard not to conclude that something has changed, that the
selection criteria for "Honored Biographees" in Marquis's Who's Who
have become--how to put it?--more democratic.
Though the number of entries in Who's Who in America has grown to over
100,000 in recent years, the publication has tried hard to convey the
impression that standards for inclusion have remained the same. Being
accepted into Who's Who is "an honor that only a select few ever
enjoy," the company boasts. Every person in the book is subjected to
"painstaking selection, research, rigorous nominee review, and thorough
editorial review." And who does the painstaking nominating and
selecting? Marquis implies that members of the publication's Board of
Advisors play a large role in the nomination process, but they don't
seem to know much about it.
"The reality is, I don't do anything," says John Fox Sullivan,
publisher of National Journal and a member of the board for the last
decade. "There is almost no communication back and forth. Once a year I
get a piece of paper asking me if I want to recommend someone. It's not
as if there's an annual retreat somewhere where we sit around and
decide who makes it this year. Or if there is, I haven't been invited."
Mindy Aloff, a dance critic whose name is also on Who's Who letterhead,
seems to have been left off the guest list, too. "They didn't give us
any guidelines for nominating people," says Aloff, who rarely forwards
names to the publication.
Then who is making the decisions? Paul Canning, the publication's
editorial director from 1992 to 1997, wouldn't give a specific answer,
though he did say that the admissions process is relatively simple.
According to Canning, in order to become an Honored Biographee in Who's
Who in America, the flagship Marquis publication, a person must meet
"qualitative and quantitative criteria." An artist, for instance, "will
have to have pieces in multiple collections at recognized museums and
have one-person shows. For Fortune 500 companies, senior vice
presidents and above are listed." Some people, said Canning, make the
cut automatically. "We have a thick binder of all the people who must
be included, like artistic directors at ballet companies in major U.S.
cities, or CEOs of Fortune 1000 companies. We look for writers on The
New York Times best-seller list. We have Nobel Prize winners, Oscar
winners."
Fair enough. But Who's Who in America also appears to contain a lot of
relatively unaccomplished people who simply nominated themselves. To
make the process of self-promotion easier, Reed Elsevier, the
publication's parent company and the owner of Lexis-Nexis, now has a
site on the Internet where would-be biographees can complete a
"biographical data form." Spaces are provided for "career history," as
well as for "awards, honors, and grants." Applicants who are
uncomfortable with sending personal information over the Web are
invited to fax their biographies to a number provided on the screen.
There's not a word about qualitative or quantitative criteria. Does
everyone who applies get into Who's Who? "I'll say a majority,"
admitted Canning, "but I can't get any more detailed than that. I think
the majority are appropriate for one of our regional or topical
publications. I think I need to leave it at that." In other words, just
about everyone who tries hard enough will get his name in print.
Donald Ray Grubbs of Portland, Texas, is proof that persistence pays
off. From 1973 to 1986, Donald Ray worked as a pipe fitter and welder
for the Pipefitters Local 195 in Beaumont. Now an employee of Longview
Inspection, a company that assesses the structural integrity of
industrial sites, Grubbs has been appearing in various Who's Who
publications for a decade or so. Only a couple of years ago, he said,
was he "elected" to Who's Who in America. "You work up the chain of
Who's Who documents," Grubbs explained. "I was in Who's Who in American
Education, Who's Who in the World, and then Who's Who in America." When
we talked to Grubbs, he had just received a letter indicating that he
had been inducted into yet another volume, Who's Who in the South and
Southwest. He sounded pleased. "I have nothing but praise to say about
them because I think they're serving a good job. People like me who
really don't get out there in the limelight, this is one of our ways of
getting a little bit of recognition. And it feels good." So good,
Grubbs said, that he has purchased a number of Who's Who products over
the years, including a commemorative wall plaque. (The plaque, he
confided, wasn't of the highest quality.) Yet despite his achievements
in the world of Who's Who, Grubbs doesn't put on airs. "I don't profess
to be a nationally recognized welding instructor," he said. On the
other hand, Grubbs pointed out, neither are a lot of other people in
the book. "Probably half of the welding staff at Ohio State University
are members of Who's Who."
Nationally recognized welding instructor or not, Donald Ray Grubbs
seems like a fairly straightforward person, which is more than can be
said for many of his fellow biographees. As most of those listed in the
book know, entries in Who's Who are mostly self-reported and largely
unchecked, making it the ideal place to tidy up an uneven educational
or work history. When Larry Lawrence, the late ambassador to
Switzerland, wanted to replace his years at Wilbur Wright Junior
College with a degree from the University of Arizona, he turned to
Who's Who. Unfortunately for Lawrence, he got greedy, giving himself a
membership in a veterans association to back up his spurious war
history and a spot on the Nobel Peace Prize Nominating Commission
before he was finally caught (though not by Who's Who) and exhumed from
his grave at Arlington cemetery. Pamela Harriman, another deceased
ambassador, never completed college, but claimed in Who's Who to have
done post-graduate work at the Sorbonne. To this day, columnist Carl
Rowan lists 22 different college degrees in his entry, none of them
identified as honorary. According to Who's Who, Rowan graduated from
three different colleges in 1966 alone, all while working as a
syndicated columnist.
Not all attempts at résumé laundering are so blatant,
though some are considerably more sinister. In 1995, someone at the
Anti-Defamation League of B'nai B'rith noticed that Willis Carto,
founder of the lunatic Liberty Lobby, was listed in Who's Who as a
"publishing executive." Although the ADL promptly notified Marquis that
Carto's publishing ventures consisted of printing anti-Semitic tracts,
his entry was not dropped until 1998. ("He never did buy the book,"
chuckles Carto's spokesman.)
In the mid-1980s, Joe Queenan, then at American Business magazine,
decided to test the Who's Who fact-checking apparatus. Queenan
submitted an application on behalf of a nonexistent magazine editor
named R.C. Webster. Webster, Queenan wrote, had graduated with a master
of fine arts degree from F&M T&A University and received doctorates
from Quaker State University and the University of Ron (Ron, France)
before moving on to edit such magazines as American Business,
Latin-American Business, The Business of Business, Your Business and
Our Business Monthly. Webster and his wife, the former Trish Abigail
Boogen, had children named Cassette, Lothar, Skippy and Boo-Boo. A
member of the Association of Men and the Bureau of People, he listed
his hobby as "managing editing." Who's Who printed most of the entry in
its following edition.
It was an embarrassing episode for Marquis, and thanks to improved
scrutiny, most of the people listed in Who's Who in America these days
almost certainly exist. But the book is still not edited thoroughly,
which means that many entries are printed at lengths curiously out of
proportion to their importance. Margaret Estelle Vorous, for example,
an elementary school librarian in Berkeley Springs, West Virginia, who
counts among her achievements being a blood donor, receives 49 lines in
Who's Who. Henry Kissinger gets only 34. Anita Dawn Sawyer of Harrison,
Arkansas, meanwhile, gets twice the space of Diane Sawyer of ABC News,
who is listed three entries down.
Still, with 105,000 biographies, there are bound to be worthwhile
tidbits buried in Who's Who in America, and there are. Who, for
instance, apart from girls who grew up in the '80s, knew that Pat
Benatar's real name was Pat Andrzejewski? Or that the rap singer Ice
Cube was born O'Shea Jackson? And it's undeniably interesting--if a
little sad--to learn that Playboy founder Hugh Hefner was first married
way back in 1949. Other "facts" in the volume make for less
scintillating reading.
Indeed, the first clue that Who's Who is a vanity publication is the
"Thoughts on My Life" feature that appears beneath some entries. This
is the part where biographees are invited to reflect upon their
achievements using their own words. It's all pretty amusing, and it
must be profitable, too, because Marquis recently decided to expand the
concept. For $150, those listed in Who's Who in America can now write
up to 200 words about themselves and their work. A 1997 direct-mail
pitch suggests that biographees use the "Enhanced Biography" option to
draft their own personal classified ads, sure to be seen by "industry
leaders and executive recruiters." "Over 22 years of progressively
responsible experience in the food service industry in key
decision-making sales and marketing roles," reads one sample entry.
"Recent accomplishments include successful product introductions into
local markets, which generated $12.3 million growth in annual
incremental sales." Perhaps "executive recruiters" really do pore over
Who's Who looking to fill highly paid CEO slots. Or perhaps not. Either
way, it's hard to see how information like this is valuable to
reference librarians, the group for whom the volume is ostensibly
written.
That is, until you notice the large number of librarians who are listed
in Who's Who in America. "We think librarians are important," explained
Paul Canning. "We think they contribute to society." They are certainly
in a position to contribute to Who's Who. The ever-growing Marquis list
now includes 20 different Who's Who volumes, including various CD-ROM
versions, many of which are updated annually. A single three-volume
edition of Who's Who in America can cost more than $500. A three-year
subscription to the entire Who's Who product line goes for $5,686.
Suddenly it becomes clear how Ruth Ferro-Nyalka, a librarian at the
Hinsdale (Illinois) public library, might have breezed through
"Marquis' unique and time-proven compilation process" to earn a spot in
Who's Who in America.
Which is not to imply that vain librarians are Marquis's only source of
income. The company won't say who buys its books, or even how many
copies it prints. "I will not elaborate on anything about Who's Who to
someone over the phone," said publisher Randy Mysel, brusquely. "A fax
won't do it, either." A call to the company's business office proves
more fruitful. Who's Who, it turns out, does a pretty good business
renting the names and addresses of its 250,000 Honored Biographees to
direct mail marketers. People who are listed in Who's Who, Marquis
assures marketers in its promotional literature, "are interested in
many types of offers," including pitches for new credit cards, magazine
subscriptions, catalogues, association memberships and "fundraising
opportunities." The entire database can be rented on computer tape for
about $22,000. Or, the woman on the phone says, the list can be broken
down by profession, sex, political affiliation or religion. There are
17,600 self-identified Catholics in Who's Who, she explains by way of
example, and 5,300 Jews.
It must be a good list, since many Honored Biographees clearly have a
weakness for ordering schlocky products through the mail. Marquis makes
certain they have plenty to buy. The company's "Reflections of Success"
catalog advertises an entire line of Who's Who-related junk, from Who's
Who lapel pins (at $52.95 plus shipping and handling, they "quietly
declare your accomplishments") to Who's Who key rings, paperweights and
crystal boxes. The home office seems to do a particularly brisk
business in commemorative wall plaques, which at close to $100 apiece
doubtless make for a profitable little sideline.
One of the latest offerings from Marquis is the Who's Who/Chevy Chase
Bank MasterCard. Cardholders are eligible for a discount on any
merchandise they buy from the Who's Who catalogue, which brings the
entire enterprise full circle. I'm not listed in any of the Who's Who
volumes, but I decided to order one anyway, mostly to see if I could. I
could. The moment my MasterCard arrived, I called Who's Who. "One
sterling lapel pin, please," I said. "I'm interested in quietly
declaring some of my achievements." "Which book are you included in?"
the woman asked. None, I said. She didn't seem fazed in the slightest.
"Well, you have to be listed," she said brightly. "But you can talk to
the editorial department about that. I'll transfer you."
<http://www.forbes.com/fyi/1999/0308/063.html>
> > One CANNOT request to be included in that Who's Who in the World
> > either. The Editors of the volume seeks out the listees...
>
> Indeed. One must be invited, as I was, as you were. I declined. You
> accepted.
So was Donald Ray Grubbs, a "World Recognized" Welder :-)
Reefy also joins Margaret Estelle Vorous, an elementary school
librarian in Berkeley Springs, West Virginia, who counts among her
achievements being a blood donor: she received 49 lines in Who's Who,
when Henry Kissinger gots only 34. And Anita Dawn Sawyer of Harrison,
Arkansas, a Junior High School Special Ed teacher, had twice the space
of Diane Sawyer of ABC News.
Most intellectuals figured out that Who's Who is a vanity pub before
they turn 25 years old. Reefy's three times that age and still
believes in Marquis, Santa, and the Easter Bunny.
> > > You not only can't show any EDUCATIONAL background in Statistics,...
>
> Neither can you. No one can: this is the Internet. All you can do is
> make unprovable claims.
Reefy can't even show that he has ever done any real applied work in
Statistics. Those papers that weren't "co-authored" consist merely of
magazine reviews of software applications and letters to editors.
Reef Fish
> illywhacker wrote:
> > Reef Fish wrote:
> > > RF> That proved that you LIED. Marquis has many Who's Who books but
> > > RF> you DO NOT get to write your own entry in Who's Who in the WORLD.
> >
> > I quote from the web page:
> >
> > "Who's Who in the World provides instant access to over 50,000
> > personal profiles-reviewed, revised, and proofread by the listees
> > themselves-providing guaranteed accuracy and immediacy."
> >
> > Providing guaranteed hagiography. Presumably one cannot actually write
> > fiction, but adding a degree of spin would certainly be possible.
The EDITORS write the citations. The Listee proofread the citations
for accuracy, as stated. Illywhacker did the LYING.
Ill> > This is most amusing. I too was asked by this defunct book
Illywhacker first called it a defunct book (and I showed the 2007
version)
which was why I said:
RF> LIE LIE LIE LIE Shameless LIE by Illywhacker
RF> TWO LIES in one line: "defunct book"? "asked"?
Now Illywhacker fabricates the hagiography after correctly
citing the Editorial policy of proofreading for accuracy.
Below are apparently written by the known FORGER who
had posted under the nasty_nass...@Yahoo.com
many times before.
Email address: nasty_nassau_grou...@yahoo.com
Who's Who of Lame 63 minutes ago
Moderate, or ignore Sep 8
Some of Reef Fish's behavior MUST NEVER AGAIN BE TOLERATED Sep 3
How to use the English Dictionary PROPERLY to find Statistical Terms
Jul 18
The same FORGER is a dedicated stalker of Reef Fish.
He also FORGED names under Large_Nasty_Grouper:
Posts in All 6 Groups -- 12 messages rec.scuba.locations -- 6
messages sci.stat.math -- 2 messages sci.stat.edu -- 1 message
rec.travel.cruises -- 1 message rec.scuba -- 1 message
rec.gambling.blackjack -- 1 message
Bob Ling is trying to smear his own name AGAIN! sci.stat.math Jul
13
Where is the best casino for Blackjact on Strip, in LV Old town?
rec.gambling.blackjack Jul 7
RTC has become SPAMMER's Capitol in Newsgroups! rec.travel.cruises
May 9
BLing BLathers onn and on... rec.scuba Mar 13
Borneo vs. Palau rec.scuba.locations Mar 2
rec,scuba.locations Mental Midget vs rec.travel.cruises Mental Midgets
rec.scuba.locations Nov 21 2005
I am not sick rec.scuba.locations Nov 15 2005
Who is this "Reef Fish" character? sci.stat.edu Oct 29 2005
Everyone is welcome to attack the SUBSTANCE of what I posted
sci.stat.math Jun 24 2005
Off topic--Buying Property in Cozumel rec.scuba.locations Jun 7
2005
Show all »
Such is the kind Illywhacker attracts.
>
> It is not only possible, but it has been proven:
>
>
> The Hall of Lame
> Tucker Carlson, 03.08.1999
>
> MICHAEL BOLANOS MADE IT INTO Who's Who in America last year.
<snip>
> a place in Who's Who in America. >
>
> Not that anybody has read Who's Who lately, or ever, at least not very
> closely. The point of Who's Who is not to read it, but to be in it.
>
> Though the number of entries in Who's Who in America has grown to over
> 100,000 in recent years,
nasty_nass...@Yahoo.com certainly spent a lot of time
digging up what he posted bashing WHO'S WHO IN AMERICA,
not realizing AMERICA is not the WORLD.
There are hundreds of publications named Who's Who.
Illywhacker probably qualified for Who's Who in Podunk County
in South West New Jersey.
There are dozens Who's Who on parts of the USA, North East,
South West, etc.
Not many make it to a WHO's WHO in the USA or Who's Who in
America, but it's out of 100,000 I am not surprised Nasty Grouper
can find a few to ridicule.
> According to Canning, in order to become an Honored Biographee in Who's
> Who in America, the flagship Marquis publication, a person must meet
> "qualitative and quantitative criteria." An artist, for instance, "will
> have to have pieces in multiple collections at recognized museums and
> have one-person shows. For Fortune 500 companies, senior vice
> presidents and above are listed." Some people, said Canning, make the
> cut automatically. "We have a thick binder of all the people who must
> be included, like artistic directors at ballet companies in major U.S.
> cities, or CEOs of Fortune 1000 companies. We look for writers on The
> New York Times best-seller list. We have Nobel Prize winners, Oscar
> winners."
>
> Fair enough. But Who's Who in America
is Who's Who in America.
Why couldn't Illywhacker or nasty_nassau_grouper find some Listee's
in WHO'S WHO IN THE WORLD to make the same ridicule?
There are 215 countries in the WORLD and not many of them are
in America.
Perhaps nasty and whacky hasn't learned their geography yet.
-- Reef Fish Bob.
Robert F Ling
> Why couldn't Illywhacker or nasty_nassau_grouper find some Listee's
> in WHO'S WHO IN THE WORLD to make the same ridicule?
Because that's already been done and Reefy can't read:
"Donald Ray Grubbs...a pipe fitter and welder...has been
appearing in various Who's Who publications for a decade
or so. Only a couple of years ago, he said, was he
"elected" to Who's Who in America. "You work
up the chain of Who's Who documents," Grubbs
explained. "I was in Who's Who in American Education,
Who's Who in the World, and then Who's Who in America."
^^^^^^^^^^^^^^^^^^^^^^^
Has Reefy been in ALL THREE of these Marquis publications TOO?
"When we talked to Grubbs, he had just received a letter
indicating that he had been inducted into yet another volume,
Who's Who in the South and Southwest."
Correction: FOUR Marquis publications!
"Yet despite his achievements in the world of Who's Who,
Grubbs doesn't put on airs... "Probably half of the welding
staff at Ohio State University are members of Who's Who."
The Ohio State WELDING staff? Must be THEIR skills at FLAMING :-)
Reef Fish
the namesless Chicken Manure, had to FORGE and hide
to make his ad homimen attacks:
Robert's Who's Who of Lame 105 minutes ago
Who's Who of Lame 4 hours ago
Moderate, or ignore Sep 8
Some of Reef Fish's behavior MUST NEVER AGAIN BE TOLERATED Sep 3
How to use the English Dictionary PROPERLY to find Statistical Terms
Jul 18
The same Chicken Manure FORGED with
large_nas...@Yahool.com in his other dozen posts:
Posts in All 6 Groups -- 12 messages rec.scuba.locations -- 6
messages sci.stat.math -- 2 messages sci.stat.edu -- 1 message
rec.travel.cruises -- 1 message rec.scuba -- 1 message
rec.gambling.blackjack -- 1 message
Bob Ling is trying to smear his own name AGAIN! sci.stat.math Jul
13
Where is the best casino for Blackjact on Strip, in LV Old town?
rec.gambling.blackjack Jul 7
RTC has become SPAMMER's Capitol in Newsgroups! rec.travel.cruises
May 9
BLing BLathers onn and on... rec.scuba Mar 13
Borneo vs. Palau rec.scuba.locations Mar 2
rec,scuba.locations Mental Midget vs rec.travel.cruises Mental Midgets
rec.scuba.locations Nov 21 2005
I am not sick rec.scuba.locations Nov 15 2005
Who is this "Reef Fish" character? sci.stat.edu Oct 29 2005
Everyone is welcome to attack the SUBSTANCE of what I posted
sci.stat.math Jun 24 2005
Off topic--Buying Property in Cozumel rec.scuba.locations Jun 7
2005
THAT's the kind of worthless SLIMES and psychopaths
Illywhacker attracks.
Whack away, Illywhacker.
-- Reef Fish Bob.
illywhacker
wrote:
> THAT's the kind of worthless SLIMES and psychopaths
>
> Illywhacker attracks.
But it is you that attracts them Reefer. By your own admission, he is
stalking you!
illywhacker;
Reef Fish
But look at ALL those people whom attracted nasty_nassau_grouper
-- they are, as I said, the kinds like YOU, illywhacker.
If he were stalking just me, then he is a pretty sorry stalker, for
having
forged only about 15 times, about once every 10,000 posts I make.
No, he is selective in sniffing out the worst SLIMES on the internet,
and use THEM as his lead-in. That's the worst of all cowards --
not only have to FORGE to make a statement of personal opinion,
but has to find someone else, such as Illywhacker, to ride behind
his posterior, under a forged name,to make HIS noise.
-- Reef Fish Bob.
illywhacker
On Oct 12, 4:54 pm, "Reef Fish" <large_nassua_grou...@yahoo.com>
wrote:
> not only have to FORGE to make a statement of personal opinion,
But Reefer, you are not Bob Ling, you know. You are merely someone who
has read his overinflated Who's Who entry.
illywhacker;
Reef Fish
illywhacker wrote:
> On Oct 12, 4:54 pm, "Reef Fish" <large_nassua_grou...@yahoo.com>
> wrote:
> > not only have to FORGE to make a statement of personal opinion,
>
Your knowledge about Bayesian statistics was exhausted in one day.
Your knowledge about Statistics and Mathematics was exhuasted a
day or two later.
Ever since, you've done NOTHING but worthless ad hominem posts.
The latest is the DUMBEST -- attacking those who are listed in
Who's Who in the World, attracting the psychopath known FORGER,
nasty_nassau_grouper.
Illiwhacker, H O W D U M B C A N Y O U G E T
???
-- Reef Fish Bob.
illywhacker
Apparently you posted the same diatribe in two threads. This is getting
out of hand. I do think this pointless conversation has run its course.
Neither of
us has written anything worth reading for a number of days, and this is
now seriously messing up the newsgroup for other people. So waddya say
we call it a day?
illywhacker;
Reef Fish
illywhacker wrote:
> Dear Bob,
>
> Apparently you posted the same diatribe in two threads. This is getting
> out of hand. I do think this pointless conversation has run its course.
> Neither of
> us has written anything worth reading for a number of days,
Speak for YOURSELF about not having written anything worth reading
for a number of days. This is MY most recent posting profile (Google)
Posts in All 7 Groups -- 207 messages sci.stat.math -- 156 messages
rec.travel.cruises -- 17 messages sci.math -- 14 messages
sci.stat.consult -- 13 messages rec.travel.air -- 3 messages
sci.stat.edu -- 3 messages rec.scuba.locations -- 1 message
MCMC, was What do you mean when you say you don't know anything about
the parameter p of a Binomial distribution? sci.stat.math 77
minutes ago
Testing hypothesis alt.sci.math.probability 3 hours ago
Testing hypothesis alt.sci.math.probability 4 hours ago
ALL pure discussions about Statistics (I cut out those responding to
the NOISE by Illywhacker and his partner FORGER nasty_nassau_grouper)
These are Illywhacker's most recent posts (all from Google profile,
without
cutting ANYTHING):
Posts in All 2 Groups -- 5 messages sci.stat.math -- 4 messages
alt.sci.math.probability -- 1 message
Let's stop this nonsense (again) sci.stat.math 15 minutes ago
Let's stop this nonsense sci.stat.math 18 minutes ago
Donald the Welder vs Bob the Fake Professor sci.stat.math 47
minutes ago
Illywhacker continues to Whack, Whack. Whack. ...and Whack some more
sci.stat.math 51 minutes ago
Illywhacker and FORGER stalker nasty_nassau_grouper sci.stat.math
69 minutes ago
Illywhacker continues to Whack, Whack. Whack. ...and Whack some more
sci.stat.math 3 hours ago
Illywhacker continues to Whack, Whack. Whack. ...and Whack some more
sci.stat.math 6 hours ago
Illywhacker caught LYING about his Who's Who invitation
sci.stat.math 6 hours ago
Illywhacker continues to Whack, Whack. Whack. ...and Whack some more
sci.stat.math 6 hours ago
Illywhacker caught LYING about his Who's Who invitation
sci.stat.math 6 hours ago
This PATTERN of me posting Statistics discussion AS USUAL while
Illywhacker and m00es posted nothing but NOISE has been ongoing for
DAYS.
Illywhacker was too BLINDED and DEAFENED his own noise to have
even noticed.
Illiwhacker> Neither of
> us has written anything worth reading for a number of days,
Illiwhacker has not written anything worth reading for MONTHS.
illywhacker
Reef Fish wrote:
> This is MY most recent posting profile (Google)
> the NOISE by Illywhacker and his partner FORGER nasty_nassau_grouper)
>
> These are Illywhacker's most recent posts (all from Google profile,
> without
> cutting ANYTHING):
Oh - that's objective.
> Illiwhacker has not written anything worth reading for MONTHS.
You are right. I have been too busy doing actual research.
illywhacker;
Reef Fish
While illywacker was the NOISIEST poster in sci.stat.,math, with his
total of 38 PURE NOISE posts, far exceeding the next top two noise
makers of m00es and Bob O'Hara, totalling 26.
This month's top posters
89 Reef Fish
22 illywhac...@free.fr
17 m00es
16 illywhacker
9 bob.oh...@nospamhelsinki.fi
8 Reef Fish
> You are right. I have been too busy doing actual research
You need to do MUCH more on learning Statistics from statisticians
-- Reef Fish Bob.
AB
by HPD I mean "high posterior density";
by "high posterior density" (e.g. at 95%) I mean an interval whose
probability under the posterior density is 95%, and whose extreme
points have the same density;
I found a solution to my problem in R, "TeachingDemos" package, "HPD"
function (e.g.: hpd(qbeta, shape1=50, shape2=250,conf=0.95)). Maybe my
problem was simpler than you thought, and I explained it in a bad
way... I'm sorry for the mess I created, please conciliate!
That' all,
thanks
AB
Reef Fish ha scritto:
> illywhacker wrote:
> > Your post contains nothing but invective, thereby demonstrating that
> > you have lost the argument but are unable to admit it, as usual.
> >
> > illywhacker;
>
> What invective? I am citing EVERY WORD I posted which you snipped:
>
> RF> I understand the difference between those two perfectly.
>
> That was a statement of FACT.
>
> RF> Perhaps you're drawing Fisher, incorrectly, into your physicists
> camp
> RF>of butchers of Bayesian statistics. Fisher can be said to be a
> NON-
> RF> Bayesian, but he is NOT a butcher of Bayesian statistics like
> RF> illywhacker.
>
> That was a statement of FACT.
>
> RF> illywhacker, give it up.
>
> That was a statement of opinion, based on illywhacter's complete
> lack of understanding about Bayesian statistics.
>
>
> RF> You don't understanding ANYTHING about Bayesian Statistics and
> RF> Bayesian inference. All you throw around are some buzz words
> RF> completely unrelated to Bayesian inference as evidence by your
> RF> plenty of misuse and abuse of those words already.
>
> That was a statement of FACT.
>
>
> Where were the invectives?
>
> Saying you don't understand ANYTHING about Bayesian Statistics
> and Bayesian inference?
>
> Saying all you did was to throw around buzz words unrelated to
> Bayesian inference?
>
> Saying that those were EVIDENCED by your misuse and abuse
> of those buzz words?
>
>
> You're were just running out of excuses for your own blunders.
>
> -- Reef Fish Bob.
illywhacker
AB wrote:
> Dear all,
> by HPD I mean "high posterior density";
> by "high posterior density" (e.g. at 95%) I mean an interval whose
> probability under the posterior density is 95%, and whose extreme
> points have the same density;
Yes, we did realize this eventually, some earlier than others, which is
why I sent you a link to the R package.
> I found a solution to my problem in R, "TeachingDemos" package, "HPD"
> function (e.g.: hpd(qbeta, shape1=50, shape2=250,conf=0.95)).
Good. I guess there was already something available in another package.
> I'm sorry for the mess I created, please conciliate!
No need to apologize: you did not create the mess, AB. RF did, as he
always does when someone dares to talk back to him. There is not much
chance of conciliation though, I am afraid: that is not the RF way.
illywhacker;
Reef Fish
AB wrote:
> Dear all,
> by HPD I mean "high posterior density";
> by "high posterior density" (e.g. at 95%) I mean an interval whose
> probability under the posterior density is 95%,
That is indeed the definition given by Wikipedia.
It is the Bayesian CREDIBLE interval which may NOT even contain
the mode of the posterior distribution!
Just as a confidence interval is defined only by the coverage or
probability content, the same is true for Bayesian credible intervals.
If the posterior mode is a sharp spike far away from the bulk of
the high density region, such as,
x
x
x
x xxxxxxxxxx
x xxxxxxxxxx
x xxxxxxxxxxxx x x
_____________________________________
then the shortest credible interval of high percent will EXCLUDE
the m0de (the highest point of the posterior density).
That is another reason why it is silly to find just the MODE when
you have an entire distribution for the unknown parameter theta.
The likelihood function (used by nonBayesians) to find just the
POINT estimate MLE because the likelihood function is NOT
a probability distribution for the unknown parameter (which is
FIXED in nonBayesian statistics). A non-Bayesian confidence
interval is based on a realization of a RANDOM INTERVAL
with a given probability of including the fixed parameter.
Thus, for a MAP user who finds the maximum posterior POINT,
the mode, it would indeed find the spike (which is very silly
indeed for a Bayesian) because it's INTERVAL coverage may
be virtually ZERO when the bulk of the coverage of the unknown
theta is elsewhere.
This should bring back, for those who read that portion of the
argument about the silliness of bring in Fisher's information
idea just to show ignorance in the prior distribution, when
Lou Thraki asked something to the effect:
Why be a Bayesian and go to all that effort to express an
uninformative prior just to find the MLE?
The picture above, together with the idea of HPD being a
CREDIBLE INTERVAL now explain fully that MAP, as mistaken
by Illywhacker to mean the mode of the posterior density, is
not only WRONG in the definition, by also WRONG in what
it accomplishes, to find only the highest point.
> and whose extreme
> points have the same density;
> I found a solution to my problem in R, "TeachingDemos" package, "HPD"
> function (e.g.: hpd(qbeta, shape1=50, shape2=250,conf=0.95)). Maybe my
> problem was simpler than you thought, and I explained it in a bad
> way... I'm sorry for the mess I created, please conciliate!
> That' all,
> thanks
> AB
Glad to see you clarify that you indeed meant HPD (as described by
Wikipedia as a credible INTERVAL) rather than the non-Bayesian
acronym of MAP which Illywhacker mistakenly assumed your HPD was,
and continued to argue that it was Bayesian.
This post should clarify for both Bayesian and non-Bayesian WHY
HPD is a good idea (as in confidence interval of coverage) or a
point estiamte such as MLE; and further clarified that MAP and
HPD are two different concepts and methods!
AB, I only wished you had come forth with you answer of what you
SOUGHT (even before you find the solution in R), because just
knowing that you were looking for the HPD rather than the non-
Bayesian MAP, would have saved days of NOISE by Illywhacker
and tons of bandwidth.
-- Reef Fish Bob.
>
> Reef Fish ha scritto:
>
> > illywhacker wrote:
> > > Your post contains nothing but invective, thereby demonstrating that
> > > you have lost the argument but are unable to admit it, as usual.
> > >
> > > illywhacker;
> >
> > What invective? I am citing EVERY WORD I posted which you snipped:
The words Illywhacker snipped included my statement that MAP is
NOT the synonym for HPD which is a Bayesian credible interval, while
MAP is what the physcists called the posterior mode.
> >
> > RF> Perhaps you're drawing Fisher, incorrectly, into your physicists
> > camp
> > RF>of butchers of Bayesian statistics. Fisher can be said to be a
> > NON-
> > RF> Bayesian, but he is NOT a butcher of Bayesian statistics like
> > RF> illywhacker.
> >
> > That was a statement of FACT.
> >
> > RF> illywhacker, give it up.
> >
> > That was a statement of opinion, based on illywhacter's complete
> > lack of understanding about Bayesian statistics..
illywhacker
Reef Fish wrote:
> now explain fully that MAP, as mistaken
> by Illywhacker to mean the mode of the posterior density
You are wrong. MAP estimation = maximum a posteriori estimation =
estimating a quantity by finding that value that maximizes the
posterior density.
> rather than the non-Bayesian
> acronym of MAP which Illywhacker mistakenly assumed your HPD was,
In your reply to the OP, the second post in this thread, you said:
RF> You use it to find the value of theta (the parameter) that
maximized the posterior
RF> density.
Thus it was you that misinterpreted the OP. In contrast, I recognized
your error long before you did, and posted a reply to the OP, the first
in the thread to mention HPD after the OP's, that suggested software
in R for HPD. I then pointed out to you several times what the OP was
asking about, but you ignored me. Then you pretended you had come up
with the idea. As usual, you cannot read and you are misrepresenting
the situtation.
> and continued to argue that it was Bayesian.
I do not argue about whether it is Bayesian, as I have said many times,
because only incompetents like you need to fixate on such trivial
points. Had you chosen to discuss whether MAP estimation is a useful
procedure, what are its drawbacks, etc., we might have had a meaningful
discussion. Instead you chose to get abusive after I corrected you by
pointing out that normalization is not necessary if you want to
maximize the posterior density.
illywhacker;
Anon.
> AB wrote:
>> Dear all,
>> by HPD I mean "high posterior density";
>> by "high posterior density" (e.g. at 95%) I mean an interval whose
>> probability under the posterior density is 95%,
>
> That is indeed the definition given by Wikipedia.
>
Incidentally, one would call that the higest posterior density interval.
Leaving off the "interval" is what caused a lot of confusion and
strife. :-(
> It is the Bayesian CREDIBLE interval which may NOT even contain
> the mode of the posterior distribution!
>
For a multi-modal distribution, one could of course construct a highest
posterior density region, which may not be a single interval. I think
this would be more in keeping with the philosophy behind the HPDI(R?!).
>
> Glad to see you clarify that you indeed meant HPD (as described by
> Wikipedia as a credible INTERVAL) rather than the non-Bayesian
> acronym of MAP which Illywhacker mistakenly assumed your HPD was,
> and continued to argue that it was Bayesian.
>
It's perhaps worth clarifying that a HPDI is only one possible credible
interval: a credible interval just has to contain the correct amount of
probability mass. One could also, for example, construct a symmetric
credible interval from the (1-p)/2 and 1-p/2 quantiles.
Incidentally, there is also a HPDinterval() function in the coda package.
Bob
--
Bob O'Hara
Dept. of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland
Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB: http://www.jnr-eeb.org
Reef Fish
illywhacker wrote:
> Reef Fish wrote:
> > now explain fully that MAP, as mistaken
> > by Illywhacker to mean the mode of the posterior density
>
> You are wrong. MAP estimation = maximum a posteriori estimation =
> estimating a quantity by finding that value that maximizes the
> posterior density.
Even with AB's clarification that he was looking for HDP, the credible
INTERVAL, and my explanation that said interval may not even contain
the maximum value of the density, you are STILL in a fog about the
entire issue?
>
> > rather than the non-Bayesian
> > acronym of MAP which Illywhacker mistakenly assumed your HPD was,
> I do not argue about whether it is Bayesian, as I have said many times,
> because only incompetents like you need to fixate on such trivial
> points. Had you chosen to discuss whether MAP estimation is a useful
> procedure, what are its drawbacks, etc., we might have had a meaningful
> discussion.
No way, LIAR.
You lied about everything, and try to change the history of the
discussion.
If you had wanted to DISCUSS anything, you would have responded to
Lou Thraki's question, and MY more precise question as to WHY any
Bayesian who has an entire posterior distribution would want to find
just the maximum point?
>
> illywhacker;
You are not only IGNORANT. You are not worthy of any Statistical
discussion. NOW that AB had thoroughly exposed your errors, and I
even explained further that what AB looked for does not even have
to contain whatever the physicists call MAP, Illywhacker is STILL
WHACKING, WHACKING, and WHACKING some more.
-- Reef Fish Bob.
illywhacker
1) I noticed that the OP was talking about intervals in my post of Oct
9. Which post of yours mentions intervals before that?
2) I then mentioned this fact to you several times in subsequent posts.
You ignored me. You finally posted about intervals much later. Which
post of yours about intervals preceded my first notification to you
that intervals were involved?
3) You mentioned the posterior mode in the second post of the thread.
No post of mine mentioning it can have occurred before that. Which post
of mine mentions posterior mode before you do?
4) You insisted that MAP estimation did not involve finding the
posterior mode in several posts. You were wrong, as citations from two
of the most respected statistical journals showed. Do you disagree with
Professor Besag that MAP estimation means finding the posterior mode?
5) You continue to insist that no one could describe MAP as 'Bayesian'
(not that I care) despite citations from two of the most respected
statistical journals that contradict you. Can you please cite one of
your Bayesian publications?
RF> No way, LIAR.
This constitutes your entire response to all these points. Together
with your bungle in elementary combinatorics in another thread, we can
conclude that you are merely an arrogant bully with the caps lock key
on.
illywhacker;
P.S.
Reef Fish wrote:
> and MY more precise question as to WHY any
> Bayesian who has an entire posterior distribution would want to find
> just the maximum point?
If you knew anything about modern uses of statistics in the analysis of
spatial processes, you would realize that the 'entire posterior
distribution' is often not readily available. MAP, which, as I am sure
you will agree, is not the greatest of techniques for many reasons,
including lack of invariance, is one way to extract information from
otherwise intractable posteriors. But you have probably never dealt
with a problem involving millions of mutually dependent variables.
Reef Fish
illywhacker wrote:
> To summarize our discussion so far, some simple questions:
>
> 1) I noticed that the OP was talking about intervals in my post of Oct
> 9. Which post of yours mentions intervals before that?
Illywhacker's initial post was Oct 7, when he mistook MAT to be HPD.
His post of Oct 9 was about 15 posts later (as I had documented in
great detail, post by post). He keeps bring back the same EXCUSE,
with the irrelevant date of Oct 9.
The rest are just more OLD excuses that are unsupported. Here's the
entire summary:
AB> by HPD I mean "high posterior density"; by "high
AB> posterior density" (e.g. at 95%) I mean an interval
AB> whose probability under the posterior density is 95%,
That is indeed the definition given by Wikipedia.
It is the Bayesian CREDIBLE interval which may NOT
even contain the mode of the posterior distribution!
That KEY POINT above was explained below, that the
HDP interval may not even have ANYTHING to do with
the maximum point of the posterior distribution.
Just as a confidence interval is defined only by the
coverage or probability content, the same is true for
Bayesian credible intervals.
If the posterior mode is a sharp spike far away from the
bulk of the high density region, such as,
x
x
x
x xxxxxxxxxx
x xxxxxxxxxx
x xxxxxxxxxxxx x x
___________________________________________
then the shortest credible interval of high percent will
EXCLUDE the mode (the highest point of the posterior
density).
That is another reason why it is silly to find just the
MODE when you have an entire distribution for the
unknown parameter theta.
The likelihood function (used by nonBayesians) to find j
ust the POINT estimate MLE because the likelihood
function is NOT a probability distribution for the unknown
parameter (which is FIXED in nonBayesian statistics).
A non-Bayesian confidence interval is based on a
realization of a RANDOM INTERVAL with a given
probability of including the fixed parameter.
Thus, for a MAP user who finds the maximum posterior
POINT, the mode, it would indeed find the spike (which
is very silly indeed for a Bayesian) because it's INTERVAL
coverage may be virtually ZERO when the bulk of the
coverage of the unknown theta is elsewhere.
========================================
This should bring back, for those who read that portion
of the argument about the silliness of bring in Fisher's
information idea just to show ignorance in the prior
distribution, when **Lou Thraki ** asked something
to the effect:
Why be a Bayesian and go to all that effort to express
an uninformative prior just to find the MLE?
========================================
The picture above, together with the idea of HPD being
a CREDIBLE INTERVAL now explain fully that MAP, as
mistaken by Illywhacker to mean the mode of the posterior
density <in Illywhacker's FIRST POST, and the 3rd post
in the thread, after AB's OP and my follow-up> is not
only WRONG in the definition, but also WRONG in what
it accomplishes, to find only the highest point.
This post is a self-contained post summarized the
SUBSTANCE that was polluted by various NOISE posted
in sci.stat.math regarding the question (and Bayesian
SUBSTANCE) of what the OP AB asked, and now
clearifies.
The entire post will be repeated if Illywhacker continues
to make his NOISE in thread thread, with the descriptive
SUBJECT.
-- Reef Fish Bob.
illywhacker
> illywhacker wrote:
> > To summarize our discussion so far, some simple questions:
> >
> > 1) I noticed that the OP was talking about intervals in my post of Oct
> > 9. Which post of yours mentions intervals before that?
>
> Illywhacker's initial post was Oct 7, when he mistook MAT to be HPD.
> His post of Oct 9 was about 15 posts later
You did not answer the question. Here is the archive record.
Oct 9: I post links to an R software package to perform HPD interval
estimation
(http://groups.google.com/group/sci.stat.math/msg/d80ede11f793153d?dmode=source&hl=en).
Oct 9: I point out to you that the OP was asking about HPD interval
estimation
(http://groups.google.com/group/sci.stat.math/msg/22c319508a32dc88?dmode=source&hl=en):
ILLY> Actually, this is not the question the OP asked. The OP asked
ILLY> about HPD intervals I believe.
Oct 10: you are still insisting that the OP is talking about
posterior mode
(http://groups.google.com/group/sci.stat.math/msg/a4e78bf6fc6795a0?dmode=source&hl=en):
RF> The OP was asking a question about finding the maximum of a
RF> posterior distribution in a Bayesian analysis.
Oct 10: I again pointed out to you that the OP was probably asking
about HPD intervals
(http://groups.google.com/group/sci.stat.math/msg/85d1cc6dab4632c1?dmode=source&hl=en).
Oct 10: I again pointed out to you that you had misinterpreted the OP's
question
(http://groups.google.com/group/sci.stat.math/msg/a03a12d2e374d823?dmode=source&hl=en).
Oct 11: I again pointed out to you that you had misinterpreted the OP's
question
(http://groups.google.com/group/sci.stat.math/msg/7d46cac3919c94bd?dmode=source&hl=en).
Oct 13: you finally realize that the OP is talking about interval
estimation
(http://groups.google.com/group/sci.stat.math/msg/ddfe8aa2bbc34f45?dmode=source&hl=en).
I answered the rest of your oft-repeated post in another thread.
illywhacker;
AB
Reef Fish ha scritto:
> AB wrote:
> > Dear all,
> > by HPD I mean "high posterior density";
> > by "high posterior density" (e.g. at 95%) I mean an interval whose
> > probability under the posterior density is 95%,
>
> That is indeed the definition given by Wikipedia.
>
> It is the Bayesian CREDIBLE interval which may NOT even contain
> the mode of the posterior distribution!
>
> Just as a confidence interval is defined only by the coverage or
> probability content, the same is true for Bayesian credible intervals.
>
> If the posterior mode is a sharp spike far away from the bulk of
> the high density region, such as,
>
> x
> x
> x
> x xxxxxxxxxx
> x xxxxxxxxxx
> x xxxxxxxxxxxx x x
> _____________________________________
> then the shortest credible interval of high percent will EXCLUDE
> the m0de (the highest point of the posterior density).
>
Anyway, I'd try to distinguish between HPD and Credible Set (CR), since
CR should be "similar" to a confidence interval, as it leaves the same
probability to the right and to the left tail, while HPD is a sort of
"level set", including all the points having a density greater than a
certain value...(and so, it may sometimes not be an interval...)
P.S.: I'm sorry for my delay, but I don't speak English very well, and
so I sometimes find it difficult to understand all the posts (as I'm
neither a statistician nor a mathematician)!
AB
Reef Fish
Yes, I've seen the description of a still farther departure of HPD
from a Bayesian credible Interval to a Credible Set.
That is NOT even a STATISTICAL estimation concept (except in very
rare special circumstances), let alone a Bayesian concept. The usual
concept of a credible interval is (as in CI) to choose the SHORTEST
interval that has the Highest probability coverage of a prescribed
probability number, such as .95 or .9. The idea of breaking such an
INTERVAL into many bits and pieces simply because of the height
of the density function is statistical perversion of no useful content,
that are dreamed up by NON-statisticians, in their own fields.
> P.S.: I'm sorry for my delay, but I don't speak English very well, and
> so I sometimes find it difficult to understand all the posts (as I'm
> neither a statistician nor a mathematician)!
>
> AB
You did VERY well. As soon as you explained what your HPD
question was, it became 100% clear that it was NOT whatever MAP
Illywhacker alledged that they were synonyms. There was absolutely
no ambiguity about what you described.
And now, as soon as you mention the alternative idea of CR, it also
became 100% clear what it was meant, and quite different from a
credible interval.
The NOISE from Illywhacker and Bob O'Hara in the entire thread
was their OWN fault of not understand what Bayesian statistics is
all about, and in their mistaken notion that the NON-Bayesian idea
of their MAP has anything to do with your HPD.
The notion of HPD IS virtually the same as a Bayesian credible
interval (assuming the shortest one) without any mention or
implication about the HIGH part of the posterior density, because
in order for the interval to contain as high a coverage probability
as possible with the shortest length, the interval NATURALLY
would seek the interval around the highest points of the
posterior distribution.
The only thing unnatural among the ideas you have mentioned is
that of the CS (credible set) simply because of the height of the
density function. I would suggest that you think about WHY you
may want such a set of disjoint sets to add up to a prescribed
probability because finding such solutions.
Thanks again for clarifying your question, which is MUCH more
Bayesian oriented CREDIBLE interval type, rather than the
mistaken notion contrived by Illywhacker from his physicists
who have no idea what Bayesian inference is.
-- Reef Fish Bob.
Anon.
No, that's the concept behind the HPDI. A credible interval is simply
an interval containing the correct amount of probability mass: obviously
there are several ways of creating such an interval (e.g. see
http://en.wikipedia.org/wiki/Credible_interval).
The idea of breaking such an
> INTERVAL into many bits and pieces simply because of the height
> of the density function is statistical perversion of no useful content,
> that are dreamed up by NON-statisticians, in their own fields.
>
I'm interested (yes, genuinely) in why you think this is a "statistical
perversion". Obviously reporting several intervals for a credible
region is a bit of a pain, but it seems to be more in keeping with the
philosophy of a HPDI, i.e. reporting the region with highest probability
mass.
Bob
--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland
Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB: www.jnr-eeb.org
Reef Fish
> Bob O'Hara
I don't doubt that you are interested, Anon Bob.
But your FREE tuition had been withdrawn, and if there's such a
thing as "probation period" to see if Bob O'Hara would learn to
keep his mouth shut when he had nothing to contribute, you had
violated that probation repeated.
So, Bob O'Hara, you'll have to sit at your dunce corner with your
Dunce Hat until the time comes on other occasions -- or you could
do some reading of the STATISTICAL literature on Interval
Estimation, and do some thinking yourself to come up with the
answer.
-- Reef Fish Bob.
Anon.
Sorry, but I honestly am: I have seen the idea of splitting the
confidence region into several intervals suggested in a well-accepted
Bayesian textbook (i.e. one used by Bayesian statisticians), so I'd like
to know what the reasons are for it being a bad idea.
st...@mimosa.csv.warwick.ac.uk
A few further comments (+ references from standard textbooks):
An "HPD region" is universally defined to be a region
(usually with pre-specified probability content) such that
the density outside the region is no higher than that within.
See e.g.
Box & Tiao "Bayesian Inference in Statistical Analysis" (1973),
page 123; Cox & Hinkley "Theoretical Statistics" (1974), page 390;
Schervish "Theory of Statistics" (1995), page 327.
In the unimodal case, if the HPD region is an interval,
then it is called an HPD interval, and is an example of
a "credible interval". [Some authors apparently define credible
interval more strictly, and demand equal probability content
in the omitted left and right tails.]
HPD regions are mainly used for summarising the posterior density
- e.g. C&H (Cox & Hinkley) say (p.390) "(c) If the parameter is
multidimensional, or if the density is multimodal, then the simplest
description of the posterior distribution may be via a series of
regions of high posterior density"
(regions which they then define to be HPD regions - so in 2-d you get
a contour map where the contours have specified probability content).
Note that C&H mention multimodality, so are specifically allowing
"HPD regions" to be disconnected. It's unfortunate that the
section in C&H is headed "Interval estimation"!
Credible intervals, as you say, are more analogous to confidence
intervals. Schervish says more about HPDs vs CIs. Schervish and
Box & Tiao also discuss the dominating measure and parameter
transformations.
Finally note that HPD regions of the marginal distributions, such as
contour plots of all 2-way marginals, can still be very poor summaries
of the overall "shape" of the higher-dimensional posterior distribution.
>P.S.: I'm sorry for my delay, but I don't speak English very well, and
>so I sometimes find it difficult to understand all the posts (as I'm
>neither a statistician nor mathematician)!
>
You express yourself very clearly. The references I've given above
may be too technical if you're neither a statistician nor a
mathematician, but all my other books that mention HPD regions
only seem to consider straightfoward univariate situations
where the region is an interval.
Regards, Ewart Shaw
--
J.E.H.Shaw [Ewart Shaw] st...@uk.ac.warwick TEL: +44 2476 523069
Department of Statistics, University of Warwick, Coventry CV4 7AL, UK
http://www.warwick.ac.uk/statsdept http://www.ewartshaw.co.uk
Will read usenet posts up to the first insult (which may be in the Subject)