Re: Statistics in Psychology?


Herman Rubin wrote:
In article <1151036033.498350.124340@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
Reef Fish <Large_Nassau_Groupen@xxxxxxxxx> wrote:


You often use MDs as an analogy for the
training required by a compotent statistician.

One cannot TRAIN a competent statistician.

That's a matter of rhetoric or semantics. You could say you cannot
train a competent medical doctor either. That's nonsense.

I agree; the ability has to be there. There is too much
training going on.

I knew my statement was ambiguous as soon as I read it after I posted
it.
What I meant to say, given Herman said that you can TRAIN a competent
medican doctors but not a competent statistician -- my statement was
meant to first point out the "rhetoric" that you can train a medical
doctor
but one might argue (on Herman's semantic and rhetoric) that you cannot

train a "competent medical doctor" -- that was what the "either" meant
to imply why Herman's statement was nonsense.

And THEN I went on to say that there is as much TRAINING required
for an APPLIED statistican as for a medical doctor -- to reach a
certain
level of competency, only using different kinds of tools.

I am glad I have the opportunity to clerity that point which was made
all too briefly in two lines so as to have lost the intended meaning
through ambiguity.

See my comment about TRAINING below.

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

You cited me correctly and also correctly interpreted my sentiment
that Statistics require as much training as a medical doctor or a
neuro-surgeon (I have also used "brain surgeon" at other times).

A surgeon needs training in the use of his tools, and
in the ability to make quick decisions based on his
knowledge of anatomy, etc.

The same with a applied statistician or a data analyst, who needs
training in the use of his STATISTICAL tools.


The tools a surgeon uses require manual dexterity.

But much of the manual dexterity is left to the assistant surgeons --
they are ROUTINE in the life and training of a surgeon.

MENTAL dexterity, on the other hand, is the quality that is
REQUIRED and ESSENTIAL to surgeons and statisticnas alike.

They
are not devisable "on the spot" in general, as they
require materials to be appropriately formed. A good
statistician can invent his tools on the spot, and does
not need training in them.

The kind of "on the spot" treatment are often in emergency
hospitals in tents at war.

NO competent statistician needs to invent his tools on the
spot (and does not need training on them) -- they are TRAINED
to deal with most of the problems as doctor are TRAINED to
deal with most of the problems patients have.

It's ONE and the SAME training needed for competency!

Only Mathematical Statisticians, who learned nothing but
meaure theory and theory of integration and some lip service
on decision theory and risks need to "devise his tools on the
spot" -- they simply did NOT RECEIVE the proper training
and education about data analysis and applied statistical
methodology, that's all. My example of an advanced Ph.D.
student at Harvard (an extreme outlier who is completely
ignorant about how to analyze data) who is otherwise
brilliant in "mathematical statistics" is an extreme example
of the TYPICAL example of a student who was educated in
a statistics department like those where Herman Rubin taught.

They are NEITHER applied statisticians nor even statisticians.

They are MATHEMATICIANS in a little special areas they
call statistics, just like "matrix algebra" or "topology" where
they only thing that matters is axiomatic and derivation of proofs
that are of no relevance to anything applied.


Also, a surgeon may have to act quickly. Rarely does a
statistician have to act quickly.

That's an irrelevant point, Herman. Doctors are NOT trained to
be speedy demons on rash decisions. Some doctors spend MONTHS
treating the same patients based on the doctor's acquired skills
through his medical training.


A competent statistician
needs to be able to apply various fields of mathematical
knowledge to the structure of the user's problem, but
the patient is not going to die while texts or computers
are being consulted. Nor is there physical skill involved.


A poorly conducted economic analysis (and there are plenty of
them) could lead to governmental policies that are tentament to,
or much worse than, killing thousands of the citizens.

What leads you to believe that these are not the usual
type of analyses?

They are indeed quite USUAL -- because the analysis were done by
people in economics and enconometrics who are NOT statisticians,
and who never had the prerequite TRAINING of being an applied
statistician!

Hence, they made a MESS of the world because their analyses are
often taken by Presidents who are even more ignorant than they are
as if they had done the correct analysis.

That kind of MISTAKES are found in economics literatures ALL THE
TIME.

I cited a couple of examples from the Journal of political Economy
which unequivacally showed that those authors did not have the
slightest idea about how the multiple regression coefficients are
interpreted. They treat them as if they were simple correlations
rather than partial correlation -- and made the SAME mistaken as
those made by the majority of the social scientists, and other areas
that call themselves science.
................

So, a USER covers an entire spectrum of CONSUMERS of
statistics -- from the top notch researcher to the low level go-fers.
As for being a good RESEARCHER in statistics, I would go as
far as saying, as I believe Kevin Thorpe said in this thread, that
a Ph.D. in statistics is neither necessary NOR sufficient!

A user does not need to understand any statistical methods,
but has to understand what probability statements mean,
and what expectations, such as expected risks, mean.

That would be the low-level consumer, but NOT the researcher.
One course may suffice, if all they are required to do is to read
and interpret some very specialized tasks, like a chicken learn to
peck at the correct symbol to get its reward.

No, even the high-level consumer. The researcher in medicine
is an excellent example; they rely on methods which make
very strong assumptions of which they are unaware, and often
they have difficulty understanding why they are not true.

We have gone a full circle back to the Harvard joint M.D. Phd
Program for MEDICAL researchers who use statistics. They
are REQUIRED to have a Ph.D degree in Statistics from Harvard
-- hardly a low level consumer of statistics.

Look at the statin studies; none of them uses the quantitative
information which is available.

Now you're pulling out more examples supporting MY point
that most medical studies are made by VERY incompetent
people in the medical profession, not even M.D.s. They are
so incompetent that some were caught CHEATING with
manufactured data because they didn't even know what REAL
data looks like and made them so regular and systematic that
anyone can tell it's a FAKE!

That's how low some medical researchers can get!

As for being a good researcher in ANYTHING, a Ph.D. is
neither necessary nor sufficient.

We ALL agreed on this one. You, Brett, myself, and Kevin,
and probably most of the others who did not explicitly say so
in this thread.

I never had a single
course in statistics, nor did I find one to audit.

And it showed. :-)

That's why you sank in the pits of measure theory and doing
mathematics as if you are doing APPLIED statistics.

Are you familiar with my earliest research?

You and I can both agree that I am NOT familiar with it because in
my entire life as a professional statistician in APPLIED statistics
and Data Analysis, having served for years as an Associate Editor
of JASA and refereed articles for dozens of journals with submitted
statistical papers -- that I have NEVER seen one single reference
of ANY paper by Herman Rubin -- and I can assure you that I've read
PLENTY of papers by mathematicians and statisticians alike who
are competent in APPLIED statistics.


Or the one
about getting estimates of a concentration with given
probability of getting a preassigned error? I have no
problem applying my mathematical knowledge, and this
includes methods of numerical analysis, where appropriate.
And it usually is.

Those mathematical statisticians who publish papers irrelevant to
application or to applied statisticis USUALLY think their papers
are relevant, because they know more mathematics and measure
theory and integration theory than anyone NEEDS to be a superb
applied statistician.

Likewise for me in matrix algebra (a few weeks in an
abstract algebra course, definitely inadequate), or in
numerical analysis, or in set theory, in all of which
I have contributed.

Its much easier to contribute in THOSE mathematics areas
as in the FIELD of Data Analysis and applied statistics which
require much TRAINING (which algebra, analysis, topology,
etc do NOT require) in dealing with DATA, and learn how to
analyze it properly, in the same manner a doctor needs to
be TRAINED on how to diagnose and treat various illnesses.

No, educated. I have had too much contact with those who
have been trained. They can no longer think.

In YOUR limited experience -- not having EVER been trained in
an applied statistics environment nor ever taken a statistics
course, your myopic view about the fine line between EDUCATION
and TRAINING is understanable.

You are like a physicist who is well familiar with Boyle's law,
Charle's Law, and other gas laws (that are taught to advanced
level SCUBA divers) but has never been near any body of
water, thinks that he can SWIM because of his superior
knowledge about physics.

As soon as one such mathemtician or physicist is put to a test,
he drowns.


If Karl Pearson or Neyman or Wald or Karlin had any
courses in statistics worth a darn, I would be highly
surprised.

But they were MUCH smarter than most MATHEMATICAL
statistics. Fisher, Savage, Mosteller, Tukey and others
were trained as MATHEMATICIANS, but they outgrew their
mathematics to become competent statisticians.

No, too many reject their mathematical knowledge, and
try to use what should not be used. A student of mine
is now doing a dissertation in which he gets a fair
approximation under fairly general assumptions on
estimating a spectral density, and mathematics was
used.

Why should anyone be surprise? ANY student of yours is
SEVERELY handicapped with any education and training
about data analysis and applied statistics.

Box-Jenkins methodology in many types of types of time
series do very well IN PRACTICE without ever estimating
any spectral density, or other mathematical constructs that
are no more realistic or deeper than the mere ARIMA(p,d,q)
models.


One thing you might not be aware of; Neyman insisted
that his students take the abstract mathematics he did
not know. One needs it to have insight.

I knew about THAT and other old tales. The ASA Annual
Meetings (of which you are unfamiliar because you are too
busy with you mathematics in the IMS) have films and interviews
about these OLD statisticians -- thereby exposing both their
strengths and weaknesses from a statistical-historical point
of view.

I would say what you cited about Neyman was one of his
WEAKNESSES. Of course, he had not really distinguished
himeslf in the Neyman-Pearson theory which is often labeled
as "classical statisticians" with all its THEORETICAL and
EMPIRICAL flaws that even non-Bayesians know -- such
as Hogg and Criag's example that one can construct a
96% confidence interval (according the the NP theory of
CI from random intervals) and then find the observed
interval to be an interval which contains the unknown
parameter 100% of the time. We are of course well familiar
with all the absurdities of the implications of Hypothesis
Testing in the N-P framework.


You are still arguing with the mentality of 40 or 50 years ago
when there were NOT departments of statistics, to teach
the essential contents of the subject as well as TRAINING
the data analysts and APPLIED researchers to do their
statistical work properly.

No, there were methods departments. It took decades to
get rid of them, and they are now coming back.

Please CITE what you mean by those "methods department"
that had been rid of, and WHAT are now coming back.


Measure theory still has to be used in applications.

Only by Mathematical statisticians, or statistical mathematicians
who are completely clueless about Data Analysis and Applied
Statistics.

-- Bob.

.

Relevant Pages

  • ebooks share lits 207
    ... Computational Statistics Statistics, Textbooks and Monographs; ... Simplified Monographs and Textbooks in Pure and Applied Mathematics; ... Nonlinear Hyperbolic Waves in Multi-dimensions Chapman & Hall/CRC ...
    (sci.med.nutrition)
  • Classical Text in Mathematics and Statistics
    ... Hardy G.H. A Course of Pure Mathematics ... Shilov G.E. Linear Algebra ... Feller W. An Introduction to Probability Theory and Its Applications ... Statistics, Volume 1: Distribution Theory ...
    (sci.math)
  • Re: Scientific books: cheap sell-out of the library
    ... >The books available from the following areas: ... >Here are sample book list in Physics: ... >Spectral Functions in Mathematics and Physics ... >Here are sample book list in Statistics: ...
    (sci.astro)
  • Scientific books: cheap sell-out of the library
    ... Scientific books: ... by Douglas N. Clark (Editor) ... Dictionary of Classical and Theoretical Mathematics ... Applied Statistics and Probability for Engineers ...
    (sci.math.symbolic)
  • Scientific books: cheap sell-out of the library
    ... Scientific books: ... by Douglas N. Clark (Editor) ... Dictionary of Classical and Theoretical Mathematics ... Applied Statistics and Probability for Engineers ...
    (sci.math.num-analysis)