Subject: Re: Corrected Sample Size
From: wpilib+@pitt.edu (Richard F Ulrich)
Date: 21 Jan 1997 21:28:55 GMT
Mark Gershman (gershmanM@ci.boulder.co.us) wrote:
: In the American Statistician (43):101-5, Kupper and Hafner presented a
: table with corrections to sample size forumulae for alpha values of 0.1
: 0.05, 0.01 (with confidence intervals (1 minus gamma) of .70, .80, .90.
: .95, .99. I am working on a forestry project where I need the corrections
: for an alpha value of 0.32 (foresters use a single standard deviate as
: their confidence level). We are using a confidence interval of .90
: Does anyone have any leads for me?
-- Compute your own, or get a stat-pack like GPOWER which will make
a personalized table of power; if your statistical test is one that
they consider.
If you are not doing a standard test, what test are you trying? -
and maybe someone can tell you more precisely where to look.
Rich Ulrich, biostatistician wpilib+@pitt.edu
http://www.pitt.edu/~wpilib/index.html Univ. of Pittsburgh
Subject: Re: Probability and Wheels: Connections and Closing the Gap
From: rhoads@sceloporus.rutgers.edu (Glenn Rhoads)
Date: 21 Jan 1997 23:07:07 -0500
bm373592@muenchen.org (Uenal Mutlu) writes:
>>Is this in C?
>It's C++
>>[bad C generator stuff deleted]
>Thanks for the info, but I doubt this being still true with the
>newer C++ compilers. Let me know if someone knows if it is also the
>case with the Borland C++ compiler v4.52 in 32-bit target mode.
>Nevertheless I'll check the workaround you gave, though I don't know
>why even 8 (!) bits should be affected by this, whereas only 1 bit is
>required for the sign...
If your random number generator is returning an integer in a range from
0 to say 2^31 - 2 (This is the most common range.), then your generator
is most likely flawed. The problem of non-random rightmost bits is
intrinsic to all linear congruential generators which is far and away
the most common type.
The problem can be removed by instead of having the generator return an
integer n from 0 to m-1, return the ratio n/m as a floating point number
x in the range 0 <= x < 1. Now the non-random part is confined to the
last few digits (for m = 2^31 - 1, you can count on about the first 15
decimal digits as being random.) You can convert x into an random integer
in the range from LOW through HIGH via the formula
p = (int)(x * (HIGH-LOW+1)) + LOW (e.g. to get an integer from 1 to 49,
use the formula p = (int)(x * 49) + 1) The last few nonrandom bits will
not effect which integer your floating point number converts to -- they
are essentially irrelevant. Many languages do this but C does not (the
C standard says that the random number generator is to return an integer!)
Used as described above, a good linear congruential generator is sufficient
for most applications. If you are doing simulations that use a LOT of
random numbers, then I would strongly recommend that you repeat the
simulation using an entirely different type of generator. Otherwise the
results really can't be trusted. The problem is that all generators are
completely deterministic and consequently, EVERY generator has some type of
problem. The application could be related in some way to the large scale
structure of the random numbers you are using and it is generally next
to impossible to determine whether this is the case ahead of time. If you
would like, I could send two different types of generators which are
both simple and fast. I could post them if there is enough interest.
-- Glenn Rhoads
Subject: Re: Is it possible???
From: ynecgan@cmc.doe.ca (Greg Neill)
Date: 22 Jan 1997 02:31:06 GMT
val (valnet2000@loop.com) wrote:
: Is it possible???
: Base - theory of Teilhard de Shardin.
:
: Creation of Hyper brain:
: Increasing speed of net ~ 10000 times.
: Direct contact from net to brain , virtual reality(?).
: Self-organization of initial seed - new evolution structure.
: Finally, new structure restricted only geometrically, by surface
: of planet and by number of involved humans (sells of Hyper brain).
: Power of Hyper Brain will excel power of human brain,
: as human brain excel power of brain cell.
: Power of Hyper Brain is INFINITE.
: Evolution approaches to omega point - crown of evolution on
: earth and , finally, in universe.
:
:
: Earth is unique place in universe - if this
: process was going somewhere else ,Omega would involve us
: already.
: ================================================
: WE ALONE IN UNIVERSE, BECAUSE WE ARE FIRST !??
: ================================================
: It puzzle me long time, please, any comment.
: Thank you.
: Val.
Val,
I think you should get together with Nancy on sci.astro and hash this
out.
Regards,
--
----------------------------------------------------------------------------
Greg Neill, | "Mystics are those who are baffled by the obvious
HNSX Supercomputers Inc. | yet possess a complete understanding of the
gneill@sx.nec.com | nonexistent." - Unknown
----------------------------------------------------------------------------
Subject: Re: Testing for normality of regression disturbance
From: barnett@agsm.unsw.edu.au (Glen Barnett)
Date: 22 Jan 1997 04:18:46 GMT
In article <32E32259.6555@ifim.sintef.no>,
=?iso-8859-1?Q?H=E5kon?= Finne wrote:
>Kmenta, in his standard econometrics text, mentions a test statistic for
>testing the normality of regression disturbances
> n*(b1/6 + ((b2 - 3)**2)/24)
>where b1 is the square of the estimate of symmetry and b2 the estimate
>of kurtosis, both based on sample values of respective moments around
>the mean. This statistic has a chi-square distribution with two degrees
>of freedom. Some call it a Lagrange multiplier test.
>
>Kmenta says that this test works well when the distribution of the
>disturbance belongs to a large class of distributions known as the
>"Pearson family", but he doesn't say more about that side of the matter.
>
>When is it unsafe to use this simple test?
>
>H=E5kon Finne
>SINTEF IFIM (Institute of social research in industry)
>N-7034 Trondheim
It depends on what you mean by unsafe.
I assume the Kmenta comment is that the test based on the asymptotic
approximation has good power against a pearson alternative.
[Note: the statistic is _asymptotically_ chi-square with 2 d.f.
In this case the distinction matters!]
Even when the data are normal, the contours of the distribution
of sqrt(b1) and b2 don't match even vaguely well with an independent
bivariate normal (i.e. they are highly non-elliptical, looking more like
a fat boomerang), even with a sample size of several hundred.
It is pointless to use an asymptotic result without having made some
investigation of when the asymptotic results really start to kick in.
The effect occurs with other distributions as well.
Consequently, the power of this test is fairly poor compared with what it
could be, even at large sample sizes; however it is quite possible that the
size of the test is still close to the nominal level for typical levels,
because the centre is in the right place; it rejects cases you'd want to
accept, and accepts cases you'd want to reject, but it may well reject
the right proportion of cases.
When a statistic is chosen that matches the contours better, power
properties are quite good against a whole range of non-normal
alternatives, being competitive with good choices like the
Anderson-Darling or Shapiro-Francia test. Of course, against distributions
with the same 3rd and 4th standardised moments as a normal, it
must always have poor power, and in fact may get worse with increasing
sample size against some of them.
So if by safe you mean "has about the right type I error rate", you may
be ok, even at moderate sample sizes. If you mean "has a low type II
error rate", you mayhave some cause for concern.
Glen
Subject: Re: Is it possible???
From: Pharaoh Chromium 93
Date: Tue, 21 Jan 1997 23:06:24 -0500
val wrote:
>
> Is it possible???
>
> Base - theory of Teilhard de Shardin.
>
>
> Creation of Hyper brain:
> Increasing speed of net ~ 10000 times.
> Direct contact from net to brain , virtual reality(?).
> Self-organization of initial seed - new evolution structure.
> Finally, new structure restricted only geometrically, by surface
> of planet and by number of involved humans (sells of Hyper brain).
>
> Power of Hyper Brain will excel power of human brain,
> as human brain excel power of brain cell.
> Power of Hyper Brain is INFINITE.
>
> Evolution approaches to omega point - crown of evolution on
> earth and , finally, in universe.
>
>
> Earth is unique place in universe - if this
> process was going somewhere else ,Omega would involve us
> already.
> ================================================
> WE ALONE IN UNIVERSE, BECAUSE WE ARE FIRST !??
> ================================================
>
> It puzzle me long time, please, any comment.
> Thank you.
> Val.
I see. Very good. Carry on.
http://alamut.alamut.org/c73/sri.htm
Subject: Re: Distribution of product of Gaussians?
From: "Robert E Sawyer"
Date: 22 Jan 1997 03:35:06 GMT
Bill Parr wrote in article ...
| In article <5brgi7$1bge@mercury.cc.uottawa.ca>, s1097716@aix2.uottawa.ca
| (Tim Creasy) wrote:
| > If X and Y are independent and have identical Gaussian distributions
| N(0,sigma)
| > what is the distribution of Z = X*Y? Or at least its variance?
| > Solutions or references on this would be much appreciated.
| >
| > Tim
|
| Here's an easy to come by solution:
| Var(Z) = E[Z*Z] - E[Z]*E[Z] (pardon the clumsy typography)
| so
| Var(XY) = E[XXYY] - E[XY]E[XY] = E[XXYY] (since X has mean 0 and X and Y
| are independent)
| = E[X*X]*E[Y*Y] = Var(X)*Var(Y)
| (independence of X & Y invoked again)
| Hope this helps.
|... ...
This could be taken even further, of course, to get all the moments of Z:
We can write X=s*N1 and Y=s*N2 where s=sigma and N1, N2 are iid N(0,1).
Then Z=(s^2)*N1*N2, and, for k=1,2,... :
E(Z^k) = (s^(2k))*E(N1^k * N2^k)
= (s^(2k))*E(N1^k)*E(N2^k)
= (s^(2k))*[E(N1^k)]^2
= (s^(2k))*[k!/{2^(k/2)[(k/2)!]}]^2 if k=even, =0 if k=odd
E.g., mean = 0
variance = s^4
skewness = 0
kurtosis = [E(Z^4)]/[E(Z^2)]^2 - 3 = 6
Subject: Re: This is impossible
From: drake.79@osu.edu (Macarthur Drake)
Date: 22 Jan 1997 06:32:30 GMT
>He was wrong,
>and they were right, but he was lucky that there was a previously
>unknown (except to the Vikings) continent between Europe and Asia to
>the east.
>Evanston Illinois
I am sorry, but you are wrong. The continent of America was
enhabited by a variety of people, so of which produced very advanced
civilization (Mayans for example) and presumably those people knew about the
continent they live on. Sorry to nit pick you, but you did just discount a
few million people who lived here BEFORE Columbus.
Subject: Re: This is impossible
From: rdadams@access4.digex.net (Dick Adams)
Date: 22 Jan 1997 02:24:20 -0500
Macarthur Drake wrote:
>> He was wrong,
>> and they were right, but he was lucky that there was a previously
>> unknown (except to the Vikings) continent between Europe and Asia to
>> the east.
>> Evanston Illinois
> I am sorry, but you are wrong. The continent of America was
>enhabited by a variety of people, so of which produced very advanced
>civilization (Mayans for example) and presumably those people knew about the
>continent they live on. Sorry to nit pick you, but you did just discount a
>few million people who lived here BEFORE Columbus.
Alas we have a whole new definition of discovery.
Subject: Re: This is impossible
From: R Mentock
Date: Wed, 22 Jan 1997 03:06:59 -0500
Macarthur Drake wrote:
>
> >He was wrong,
> >and they were right, but he was lucky that there was a previously
> >unknown (except to the Vikings) continent between Europe and Asia to
> >the east.
> >Evanston Illinois
>
> I am sorry, but you are wrong. The continent of America was
> enhabited by a variety of people, so of which produced very advanced
> civilization (Mayans for example) and presumably those people knew about the
> continent they live on. Sorry to nit pick you, but you did just discount a
> few million people who lived here BEFORE Columbus.
He mentioned this in his original post.
--
D.
mentock@mindSpring.com
http://www.mindspring.com/~mentock/index.htm
Subject: Re: Question, PDF of A + B
From: John Hudson
Date: Wed, 22 Jan 1997 08:27:32 +0000
amcorso@aol.com wrote:
>
> any closed form, or semi closed form fudge, of the distribution of the SUM
> of two jointly lognorma variables with known variance
the classical reference is S.S.Schwartz and Y.S.Yeh: Bell system tech
J., 61, No. 7, Sept. 1982, pp.1441-1462
also Beaulieu N.C. et.al.: IEEE Trans. Comms, COM-43, Dec. 1995,
pp.2869-2873 for a good reference list.
JEH
Subject: ALL-RUSSIAN CONFERENCE , TAMBOV, MAY 12-15, 1997
From: "Anrey V. Bogoslovsky"
Date: 22 Jan 97 17:11:07 GMT
International Informatization Academy
Russian Assotiation for Engineering Education
Tambov Military Air Engineering College
CALL FOR PAPERS
The 5th all - Russian scientific and technical conference
at Tambov Military Air Engineering College
May 12 - 15, 1997
Tambov, Russia
This conference is devoted to various aspects of increasing of information
processing means and methods efficiency and will cover the following
technical areas at the committees:
1. Methods and devices of digital information processing.
2. Discrete-analogue methods and devices of multidimentional information
processing in real time.
3. Optimal methods and divices for receiving, processing and forming of
signals and fields.
4. Matematical support of computers in evalution, identification and
control problems.
5. Artificial intelligence systems.
6. Safety of fligts: technical diagnosis, forecasts, durability,
nondestructive check of quality and condition.
Working languages of the conference are Russian and English.
SUBMISSION
One copy of paper should be received by March 1, 1997 at the address:
Russia, 392006 Tambov-6, TVVAIU, Research dept., Goncharuk N.M., e-mail:
CONF97@tmaec.ru
The form of the manuscript should be as follows:
1. Cover page includes title, author(s) , scientific degree(s), positions,
postal address, e-mail address, phone number, number of committee chosen.
2. The paper should be no more than 2 pages (12-point fonts, 2 intervals)
including text, figures and references.
3. Materials are published before the beginning of the conference.
LOCATION
Tambov is located 500km to the South-East of Moscow and can be reached from
Moscow by car, train (Paveletcky railwaystation), air (Bikovo airport).
REGISTRATION FEE
1. For citizens of Russia: fees are to be transferred to the account
000141867, GRKTc TcB, MFO 046850001, INN 6833008326, Tambov, receiver
Voinsky (for juridicial persons) and to 392006, Tambov-6, TVVAIU, account
000141867, GRKTc TcB, MFO 046850001, INN
6833008326 (for physical persons). State the name of the participant in
the postal order.
2. For foreign participants: fees are taken upon arrival.
3.The registration fee is 120 000 rubles or 20 US$
ADDRESS
Russia, 392006 Tambov-6, TVVAIU
Goncharuk N.M.
E-mail: CONF97@tmaec.ru
tel/FAX + 0752 226530
tel + 0752 226505
tel + 0752 226505
Subject: Re: Is it possible???
From: stefanj@io.com (Stefan E. Jones)
Date: 22 Jan 1997 10:25:25 -0600
In article <32E53E36.730@loop.com>, val wrote:
>Is it possible???
Yes. But the more important question is "is it likely?" Better yet, is
it a falsifiable hypothesis that can be tested?
> Base - theory of Teilhard de Shardin.
That's de Chardin.
> Creation of Hyper brain:
>Increasing speed of net ~ 10000 times.
>Direct contact from net to brain , virtual reality(?).
>Self-organization of initial seed - new evolution structure.
>Finally, new structure restricted only geometrically, by surface
>of planet and by number of involved humans (sells of Hyper brain).
Simply increasing the speed and easy of communication will not spawn
a super entity. The human brain (and other organs) are not great masses
of undifferentiated cells. They are highly organized. They are the result
of millions of years of evolution, of responding to outside pressures and
adaption through trial and error.
Self-organization is not magic. There must be some sort of stimulus or
reason to make this thing sentient.
>Power of Hyper Brain will excel power of human brain,
>as human brain excel power of brain cell.
>Power of Hyper Brain is INFINITE.
Wow! Have you been reading Dr. Bronner's soap labels? (100% PURE! ALL
ONE! ESSENE SCROLLS PREDICT SUPERBRAIN! PURE SOAP CURES PILES, PREVENTS
UNNECESSARY CAPITALIZATION!)
>Evolution approaches to omega point - crown of evolution on
>earth and , finally, in universe.
Evolution has no direction, meaning, or purpose. de Chardin was attempting
to meld christian theology and evolutionary thought. Very interesting stuff,
but not necessarily true.
>Earth is unique place in universe - if this
>process was going somewhere else ,Omega would involve us
>already.
There's no guarantee that "this process" would happen elsewhere. Other
Sapient species would almost certainly have different ways of looking
at things.
> ================================================
> WE ALONE IN UNIVERSE, BECAUSE WE ARE FIRST !??
> ================================================
>
>It puzzle me long time, please, any comment.
It sounds like you've stumbled upon a variant of Fermi's paradox.
I'll let someone else tackle that.
--Stefan
--
+-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-+
SeJ@aol.com ~ sjones@andrew.cmu.edu ~ stefanj@io.com
http://www.ini.cmu.edu/~sjones/
CHARGES APPLIED FOR UNSOLICITED COMMERCIAL EMAIL!
Subject: Re: This is impossible
From: nyikos@math.scarolina.edu (Peter Nyikos)
Date: 22 Jan 1997 17:56:18 GMT
CC: Drake, because I've joined this thread quite late.
rdadams@access1.digex.net (Dick Adams) writes:
>Macarthur Drake wrote:
>> This messege is to provoke a serious scientific debate.
>> I am an engineer, no biologist, astronomer or statictician or anything,
>> but something puzzles me. I am sure you are aware of the Late Dr.
>> Sagan's quote " extraordinary claims require extraordinary proof "
>> with regards to extraterrestrial life, UFOs etc. I have also heard
>> people say that the discovery of life on another world would be the
>> greatest discovery in human history.
>> I beg to differ with both of these ridiculus statments.
>> [snip]
>> I would appriciate any math or stats expert to comment on the
>> chances that we are alone in the entire universe. I bet that s/he'd
>> say that it is statistically impossible for us to be alone, so what's
>> the big deal we know that life is there, just a matter of time 'til
>> we find it....or them us!
"Statistically impossible" is either a redundancy or an oxymoron
and needs to be defined carefully before it can be commented on.
As a mathematician with a wife who worked eight years in biochemistry,
I would add that we just know too little about the probability of
life having arisen spontaneously on earth to be able to
estimate the chances that we are alone in the universe.
your question. Nobel Laureate Francis Crick has written at length
of the dearth of hard data along these lines in _Life Itself_.
Check it out--you will be fascinated and intrigued by his theory
of directed panspermy.
Richard Dawkins has written at length about the odds and how little
we know about them in _The Blind Watchmaker_, but he is a far
less logical person than Crick and suffers a disastrous lapse
in logic when he claims that if life arose only once in the
universe, it had to be on earth. Had he said "intelligent life"
instead of just "life" he would have been on firmer ground,
but as it is he seems blissfully unaware of Crick's book or
theory.
>> [snp - again]
>Extraordinary claims DO REQUIRE extraordinary proof.
What makes the two claims above extraordinary is their specificity.
Most scientists would not call the following claim extraordinary:
>Life elsewhere in the universe is probable;
Only a claim as unequivocal as the ones listed by Sagan
is extraordinary. Drake's claim above falls into that category.
> finding it may well
>be the greatest discovery ever made up to that time.
Now THAT's a non-extraordinary claim! ;-)
Peter Nyikos -- standard disclaimer --
Professor, Dept. of Mathematics
University of South Carolina
Columbia, SC 29208
Subject: Re: ANN: some new newsgroups created: alt.sci.math.combinatorics and more...
From: wpilib+@pitt.edu (Richard F Ulrich)
Date: 22 Jan 1997 16:00:27 GMT
Stany De Smedt (sdesmedt@is1.vub.ac.be) wrote:
: In article <32e58113.45072288@news.muenchen.org>, bm373592@muenchen.org
: (Uenal Mutlu) wrote:
: > Some new, specialized, math newsgroups were created:
: >
: > alt.sci.math.design_theory
: > alt.sci.math.probability
: > alt.sci.math.statistics.prediction
: > alt.sci.math.combinatorics
: > alt.sci.math.galois_fields
:
: This seems a great idea to me. Is it not possible to create more of them ?
Anybody can create groups, but the trickier part may be to get
distributors to distribute them. There is a process that involves
announcements of a proposed group, discussion, and collection of votes
- at least for the "sci." hierarchy.
My distributor (University of Pittsburgh) is somewhat selective about
what it distributes in various hierarchies. We do get a large number
of "alt." groups, but certainly not all of them - I have read that some
distributors delete that whole hierarchy, so you might think twice
about replacing a general "sci." group with a particular "alt." group.
About circulation: my software will only post to what it recognizes
as valid groups, and I had to delete half the Newsgroups listed for
"Distribution: " by the original announcement, just to make this post.
I never read the *proposal* to create any of these groups named above,
so they were not discussed or voted on in the usual .stat newsgroups.
Maybe that is a distinction between "alt." groups, and some others? -
I don't know. I do suspect that I might never receive any of the
above unless I put in a special request to have them available.
By the way, those sound to me like rather narrow topics, and I probably
would not read any of them.
Rich Ulrich
Subject: Re: ANN: some new newsgroups created: alt.sci.math.combinatorics and more...
From: Clay Helberg
Date: Wed, 22 Jan 1997 12:38:06 -0600
Richard F Ulrich wrote:
>
> Stany De Smedt (sdesmedt@is1.vub.ac.be) wrote:
> : In article <32e58113.45072288@news.muenchen.org>, bm373592@muenchen.org
> : (Uenal Mutlu) wrote:
>
> : > Some new, specialized, math newsgroups were created:
> : >
> : > alt.sci.math.design_theory
> : > alt.sci.math.probability
> : > alt.sci.math.statistics.prediction
> : > alt.sci.math.combinatorics
> : > alt.sci.math.galois_fields
> :
> : This seems a great idea to me. Is it not possible to create more of them ?
>
> Anybody can create groups, but the trickier part may be to get
> distributors to distribute them. There is a process that involves
> announcements of a proposed group, discussion, and collection of votes
> - at least for the "sci." hierarchy.
>
> I never read the *proposal* to create any of these groups named above,
> so they were not discussed or voted on in the usual .stat newsgroups.
> Maybe that is a distinction between "alt." groups, and some others? -
Yes, that is indeed the distinguishing feature of the alt.*
hierarchy--no Call for Discussion (CFD) or Call for Votes (CFV) are
required. Just define the group, and off you go. That is why there are
so many bizarre alt.* groups (perhaps one of the bizarrest is
alt.sexy.bald.captains, apparently some sort of Patrick Stewart fan
club).
You also make a good point about those groups not being available at
some sites. That is definitely a potential drawback, especially since
this may have the unfortunate side effect of filtering out some of the
brightest potential participants.
> By the way, those sound to me like rather narrow topics, and I probably
> would not read any of them.
>
> Rich Ulrich
Agreed--however, I suspect that those groups will have a smallish core
of participants, more analagous to a roundtable discussion than the more
mainstream groups (which might be analagous to a conference).
--Clay
--
Clay Helberg | Internet: helberg@execpc.com
Publications Dept. | WWW: http://www.execpc.com/~helberg/
SPSS, Inc. | Speaking only for myself....
Subject: Re: Probability and Wheels: Connections and Closing the Gap
From: bm373592@muenchen.org (Uenal Mutlu)
Date: Wed, 22 Jan 1997 17:31:16 GMT
On 14 Jan 1997 23:24:01 -0500, rhoads@sceloporus.rutgers.edu (Glenn Rhoads) wrote:
>Typically, programmers use the formula x = (n mod 49) + 1 to convert
>n, the number returned by pseudo-random number generator, to the
>number x of the desired range. This is a bad thing to do in C.
and, On 21 Jan 1997 23:07:07 -0500, rhoads@sceloporus.rutgers.edu (Glenn Rhoads) wrote:
>If your random number generator is returning an integer in a range from
>0 to say 2^31 - 2 (This is the most common range.), then your generator
hmmm. mine returns in the range from 0 to RAND_MAX, which is defined
as being 2^15 - 1 = 32767; cf. below.
>is most likely flawed. The problem of non-random rightmost bits is
>intrinsic to all linear congruential generators which is far and away
>the most common type.
>
>The problem can be removed by instead of having the generator return an
>integer n from 0 to m-1, return the ratio n/m as a floating point number
>x in the range 0 <= x < 1. Now the non-random part is confined to the
>last few digits (for m = 2^31 - 1, you can count on about the first 15
>decimal digits as being random.) You can convert x into an random integer
>in the range from LOW through HIGH via the formula
>p = (int)(x * (HIGH-LOW+1)) + LOW (e.g. to get an integer from 1 to 49,
>use the formula p = (int)(x * 49) + 1) The last few nonrandom bits will
>not effect which integer your floating point number converts to -- they
>are essentially irrelevant. Many languages do this but C does not (the
>C standard says that the random number generator is to return an integer!)
>
>Used as described above, a good linear congruential generator is sufficient
>for most applications. If you are doing simulations that use a LOT of
>random numbers, then I would strongly recommend that you repeat the
>simulation using an entirely different type of generator. Otherwise the
>results really can't be trusted. The problem is that all generators are
>completely deterministic and consequently, EVERY generator has some type of
>problem. The application could be related in some way to the large scale
>structure of the random numbers you are using and it is generally next
>to impossible to determine whether this is the case ahead of time. If you
>would like, I could send two different types of generators which are
>both simple and fast. I could post them if there is enough interest.
Ok, I see the important point you're saying. I use the following macro
or the corrosponding C++ inline function from the stdlib.h of the the
compiler vendor. It's the "random(num)" which I use. There, it should
have been taken into account what you're saying. Please let me know
if you think the below is ok or not ok. I like standards, and won't use
any other method (since then it IMHO would be much more doubtful), and I
also would avoid the use of floating point for this.
(Ok, there is one other thing I sometimes use: the "linear congruential
random number generator" from Knuth, which in each series produces
each number exactly once (seed'ing as usual produces different 'series')
I would say, that these two types of RNGs must be seperated from each other;
ie. the usual rand() is a true pseudo-RNG, but not a true
linear_congruential-RNG (since rand() usually produces dupes during the run
of a series, but the other not), and for such a simulation the second
should IMO not be applicable).
Below is the "random(num)" I use in the simulation program, which in turn
uses the standard rand():
/* rand() implementation in the standard C/C++ library of the compiler I use.
The "random(num)" macro/func below is from the stdlib.h:
'int' can be 16 or 32 bits in size
'long' is at least 4 bytes long
rand() will give a number from 0 to 32767
random(num) will give a number from 0 to num-1; num can be > 32767
*/
/* here, rand() in the stdlib always returns a number from 0 to 32767 (= 0x7FFFU)
#define RAND_MAX 0x7FFFU
#ifdef __cplusplus /* C++ inline func */
inline int random(int num)
{ return (int) (((long) rand() * num ) / (RAND_MAX + 1)); }
#else /* C macro */
#define random(num) \
(int) (((long) rand() * (num)) / (RAND_MAX + 1))
#endif
But, if you're saying the rand() itself is buggy, then I'm indeed lost,
but I think you mean (cf. above) the _usage_ of the rand(), don't you?
But it's easy to test the randomness of all RNG's: ie. the
"Chi-Square Goodness Of Fit"-Test will do this; it's easy and
simple to do; if someone wants C/C++ code of this let me know.
-- Uenal Mutlu (bm373592@muenchen.org)
* Math Research * Designs/Codes * SW-Development C/C++ * Consulting *
Loc: Istanbul/Turkey + Munich/Germany
Subject: Re: Optimization and Combinations
From: wpilib+@pitt.edu (Richard F Ulrich)
Date: 22 Jan 1997 20:14:53 GMT
THORK (ChristopherJ.Siegle) wrote:
: Hello Everyone.
:
: Does anyone know of a procedure that can calculate the optimal
: combination of variables based on the value of an independent
: variable? I have a matrix of 13 BY 13 variables. From that
: matrix, I would like to know what combination produces the
: highest average from an independent variable.
:
: I guess this would be a probability matrix of 13! [factorial].
:
- I find this request to be pretty close to unintelligible.
But if someone is looking for 13-factorial combinations of things,
then they might be looking for CHAID (automatic interaction
detection). If you have 10,000 cases or so in your sample, then you
might get some believable, useful results; or so I remember someone
recommending.
Hope this helps. Else, you might indicate what you mean by "optimum"
or what you mean by "matrix", i.e., what is the content of one cell?
Rich Ulrich, biostatistician wpilib+@pitt.edu
http://www.pitt.edu/~wpilib/index.html Univ. of Pittsburgh
Subject: Re: Probability and Wheels: Connections and Closing the Gap
From: barnett@agsm.unsw.edu.au (Glen Barnett)
Date: 22 Jan 1997 21:42:44 GMT
In article <32e64303.94727220@news.muenchen.org>,
Uenal Mutlu wrote:
>But it's easy to test the randomness of all RNG's: ie. the
>"Chi-Square Goodness Of Fit"-Test will do this; it's easy and
>simple to do; if someone wants C/C++ code of this let me know.
>
This is not so. The chi-square test allows you to check that the
shape of the distribution is right, and it is not a particularly
good test against the more interesting alternatives. What it
doesn _not_ check for is randomness. The series can be decidedly
nonrandom, such as:
.00 .01 .02 .03 ..... .99 .00 .01 ...
but a chi-square test with, say 10 or 20 or even 100 cells,
cannot pick this up, since it doesn't look at the order of the
values, only the shape of the distribution.
Glen
Subject: Re: This is impossible
From: casanova@crosslink.net (Bob Casanova)
Date: Wed, 22 Jan 1997 22:50:58 GMT
On 22 Jan 1997 17:56:18 GMT, in sci.skeptic, nyikos@math.scarolina.edu
(Peter Nyikos) wrote:
>
>Richard Dawkins has written at length about the odds and how little
>we know about them in _The Blind Watchmaker_, but he is a far
>less logical person than Crick and suffers a disastrous lapse
>in logic when he claims that if life arose only once in the
>universe, it had to be on earth.
I think you missed the obvious point to this quote, since it's
essentially self-fulfilling. Since we *know* that life exists here,
*if* it only arose once it was (and therefore "had to be") here. Or do
you think I've missed a deeper implied meaning?
>Peter Nyikos -- standard disclaimer --
>Professor, Dept. of Mathematics
>University of South Carolina
>Columbia, SC 29208
>
>
>
>
>
(Note followups, if any)
Bob C.
"No one's life, liberty or property is safe while
the legislature is in session." - Mark Twain
