| |
 |
|
|
Science Forum Index » Statistics - Math Forum » I WONDER...
Page 1 of 1
|
| Author |
Message |
| Luis A. Afonso... |
 Posted: Tue May 20, 2008 11:45 am |
|
|
|
Guest
|
I WONDER
___1___I wonder if Jack Tomsky / John Smith are working on to prove a coin’s fair trough flipping it
_________Test Hypotheses
_________H0 : p(heads on) = ½
_________H1 : p(heads on) =/ ½
___2___ They didn’t say if to Test Hypotheses by Monte Carlo simulations is (YES or NOT) a truthful method.
___3___ They are quiet about if one can prove, via test hypotheses, two parameters are equal or one only can assert that there’s not sufficient evidence to reject it (if so).
___4___They are mute about the reason they present the confidence interval for as___xs +/- W____given that W is the semi-amplitude of the interval containing 3´800´000 out of 4´000´000 (95%) simulated sample medians and xs the observed value of the single sample median.
___5___ They didn’t enlighten us what is the reason by which the Welch Test shouldn’t converge to the Student one in the case that the Normal Populations have equal variances.
___6___ We don’t know why the Jarque-Bera test _a) must have critical values monotonically increasing
_b) they mustn’t be greater than the Chi-square, 2 df, critical value taking in consideration the respective alpha.
__7___ Having stated that I’m a full of errors programmer, using an wrong language, I obtained exactly the same critical values published in Literature concerning the Lilliefors´s Test and Jarque-Bera too.
__8___ Taking into account the above, why they did not correct the critical values I presented here through a program that: a) is error – free, b) was obtained by the chosen the adequate Programming Language.
__9__ Why they provide any explanation one must prefer R (an object – oriented Language) than Basic to perform the calculations I’ve made.
__10__ I wonder why the DUO uses systematically rude language substituting logic argumentation: Are they really persuaded that the Readers will believe in preference on those they are able to make more noise, or by the rationally shown expertise?
Luis Amaral Afonso, the Questioner. |
|
|
| Back to top |
|
| John Smith... |
| Posted: Tue May 20, 2008 12:58 pm |
|
|
|
Guest
|
Quote: I WONDER
___1___I wonder if Jack Tomsky / John Smith are
working on to prove a coin’s fair trough flipping it
_________Test Hypotheses
_________H0 : p(heads on) = ½
_________H1 : p(heads on) =/ ½
"no one can tell if 8/13 is greater than 5/13" This is a statement of which Luis Amaral Afonso is extremely proud.
THREAD: Jarque-Bera test: confidence intervals for normal data THREAD STARTED: Mar 7 2007, 12:16 pm
Quote: ___2___ They didn’t say if to Test Hypotheses by
Monte Carlo simulations is (YES or NOT) a truthful
method.
In Adumbostistics code for Acceptance-Rejection calculations, Luis Amaral Afonso managed to
(1) confuse probabilities with quantiles,
(2) confused cdfs with inverse cdfs,
(3) set the scale parameter of an exponential equal to zero,
(4) couldn't calculate the log correctly,
(5) and came to the remarkable conclusion that the 0.765 quantile is not unique, but is satisfied by all x between zero and infinity.
As pointed out by Our Moderator, Luis Amaral Afonso managed to do all this in just one "program". While any of these
"errors" would invalidate a program in traditional statistics, in Adumbostistics they are permissible.
THREAD: The J-B test AGAIN THREAD STARTED: Jan 25, 4:32 pm
Quote: ___3___ They are quiet about if one can prove, via
test hypotheses, two parameters are equal or one only
can assert that there’s not sufficient evidence to
reject it (if so).
it is correct to reject the null hypothesis when p = 0.99868, especially when
t = 3.484 has a significance level of 0.13
THREAD: Difference on means THREAD STARTED: Aug 22 2007, 1:09 pm
Quote: ___4___They are mute about the reason they present
the confidence interval for as___xs +/- W____given
that W is the semi-amplitude of the interval
containing 3´800´000 out of 4´000´000 (95%) simulated
sample medians and xs the observed value of the
single sample median.
5.76E6 < 1E6
THREAD: An RNG test by the DKW formula THREAD STARTED: Jul 14 2007, 5:56 pm
***
Quote: ___5___ They didn’t enlighten us what is the reason
by which the Welch Test shouldn’t converge to the
Student one in the case that the Normal Populations
have equal variances.
The df for the Welch statistic converges to a constant and not a random variable
THREAD: Welch formula doesn´t converge THREAD STARTED: Feb 15, 4:09 pm
Quote: ___6___ We don’t know why the Jarque-Bera test _a)
must have critical values monotonically increasing
_b) they mustn’t be greater than the Chi-square, 2
df, critical value taking in consideration the
respective alpha.
p-values and significance levels are the same thing
THREAD: Difference on means THREAD STARTED: Aug 22 2007, 1:09 pm
Quote: __7___ Having stated that I’m a full of errors
programmer, using an wrong language, I obtained
exactly the same critical values published in
Literature concerning the Lilliefors´s Test and
Jarque-Bera too.
Stirling's Formula is 0 = 0, which follows after taking the limits in the Adumbostistics version of Stirling's Formula
THREAD: Two birthday probability problems THREAD STARTED: Feb 27, 5:40 pm
Quote: __8___ Taking into account the above, why they did
not correct the critical values I presented here
through a program that: a) is error – free, b) was
obtained by the chosen the adequate Programming
Language.
It is not necessary to use the correct formula in any program that you write.
Our Moderator showed that the reason Luis Amaral Afonso's answers come out wrong is that he used the wrong statistic. The correct statistic is
t = (Xbar-Ybar)/[s*sqrt((1/Nx)+(1/Ny))],
where
s = [Sum(Xi-Xbar)^2 + Sum(Yi-Ybar)^2]/(Nx+Ny-2).
Then under H, s^2 is proportional to a chi-square and t has Student's t distribution with Nx+Ny-2 degrees of freedom.
The square of your denominator is a linear combination of chi-squares, which is not a chi-square unless Nx=Ny.
That's why your statistic does not have Student's t distribution.
THREAD: ONLY STUPID PEOPLE DOESN´T CHEK ITS BELIEFS. THREAD STARTED: Dec 25 2007, 10:29 am
Quote: __9__ Why they provide any explanation one must
prefer R (an object – oriented Language) than Basic
to perform the calculations I’ve made.
__10__ I wonder why the DUO uses systematically rude
language substituting logic argumentation: Are they
really persuaded that the Readers will believe in
preference on those they are able to make more noise,
or by the rationally shown expertise?
To be a statistical programmer, it is not even necessary to be a competent programmer, and it is okay to be unable to learn any new languages. Luis Amaral Afonso admits: " I´m a pitiable programmer using an old, ineffective and unfashionable language (BASIC). Unfortunately it is the only I hadlearned."
THREAD: Kurtosis, skewness and bootstrapping THREAD STARTED: Feb 23 2005, 9:36 am
Quote:
Luis Amaral Afonso, the Questioner.
The answers are the principles of Adumbostistics! |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Tue May 20, 2008 3:30 pm |
|
|
|
Guest
|
CONFIDENCE INTERVALS
A common * cast * on finding Confidence Intervals of a parameter T is such that one starts with the observed value in the sample, To , the aim being to find an interval where E(T) lies with a chosen probability p.
Considering the Sample Statistics W, which Mathematical Formula is KNOWN
______W= (To – E(T)) / s _____________(0)
It´s supposed that the sample statistics W follows a known Distribution.
Let be u, v such that, with a previously select alpha value:
_____p(W<u) = F(u) = alpha/2______________(1)
_____p(W<v) = F(v) = 1-alpha/2____________(2)
Then the interval
I(alpha) = [To + s*F(u) , To + s*F(v)]_________(3)
contains exactly with the probability p = 1 - alpha the (unknown) parameter value E(T).
THAT IS: IF WE REPEAT A LARGE NUMBER OF TIMES THE RANDOM EXPERIMENT (TRIAL) we will find that asymptotically we shall find the Population Parameter Value with probability p. (This is the principle of Monte Carlo Method frequency <> robability).
Here, asymptotically, has a precise meaning: infinite number of trials.
If W is symmetrical about ZERO then, necessarily, the quantiles (1) and (2) are symmetrical and one have F(u) = - F(v) as it happens for the Normal Standard. The interval I(alpha) then is represented by To +/- s*F(u).
HYPOTHESES TESTS
In terms of Hypotheses test, the problem is very similar however has an * inverse* feature.
Supposing (at the start)
_______H0: T = c and H1: T =/ c
Our goal: to find if we should reject H0 or, alternatively, one have no sufficient evidence to do so.
Choosing alpha the statistics is compared with the u and v Critical Values (see above):
___________W0= (To – c) / s ____________
Therefore two cases 0) or 1) could happen
__A)__________u <= W0 <= v
__B)__________W0 < u , or else , W0 > v
The Decision is:
_____0) fail to reject H0, one can state that T has not the value c,
_____1) reject H0, the parameter T has not the value c.
This is the classic way.
*************************
How about Monte Carlo?
Firstly the W (T, data) could be unknown in most cases, or based on coarse approximations if sample sizes are not large, or even very troublesome to derive analytically. Therefore the analytical (classical) way shall not advise. For example
___The comparison between two normal Populations, variances not equal one another (several not satisfactory approximate tests).
___The median of Cauchy samples (troublesome)
___The Liliefors Test (unknown, I presume)
___The Jarque-Bera GOF test, even asymptotically a complete fiasco, it DOESN´T CONVERGE TO THE CHI-SQUARED (the IMBECILE TWO, Jack Tomsky - John Smith say it does but is unable to defend his point of view).
This last two was feasible even using a Personal Computer, Readers see my posts to be aware.
Monte Carlo procedure is a synthetic procedure: one simulate the entire Population of the Test Statistics following the model under study, supposing H0 true, and compare W0 with the critical values (Empirical Distribution Function quantiles, taking into account the alpha values chosen).
The procedure has several vantages.
___On condition that we are able to simulate the samples is exact (in terms of Model)
___ No need to know the expression of W(data).
Drawbacks
___ Is time-consuming
MORE VANTAGES:
___ Put the Classical Gang furious.
___And only say nonsense when try to criticise a method to which they must be aware (is a 40 years very used by the true Statisticians all World procedure) but in fact these Dinosaurs are complete ignorant about.
___ The Decision depends only from the quantiles, have not the I(alpha) ____(3)_____and so the IMBECILES STAY COMPLETELY CONFUSED : GIVEN THAT THE UNIVERSITY DID NOT TRAINED THEM TO **** THINK **** THE RESULT IS WHAT IS EXPECTED: THEY KICK AS DONKEY (THEY ARE) WHEN THEY MET A SERMON NOT THE SAME THEY HAD HEARED AS CHILDREN.
Nowadays even the ROBOTS have a little capacity to learn. Has not the IMBECILE DUO ENVIE?
*************************************
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Tue May 20, 2008 4:03 pm |
|
|
|
Guest
|
THE DUO IS A LIAR TEAM
Posted: Feb 13, 2008 7:20 PM
Dear Readers
The strangest episode had occurred here. The * late * (I hope so ?) Bob Ling had made the same BLUNDER Jack Tomsky do: eliminating the term 1/nX + 1/nY from the formula relative to testing the * equality * of two independent Population means. GIVEN THAT it´s unbelievable that such a monstrosity could even be printed impossibility hey own the same edition textbooks. The only possibility is that Tomsky had inherit from Bob such a * qualified * notes. The only possibility LEFT.
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| John Smith... |
| Posted: Tue May 20, 2008 4:11 pm |
|
|
|
Guest
|
Pretty funny coming from a man who says,
It is possible for the same quantity to be both a pair of critical values and a confidence interval.
THREAD: How to simulate Cauchy samples THREAD STARTED: May 8, 2:28 am |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Tue May 20, 2008 6:12 pm |
|
|
|
Guest
|
*** "no one can tell if 8/13 is greater than 5/13" This is a statement of which Luis Amaral Afonso is extremely proud.***
*****************
A heehaw more originated from the guest dunkey.
It is the way pseudo-statisticians (or Mathematicians) misunderstand RANDOMNESS: I repeated several times that
____The symbol 8/13 was shorthand of 8 successes in 13 trials,
____The expression 8/13 > 5/13 (0.05) indicates that the left Population has, at 5% significance level, a proportion greater than that the Right one.
_____There is not my fault that apes - humanoids do not get the difference between a codified representation of a Hypotheses Test RESULT and a Numerical Fraction. Mathematicians are the people less able to see the difference between Maths and Probability, or else, between Deterministic Phenomena and Random Events. (Ce n´est pas ma faute, pas du tout).
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| John Smith... |
| Posted: Tue May 20, 2008 6:30 pm |
|
|
|
Guest
|
That's pretty funny coming from a guy who thinks that
************************************************
5.76E6 < 1E6
THREAD: An RNG test by the DKW formula THREAD STARTED: Jul 14 2007, 5:56 pm |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Wed May 21, 2008 12:50 am |
|
|
|
Guest
|
Paranoia
In the *Alice in the Wonderland* world of US Academy seems perfectly admissible that a pedantic nobody in Statistics has the abominable (at least in Europe) gall to joint his declassified name to a serious Theorem, as was the Dvoretzky - Kiefer - Wolfowitz outrageously (for Science) re-baptized Tomsky - Dvoretzky - Kiefer - Wolfowitz (TDKW) inequality !!! That’s the limit! Besides (he admit) he had never before met such a result !!!.
I ask: is this the American way to respect intellectual property? Fair? In a word: decent? No surely, definitively is pure, irrefutable, PARANOIA.
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Wed May 21, 2008 3:08 am |
|
|
|
Guest
|
I said
*** ___4___They are mute about the reason they present the confidence interval as:
_______________xs +/- W____
given that W is the semi-amplitude of the interval containing 3´800´000 out of 4´000´000 (95%) simulated sample medians and xs the observed value of the single sample median.***
Comment
Because WORMTOMSKY adds the observed value to the semi-amplitude of the interval (of simulated data) containing (by construction) 95% of all values it follows that
____or WORMTOMSKY has the EXTRAORDINARY chance that xs fails exactly in its centre, OR
____95% Probability turns up other unknown prob. >0.95 !!!!!!!!!!!!!!!
__WORMTOMSKY EQUATION
________ 0.95 = 0.95 + X
WHERE X IS *** ANY VALUE we want *** IN THE INTERVAL [0, 0.05].
The Nobel Prize is truly a very weak compensation for such a FIND!
Haven´t you ENVY of the new generation ROBOT able to show a LIMITED CAPACITY TO LEARN?
Even limited was a blessing to you, sure.
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| Luis A. Afonso... |
| Posted: Wed May 21, 2008 3:50 am |
|
|
|
Guest
|
I said
*** ___4___They are mute about the reason they present the confidence interval as:
_______________xs +/- W____
given that W is the semi-amplitude of the interval containing 3´800´000 out of 4´000´000 (95%) simulated sample medians and xs the observed value of the single sample median.***
Corrected Comment
Because WORMTOMSKY adds the observed value to the semi-amplitude of the interval (of simulated data) containing (by construction) 95% of all values it follows that
____or WORMTOMSKY has the EXTRAORDINARY chance that xs fails exactly in its centre, OR
____95% Probability turns up other unknown prob. >0.95 or < 0.95 whichever !!!!!!!!!!!!!!!
Haven´t you ENVY of the new generation ROBOT able to show a LIMITED CAPACITY TO LEARN?
Even limited was a blessing to you, sure.
Luis Amaral Afonso (The Moderator Destroyer) |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Thu Jan 08, 2009 2:55 am
|
|