Re: Adumbostistics: a case study

On 25 Maio, 22:03, John Smith <jsmith_...@xxxxxxxxxxx> wrote:
Perry,

In Adumbostistics, it is not necessary to specify null and alternative hypotheses.  A case in point is given by Luis Amaral Afonso in THREAD: An RNG test by the DKW formula  THREAD STARTED:  Jul 14 2007, 5:56 pm

He makes a complicated simulation whose goal is to determine if a sequence of random numbers is random.  He arbitrarily chooses a test statistic (without reference to a null hypothesis) and arbitrarily (without reference to a tail area) chooses a "critical value" of 1E6.  

He says right there in the first post (where "LAST VALUE" is his "test statistic"): "SINCE THIS LAST VALUE IS LESS THAN 1E6 we conclude that there is no EVIDENCE that the *numbers FACTORY* is working erratically."

When I asked him whether 1E6 corresponds to a 1% critical value, he could not respond.  Eventually he admitted that the choice of 1E6 is something he made up.  But he still insisted that what he was doing constituted a statistical test.

So, Perry, this is an example of Adumbostistics.  You don't need null and alternative hypotheses, nor do you need true critical values.  Isn't it amazing that all this has been invented by Luis Amaral Afonso???  

John ` Smith

I´m Luis Afonso
This guy is lying . My checking procedure is based on
the DKW inequality , a result that is clled the Fundamental Theorem of
Statistics and that he didn´t met before tell him. last year. He don´t
know, he refuses to learn a very simple matter: the distribution
function of whatever statistical test is liable to obtain
by simulation. The DKW (please, check) uniquality say
us how close are this simulated Distribution from the
rigorous one (if this exist). Function of the number of thse
simulations (NS) and the maximum (absolute) difference , d, one can
know the maximum probability p
the difference occurs. My *procedure test* is based on this 1960´s DKW
theorem. In consequence I evaluate the minimum N that a given
difference occurs,
fixing d and p, N is evaluted, As I know U(0,1) fractiles I´m able to
compare N with ND (which´s arbitrary, and that the two IMBECILES make
confusion with a sample size!!!) , As long as I attain a d when N<NS
the machine is behaving properly
Luis
.

Relevant Pages

  • Testing for non-correlated binormality of data
    ... I'd like to know an appropriate statistical test for finding out ... I have a value created by a computer simulation. ... counts the number of particles in each square on the plane for squares ... I know the answer to the diffusion equation, ...
    (sci.stat.math)
  • Testing for product of Gaussians
    ... I'd like to know an appropriate statistical test for finding out ... I have a value created by a computer simulation. ... counts the number of particles in each square on the plane for squares ... I know the answer to the diffusion equation, ...
    (sci.math)