The Math Forum

Search All of the Math Forum:

Views expressed in these public forums are not endorsed by NCTM or The Math Forum.

Math Forum » Discussions » sci.math.* » sci.stat.math

Notice: We are no longer accepting new posts, but the forums will continue to be readable.

Topic: Goodness of fitt : the Population´s "Fingerprints"
Replies: 2   Last Post: Oct 13, 2012 10:19 PM

Advanced Search

Back to Topic List Back to Topic List Jump to Tree View Jump to Tree View   Messages: [ Previous | Next ]
Luis A. Afonso

Posts: 4,758
From: LIsbon (Portugal)
Registered: 2/16/05
Goodness of fitt : the Population´s "Fingerprints"
Posted: Oct 10, 2012 12:22 PM
  Click to see the message monospaced in plain text Plain Text   Click to reply to this topic Reply

The Population?s ?Fingerprints?

Here the ?drama? is played by two characters: The team of interval?s tests, with their probabilities to ?capture? the test, no significance, noted 0, or outside, noted 1, by the other hand the Population i.i.d. random samples under test. Therefore, for example, the output/result symbol [011] denotes that the first test is inside and the second and third is no significant. The interval?s scorers, here 3, could be as large as we want, the outputs have an unlimited number of symbols as [0100?10]. This shows that the second score is significant (outside the respective interval), the first, third and fourth no significant.
Given an observed output we can, by Monte Carlo simulating, to evaluate how likely it is given a proposed Population.
We only intend to give an example based on the JB test and the Skewness and Excess Kurtosis parameters for the Populations Normal, Uniform, Gambel (0,1) and (1,2), family CDF = exp(-exp ((A-x)/B)), which inverted give directly x = A - B*log (-log (CDF)), used to simulate i.i.d. samples.

Analysis the Table below we can, for example, to state that a Uniform 90-sized sample ?cannot? show [000] or [110], probability 0.011 for this issue. Therefore if we have the chance to observe [000] or [110] we immediately exclude Uniformity, of course. On contrary the Normal Distribution is quite likely 46.5 + 20.3= 66.8%.


Table: Critical Values: Skewness, S, and Excess Kurtosis, k, 2.5% significance level, JB test 5%, for sample sizes 60 (10) 100, normal data, and U(-3, 3) obtained from 1 million samples each, JB, 4 million.

_Normal Data

_______ JB(.975)_____S(2.5%)________k(2.5%)_____

__60_____7.75____[-0.696, 0.696]___[-0.949, 1.622]__
__70_____7.78____[-0.645, 0.645]___[-0.902, 1.521]__
__80_____7.86____[-0.604, 0.604]___[-0.860, 1.419]__
__90_____7.95____[-0.572, 0.572]___[-0.823, 1.362]__
_100_____7.95____[-0.543, 0.543]___[-0.792, 1.291]__

When we test normality , using the intervals above, and data are Normal/Uniforme/Gumbel(0,1)/Gumbel(1,2) we obtain the following results (0 denoting inside, 1 outside):


_size=_____60____ 70_____80____90____100__
__[000]__ 0.657__0.591__0.524__0.465__0.405__N
___________0_____0_____ 0______0_____0____G(1,2)
__[001]__ 0.029__0.028__0.029__0.029__0.028__
___________0_____0_____ 0______0_____0____
__[011]____ 0_____0_____0______0______0____
__[100]__0.001__0.001___ 0______0______0____
________ 0.001__0.001__0.001___ 0______0____
__[101]____ 0_____0_____0______0______0____
__[110]_ 0.132__0.167__0.201__0.237__0.282__
________0.265__0.304__0.332_ 0.355__0.368___
__[111]_ 0.054__0.059__0.063__0.065__0.070__
________0.303__0.342__0.382_ 0.415__0.450___
________0.694__0.771__0.830_ 0.877__0.911_


It?s irrelevant the parameters exact values for these tables construction. In fact, the ways S and k are defined are automatically ?standardized? in what concerns the normal Population mean value and standard deviation. Otherwise one was compelled to adjust the estimations obtained from the real sample to the estimated ones by Monte Carlo, and the Table would loss generality.

Luis A. Afonso

Point your RSS reader here for a feed of the latest messages in this topic.

[Privacy Policy] [Terms of Use]

© The Math Forum at NCTM 1994-2018. All Rights Reserved.