Re: Statistical Ranking for Non-Normal Populations

From: Peter Hach (Willy.Yoda_at_gmx.net)
Date: 10/15/04 

Date: Fri, 15 Oct 2004 21:07:30 +0000 (UTC)

>
>Again, thanks for the problem! Now I am also interested in the
>practical aspects of this data, like what is it about, what
>is the use, why sampling is expensive... or is it only a scenario?
>

Very interesting - many thanks for the long answer!

This is not only a scenario; but a real problem related to statistical
ranking of configrations. I want to run a number of different tasks
t1...tn on a computer system for which I have a number of different
configurations C1,...,Cm; running each task to get the real time taken
is expensive, and they may vary in cost significantly.

The main reason why I was looking for different theory than what
people have used in statistical ranking is that here it seems that
generally most papers assume that one samples from a normal and while
I'm willing to sample sufficiently often for the CLT to hold for the
sample mean, sampling sufficiently often to get a number of means (for
batches of samples) that correspond to a sample from an (approximate)
normal distribution seems excessive (and since the distribution of the
X[i] is non-normal still vulnerable to outliers).

Since my background in statistics is limited, my intuition may be
wrong, or I may have simply not looked at the right sources, however.

Relevant Pages

  • Misuse of statistics
    ... The use of the normal distribution as a model for ... are those of the Statistics Department or of Purdue University. ... Herman Rubin, Department of Statistics, Purdue University ...
    (misc.health.diabetes)
  • Re: Intensity of selection and the Price equation - continued(1)
    ... do statistics, since there are no controlled experiments available to ... Even a malicious Nature, rolling ... that arises from probability theory (and ... A normal distribution is ...
    (sci.bio.evolution)
  • Reef Fish Statistics for Dummies: Applied Simple Regression
    ... Statistics to arrive at the "Simple Regression" topic, ... Yin the model comes from a Normal distribution with mean ... That was where "Lesson #1" was ...
    (sci.stat.math)
  • Re: most probable value of estimator of k-th central moment
    ... statistics. ... this problem even for a normal distribution has been of ... normalized third and fourth moments/ as skewness and kurtosis. ... There also has been some work on the higher momensts (see ...
    (sci.stat.math)
Loading