OFF: Re: HW: Tour Shirts

Paul Mather paul at GROMIT.DLIB.VT.EDU
Wed Apr 28 12:23:34 EDT 2004


On Tue, Apr 27, 2004 at 07:18:32PM +0100, M Holmes wrote:

[[Observations about the UK political parties and UK election.]]

I guess I wasn't thinking about UK politics in specific, which is why
I used "Party A" and "Party B."  (I realise I should have generalised
it to "Party 1 ... Party N.")  But, I agree with you that if you can
be guaranteed that the parties do not differ appreciably from each
other in their policies, then it doesn't make sense to spend time
becoming informed about them, and a random choice is about as useful a
strategy as any.  (That raises a chicken-and-egg problem: how do you
arrive at that state of knowledge that there are no essential policy
differences.:)

But, I would caution against using appeals to the "average" in such
logic, especially when directed towards natural/sociological
phenomena.  Such distributions are often not Normal (i.e., Gaussian),
which aligns naturally with our popular notion of "fairness" and
"balance."  With a statement like "assume the average wage is 20K," I
would be much happier if you'd at the very least used the median.
Incomes and wealth are notoriously Pareto in their distribution, and
the mean is often well above the median because of the high skew in
the distribution.  The same goes for natural phenomena such as
popularity, etc.  Things are more naturally highly-skewed than people
appreciate, and most of the variance in the distribution is actually
concentrated right at (and contributed by) the extreme upper tail.
(One effect of this is that a relatively tiny number of such outliers
have a tremendous bias effect upon the whole.)

Those of us needing to do file size and access modelling know it's a
real pain to generate accurately those upper tails for the purposes of
creating synthetic large-scale collections and workloads. :-(

Cheers,

Paul.

e-mail: paul at gromit.dlib.vt.edu

"Without music to decorate it, time is just a bunch of boring production
 deadlines or dates by which bills must be paid."
        --- Frank Vincent Zappa



More information about the boc-l mailing list