h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

"There are three kinds of lies... lies, damn lies, and statistics."

And he was clever, not just a ranting idiot who didn't make honours because of statistics. (thats me by teh way)

I love stats - like how many / how often / where when and why.
Where in cyberspace are these things listed. As a dumb example :-
How many curry houses are there in the UK ? So you get the idea? Stats about any and everything - where are they???

I don't mind figures. You know, out of 100 people interviewed 30 of them though this. And even percentages are fine. Its when you start doing things like averaging or T tests with degrees of error. When I studied statistics I found that all tehy did was make clean figures look way more ambiguous and also found that statistics coudl be used to make your findings work out the way you wanted regardless of teh results (One of our books was actually called "How to lie with statistics")

Which is odd because it is impossible for a statistic to lie.

Its quite possible to lie about what one means though.

A statistic can't lie. But it can make stuff damned ambiguous

No a statistic is simply the result of applying a mathematical formula to a set of numbers. It is entierly true (in itself). It is only the attempt to explain the 'meaning' of that statistic that can lie.

Z

My point is neat figures rarely need explaining, ne is bigger thna teh otehr, one is smaller. You can draw your conclusions. Statistics are an easy tool to add ambiguity but performing unnececary mathematics on a perfictly readable figure.

Statistics is a flawed way to express reality,cos it transcends that reality. When we,say, average things that are of themselves the results of our observations(and making observations is already a leap from reality)we cause several 'mistakes' to crop up:
1)multiple observations are being degenerated into one simple number,which is nowhere to be seen.
2)This number doesn't correspond to reality,and therefore is not representative of it,because its deduction is not only phylosophically incorrect but is a matter of purely mathematical speculation.It is well known that there are several types of mean numbers: arithmetic mean, weighted mean,chronologic mean,and geometrical mean,as far as I remember the latter allows to get the smallest average number. Choosing the type is pupely a matter of taste,(or regulatory provisions ,that is ditto).
Which thing would you prefer- paraboloic regression curve or hyperboloic? It is a matter of taste too, since coefficents on which your judgement is based are of dubious value. Well, you minimize the margin of error, but the error is always there, whilst the reality has no room for error, we base our satistics on factual happenings, how can there be the room for error? why should we string our facts on that polished well-directed curve,while the only thing this curve does is to string us along. The error is our error in choosing the curve, or in our choice to use 'statistics', not the factual error.
We suppose our "statistics" to be impeccable,thinking that it is the inferior reality that dares to deviate itself from preordained tracks.Why such a word choice then?
What the statistics does is often to give us (often false) reassurance that this or that event is likely to happen with that or that probability, based on our past experience only subliminally allowing the changes taking place. We assume probability(as is the case with disribution curves ) where is nothing but pure facts.
This mean numbers are truly mean.
All things said,it is worthwhile to say that statistics is truly ingenious instrument to simplify the reality and as such really helps us to get to grips with the dificulties afforded by that.It is great to have scientifically approved immidiacy of perception.But all perceptions are somehow false.

I think I agree with you, or at least what I understood. Basically, raw data is clear, and a statistic generalises it, adding in a margin of error.