Radford Neal wrote:
>
> >Radford Neal wrote:
> >> I presume that the people making such models are interested in whether
> >> or not the poor or good performance of a country might be due to
> >> controllable factors such as organization, training facilities, etc.
> >> In other words, they want to know if they could be doing better, given
> >> the resources available. So it makes perfect sense to include
> >> population and GDP as explanatory variables, but NOT type of
> >> organization of the Olympic Committee, or type of training facility
> >> used. However, the climate should indeed be included as an
> >> explanatory variable, if it is thought that it might be important.
> >> There will of course be random noise, though I'd think that many
> >> injury problems might be attributable to the training regime used, or
> >> to sending athletes to the games who shouldn't have been selected to
> >> go because of their injuries.
>
> Paige Miller <[EMAIL PROTECTED]> wrote:
> >Hey Radford, why wouldn't you want to include "type of training
> >facility" in the model? If it is a useful predictor variable, then you
> >have a "controllable factor" -- in other words, it may tell you that
> >training facility type A results in more medals than type B, so your
> >country should start building facility type A and stop building
> >facility type B.
>
> Well, this depends partly on what you're going to look at after you
> fit your model. If - as in the original post - you're just going to
> look at the residuals and say "country A underperformed - they must be
> doing something wrong", then you don't want to include controllable
> variables in the regression. It would, for example, be silly to say
> "on the other hand, country B did better than expected, no need to
> change anything there", if what the positive residual for country B is
> actually saying is that country B did better than expected given that
> they are using a bad training regime, have a disorganized Olympic
> Committee, and conscript winning athletes into the army for ten years.
>
> On the other hand, if you are going to look at the regression
> coefficients too, then you might hope to conclude things about which
> training facilities are best, etc. However, you will have the usual
> problems in trying to make causal conclusions from observational data.
> You will also have the problem that some training facilities cost more
> than others. It's no use telling the Ugandans that they would have
> won more medals if only they'd had a sophisticated training facility
> that they can't afford.
Ah, good points. I agree that the terms you might want to put into the
model depend on what purpose you intend to use this analysis for. I do
think that the Ugandans need to spend more money on Winter Olympics
events, as they are clearly under-performing there ... no wait ...
wrong model ...
But the whole idea that you can predict the performance of individuals
(or teams) based upon national statistics such as GDP and population
even though there may be correlations. I doubt that there is a causal
connection.
There might be a more causal connection between type of training
facility and Olympic medals won and so I would personally want to look
at the regression coefficients, keeping in mind your advice that it
may be tough to draw conclusions about regression coefficient from
observational data (and of course there are many many other factors in
addition to type of training facilities).
--
Paige Miller
Eastman Kodak Company
[EMAIL PROTECTED]
"It's nothing until I call it!" -- Bill Klem, NL Umpire
"Those black-eyed peas tasted all right to me" -- Dixie Chicks
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
