Re: negative adjusted r square
From: G Robin Edwards (robin.edwards_at_argonet.co.uk)
Date: 11/17/04
- Next message: Phil Sherrod: "Re: Surprising Pattern of Florida's Election Results"
- Previous message: Zeno Franco: "Sorry, posted twice yesterday"
- In reply to: joseph: "negative adjusted r square"
- Messages sorted by: [ date ] [ thread ]
Date: Thu, 18 Nov 2004 00:47:10 +0100
In article <vz5lh555zn5n@legacy>,
joseph <josephFrank1969@hotmail.com> wrote:
> I am reciving a negative adjusted R square when running my regerssion.
> Is it possible or I am doing something wrong? If it is possible what
> is the interpretation of it
If you look at the t statistic(s) for your parameter estimate(s) you
will find value(s) less than 1. What this means is that the act of
suggesting the model uses more information than it reveals. You'll
also find that the residual standard deviation is greater than the original
standard deviation of the data.
If you look at the probability value for the t statistic you'll see that it is exactly that for the correlation coefficient between the dependent and independent variables (for simple regression only, of course).
Whatever statistics you care to look at you can be pretty confident
that your model does not adequately describe your data. There is a
choice of "solutions". More data, better data, a "better" model, or any
combination of these.
It happens to me very frequently, especially when I'm looking at data
that are alleged to relate to "global warming". In my case the data
are generally "givens", and there are usually no more to be had, so the
only solution is a better model, for which a sustainable rationale must
be available. And at this stage I worry about inferential matters.
It's a hard life.
- Next message: Phil Sherrod: "Re: Surprising Pattern of Florida's Election Results"
- Previous message: Zeno Franco: "Sorry, posted twice yesterday"
- In reply to: joseph: "negative adjusted r square"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
