Re: Degree of Freedom
- From: sagar <arijit14@xxxxxxxxx>
- Date: Wed, 16 Jul 2008 21:57:48 -0700 (PDT)
On Jul 16, 11:02 pm, RichUlrich <rich.ulr...@xxxxxxxxxxx> wrote:
On Wed, 16 Jul 2008 11:34:08 EDT, Jack Tomsky <jtom...@xxxxxxxxxxxxx>
wrote:
Hi,
I've two questions. First, what is degree of freedom
(DoF)? Second,
why it is important in the regression analysis.
Suppose in two similar
type of regression models one regression model I've
10 DoF & another
I've 15 DoF. How more DoF affect the results.
Thank you.
sagar.
If you have a (multiple) regression with k factors (possibly including
an intercept) and N observations, the degrees of freedom for
estimating the variance is N-k.
To perform a statistical test that an individual regression
coefficient beta_i is zero, say, you divide the estimate by its
corresponding standard error and you end up with a t test with N-k
degrees of freedom.
If you're testing several regression coefficients simultaneously, you
end up with an F test where the degrees of freedom in the denominator
is N-k and the degrees of freedom in the numerator is the number of
coefficients tested. All things being equal, the higher the degrees
of freedom, the more accurate will be the estimates.
Jack
Also -- More d.f. in the numerator says that you are using
more predictors, and that is why you expect more accuracy -
for the fit, if not for the effective prediction in the future.
R^2 never decreases. Also, more d.f. in the denominator says
that your sample size is larger, providing more accuracy (all things
being equal).
When the d.f. in the denominator is not large, then the
increase in the number of predictors can come at the cost
of "overfitting". The "Adjusted R^2" is an estimate of what
your equation is worth, accounting for the fact that the
expected R^2 for random predictors is equal to the number
of predictors divided by the total d.f.
--
Rich Ulrich
Thanks a Lot to both Jack & Rich Ulrich for clarifying my doubts.
.
- References:
- Degree of Freedom
- From: sagar
- Re: Degree of Freedom
- From: Jack Tomsky
- Re: Degree of Freedom
- From: RichUlrich
- Degree of Freedom
- Prev by Date: Re: paypal wholesale Diesel shirt Bape jeans(paypal accept)(www super-
- Next by Date: Re: Estimate if two bivariate sets are statistically different
- Previous by thread: Re: Degree of Freedom
- Next by thread: Displaying multivariate data
- Index(es):
Relevant Pages
|
