[R] ref.df in mgcv gam

[Simon Wood]

These are a bit of a throwback really, and are not very useful - they 
are reference degrees of freedom used in computing test statistic and 
the p-values, but since the null distributions are non-standard the 
reference DoF is not very interpretable.

best,

Simon

On 27/03/2019 13:58, Susan Elias wrote:
> Hello Group.
>
> Regarding degrees of freedom in GAM models fit by mgcv, I do understand the
> estimated df, but what does "ref.df" in the summary output mean?
>
> I didn't see an explanation in Wood (2017) or find it in the function
> code.  The answer may be on page 222 of
>
> Wood SN. On p-values for smooth components of an extended generalized
> additive model. Biometrika 2013;100(1):221-228
>
> or thereafter, but I am not a statistician.  I could use help understanding
> what the reference degrees of freedom are.
>
> Thank you,
>
> Susan
>
>
>
>
-- 
Simon Wood, School of Mathematics, University of Bristol, BS8 1TW UK
https://people.maths.bris.ac.uk/~sw15190/