Dear Jade
It seems to me that there are several issues here.
1 - if you fit a separate parameter for each heaping day you efectively
remove them from the model altogether. If they do carry meaningful
information then that is undesirable.
2 - if the reason that there ae so many 15s is because
Hi Michael,
I don't think my reply to your email came through to the list, so am
resending (see below). Problems with subscription have now hopefully
been resolved. Apologies if this is a double posting!
On Thu, 9 Jun 2022 at 15:27, jade.sho...@googlemail.com
wrote:
>
> Hi Michael,
>
> Thanks fo
Dear Jade
Do you really need to fit a separate parameter for each heaping day? Can
you not just make it a binary predictor or a categorical one with fewer
levels, perhaps 14 (for heaping in each year) or 12 (for each calendar
month). I have no idea whether that would help but it seems worth a
I would not worry too much about high concurvity between variables like
temperature and time. This just reflects the fact that temperature has a
strong temporal pattern.
I would also not be too worried about the low p-values on the k check.
The check only looks for pattern in the residuals whe
;
> -Original Message-
> From: R-help On Behalf Of jade.shodan--- via
> R-help
> Sent: Sunday, June 5, 2022 3:02 PM
> To: r-help@r-project.org
> Subject: [R] High concurvity/ collinearity between time and temperature in
> GAM predicting deaths but low ACF. Does this matter?
>
Subject: [R] High concurvity/ collinearity between time and temperature in GAM
predicting deaths but low ACF. Does this matter?
[External Email]
Hello everyone,
A few days ago I asked a question about concurvity in a GAM (the anologue of
collinearity in a GLM) implemented in mgcv. I think my
6 matches
Mail list logo
