[R] Question about GAMs
David Winsemius
dwinsemius at comcast.net
Thu Oct 13 18:37:40 CEST 2011
On Oct 13, 2011, at 11:20 AM, pigpigmeow wrote:
> I'm confused...
> I type ..
> predict.gam(ozonea,type=s(ratio,bs="cr"))
>
> pressure maxtemp s(avetemp) s(ratio)
> 1 -0.0459102290 -0.185178463 0.263358446 -0.164558673
> 2 -0.0286464652 -0.194731320 0.199315027 0.727823293
> 3 0.0478073459 -0.013227033 0.002228896 0.342373202
> 4 -0.0089164494 0.082301539 -0.037331159 -0.067260889
> 5 0.0675373617 0.024984396 -0.047067786 -0.357569389
> 6 0.0823348735 0.101407254 -0.075884852 -1.485036738
> 7 -0.0977015204 0.177830112 -0.094755158 0.236575309
> 8 -0.0903027645 0.225594398 -0.113346667 0.435141242
> 9 0.0206785742 0.187382969 -0.066346157 -0.256133513
> 10 -0.1371615520 0.101407254 -0.131656887 0.145057584
> 11 -0.0477674066 -0.181001505 0.260279546 0.180513043
> 12 -0.0921599421 -0.009050075 -0.020511366 0.281470433
> 13 0.0681464361 -0.219212934 0.335348247 0.270813178
> ......
>
> I want to show s(ratio,bs="cr") term, and show the warning message
> Warning message:
> In predict.gam(ozonea, type = s(ratio, bs = "cr")) :
> Unknown type, reset to terms.
>
> I don't understand what does it mean.
>
> By the way, i use log-link function,
> should I convert log-link function of newozone to fitted value of
> newozone?
> 1. log(newozone) - s(ratio,bs="cr") = x
> then exp(x)
Perhaps. I do not see 'newozone' defined anywhere above and can only
speculateo speculate. I'm pretty sure Wood's uncopied comment was
intended to remind you that differences in the log (regression) scale
are equivalent to ratios on the measured scales.
So either :
exp( diff(coef, log(measured_or_external_baseline) ) )
# OR
exp(coef)/measured_or_external_baseline
How that is properly calculated in pkg:mgcv is a separate issue and
Simpson has suggested you read the help pages more carefully.
>
> 2. exp(newozone) - s(ratio,bs="cr") =X
> then x
Probably incorrect.
>
> which one is correct?
> so confused
>
David Winsemius, MD
West Hartford, CT
More information about the R-help
mailing list