My GAMM model is to find drivers of species richness in forests is
gamm1<- gamm(Total Species Richness ~ fROT + s(PH) + s(LOI) + ASP +
s(SQRT_ELEV) + CANCOV + s(SQRT_TOTCWD) + s(WELLF) + s(WELLN) +
s(OLDWDLD) + s(DISTWOOD) + s(Annprec), random=list (fSITE =~1), family
= poisson, data = BIOFOR3)
My issues are that the validation graphs are using methods I'm
unfamiliar with e.g. square rooting values and Pearson's residuals.
I'm following as best I can recommendations in Wood 2006 and Zuur 2009
but have some (I think) basic questions on graphical validation:
1. plot sqrt transformed fitted values vs square root transformed
observed values - should be straight line.
I presume the observed values mentioned above refer only to observed
RESPONSE variable
2. Pearson residuals vs. sqrt fitted values (should form band with no
patterns)
I'm totally unfamilar with using Pearson residuals. Until I can see
example of a band with no patterns I can't really say if my output is
OK! Could this band also be described as a cloud of points with no
pattern?
I have 2 factors and 2 unsmoothed explanatory variables in my model ?
what residual graphs are suitable for these? Is it the "traditional"
ones of normalized residuals vs raw explanatory variables?
Also for a lot of other model checking residuals are plotted against
explanatory variables..what?s appropriate residuals to plot
vs.smoothed explanatory variables for a GAMM?
3. Raw residuals (observed vs fitted) vs. sqrt fitted values (should
show clear cone)
But my output doesn't show clear cone and I don't know what action to
take to rectify this issue
Infact graphs 2. and 3. are the exact same.
Some help with this would be much appreciated.
Thanks,
Karen
The only examples I have for graph output is Wood 2006 Fig 6.11
