[R-sig-Geo] question about regression kriging

[Tomislav Hengl]

Sorry, I meant "UK variance of the interpolated logits".

I guess the solution is to either:

(A)
1. Run 100 SG simulations using logits;
2. Back-transform the values to 0-1 scale;
3. Determine the variance per grid node (now in 0-1 scale).

Running simulations could also be rather time-consuming (I often give up running SG simulations in
gstat with multiple raster maps as predictors -- memory limit problems, too time-consuming).

(B)
1. Determine the upper and lower 1-s confidence intervals -- 2 maps (in logit scale);
2. Back-transform the maps and derive the difference in 0-1 scale (estimation error);

But I guess that a method to directly back-transform the UK variance from logit scale to 0-1 scale
does not exist.

Tom Hengl

-----Original Message-----
From: Edzer Pebesma [mailto:edzer.pebesma at uni-muenster.de] 
Sent: woensdag 9 april 2008 14:48
To: Tomislav Hengl
Cc: r-sig-geo at stat.math.ethz.ch
Subject: Re: [R-sig-Geo] question about regression kriging

Tomislav Hengl wrote:
> PS: How do you back-transform the GLM prediction variance to original scale (I was not aware of
> this, apologies)? A reference would do.
>   
1. The white book, "Statistical Models in S"
2. The source of predict.glm (not for the faint of heart)
3. McCullogh & Nelder (I guess it's in there, but my copy is still in a 
box!)
--
Edzer