## AI-GEOSTATS: question concerning confidence intervals

Expand Messages
• Dear members of the mailing - list, I deal with the interpolation method kriging. My question is how to get reliable confidence intervals if the data doesn t
Message 1 of 2 , Feb 7, 2003
• 0 Attachment
Dear members of the mailing - list,

I deal with the interpolation method kriging. My question is how to get
reliable confidence intervals if the data doesn't have a gaussian
distribution.
The common way to interpolate non-gaussian data is to transform it and
to estimate the variogram of the transformed data.
But how can I receive confidence intervals for the original data then
since they are not transformable.

I would be very thankful for useful answers and hints on literatur.

sincerely yours Antje Müller

[Non-text portions of this message have been removed]
• HI Antje: I am working with OK trying to get confidence intervals, Deutsch and Journel (1997) discuss on page 125 the calculation of Error Simulation , as I
Message 2 of 2 , Feb 7, 2003
• 0 Attachment
HI Antje: I am working with OK trying to get confidence intervals, Deutsch and Journel (1997) discuss on page 125 the calculation of "Error Simulation", as I understand the confidence intervals are calculated using the model, not the data values. It seems the solution is : "... a nonconditional simulation to generate the simulated erors and two krigings to generate the actual estimated values and unique estimated values..." page 128..
It looks that another alternative is to generate an histogram of the model using the program trans.exe, this is for a gaussina distribution with unit mean and variance...
Also, Isaaks and Srivasta (1989) discuss confidence intervals for OK and BK on Chapter 20....It looks like is a large process without the adecuate algorithm...
Hope, this helps... I am still trying
Sandra

Antje M�ller <anmuller@...-muenster.de> wrote:Dear members of the mailing - list,
I deal with the interpolation method kriging. My question is how to get reliable confidence intervals if the data doesn't have a gaussian distribution.
The common way to interpolate non-gaussian data is to transform it and to estimate the variogram of the transformed data.
But how can I receive confidence intervals for the original data then since they are not transformable.
I would be very thankful for useful answers and hints on literatur.
sincerely yours Antje M�ller

---------------------------------
Do you Yahoo!?