'Log' design matrix function

[geoffwah]

Hi All

Just a quick query. What does MARK do when one is using the 'Log' design matrix function to transform an individual covariate, and the covariate has zero values?

I've just run and occupancy analysis with the above scenario, and didn't have any problems....

Much appreciate any advice.

Regards

Geoff

[gwhite]

Geoff:
The exact code is:

LOG(MAX(MINLOG,X(1)))

where MINLOG was equal to 1.E-200, but I just changed it to 1.E-300 in the latest update put on the web. The change was made because for REALLY large analyses, you can get probabilities from certain encounter histories that were smaller than 1.E-200. Changing the value of MINLOG helps this problem, although there can still be problems.

A value of MINLOG smaller than 1.E-300 causes numerical errors.

Gary

[geoffwah]

Hi Gary

Much thanks for this information. I'm afraid however, that I don't quite follow! Would you mind expanding on the code?

Apologies

Geoff

[cooch]

Hi Gary

Much thanks for this information. I'm afraid however, that I don't quite follow! Would you mind expanding on the code?

Apologies

Geoff

Look at the code (even if you don't 'do FORTRAN'), it relatively straightforward:

LOG(MAX(MINLOG,X(1)))

It looks at the value (X), compares it with MINLOG (which is set internally as a constant), and then takes the log of whichever is the larger of the two. If the value of X is 0, then MINLOG will be larger. MINLOG is set to a very small number (1.E-300; about as close to zero as you might want, without being 0), but is not zero, so that LOG(MINLOG) is defined.

In other words, if X=0, then you 'trick' things by substituting a value that is really close to 0.

[geoffwah]

Ahh I see! Much thanks for the explanation Evan.

Cheers

Geoff

[cooch]

Ahh I see! Much thanks for the explanation Evan.

Cheers

Geoff

You're welcome.