new MARK build (v. 4.3) | online now...

[egc]

Greetings -

Gary has just posted a 4.2 -> 4.3 upgrade. You can download the setup file from Gary's site, or here at

http://www.phidot.org/software/mark/download/


Notable changes include

a. A message asking the user if they really want to exit out of MARK has been added because some users are accidentally clicking the red exit box in the upper right corner of the MARK application. If you feel this message is not needed, you can turn it off in the File | Preferences menu. Personally, I recommend you look before you click!

b. The design matrix is now scaled internally so that you do not have to use the "Standardize Individual Covariates" to scale covariates to obtain numerical convergence. Hence, the only reason to use the standardize option is to have the mean of the covariates equal to zero.

c. averaging, i.e., this box is checked when you start the model averaging dialog. If you want the original equation, uncheck this box.

d. The absolute value link function has been added to the list of possible link functions. The absolute value link is handy for closed captures models when the estimated population size is close to M(t+1) because the parameter is counted as being estimated. In contrast, the default log link will not count such parameters as being estimated because the beta value is approaching negative infinity. The absolute value link function works particularly well with the Coulombe closed captures example distributed with MARK. To access the absolute value link function for population estimation, you must use the "Parm. Specific" link option.

[egc]

b. The design matrix is now scaled internally so that you do not have to use the "Standardize Individual Covariates" to scale covariates to obtain numerical convergence. Hence, the only reason to use the standardize option is to have the mean of the covariates equal to zero.

Gary sent me a note with some further details on how this is handled -

"The scaling is done by determining the maximum absolute value of
entries in each column of the design matrix, and then dividing each column by its max. This results in each column scaled to between -1 and 1. Then, the beta estimates have to be divided by this same max (which is counter-intuitive until you actually write out the formulas), and the VC matrix of the beta estimates divided appropriately by these maximum values. The beta values reported are like the original design matrix would suggest, but all the internal calculations are done with the transformed beta values."

So, all of the beta values, and the V-C matrices that MARK reports are the 'back-scaled' values - so, in effect - you don't need to do anything to be able to use these values directly.