SE of zero

[darryl]

David,
I agree with what Bret and Dave are saying in terms of the number of covariates you're trying to fit to the model, especially if you're trying to fit models with all of them at once. If you're only doing 1 or 2 at a time, perhaps it's not so bad. That said, I noted for a couple of your p(.) models you have an estimate of exactly 0.5 and a SE of 0. Have you double checked you're design matrix that you actually have a '1' somewhere along the row(s) for p. To me it looks like you may not have.

A SE of 0 with a parameter estimate of 1 is not surprising as 1 is on the boundary of allowable values for that parameter. I'm sure that's been covered here on the forum before and/or is in the MARK book (Evan will chip in any time telling giving us the exact location).

The other thing to remember about MARK is that when you have covariates in the model, the 'real estimates' only apply to 1 particular combination of covariates, not to the entire sample. From memory, I think the default uses the covariate values from the first site in the data. Not sure if you can get MARK to give you a value for every site, but if you wanted that you could; 1) do it by hand; 2)probably do it from within R / RMARK; or 3) use PRESENCE.

Cheers
Darryl

[David]

i agree. i should take some predictors out.
about standardizing the covariates as Bret mentioned.
according to chap. 2 (p. 44) of “Program MARK- A Gentle Introduction” Evan Cooch & Gary White (Eds):
"...In this case, you should scale the values of weight to be from 0.1 to 0.5 by multiplying each ‘weight’ value by 0.0001. In fact, MARK defaults to doing this sort of scaling for you automatically (without you even being aware of it). This ‘automatic scaling’ is done by determining the maximum absolute value of the covariates, and then dividing each covariate by this value."
so, i am a little bit confuse- do i need to standardize covariates or not?

[darryl]

MARK does it automatically, so you shouldn't have to, provided the automatic procedure is working ok. Before you run the model I believe there's an option that allows you to turn the rescaling off so you can see if it makes any difference to your results.

Have you checked the design matrices of the 'problem' models? What are the estimates of the beta parameters like?

Darryl

[cooch]

i agree. i should take some predictors out.
about standardizing the covariates as Bret mentioned.
according to chap. 2 (p. 44) of “Program MARK- A Gentle Introduction” Evan Cooch & Gary White (Eds):
"...In this case, you should scale the values of weight to be from 0.1 to 0.5 by multiplying each ‘weight’ value by 0.0001. In fact, MARK defaults to doing this sort of scaling for you automatically (without you even being aware of it). This ‘automatic scaling’ is done by determining the maximum absolute value of the covariates, and then dividing each covariate by this value."
so, i am a little bit confuse- do i need to standardize covariates or not?

See pp. 9-11 or so in Chapter 11, where standardization (of individual covariates) is discussed.

[David]

Darryl-
The estimates of the beta parameters in the 'problem' models-
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.5000000 0.0000000 0.5000000 0.5000000
2:Psi 0.5446533 0.1435069 0.2778706 0.7880531

LOGIT Link Function Parameters of {p(.)psi(var5,var6,var7,var8)}
95% Confidence Interval
Parameter Beta Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1: -2.3360286 2.2025599 -6.6530461 1.9809890
2: 0.0177176 0.0074961 0.0030252 0.0324100
3: -0.0744158 0.0922367 -0.2551997 0.1063681
4: -0.0019808 0.0117222 -0.0249563 0.0209946
5: -0.0023183 0.0013708 -0.0050051 0.3684832E-0

and
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:p 0.5000000 0.0000000 0.5000000 0.5000000
2:Psi 0.5314491 0.1309893 0.2880092 0.7607868

LOGIT Link Function Parameters of {p(.)psi(var5,var6,var7)}
95% Confidence Interval
Parameter Beta Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1: -2.4802548 1.8465242 -6.0994424 1.1389328
2: 0.0120601 0.0059287 0.4398550E-003 0.0236803
3: -0.0735878 0.0837842 -0.2378048 0.0906292
4: -0.0026495 0.0106447 -0.0235131 0.0182142