Variance component estimation for demographic model

[HenSH]

Hi,
So I am currently working on a stochastic demographic model. I want to do some variance component estimation for my preferred model in Program Mark so I can input it into the Leslie Matrix Model and can't seem to figure it out. So for calf survival in I retrieve the preferred model (phi(period)p(.), which showed a decline in survival occurring in 2002. I go to output/specific model output/variance components/beta parameter estimates. Then with the phi(period)p(.) model there are 5 parameter indices. 1:phi,2:phi,3:phi,4:p,5:p. Parameter 1=1994-2001, 2=2002-2011, 3=fixed, 4=constant recruitment, 5=fixed. If I am interested in the variance component estimation for 1994-2001 I assumed I would have selected 1 and 4. I then click ok, and it gives a warning saying I only selected 2 parameters. I click ok and it crashed Mark.
I have tried this with the Adult survival dataset which has a phi(.)p(.) preferred model and the same thing happens. Can anyone tell me what I am doing wrong.
Thanks.

[cooch]

Hi,
So I am currently working on a stochastic demographic model. I want to do some variance component estimation for my preferred model in Program Mark so I can input it into the Leslie Matrix Model and can't seem to figure it out. So for calf survival in I retrieve the preferred model (phi(period)p(.), which showed a decline in survival occurring in 2002. I go to output/specific model output/variance components/beta parameter estimates. Then with the phi(period)p(.) model there are 5 parameter indices. 1:phi,2:phi,3:phi,4:p,5:p. Parameter 1=1994-2001, 2=2002-2011, 3=fixed, 4=constant recruitment, 5=fixed. If I am interested in the variance component estimation for 1994-2001 I assumed I would have selected 1 and 4. I then click ok, and it gives a warning saying I only selected 2 parameters. I click ok and it crashed Mark.
I have tried this with the Adult survival dataset which has a phi(.)p(.) preferred model and the same thing happens. Can anyone tell me what I am doing wrong.
Thanks.

You need to have a thorough read of Appendix D. While reading, work through the examples (especially example D.4.3). Several issues will emerge if you do -- not the least of which is you don't seem to have nearly enough parameter space (number of periods) to estimate much of anything.

[DTI]

Hi there - I'm working through Appendix D right now, but am unable to find the example files used (E.g. binomial-example.dbf / binomial-example.fpt). I've installed the most recent version of MARK and have downloaded the supplemental files from the MARK book website, but still can't find them.

Is there somewhere else that they need to be downloaded from, or am I just completely incapable of properly searching my computer?

[HenSH]

Hi,

Thank you for pointing me towards Appendix D. I too was unable to find the examples for Appendix D. However I worked though it anyway and have sorted out my issues. Basically I needed to be working with my phi(t)p(.) model and selecting the appropriate years.

Thanks so much.

[cooch]

Hi there - I'm working through Appendix D right now, but am unable to find the example files used (E.g. binomial-example.dbf / binomial-example.fpt). I've installed the most recent version of MARK and have downloaded the supplemental files from the MARK book website, but still can't find them.

Is there somewhere else that they need to be downloaded from, or am I just completely incapable of properly searching my computer?

No, they're there. Look again (re-download first).

[cooch]

Hi,

Thank you for pointing me towards Appendix D. I too was unable to find the examples for Appendix D. However I worked though it anyway and have sorted out my issues. Basically I needed to be working with my phi(t)p(.) model and selecting the appropriate years.

Thanks so much.

I'm not sure you understand. First, in the example I mentioned in the Appendix, the years are categorized by (water) level -- however, the moments RE approach does not give you an estimate of for each level. Regardless (and this is important), the example was constructed to have a minimum of 10 years per level. If you don't have a lot of years (say, at least 7, but preferably 10 or more), then constructing the estimate is iffy at best. See point (4) on p. D-44. In fact, make sure you thoroughly read all the points in that list (D43 to D45).

Also, for a valid estimate of the process variance, you should not be generating it from a reduced model -- phi(t)p(t) would be better. This is also discussed in various places in the appendix.

[DTI]

No, they're there. Look again (re-download first).

Ah, excellent. They were there once I re-downloaded. Thanks.

[cooch]

Also, for a valid estimate of the process variance, you should not be generating it from a reduced model -- phi(t)p(t) would be better. This is also discussed in various places in the appendix.

Seems I lie (or in the current political vernacular, I 'mis-spoke'). The general point that you should not constrain/reduce parameter structure for (say) encounter probability (i.e., you should not use phi(t)p(.), but rather phi(t)p(t), even if the former is more parsimonious), was in an earlier version of the chapter, but seems to have been edited away at some point. I'll re-introduce the basic text in the next revision.

Basically, the point is this -- if you constrain p, you force additional variation onto phi, which affects your estimate of true process variance sigma^2 for phi.