Supply intial values orig parameterization

[Cris]

Dear forum members,

I’m running a two-species analysis and I have convergence problems. Thus, as I have learnt at the forum, I have tried alternate parameterization. However, I have to use the original parameterization because the parameters from the orig parameterization is a function of my covariates but I do not know how can I supply the initial values?
I have obtained these results from the alternate parameterization:
Untransformed Estimates of coefficients for covariates (Beta's)
estimate std.error
A1 ccupancy psiA 1.524753 (3.347971)
A2 ccupancy psiB 0.634738 (4.198415)
A3 ccupancy psiBa -21.725555 (0.000000)
A4 ccupancy psiADren 0.207483 (4.126084)
A5 ccupancy psiAPinf 1.258290 (1.773080)
B1 :detection pA[1] 1.361864 (7.330402)
B2 :detection pB[1] -0.212826 (0.000064)
B3 :detection rA[1] -0.777530 (16.922650)
B4 :detection rB[1] -2.528118 (7.065942)

Should I copy these Beta's values or I need to edit or transform them?

Thanks a lot!

[jhines]

Cris,

The PsiA parameter is the same for both parameterizations, so you could use the final estimates from one parameterization as starting values in the other.

For PsiB, I suggest using values which are between the values from psiB1, and psiB2 (as psiB=psiA*psiB1+(1-psiA)*psiB2).

Since PsiBa=(1-psiA)*psiB2, you could work out a value for psiBa using estimates from the alt-parameterization, then back-transform to a beta value (beta=log(x/(1-x))). If psiA or psiB2 are dependent on covariates, it's more complicated.

Cheers,

Jim

[Cris]

Cris,

The PsiA parameter is the same for both parameterizations, so you could use the final estimates from one parameterization as starting values in the other.

For PsiB, I suggest using values which are between the values from psiB1, and psiB2 (as psiB=psiA*psiB1+(1-psiA)*psiB2).

Since PsiBa=(1-psiA)*psiB2, you could work out a value for psiBa using estimates from the alt-parameterization, then back-transform to a beta value (beta=log(x/(1-x))). If psiA or psiB2 are dependent on covariates, it's more complicated.

Cheers,

Jim

Thanks Jim for your quick answer,
but the occupancy of the model is dependent on a categorical covariate (3 types of sites where I could find both species), then I have 3 different values for psiA. I feel very lost...I have no idea which initial values use, is necessary to supply all or just some of them?
Thanks again,
cris

[jhines]

Cris,

If you don't specify initial values, PRESENCE will use 0.0. This works most of the time, but when it doesn't, it's usually because the beta parameter estimate is not close to zero. If you specify initial values, you have to start with the first one, but you don't need to list all of them. If you only wanted to give an initial value for the 3rd parameter, you could enter zero for the first two, then the value for the 3rd, and not enter the rest of them.

When I think I need to give initial values, I usually try final values from similar models that worked first. If I don't have a similar model, I try to guess initial values where my 'guess' is usually something like -2 if I think the parameter is small, or 2 if I think the parameter is large. You don't have to be exact, but this little bit of information is frequently enough to get the optimization routine going to the maximum likelihood value.

Cheers,

Jim