Longitudinal modeling with extended twin design

Hey,

I have a couple of questions regarding my research and I would really appreciate getting some feedback if my undertaking is reasonable.
I am investigating stability of a political phenotype in a two-wave panel design. I'm including not only twins in my analysis, but also one non-twin full sibling. The scripts are standing and working (verified with replication).

I further partition the C into the twin environment (only shared between twins) and C (environment shared by twins and their non-twin sib). So my base model will be ATCE.
My first question here is: Am I free to estimate dominant effects in these models: ADTE, ADCE? In fact, is there any combination of ADTCE which would not work (short of one parameter)?
I am not really expecting dominance, but still I want to compare it to the base model in the appendix (model fit, AIC).
Is there any data on how many full sibs need to be available for it making sense to estimate four parameters? (full sibs are available for about 25-50% of the data). Still I am aware that I gain power by just including them, even if only three parameters are estimated. Are there any recommended papers besides Posthuma (2002) which specifically focus on this design and you recommend? Haven't been able to successfully come up with much.

First I do classic univariate analysis, investigating both waves separately. I plan on reporting both saturated and reduced models because I have some inconsistency in the results. One cohort (I investigate cohorts, separately) has the following results: wave 1: A: 50%, E: 50%; wave 2: A: 25%, T: 25%, E: 50%. It does not make any sense for T to appear in the second wave if it was not present before. Even more surprising: the 25 % T disappears entirely after further model reducing, which I find hard to believe as it is quite sizable to drop out. Also interesting: the A only barely wins the contest. That is, the model is extremely short of dropping the A and just keeping T and E. I'm really confused and don't really know what to do with these results. Performing the analysis without the full-sib does produce very similar results. Now the other cohorts provide also similarly confusing results, though to lesser degree. There seems to be a C/T present in all cohorts in wave 2, but not/only limited in wave 1. In some cohorts the T drops out after model reduction, in others it does not. Due to the phenotype at hand, I would not entirely preclude the C/T being activated through some sort of environmental stimuli (election) --> context, therefore breaking the pattern where C/T are usually found. However, again, the model reduction provides mixed evidence for the C being consistent or not. How could I resolve or present this?

Following univariate analysis, I switch to a bivariate Cholesky model (also including non-twin sib). Now my main question here is: Should I drop paths beforehand which were dropped in the univariate case (have seen this in the literature). Or consider the full model as the base model no matter what, even if now some sizable C/T remains after model reduction (therefore to some extent contradicting univariate findings), instead of excluding T beforehand? I tend to consider the T, as I'm aware that through bivariate analysis, more covariances may be able to better detect the C/T.

Any help would be much appreciated! Particularly what you would like to see as a reviewer.

Thanks a lot.