I've been using twin models for a few years now and this is my first time running into this problem.
I am interested in testing whether emotion-specific processing is heritable. To do this, I've tried using a difference score (and am estimating whether the difference in processing between emotion X and emotion Y is heritable) and I've also tried taking a residual from regressing one emotion on another.
I've noticed that my ICCs are pretty close to zero and sometimes MZ ICC is negative while DZ ICC is positive, and sometimes it's the opposite. In most cases, the E model is the best fit in line with the ICCs not showing any strong correlations between twins. However some emotions do have significant A and/or C components. I have 400 participants with about 2/3 being MZ twins. Same-sex pairs only and no non-twin siblings.
As the title says, I'm getting negative ACE estimates where the A and C component cancel each other out. For example, one ACE model result is 0.41, -0.41, 0.99 for the ACE components, respectively. Another model is 0.70, -0.48, 0.78.
When the AE or CE models are the best fit, usually I end up with two positive components. But one significant AE model results in A = -0.18 and E = 1.18. The full ACE model for this was originally -0.35, 0.15, 1.20 and the ICCs were -0.16 for MZ twins and -0.02 for DZ.
I get similar results regardless of whether I use the difference score or the residual score, and I've used the same script multiple times without this problem (and I got the script from a behaviour genetics course I took). This is however my first time trying to assess heritability using a difference score or residual. I've also tried the ADE model and get the same pattern still, but the ACE models fit better.
What I want to know is whether negative ACE components are valid and if so, how to interpret them? If it's more likely to be a model misspecification I can share my script, but I didn't get any warnings. If there are any other posts on this I would love to read it as I couldn't find any using the search.
Thank you very much for your help!