I was wondering, how similar or different is umxACEv as compared to umxSexLim (the closest of umxSexLim being the homogeneity choice)? Are they both using the direct variance approach (I know it says for umxSexLim it says it is based on a correlated factors model)? If they are in fact different in that sense, is there a tradeoff between them in that sense (correlated factors vs direct variance)? Definitely would be helpful to know!

Thanks as always!

I guess I am mainly curious about how (if it at all even could be compared) a univariate ACE direct variance approach would relate to a homogeneity related umxSexLim model. I feel like both have their benefits (you can vary the mean in males/females in umxSexLim), but in univariate ACEv, it's inherently not as biased from my understanding. Would that be correct? And if so, is it kind of up to debate which may be preferred in a given context in general, or are they best not compared (since it may not be apples to apples)?

Hi there,

`umxSexLim`

is Cholesky-based. So at the end, you get out a umxACE model, not a umxACEv model.Okay, I see, that makes sense. Thanks!

Do you know how beneficial it is to actually let the means vary when it comes to males vs females (the homogeneity model) vs not allowing that to happen?

I guess there might be a more specific to OpenMx version of the direct variance approach with altering the means for males vs females (or at least, something along those lines) but may be a bit more tricky for implementation.

I guess a related question is--currently I have data which has a relatively equal # of male and female pairs (not significant difference), and I do not include it as a covariate for ACEv (I only stumbled upon the sex limitation model recently)--so am just wondering how notable it might be to allow the mean sexes to vary (and as that isn't implemented in umx yet with the direct variance approach, but would need to be implemented through other means) if I am most interested in the group level (homogeneous) variance result?

umx functions generally result in an mxModel. Once you understand how these are specified, it is pretty easy to change the fitted model (fix, free parameters, or stop them being constrained equal to each other). umx has a possibly easier to use version, iirc, but in any case, you need to inspect the model in R to figure out which of its constituent models and matrices have the relevant parameters in them. The learning slope gets a bit steeper, but given R expertise it's really not bad.

Thanks again! I see. I will have to look into it more carefully in this case--there does appear to be umxSetParameters() which seems to parallel omxSetParameters(). I would be curious if you could actually alter the means between just the sexes in this case, which I will have to look into. Hopefully it is relatively straightforward after spending some time on it.