I'm trying to estimate a twin model that has two groups, but I need to estimate the effect of a covariate on y1 and y2. Specifically, I'm trying to estimate example 5.18 in the new user's manual. I'm trying to augment this model so that it is a MACE model. When I just add the covariate on y1 and y2, I can't seem to identify the model. That is, the model will not converge with a chi-square. Any thoughts as to why and how I can identify my model?
For CFA (perhaps also multiple group CFA), and when using a sample of twins, both MZs and DZs, is it possible to use the stratification option along with type = complex to handle nonindependence of twins in sample?
The CLUSTER option deals with non-independence of observations not STRATIFICATION. Non-independence of twins is already handled in ACE twin models because the two twins are analyzed together in a multivariate model that models the non-independence.
Thanks Linda. I haven't considered the ACE approach because I've not seen examples with twin data using latent variables, especially more than two LVs. In my case I am also doing longitudinal CFA and SEM.
Dan R. posted on Friday, September 18, 2015 - 9:06 am
I'm trying to extract factor scores (for individuals) from latent variables created from cross-sectional data that includes family clusters with MZ & DZ twin pairs and siblings within those families. To make it a little more interesting, there are also some singletons. The intention is to create factor scores that control/adjust for the non-independence of not only twins but the family structures that the observations are nested within. I'm not interested (I don't think) in factor structure differences on within and between levels, just a TYPE = COMPLEX model. I've create variables for 'family' and 'zygosity', but I'm not sure if CLUSTER = family zygosity; is what I'm looking for.
Would saving the between level factor scores from a model using the following commands accomplish those goals?
TYPE = TWOLEVEL COMPLEX; CLUSTER = family zygosity;
Thank you for any clarification of these commands that you can provide, or relevant resources to learn more.
I think you want to have Cluster=family, so not include zygosity because the usual twin approach has zygosity as a Grouping variable, that is, a fixed mode of variation. See our UG for twin examples.
Now, if you use Type=Complex with this you will get the same factor scores as without Complex because Complex affects only the SEs of the model parameters.
If you do a Type=Twolevel analysis you would get different within-level factor scores than in a Type=Complex or regular analysis. The between-level would refer to the family component and not the twin component.
Dan R. posted on Sunday, September 20, 2015 - 11:59 am
Thank you, Bengt. Would Cluster=family take each families' clustering/similarity into account (i.e., on an individual family level), or does it do this by findings an average degree of similarity within clusters (an ICC for example) and applying that to individual's data when they come from a clustered group? I guess I'm just looking for more information on how the cluster function works and what it does to the results. I couldn't find much beyond how to use the cluster function in the UG (though I could have missed a relevant section), so some direction to more technical details would be greatly appreciated. Thanks!
Hello! I am conducting analyses similar to what Dan R. posted about - with twins, siblings, and singletons. I simply want to adjust the SEs for the nonindependence in the analyses as opposed to attempt a multilevel analysis (using type = complex). This is not meant to be a genetic study.
I see that it was advised to simply have a cluster variable for family. My concern is that this would ignore the nonindependence of twins nested within some of those family units. Above, it was advised against having a twin cluster variable. Is there anything in Mplus prohibiting having two clustering variables, one for twins/triplets and one for families (outside of it not being the typical approach to twin data)? Thanks!
Cluster=Family and Type=Twolevel or Type=Complex takes into account a common family influence. It sounds like you want to additionally allow for twin influence (so capturing their additional correlation)? Maybe then you want to handle the correlations via a single-level, wide analysis where each family (each row in the data) has variables for twins, siblings, and singletons - with missing data for these variables for families that don't one of these types. I am not familiar with that modeling so you may want to ask on a genetic/twin modeling web site.