This forum is designed for general questions about how to use OpenMx. If you can't find another place where your question fits, then this is the place to be!
Modification indices (MIs) could be computed from the estimated first and second derivatives. A bit of tinkering with matrix algebra might yield the MIs.
In other software packages, MIs do not, as far as I know, take into account possible equality constraints between elements of the matrices. It would be a neat feature if OpenMx could do this, because MIs are useless for behavior genetic models in which it is meaningless to free up, e.g., a path from genotype to phenotype for one individual but not to do so for their relative at the same time.
This is something that I'd like to see addressed in some way. I'm not sure I like "modification indices", though.
I'd rather see something implemented like Taehun Lee's and Bud MacCallum's recent work on parameter influence. The idea there is to estimate a value for how influential a parameter is. That way we are gaining understanding about how much a model would change if a parameter were gone (or added), and also gaining understanding as to what proportion model fit is changed when a parameter is forced away from its optimal value by some small (either proportional or absolute) value.
Yes, however, I am not sure what that brings over and above a likelihood-based confidence interval (or a scatterplot matrix of bootstrap parameter estimates).
In my understanding, the likelihood-based confidence interval methods tell about single parameter moves while the fungible parameter method asks about whether confidence intervals covary. Bootstrapping tells more about how the data relates to the model mixed with the structure of the model itself. Fungible parameters, as I understand it, are designed to tell about off-axis parameter instability in the structure of the model.
Standardized can be computed from the output: see the twin scripts for an example, where variance components are standardized against the sum of variance.
Passed along on behalf of a another user: Is there a way to get standardized estimates and modification indices from the output of mxRun?
Did you find an answer to this question? I'm especially interested in the modification indices.
Modification indices (MIs) could be computed from the estimated first and second derivatives. A bit of tinkering with matrix algebra might yield the MIs.
In other software packages, MIs do not, as far as I know, take into account possible equality constraints between elements of the matrices. It would be a neat feature if OpenMx could do this, because MIs are useless for behavior genetic models in which it is meaningless to free up, e.g., a path from genotype to phenotype for one individual but not to do so for their relative at the same time.
This is something that I'd like to see addressed in some way. I'm not sure I like "modification indices", though.
I'd rather see something implemented like Taehun Lee's and Bud MacCallum's recent work on parameter influence. The idea there is to estimate a value for how influential a parameter is. That way we are gaining understanding about how much a model would change if a parameter were gone (or added), and also gaining understanding as to what proportion model fit is changed when a parameter is forced away from its optimal value by some small (either proportional or absolute) value.
Yes, however, I am not sure what that brings over and above a likelihood-based confidence interval (or a scatterplot matrix of bootstrap parameter estimates).
In my understanding, the likelihood-based confidence interval methods tell about single parameter moves while the fungible parameter method asks about whether confidence intervals covary. Bootstrapping tells more about how the data relates to the model mixed with the structure of the model itself. Fungible parameters, as I understand it, are designed to tell about off-axis parameter instability in the structure of the model.
Standardized can be computed from the output: see the twin scripts for an example, where variance components are standardized against the sum of variance.