Hello,
We are just beginning to work with OpenMx to study behavior in relation to genetic and environmental components in twins. In our dataset we have a fairly large number of missing values (checks showed they are missing at random). Normally we would opt for multiple imputation in SPSS and run our analyses on the separate imputed datasets, while also presenting a pooled analyses. But because of the behavioral genetics component, we want to use OpenMx for our analyses.
I read elsewhere on this Forum a very brief mention of reading in multiple files and running the same mxModel on each, but does OpenMx also provide any possibility of running a pooled analysis of all the separate imputed datasets?
Thanks in advance!
Jizzo
Printer-friendly version
OpenMx lacks built-in features for multiple imputation. But, there's a good chance you won't need multiple imputation, since full-information maximum-likelihood (FIML) is likewise robust to missingness-at-random; see Rubin (1976) and Schafer & Graham (2002). If you have missing observations on any regression covariates, avoid modeling the covariates as "definition variables" that define a conditional model for the phenotypic means. Instead, incorporate the regression into the model for the phenotypic covariance and thereby model the joint distribution of phenotypes and covariates, either as in Schwabe et al. (2016), or by setting up your model using MxPaths; you may find features in R package 'umx' useful.
Thanks for you comment. We'll take a look at this method.
FWIW, OpenMx does FIML whenever you analyze raw data, with
mxFitFunctionML().