umx stands for "user" mx function, and is a library of helper functions for doing Structural Equation Modeling in OpenMx.

You can install the most recent version from github as follows

install.packages("devtools")
library("devtools")
install_github("umx", username = "tbates")
library("umx")
?umx
# on windows you might need
# install_github("umx", username = "tbates", args = "--no-multiarch")
# On the newest version of devtools, this will work
# install_github("tbates/umx")

The umx package has a range of useful helpers to help you get things done in SEM. This list is not comprehensive: install the package to learn more

1. Building Models
   • umxStart() # Add sane start values to a model: very helpful
   • umxLabel() # Add labels to paths: Labels allow you to set, equate, and drop paths by label!
   • umxLatent() # Helper for building formative and reflective latent variables from their manifest indicators
2. Run models
   • umxRun() # Use in place of mxRun to: set labels, starts, compute saturated for raw data, run model until it returns green
   • umxReRun() # Modify a model (drop paths etc), run, and even return the comparison all in 1 line
3. Reporting output
   • umxSummary(model) # Get a brief summary of model fit, similar to a journal report (Χ², p, CFI, TLI, & RMSEA)
   • umxPlot(fit1, std=T, precision=3, dotFilename="name") # Create a graphical representation of a RAM model (outputs a GraphViz file)
   • umxTime(fit1) # Report the time taken by a model in a compact friendly, programable format
4. Modify models
   • umxMI() # Report the top n modification indices
   • umxAdd1() # add parameters and return a table of the effect on fit
   • umxDrop1() # Drop parameters and return a table of the effect on fit
   • umxGetParameters(model, regex = "as_r_2c_[0-9]", free = T) # A powerful assistant to get labels from a model. like omxGetParameters but uses regular expressions.
   • umxReRun() # re-run a model: Quickly drop or free parameters, rename the model, and re-run...
5. Data and package helpers
   • umxHcor(data, use = "pairwise.complete.obs") # Compute appropriate pair-wise correlations for mixed data types.
   • umx_lower2full(lower.no.diag, diag=F, byrow=F) # Create a full matrix from a lower matrix of data
   • umxUpdateOpenMx(bleedingEdge = FALSE, loadNew = TRUE) # Update the OpenMx package
   • umxSaturated(model) *# Create a saturated model when raw data are being used. *
     • summary(model, SaturatedLikelihood = model_sat$Sat, IndependenceLikelihood = model_sat$Ind)
     • nb:* Saturated solutions are not computable for definition variables and some other models.

Feel free to use, and improve: Log suggestions here using the Github comments, wiki, or git.

• Tim Bates
• Michael Culbertson (modification index code (based on functions in John Fox's SEM package))
• Ryne Estabrook (code to Standardize RAM models)