mxMI {OpenMx}R Documentation

Estimate Modification Indices for MxModel Objects


This function estimates the change in fit function value resulting from freeing currently fixed parameters.


mxMI(model, matrices=NA, full=TRUE)



An MxModel for which modification indices are desired.


Character vector. The names of the matrices in which to search for modification


Logical. Whether or not to return the full modification index in addition to the restricted.


Modification indices provide an estimate of how much the fit function value would change if a parameter that is currently fixed was instead freely estimated. There are two versions of this estimate: a restricted version and an full version. The restricted version is reported as the MI and is much faster to compute. The full version is reported as MI.Full. The full version accounts for the total change in fit function value resulting from the newly freed parameter. The restricted version only accounts for the change in the fit function due to the movement of the new free parameter. In particular, the restricted version does not account for the change in fit function value due to the other free parameters moving in response to the new parameter.

The algorithm respects fixed parameter labels. That is, when a fixed parameter has a label and occurs in more than one spot, then that fixed parameter is freed in all locations in which it occurs to evaluate the modification index for that fixed parameter.

When the fit function is in minus two log likelihood units (e.g. mxFitFunctionML), then the MI will be approximately chi squared distributed with 1 degree of freedom. Using a p-value of 0.01 has been suggested. Hence, a MI greater than qchisq(p=1-0.01, df=1), or 6.63, is suggestive of a modification.

Users should be cautious in their use of modification indices. If a model was created with the aid of MIs, then it should always be reported. Do not pretend that you have a theoretical reason for part of a model that was put there because it was suggested by a modification index. This is fraud. When using modification indices there are two options for best practices. First, you can report the analyses as exploratory. Document all the explorations that you did, and know that your results may or may not generalize. Second, you can use cross-validation. Reserve part of your data for exploration, and use the remaining data to test if the exploratory model generalizes to new data.


A named list with components


The restricted modification index.


The full modification index.


A list of models with one additional free parameter


Sörbom, D. (1989). Model Modification. Psychometrika, 54, 371-384.

The OpenMx User's guide can be found at


# Create a model
manifests <- names(demoOneFactor)
latents <- c("G")
factorModel <- mxModel("One Factor",
      manifestVars = manifests,
      latentVars = latents,
      mxPath(from=latents, to=manifests),
      mxPath(from=manifests, arrows=2),
      mxPath(from=latents, arrows=2,
            free=FALSE, values=1.0),
      mxPath(from = 'one', to = manifests),
      mxData(observed=cov(demoOneFactor), type="cov", numObs=500,
             means = colMeans(demoOneFactor)))
#No SEs for speed
factorModel <- mxOption(factorModel, 'Standard Errors', 'No')
factorRun <- mxRun(factorModel)

# See if it should be modified
# Notes
#  Using full=FALSE for faster performance
#  Using matrices= 'A' and 'S' to not get MIs for
#    the F matrix which is always fixed.
fim <- mxMI(factorRun, matrices=c('A', 'S'), full=FALSE)
round(fim$MI, 3)
plot(fim$MI, ylim=c(0, 10))
abline(h=qchisq(p=1-0.01, df=1)) # line of "significance"

[Package OpenMx version 2.7.16 Index]