I am running a trivariate Cholesky decomposition using a script from the workshop. This script did not include CIs but I included it from another source. However, while the CI script allows me to run CIs for the genetic and environmental correlations, I can't figure out how to modify the script so I can run CIs for my standardized path estimates. The portion of my script that runs the CIs for correlations is below:
corComp_A <- mxAlgebra(solve(sqrt(ACE.IACE.A))%%ACE.A%% solve(sqrt(ACE.IACE.A)), name="corComp_A")
corComp_E <- mxAlgebra(solve(sqrt(ACE.IACE.E))%%ACE.E%% solve(sqrt(ACE.IACE.E)),name="corComp_E")
Add GE correlations to the model, and ask the model to report CIs when it runs
CImodel <- mxModel(multAEFit, corComp_A, corComp_E, mxCI(c("corComp_A", "corComp_E")))
Run the model
CImodelFit <- mxRun(CImodel, intervals=TRUE)
CIAEsumm <- summary(CImodelFit)CI
CIAEsumm
Any suggestions?
try
summary(CImodelFit)$CI
instead of
CIAEsumm <- summary(CImodelFit)CI
Currently, the CI summary shows 0 for paths fixed to zero.
I wonder if paths that are fixed might be better shown with NA in the upper and lower bound? i.e., the CI was not estimated at zero, it was unable to be moved from zero (or whatever the path was fixed at).
Perhaps they should even be suppressed from the output, perhaps as an option? "suppress.fixed.paths=F" That would also clean out all the empty upper triangles of type=Lower matrices from the output.
I wonder too if CIs which include zero might helpfully be marked with an asterisk or something in a new column showing that it is not significant at the Confidence level requested: that might be a 2.0 feature, but with these features, the output would be very easy to interpret.
Having programmed such things myself, there is a nice CPU saving when trying to compute the CI of a fixed matrix element. In reality, it isn't NA, it is what it is (the current value) so I don't support changing this behavior.
Marking those whose CI's overlap with zero with an asterisk seems contrary to the usual behavior - * for significant, ** for more significant, *** for really f***ing significant. Also, although zero is a very common marker for significance, it isn't the only one (we might want to know if a genetic correlation is sig different from 1, e.g.). Still, zero is the usual milepost, so perhaps an NS marker would be ok.