search

tbates / umx

Making Structural Equation Modeling (SEM) in R quick & powerful
https://tbates.github.io/
44 stars 17 forks source link

Ensure it is trivial to do GSEM with umxRAM #146

Open tbates opened 3 years ago

tbates commented 3 years ago

Consider wrappers for the GWAS pipeline also. https://github.com/MichelNivard/GenomicSEM/wiki

mcneale commented 3 years ago

Yes, there are some weirdnesses with Lavaan or OpenMx WLS(MV) implementations that make them disagree. Others have reported inconsistencies between Lavaan and MPlus. At least we can triangulate a bit - and with two of them being open source we can actually figure out WHY they differ if they do.

For more complicated modeling, OpenMx will be faster than Lavaan due to using C++ instead of R for computation & optimization. However, for WLS, where the weight matrix has to be inverted only once, most of the heavy lifting has already been done, with WLS model-fitting proceeding very rapidly.