umxEFA does not return loadings

[AgnerF]

Thanks for maintaining this code and for fixing issue #103. I am impressed how well umxEFA solves problems with missing values and how it can do a factor analysis with more variables than observations. It takes hours to calculate in my case, even without the summary, but the result is perfect.

Anyway, two new bugs have appeared since last time:

1. umxEFA returns the data matrix rather than the loadings unless you specify loadings="model".
2. umxEFA does not work with rotation methods from GPArotation, e.g. quartimax. The error message says "Error in t.default(rm) : argument is not a matrix".

[tbates]

I pushed a commit to umxEFA last week so that it returns scores by default when requested (used to require user to also set return = "loadings") this and other edits might have fixed your issue. I'm can't replicate the issue with quartimax:

x = umxEFA(mtcars[, myVars], factors = 2, rotation = "quartimax", return = "loadings")
[1] "Rotation results"
Orthogonal rotation method Quartimax converged.
Loadings:
         F1     F2
mpg  -0.893  0.135
disp  0.914 -0.140
hp    0.756 -0.530
wt    0.968  0.177
qsec -0.327  0.833

Rotating matrix:
      [,1]   [,2]
[1,] 0.989 -0.149
[2,] 0.149  0.989

[mcneale]

Could you return SEs of the scores as well? With the rubbish scales many of us have to use/analyze, the SEs can vary a lot, which has knock-on side effects of subsequent analyses of the factor scores.

[tbates]

Could add a note to people who want the full-Monty to use mxFactorScores which returns this attributed 3-D matrix:

 num [1:50, 1, 1:2] -0.167 -0.106 -2.137 -0.157 1.106 ...
 - attr(*, "dimnames")=List of 3
  ..$ : chr [1:50] "1" "2" "3" "4" ...
  ..$ : chr "G"
  ..$ : chr [1:2] "Scores" "StandardErrors"

Is there also an example paper doing something with SEs, I could cite that.