umx!

Factor scores with missing data!

It is often useful to obtain scores for subjects on the latent variables in a model.

A familiar method is scores from fact.anal. This has several limits: It’s limited to factor models and can’t handle missing data. This tutorial covers both cases briefly, begining with umxEFA.

Factor scores from factor analysis

umx provides umxEFA to conduct exploratory factor analysis, and this supports returning scores for subjects. Unlike fact.anal which doesn’t allow missing data, umxEFA support scores even in data with missing cells.

Like fact.anal, to get scores from umxEFA you simply set scores = "Regression". umxEFA supports alternative methods for computing them.

Here’s a quick 2-factor model of some variables in the built-in mtcars data set:

myVars = c("mpg", "disp", "hp", "wt", "qsec")
x = umxEFA(mtcars[, myVars], factors =   2, rotation = "promax", scores= "Regression")
x

Factor scores from arbitrary RAM models.

require(umx)
data(demoOneFactor)

manifests = names(demoOneFactor)
m1 = umxRAM("OneFactor", data = demoOneFactor,
	umxPath("G", to = manifests),
	umxPath(v.m. = manifests),
	umxPath(v1m0 = "G")
)

We can compute factor scores thus:

fs = umxFactorScores(m1, 'Regression')

Create some missing data and compute factor scores

All the data-points of demoOneFactor are present. Let’s delete 10% from each column at random:

df = demoOneFactor
n = dim(df)[1];
for (i in dim(df)[2]) {
	df[sample(n, .1 * n), i] = NA
}
all(complete.cases(df))

Run again with the missing data:

# 1. update model with new data
m2 = mxModel(m1, mxData(df, type = "raw"))
# 2. score factors
fs = umxFactorScores(m2, 'Regression')
umx_round(fs, digits=3)

Just saved yourself $1000 buying Mplus for factor scores with missing data!