Singular matrix error when summarizing structural equation model fit

Question

I have been analyzing some data in R using the lavaan package for maximum likelihood estimation of missing values. I ran the model some days ago and everything worked just fine.

Then I was a bit unlucky and spilled some water over my laptop and had to get a new one. I re-installed R and, since then, I can no longer run the model because it gives me the following error message:

Error in solve.default(X[[i]], ...) : 
  Lapack routine dgesv: system is exactly singular: U[13,13] = 0

Apparently this cannot be solved because the matrix cannot be inverted. However, as I said, before the water-spilling incident everything worked just fine.

If I leave out the maximum likelihood code (see below), then everything works just fine, but cases that I don't want to exclude are excluded. Can you help me?

---

Approach 1 with maximum likelihood estimation of missing values, which does not work:

sem.model <- '# Measurement Model:
              Variable1 =~ Item1 + Item2 + Item3
              Variable2 =~ Item4  + Item5 + Item6
              Variable3 =~ Item7 + Item8 + Item9 + Item10
              Variable4 =~ Item11
              Variable5 =~ Item12

              # Structural Model
              outcomevariable ~ Variable1 + Variable2 + Variable3 + Variable4  + Variable5 + 
                  Item13 + Item14 + Item15 + Item16  + Item17 + Item18'

This first step works:

sem.model.fit <- sem(model = sem.model,
                     data = dataSRL,
                     estimator = "MLR", 
                     missing = "ML")`

This second step gives me the solve.default error quoted above:

summary(sem.model.fit,
        fit.measures = TRUE,
        standardized = FALSE,
        rsquare = TRUE, 
        modindices = TRUE)

Approach 2 without maximum likelihood estimation of missing values, but listwise deletion (this works):

sem.model <- '# Measurement Model:
              Variable1 =~ Item1 + Item2 + Item3
              Variable2 =~ Item4 + Item5 + Item6
              Variable3 =~ Item7 + Item8 + Item9 + Item10
              Variable4 =~ Item11
              Variable5 =~ Item12

              # Structural Model
              outcomevariable ~ Variable1 + Variable2 + Variable3 + Variable4 + Variable5 + 
                  Item13 + Item14 + Item15 + Item16  + Item17 + Item18'

## first step
sem.model.fit <- sem(model = sem.model,
                     data = dataSRL,
                     estimator = "MLR") # deleted some code HERE

## second step
summary(sem.model.fit,
        fit.measures = TRUE,
        standardized = FALSE,
        rsquare = TRUE, 
        modindices = TRUE)

This works for some reason?

Answer 1

good question! I ran into this same problem the other day. The upgrade to lavaan .6-13 is causing this. I have a free two-day seminar on path analysis in lavaan and we had to update it to address this problem. You can get access to the fix and the materials here in case you're interested (it's all free of charge):

https://instats.org/seminar/path-analysis-with-interactions-and-indi1124

Good luck!