`layout_matrix()` can make `semPaths()` fail with `non-numeric argument to binary operator`

[jorgesinval]

Hi, thanks for semptools.

I ran into an issue when using a layout created by layout_matrix(). After subsetting to latent nodes, passing the same layout to semPaths() throws an error.

# packages
library(lavaan)
library(semPlot)
library(semptools)

model <- "
# measurement model
visual  =~ x1 + x2 + x3
textual =~ x4 + x5 + x6
speed   =~ x7 + x8 + x9
g       =~ visual + textual + speed

# latent correlations
visual ~~ textual + speed
textual ~~ speed

# latent regressions
visual + textual + speed ~ g
"

# fit the model
fit <- sem(model, data = HolzingerSwineford1939)

# alternative coordinate layout
m2 <- layout_matrix(
  g       = c(3, 1),
  visual  = c(1, 2),
  textual = c(2, 2),
  speed   = c(4, 2)
)

plot_model <- semPlotModel(fit)

plot_model <- keep_nodes(
  object = plot_model,
  nodes = m2[!is.na(m2)])

# draw the diagram
semPaths(object = plot_model,
  layout = m2)

Error

Error in `l[, 1] - min(l[, 1])`:
! non-numeric argument to binary operator

I believe this error is not related to semptools, but rather to semPlot.

[sfcheung]

I believe you want to draw only the latent variables. There is no need to use keep_nodes(). You can use structural = TRUE when calling semPaths().

But you are right, the error is due to semPaths(). Not sure if this is a bug or intentional, but I recall encountering this problem before. Somehow, the layout matrix sometimes cannot have only two columns. The following matrix will work, adding one empty columns:

library(semptools)
m2 <- layout_matrix(
  g       = c(3, 1),
  visual  = c(1, 3),
  textual = c(2, 3),
  speed   = c(4, 3)
)
m2
#>      [,1] [,2] [,3]     
#> [1,] NA   NA   "visual" 
#> [2,] NA   NA   "textual"
#> [3,] "g"  NA   NA       
#> [4,] NA   NA   "speed"

^{Created on 2026-04-04 with reprex v2.1.1}

One more thing. The names in the layout matrix must match the labels of the nodes in the output of semPaths(). Therefore, you also need to add nCharNodes = 0, or use the abbreviated labels generated by semPaths() when using the layout matrix. This is a requirement of the layout argument of semPaths, not a limitation of semptools.

Please find below a modified version of your code:

# From https://github.com/sfcheung/semptools/issues/204

# packages
library(lavaan)
#> This is lavaan 0.6-21
#> lavaan is FREE software! Please report any bugs.
library(semPlot)
library(semptools)

model <- "
# measurement model
visual  =~ x1 + x2 + x3
textual =~ x4 + x5 + x6
speed   =~ x7 + x8 + x9
g       =~ visual + textual + speed

# latent correlations
visual ~~ textual + speed
textual ~~ speed

# latent regressions
visual + textual + speed ~ g
"

# fit the model
fit <- sem(model, data = HolzingerSwineford1939)
#> Warning: lavaan->lav_model_vcov():  
#>    Could not compute standard errors! The information matrix could not be 
#>    inverted. This may be a symptom that the model is not identified.

# alternative coordinate layout
m2 <- layout_matrix(
  g       = c(3, 1),
  visual  = c(1, 3),
  textual = c(2, 3),
  speed   = c(4, 3)
)

semPaths(object = fit,
  nCharNodes = 0,
  layout = m2,
  structural = TRUE)

^{Created on 2026-04-04 with reprex v2.1.1}

If nCharNodes is not used, the labels will be automatically abbreviated. The layout matrix will work if the abbreviated labels are used:

m3 <- layout_matrix(
  g       = c(3, 1),
  vsl     = c(1, 3),
  txt     = c(2, 3),
  spd   = c(4, 3)
)

semPaths(object = fit,
  layout = m3,
  structural = TRUE)

Hope this helps.

[jorgesinval]

Thank you! I opened an issue in the semPlot repo.