Lapack routine dgesv: system is exactly singular: U[51,51] = 0

[ryan chesnut]

Hello lavaan community,

I'm encountering the following error message when estimating a cross-lagged panel model using version 0.6-13: Lapack routine dgesv: system is exactly singular: U[51,51] = 0.

I've previously estimated this model using version 0.6-12 with no issues. I've downgraded to version 0.6-12 and the model estimates again seemingly fine. As always, I appreciate the help this group provides.

Here is my model syntax:

t2t5distress.pf.clpm5 <-'
#measurement model
t2dis=~t2ptsdsum+f2*t2depsum+f3*t2anxsum
t3dis=~f1*t3ptsdsum+f2*t3depsum+f3*t3anxsum
t4dis=~f1*t4ptsdsum+f2*t4depsum+f3*t4anxsum
t5dis=~f1*t5ptsdsum+f2*t5depsum+f3*t5anxsum
t2pf=~f4*t2parent_func1r+f5*t2parent_func2r+f6*t2parent_func3r+f7*t2parent_func4r+f8*t2parent_func5r
t3pf=~f4*T3PARENT_FUNC1+f5*T3PARENT_FUNC2+f6*T3PARENT_FUNC3+f7*T3PARENT_FUNC4+f8*T3PARENT_FUNC5
t4pf=~f4*T4PARENT_FUNC1+f5*T4PARENT_FUNC2+f6*T4PARENT_FUNC3+f7*T4PARENT_FUNC4+f8*T4PARENT_FUNC5
t5pf=~f4*T5PARENT_FUNC1+f5*T5PARENT_FUNC2+f6*T5PARENT_FUNC3+f7*T5PARENT_FUNC4+f8*T5PARENT_FUNC5
t2self=~t2mi3+t2mi4+t2mi5+t2mi6
t2oth=~t2mi1+t2mi2+t2mi7+t2mi8+t2mi9
t1combat=~t1expos1r+t1expos2r+t1expos3r+t1expos4r+t1expos5r+t1expos6r+t1expos7r+t1expos8r+t1expos9r
#correlated residuals
t2ptsdsum~~t3ptsdsum+t4ptsdsum+t5ptsdsum
t3ptsdsum~~t4ptsdsum+t5ptsdsum
t4ptsdsum~~t5ptsdsum
t2depsum~~t3depsum+t4depsum+t5depsum
t3depsum~~t4depsum+t5depsum
t4depsum~~t5depsum
t2anxsum~~t3anxsum+t4anxsum+t5anxsum
t3anxsum~~t4anxsum+t5anxsum
t4anxsum~~t5anxsum
t2mi1~~t2mi2; t2mi3~~t2mi4; t2mi5~~t2mi6; t2mi7~~t2mi8+t2mi9
t2parent_func1r~~T3PARENT_FUNC1+T4PARENT_FUNC1+T5PARENT_FUNC1
T3PARENT_FUNC1~~T4PARENT_FUNC1+T5PARENT_FUNC1
T4PARENT_FUNC1~~T5PARENT_FUNC1
t2parent_func2r~~T3PARENT_FUNC2+T4PARENT_FUNC2+T5PARENT_FUNC2
T3PARENT_FUNC2~~T4PARENT_FUNC2+T5PARENT_FUNC2
T4PARENT_FUNC2~~T5PARENT_FUNC2
t2parent_func3r~~T3PARENT_FUNC3+T4PARENT_FUNC3+T5PARENT_FUNC3
T3PARENT_FUNC3~~T4PARENT_FUNC3+T5PARENT_FUNC3
T4PARENT_FUNC3~~T5PARENT_FUNC3
t2parent_func4r~~T3PARENT_FUNC4+T4PARENT_FUNC4+T5PARENT_FUNC4
T3PARENT_FUNC4~~T4PARENT_FUNC4+T5PARENT_FUNC4
T4PARENT_FUNC4~~T5PARENT_FUNC4
t2parent_func5r~~T3PARENT_FUNC5+T4PARENT_FUNC5+T5PARENT_FUNC5
T3PARENT_FUNC5~~T4PARENT_FUNC5+T5PARENT_FUNC5
T4PARENT_FUNC5~~T5PARENT_FUNC5
#correlated disturbances
t2self~~t2oth+t2dis+t2pf
t2oth~~t2dis+t2pf
t2dis~~t2pf
t3dis~~t3pf
t4dis~~t4pf
#LV variances
t2dis~~1*t2dis
t3dis~~NA*t3dis
t4dis~~NA*t4dis
t5dis~~NA*t5dis
t2pf~~1*t2pf
t3pf~~NA*t3pf
t4pf~~NA*t4pf
t5pf~~NA*t5pf
#structural model
t2self~t1combat+T1GENDER+t1paygrade2+t1deptimer+t2numkids+navy+marines+af+ngr+t2secondmarr+t2nevermarried+t2sdw+race
t2oth~t1combat+T1GENDER+t1paygrade2+t1deptimer+t2numkids+navy+marines+af+ngr+t2secondmarr+t2nevermarried+t2sdw+race
t2dis~t1combat+T1GENDER+t1paygrade2+t1deptimer+t2numkids+navy+marines+af+ngr+t2secondmarr+t2nevermarried+t2sdw+race
t2pf~t1combat+T1GENDER+t1paygrade2+t1deptimer+t2numkids+navy+marines+af+ngr+t2secondmarr+t2nevermarried+t2sdw+race
t3dis~c1*t2dis+a1*t2self+a2*t2oth+b1*t2pf
t4dis~c2*t2dis+c3*t3dis+a3*t2self+a4*t2oth+b2*t2pf+b3*t3pf
t5dis~c4*t2dis+c5*t3dis+c6*t4dis+a5*t2self+a6*t2oth+b4*t2pf+b5*t3pf+b6*t4pf
t3pf~c7*t2pf+b7*t2dis+a7*t2self+a8*t2oth
t4pf~c8*t2pf+c9*t3pf+b8*t2dis+b9*t3dis+a9*t2self+a10*t2oth
t5pf~c10*t2pf+c11*t3pf+c12*t4pf+b10*t2dis+b11*t3dis+b12*t4dis+a11*t2self+a12*t2oth
#Indirect Effects
t2s.t3d.t4p := a1*b9
t2o.t3d.t4p := a2*b9
t2s.t4d.t5p := a3*b12
t2o.t4d.t5p := a4*b12
t2s.t3d.t4d.t5p := a1*c3*b12
t2o.t3d.t4d.t5p := a2*c3*b12
t2s.t3p.t4p := a7*c9
t2o.t3p.t4p := a8*c9
t2s.t4p.t5p := a9*c12
t2o.t4p.t5p := a10*c12
t2s.t3p.t4p.t5p := a7*c9*c12
t2o.t3p.t4p.t5p := a8*c9*c12
t2s.t3d.t4p.t5p := a1*b9*c12
t2o.t3d.t4p.t5p := a2*b9*c12
t2d.t3d.t4p := c1*b9
t2d.t3d.t4d.t5p := c1*c3*b12
t3d.t4d.t5p := c3*b12
t2d.t3p.t4p := b7*c9
t2d.t3p.t4p.t5p := b7*c9*c12
t3d.t4p.t5p := b9*c12
'
t2t5distress.pf.clpm5.fit <- sem(t2t5distress.pf.clpm5,spss1,estimator="MLR",missing="FIML.x",std.lv=TRUE)
summary(t2t5distress.pf.clpm5.fit,fit.measures=TRUE,standardized=TRUE,rsquare=TRUE)

Thanks,

Ryan

[Terrence Jorgensen]

If you can provide the data for a reprex, it could be useful to open an issue on GitHub. 

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[Yves Rosseel]

This is most likely due to a bug in 0.6-13 (only) that occurs if you have:
- missing = "ml" (or "fiml", or "fiml.x")
- exogenous observed 'x' covariates in the model
- fixed.x = TRUE (which is the default)

and you request the default fit measures. The error occurs in a new
piece of code that computes the 'robust' RMSEA in case of incomplete data.

As a workaround, either set fixed.x = FALSE, or downgrade to 0.6-12:

library(remotes)
install_version("lavaan", "0.6-12")

or install the github version where this bug has been fixed:

library(remotes)
install_github("yrosseel/lavaan")

A new version of lavaan (0.6-14) including this fix will probably be
released in a few weeks.

Yves.