cfa with binary categorical variables and missing data

150 views

Skip to first unread message

Elena Murray

unread,

Apr 2, 2019, 10:22:41 AM4/2/19

to lavaan

Dear all,

i am new to lavaan, so please excuse my beginners questions: I am trying to run a CFA on binary categorical (Yes/no) data. I also have some missing data, and want to avoid listwise deletion.

As en estimator i used WLSMV (DWLS), which wont let me use "missing = fiml".

My command looks like this for the moment:

fit <- cfa(model1, data = SOMSE, estimator = "WLSMV", missing="pairwise")

summary(fit, fit.measures = TRUE, standardized = TRUE)

If i use use missing = "pairwise" i receive the error msg:

Error in nlminb(start = start.x, objective = objective_function, gradient = GRADIENT, :

NA/NaN gradient evaluation

Any idea what i can do now? Is there any way i dont have to run multiple Imputation?

Thank you very much already!

PD

unread,

Apr 3, 2019, 6:06:32 AM4/3/19

to lavaan

fiml only works with the ml estimator

Did you order your 'items' for the WLSMV estimator?

The error means that something went wrong in the objective or gradient function, try listwise instead. Pairwise deletion can lead to nonpositive definite matrices

Terrence Jorgensen

unread,

Apr 4, 2019, 9:40:19 AM4/4/19

to lavaan

Is there any way i dont have to run multiple Imputation?

missing="pairwise" assumes MCAR, which is quite a strict assumption. Multiple imputation (facilitated by the runMI() functionality in semTools) only assumes MAR data.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

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