Pairwise deletion option not working for CFA, WLSMV

[curti...@gmail.com]

Hi everyone,

I'm trying to fit a CFA right now with WLSMV, and wanted to use the pairwise deletion method of handling missing values.

Unfortunately, it hasn't been working for me. Might anyone be able to diagnose the error source?

I've included the code snippet and error message below.

From what I understand back in 2013 (http://lavaan.ugent.be/history/dot5.html), pairwise deletion was implemented.

Thanks for your help in advance.

# Case B: trying missing="pairwise"

resultsPwise <- cfa(ctlModel, dataset, estimator = "WLSMV", missing="pairwise")

summary(resultsPwise, fit.measures = TRUE)

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

Error output

> resultsPwise <- cfa(ctlModel, dataset, estimator = "WLSMV", missing="pairwise")

Error in nlminb(start = start.x, objective = minimize.this.function, gradient = GRADIENT,  : 

  NA/NaN gradient evaluation

In addition: Warning message:

In lav_samplestats_from_data(lavdata = lavdata, missing = lavoptions$missing,  :

  lavaan WARNING: number of observations (219) too small to compute Gamma

[Yves Rosseel]

On 06/30/2018 11:22 PM, curti...@gmail.com wrote:
> Error in nlminb(start = start.x, objective = minimize.this.function,
> gradient = GRADIENT,  :
>   NA/NaN gradient evaluation

Oh dear. The optimizer was lost in (parameter) space. This is most
likely due to this:

>   lavaan WARNING: number of observations (219) too small to compute Gamma

It looks like Gamma is unstable. I have recently found a way to
'stabilize' Gamma in such cases, that may or may not help in this case.
Could you send me your data + script? It would help me to test this.

Yves.

[Tim Cadman]

Hi Yves,

I've had exactly the same error message trying to use pairwise deletion with WLSMV:

> pairwise.fit <- cfa(model, data = data, estimator="WLSMV", missing="pairwise")

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

  NA/NaN gradient evaluation

In addition: Warning message:

In lav_samplestats_from_data(lavdata = lavdata, missing = lavoptions$missing,  :

  lavaan WARNING: number of observations (523) too small to compute Gamma

Did you make any progress with this?

best,

Tim