The optimizer (NLMINB) claimed the model converged, but not all elements of the gradient are (near) zero

Luana Yang

unread,

Mar 1, 2020, 8:37:35 PM3/1/20

to lavaan

Dear all,

I use Lavaan to build a sem model, but the fit measures not available if model did not converge. How can I deal with it ?

Here is the syntax I am running:

model1 <-'

## Lavaan latan variables

UI =~ PIS+NL

T =~ AT+MTCM+MTWM

P =~ AP+PWM+PCM

## Lavaan regression

amp~ UI+DEM+NPP+T+P+HI

NPP~ P+T+UI+DEM

HI+T+P~ UI

NL ~ PIS

## Lavaan covariance

T~~P

'

fit1 <- sem(model1, data=dataNOR, estimator="MLR",std.lv=TRUE)

The results:

Warning message:

In lav_model_estimate(lavmodel = lavmodel, lavpartable = lavpartable, :

lavaan WARNING: the optimizer (NLMINB) claimed the model converged,

but not all elements of the gradient are (near) zero;

the optimizer may not have found a local solution

use check.gradient = FALSE to skip this check.

lavaan 0.6-5 did NOT end normally after 213 iterations

** WARNING ** Estimates below are most likely unreliable

Estimator ML

Optimization method NLMINB

Number of free parameters 34

Number of observations 229118

Model Test User Model:

Test statistic NA

Degrees of freedom NA

Parameter Estimates:

Information Observed

Observed information based on Hessian

Standard errors Robust.huber.white

Latent Variables:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

UI =~

PIS 0.023 NA 0.023 0.316

NL 0.044 NA 0.044 0.526

T =~

AT 0.207 NA 0.211 1.026

MTCM 0.202 NA 0.206 0.963

MTWM 0.138 NA 0.141 0.920

P =~

AP 0.133 NA 0.134 1.126

PWM 0.075 NA 0.075 0.553

PCM 0.056 NA 0.056 0.646

Regressions:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

amp ~

UI -0.021 NA -0.021 -0.127

DEM 0.065 NA 0.065 0.042

NPP 0.382 NA 0.382 0.368

T 0.033 NA 0.033 0.206

P 0.049 NA 0.050 0.307

HI 0.027 NA 0.027 0.030

NPP ~

P 0.090 NA 0.091 0.583

T 0.002 NA 0.002 0.010

UI 0.020 NA 0.020 0.131

DEM 0.141 NA 0.141 0.094

HI ~

UI 0.027 NA 0.027 0.146

T ~

UI -0.201 NA -0.197 -0.197

P ~

UI -0.090 NA -0.089 -0.089

NL ~

PIS 0.447 NA 0.447 0.384

Covariances:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

.T ~~

.P -0.029 NA -0.029 -0.029

Variances:

Estimate Std.Err z-value P(>|z|) Std.lv Std.all

.PIS 0.005 NA 0.005 0.900

.NL 0.003 NA 0.003 0.449

.AT -0.002 NA -0.002 -0.052

.MTCM 0.003 NA 0.003 0.073

.MTWM 0.004 NA 0.004 0.154

.AP -0.004 NA -0.004 -0.267

.PWM 0.013 NA 0.013 0.694

.PCM 0.004 NA 0.004 0.582

.amp 0.015 NA 0.015 0.573

.NPP 0.016 NA 0.016 0.649

.HI 0.032 NA 0.032 0.979

UI 1.000 1.000 1.000

.T 1.000 0.961 0.961

.P 1.000 0.992 0.992

R-Square:

Estimate

PIS 0.100

NL 0.551

AT NA

MTCM 0.927

MTWM 0.846

AP NA

PWM 0.306

PCM 0.418

amp 0.427

NPP 0.351

HI 0.021

T 0.039

P 0.008

Warning message:

In .local(object, ...) :

lavaan WARNING: fit measures not available if model did not converge

Thanks for your time.

Jeremy Miles

unread,

Mar 1, 2020, 8:52:18 PM3/1/20

to lav...@googlegroups.com

One thing to do is to try different starting values.

However, I think this is a weird looking model. You've got a lot of variables, and very few covariances. (I didn't spend the time to draw out the path diagram, but if you did, it would be helpful).

Some of your variances are small - the usual advice is to make sure that the variances of your variables are within an order of magnitude of one another, or you run into precision estimates.

Finally, you have some negative variances - variances can't be negative, so whether you have fit indices or not, you know that this model is wrong.

Hope that helps,

jeremy

Luana Yang

unread,

Mar 1, 2020, 9:22:17 PM3/1/20

to lavaan

Dear jeremy,

Thanks for your help. I know this model have some problems, but how can I make the variances of my variables are within an order of magnitude of one another?

在 2020年3月2日星期一 UTC+8上午9:37:35，Luana Yang写道：

Jeremy Miles

unread,

Mar 1, 2020, 9:32:07 PM3/1/20

to lav...@googlegroups.com

If the variance is small, multiply it by 10.

If it's large, divide it by 10.

J

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/59182d0f-e486-47e9-aade-76b5017331a9%40googlegroups.com.

Luana Yang

unread,

Mar 1, 2020, 10:06:32 PM3/1/20

to lavaan

（I don't know if my reply was sent successfully, so I sent it again.The variance of the AT was -0.052, and that of PIS was 0.9(I don't know if I should look at the variance after normalization). Should I multiply AT in the raw data by 10 and divide PIS by 10?

在 2020年3月2日星期一 UTC+8上午10:32:07，Jeremy Miles写道：

To unsubscribe from this group and stop receiving emails from it, send an email to lav...@googlegroups.com.

Reply all

Reply to author

Forward

Message has been deleted