Standard Errors don't match AMOS

[David Simmonds]

Hi Lavaan Experts,

I am new to Lavaan and have been comparing AMOS models (on which I learned SEM) with lavaan.

Everything matches, in terms of general model fit. Covariances and Estimates match fine. 

But the Standard Errors and Z-scores (Critical Ratios?) are different. The Standard Errors in lavaan seem to be lower by about 30% and the z-scores higher by about 25%.

Please see below.

Any help, tips, guidance, etc will be greatly appreciated.

Thanks,

David

***********************************************************************************

AMOS

=======

=======

			Estimate	S.E.	C.R.	P	Label
Perceived_Length			2.127	.398	5.343	***	par_5
e1			.035	.010	3.602	***	par_6
e2			.091	.018	5.010	***	par_7
e3			.161	.034	4.706	***	par_8
e4			.020	.006	3.292	***	par_9
e5			.047	.010	4.729	***	par_10

			Estimate	S.E.	C.R.	P	Label
Judge1	<---	Perceived_Length	1.000
Judge2	<---	Perceived_Length	.795	.030	26.359	***	par_1
Judge3	<---	Perceived_Length	1.368	.043	32.009	***	par_2
Judge4	<---	Perceived_Length	.828	.019	43.698	***	par_3
Judge5	<---	Perceived_Length	.730	.023	31.750	***	par_4

***********************************************************************************

lavaan

=========

Latent Variables:

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

  Perceived_Length =~                                    

    Judge1               1.000                           

    Judge2               0.795    0.023     34.344      0.000

    Judge3               1.368    0.033     41.634      0.000

    Judge4               0.828    0.015     56.668      0.000

    Judge5               0.730    0.018     41.308      0.000

Variances:

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

    Perceivd_Lngth   2.105       0.303       6.954    0.000

   .Judge1                0.035       0.007       4.741    0.000

   .Judge2                0.090       0.014       6.524    0.000

   .Judge3                0.159       0.026       6.147    0.000

   .Judge4                0.020       0.005       4.290    0.000

   .Judge5                0.046       0.008       6.168    0.000

[Christian Arnold]

Hi, at first glance it looks like you are using the MLR or MLM estimator with lavaan. Make sure that you estimate with ML so that the results are comparable. A tip: if you want to compare Amos and lavaan results, it makes sense to run lavaan in EQS emulation mode. Best, CA

Am 07.10.18, 07:02, David Simmonds <davidms...@gmail.com> schrieb:

--
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 post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

[car...@web.de]

Ah, sorry. I didn't see the attachments. The chisquare values are not identical either. You are using the wrong matrix. Best, CA

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

[David Simmonds]

  Hi Christian

Thanks very much for your feedback. It's greatly appreciated  I do believe the matrix is the correct one though, because the estimates are the same. There are several matrices which tiny variations but they don't make a difference to the standard errors. 

Which one do you believe I should be using?

I tried to use ML for estimation but it gives me this error: Error in Gamma[[g]] %*% WD : requires numeric/ complex matrix/vector arguments

Best Regards,
David Simmonds

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

To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

--

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.

[Yves Rosseel]

Are you sure N=100? The SE from AMOS seem to match the values you would
get if N=60.

Yves.

[car...@web.de]

Hi Yves et al., the model implied covariance matrix was used as input for lavaan. Best, CA

[David Simmonds]

Hi Christian,

 I'm using examples from my SEM course in grad school. I have the correlation matrices but not the covariates matrices. That's why I'm using the implied covariances from the Amos output as input to lavaan as input. I guess that's correct?

Best Regards,
David Simmonds, PhD

On Sun, Oct 7, 2018, 2:27 PM <car...@web.de> wrote:

Hi Yves et al., the model implied covariance matrix was used as input for lavaan. Best, CA

--

[David Simmonds]

Yves Himself !! 

How's it going, Rockstar of R?  

Thanks for jumping in. Great package you've written - lavaan. 

I feel awkward. LOL. You are right. I corrected it and made N=60 and voila!! 

  Perceived_Length =~                                    

    Judge1               1.000                           

    Judge2               0.795    0.030   26.603    0.000

    Judge3               1.368    0.042   32.249    0.000

    Judge4               0.828    0.019   43.895    0.000

    Judge5               0.730    0.023   31.997    0.000

Variances:

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

    Perceivd_Lngth    2.091    0.388    5.387    0.000

   .Judge1            0.035    0.010    3.672    0.000

   .Judge2            0.089    0.018    5.054    0.000

   .Judge3            0.158    0.033    4.762    0.000

   .Judge4            0.020    0.006    3.323    0.001

   .Judge5            0.046    0.010    4.778    0.000

 

[David Simmonds]

Yves, Christian,

One last question so this whole thing is not a waste of time, unless I should post it in another forum.

When I try to specify the estimator, I get this message:

 SEM_Model <- sem(SEM_ModelText, sample.cov = Covariance_Matrix, estimator="MLM",sample.nobs = 60)

Error in Gamma[[g]] %*% WD : 

  requires numeric/complex matrix/vector arguments

If I need to post in another thread I understand. I feel like a jerk for wasting everyone's time but I do appreciate all your help!!

[Yves Rosseel]

On 10/7/18 9:21 PM, David Simmonds wrote:
> Yves, Christian,
>
> One last question so this whole thing is not a waste of time, unless I
> should post it in another forum.
>
> When I try to specify the estimator, I get this message:

> * SEM_Model <- sem(SEM_ModelText, sample.cov = Covariance_Matrix,
> estimator="MLM",sample.nobs = *60)
> /Error in Gamma[[g]] %*% WD : /
> /  requires numeric/complex matrix/vector arguments/

MLM (and MLR) require a full dataset; they do not work with sample
statistics.

Yves.