How to interpret standardized coefficients for latent variable in CFA?

82 views
Skip to first unread message

Guilherme Parreira

unread,
Nov 20, 2020, 8:12:51 AM11/20/20
to lavaan
Hello there!

I am fitting a CFA model with the following syntax:

    cfa_esp <- ' Tarefa =~ Coop. + Esf.Melh. + PapelImp. 
                 Ego =~ Pun.Erros + Recon.Desig. + Riv.Membr.'
    m0_cfa <- cfa(cfa_esp, data =data, orthogonal = F, estimator = "WLSMV")

And I obtain these results from the coefficients:

    summary(m0_cfa, standardized = T)$PE

    Latent Variables:
                             Estimate  Std.Err  z-value  P(>|z|)   Std.lv  Std.all
      Tarefa =~                                                                   
        Coop.                   1.000                               0.688    0.891
        Esf.Melh.               0.639    0.042   15.170    0.000    0.439    0.793
        PapelImp.               0.677    0.043   15.678    0.000    0.466    0.797
      Ego =~                                                                      
        Pun.Erros               1.000                               0.262    0.403
        Recon.Desig.            4.872    1.514    3.218    0.001    1.277    1.388
        Riv.Membr.              1.069    0.221    4.834    0.000    0.280    0.307


How can I interpret `Std.all` column?

This is my try for each estimate column:

+ A unit increase in `Tarefa` implies in .639 units increase in `Esf.Melh.`;
+ One standard deviation increase in `Tarefa` implies in .439 units increase in `Esf.Melh.`
+ The correlation of `PapelImp.` and `Tarefa` is .797 (since we are building latent variable, Std.all is interpreted as correlation);

This is what I thought it was right. However, the Std.all coefficient is 1.388 for `Ego` and `Recon.Desig.`. So, how can I interpret these coefficients?

Thanks
Obs.: I also made this question on stats.stackexchange. Thanks

Guilherme Parreira

unread,
Nov 20, 2020, 9:11:44 AM11/20/20
to lavaan
In fact, this is my full model:
cfa_esp <- '# Definição das Variáveis latentes (CFA)

             # Criação das Variáveis latentes
             Tarefa =~ Coop. + Esf.Melh. + PapelImp. 
             Ego =~ Pun.Erros + Recon.Desig. + Riv.Membr. 
             TrabalhoEmEquipe =~ PREPARACAO + EXECUCAO + AVALIACAO + AJUSTES + GME
             
             SatisfacaoNoEsporte =~ TreinInstr + TratPessoal + DesempenhodaEquipe + DesempenhoIndividual

             # Regressão
             TrabalhoEmEquipe ~ Tarefa + Ego
             SatisfacaoNoEsporte ~ Tarefa + TrabalhoEmEquipe + Ego
             
             # Correlação
             Tarefa ~~ Ego
           '
But I don't think that this makes difference in the interpretation.

Best Regards

Patrick (Malone Quantitative)

unread,
Nov 20, 2020, 9:19:45 AM11/20/20
to lav...@googlegroups.com
A standardized loading over 1 is an impermissible solution, but judging the large standard error on the unstandardized coefficient it may not be significantly over 1.

That said, the standardized loadings on Ego are widely disparate, especially with only three items. You may have an empirical underidentification issue. I would go back and examine the (polychoric?) correlation matrix, as well as the residual correlation matrix from the CFA.

--
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/7da4c96c-12e8-464f-a052-aa8cbca482f2n%40googlegroups.com.


--
Patrick S. Malone, Ph.D., Malone Quantitative
NEW Service Models: http://malonequantitative.com

He/Him/His

Guilherme Parreira

unread,
Nov 20, 2020, 12:47:29 PM11/20/20
to lavaan
Thank you Mr. Malone.

Hum, I see. I thought that the coefficients in SEM were constrained to vary in the correct scale (in this case, I thought that std.all was forced to be between [-1;1]), but they are not. Interesting. So, this is just a sign that my model is not running fine... OK

But my interpretations are right (if those coefficients were reliable?)?

Tks

Patrick (Malone Quantitative)

unread,
Nov 20, 2020, 1:24:39 PM11/20/20
to lav...@googlegroups.com
Look into "Heywood case" where a residual variance is negative. You haven't said explicitly, but you very likely have a negative residual variance for Recon.Desig. You should look into your data and results to see how concerned to be.

Also, you're dealing with factor loadings, which are a kind of regression coefficient, but usually aren't discussed in the same language. I'm also not sure why you changed from regression language in your first two points to correlation language in your third, which is incorrect.

Pat

Guilherme Parreira

unread,
Nov 20, 2020, 1:35:06 PM11/20/20
to lavaan
Thank you Mr. Pat.

I do "have a negative residual variance for Recon.Desig", thanks. I will review it in my variable construction.

Ok, thank you. I will correct the interpretation from the third one:.

+ A unit increase in `Tarefa` implies in .639 units increase in `Esf.Melh.`;
+ One standard deviation increase in `Tarefa` implies in .439 units increase in `Esf.Melh.`
+ One standard deviation increase in `Tarefa` implies in .797 standard deviation increase in `PapelImp.`

Thank you so much! Saved my weekend
Reply all
Reply to author
Forward
0 new messages