#### Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

# Where to find the covariance of latent variables in JASP SEM analyses?

Trying to find this but can only see the covariances of the items of the variables, not the covariances of the latents in the SEM model. Can you help me find this? Thanks!.

• edited October 2020

Hi @Liesbeth,

Thank you for this discussion. You are right. I see that the covariances of the latent variables are not presented in the output panel (in JASP version), and there only exist variances of the latent variables. I will ask @evankesteren to solve this issue!

Cheers,

Ihnwhi

• edited October 2020

Hi @Liesbeth,

With the political democracy data (stored in the JASP SEM data archive), I performed SEM analysis with three latent variables (ind60, dem60, and dem65) where dem65 is predicted by ind60 and dem60. That is, dem65 is an outcome latent variable (i.e., endogenous latent variable), and ind60 and dem60 are explanatory latent variables (i.e., exogenous latent variables). I used the lavaan syntax below.

# latent variables

ind60 =~ x1 + x2 + x3

dem60 =~ y1 + y2 + y3 + y4

dem65 =~ y5 + y6 + y7 + y8

# regressions

dem65 ~ ind60 + dem60

# residual covariances

y1 ~~ y5

y2 ~~ y4 + y6

y3 ~~ y7

y4 ~~ y8

y6 ~~ y8

When I applied the syntax, I could see the estimate of the covariance between two latent variables (the covariance among latent variables should be expressed as "factor1 ~~ factor2", for example). To be specific, I could see the covariance between the two explanatory latent variables (ind60 and dem60) in the output panel. I emphasized the corresponding part with the black rectangle.

In this example, the covariance between the two latent variables, ind60 and dem60, is 0.665. When I tried to predict a certain latent variable with other latent variables, covariance is estimated. This is because, the covariance between exogenous variables is estimated, by default. If the covariance is still not estimated when there are even exogenous latent variables, please let us know!

You can also find the extensive JASP SEM tutorial at https://jasp-stats.org/2018/07/03/how-to-perform-structural-equation-modeling-in-jasp/.

Cheers,

Ihnwhi

• Thank you! I'll look into it!

• Can I just check, I had given variables permission to correlate, so when I put in

latentfactor 1~~ latentfactor2

this should not change any of the fit indices or parameter estimates, I assume?

• I noticed when I do that, there is one Df less, and the estimate is always 1.000. Is there another way that does not change the other estimates, as it seems when I put this in this way, it fixes the covariance to 1? I would be looking for a way to know the covariance without fixing the pathway if that is at all possible. Thanks in advance for your help with this!

• Hi @Liesbeth,

Thank you for your follow-up questions!

Yes, you can specify covariance among latent variables as latentfactor1 ~~ latentfactor2

In SEM, covariances among the exogenous variables (exogenous variables refer to variables that predict other variables) are automatically estimated! Otherwise, you should specify the covariance, and this results in a loss of DF as you observed.

I am not sure why the estimate becomes always 1 although you did not explicitly specify the covariance as latentfactor1 ~~ 1 * latentfactor2. Could you share parts of your model syntax for me to take a closer look at?

Cheers,

Ihnwhi