StdYX vs. estimated correlations
Message/Author
 Ferdinand Gruebler posted on Sunday, August 24, 2008 - 5:44 am
Hi,

I was wondering what the difference is between the standardized coefficients given in the StdYX-row by using the STANDARDIZED-command in the output line and the estimated correlation matrix you can get by using the Tech4-command in the output line.

I thought standardized coefficient in a linear regression model and though in a latent regression model can be interpreted like correlations.

But the StdYX- and estimated correlation-values differ between the same latent variables ...

What value should I take then?
What is the difference between stdYX and "estimated correlations"?
 Linda K. Muthen posted on Sunday, August 24, 2008 - 10:39 am
With one covariate, these are the same. With more than one, they are not. You would use standardized if you want standardized coefficients and TECH4 if you want correlations.
 Ferdinand Gruebler posted on Sunday, August 24, 2008 - 10:48 am
hum, ok, thanks.

But I have one more question:

I have empirical Data and a model.
And I used STANDARDIZED and TECH4.

Now I have a stdYX Value of -0.307 from an independent on an dependent latent variable.

The estimated correlation matrix says me now that these two variables correlate by r=.354.

But how could that be?

How can the standardized regression coefficient be negative while the correlation is positive?
 Bengt O. Muthen posted on Sunday, August 24, 2008 - 11:01 am
Is that independent latent variable the only variable that influences the dependent latent variable you mention?
 Ferdinand Gruebler posted on Sunday, August 24, 2008 - 11:10 am
No, there are six latent variables who have an influence on the dependent latent variable.

So, I understand that the coefficient is not the same as the correlation when there are more than one independent variable.

But I thought that if the coefficient is negative the correlation can not be positive, can it?
 Bengt O. Muthen posted on Sunday, August 24, 2008 - 11:39 am
Yes, it can. Because the correlation consists of several parts, also going through the other independent variables, where their correlations influence the correlation. Linear regression books probably have examples of this.