donata posted on Wednesday, August 25, 2004 - 3:12 pm
I have 10 variables for a two factor CFA. Two of the variables are negative skewed (-1.8). Is it possible to tranform this variable using LN and then use them for the CFA? I know that the covarianca matrix would change. Is there any other solution?
thanks in advance!
bmuthen posted on Wednesday, August 25, 2004 - 3:26 pm
You can transform the variable using Define. But you can also leave it as is and simply use non-normality robust estimation using the estimator MLR. The choice is affected by what your assumption is regarding the relationship between the factor and this indicator - is it linear or not?
donata posted on Thursday, August 26, 2004 - 8:59 am
My assumption is that the relationship between the factor and the indicator is linear.
But tranforming the variable the variance will change and as a consequence also the covariance matrix. Does it make any problem?
Thanks a lot.
bmuthen posted on Thursday, August 26, 2004 - 11:59 pm
If you assume that you have a linear relationship without the transformation, I would not transform. Instead, I would either use a non-normality robust analysis (MLR or MLM), or use censored or two-part modeling.
If you assume that you get closer to a linear relationship after transforming, then choose that approach. Do the transform in Define. Yes the covariance matrix will change but that should not be a problem.
Dear Dr. Muthens: "Centering" command is to ask Mplus to do the transformation: newx=x-mean(x). Is there a similar simple command to request Mplus to generate standardized scores for the indicator variables automatically? I mean, do we have to use "define" function like the following, e.g. define: newx1=(x1-mean(x1))/std(x1).