Linda, in 2009 your responded to a question by saying that Mplus does not do Principal Components Analysis. Is that still true?
I didn't see it mentioned in the Mplus 7.0 Users guide.
By principal components analysis I mean extraction of orthogonal components that are weighted combinations of observed values. The components explain 100% of the observed variance, and one can say how much variance is accounted for by each component.
If Mplus does do Principal Components Analysis now, could you point me to the Mplus syntax?
Yes, this is still true. Mplus does not do Principal Component Analysis.
Jan Ivanouw posted on Monday, October 16, 2017 - 1:46 pm
While I am not advocating the use of Principal Conponents analysis, some psychological research rely heavily on this method. In order to estimate the amount of error stemming from this method, I would like to examine some data sets with both PC and EFA. Now, I wonder if I can mimick PC by fixing residual variance to 0, forcing all variance to be used in the definition of the factors?
Maybe that works, keeping the factors uncorrelated. But I wonder if those zero residual variances will lead to a non-pos def covariance matrix.
Jan Ivanouw posted on Tuesday, October 17, 2017 - 11:15 am
Thank you for you answer.
I tried with a CFA. You're right about the covariance matrix. However, it works when I fix the residual variances at a very small amount.
Is it possible to try out the same with EFA or ESEM. I tried, but - for EFA it seems that I cannot fix error variance, - for ESEM I could not get it to accept specification either of orthogonal factors or fixed error variances.
Did I just not do it right, or is it impossible in Mplus? How would the commands look like?
I think you can do it in ESEM by choosing the right rotation procedure and by fixing residual variances to low values. But on the whole, I'm not sure it is worth trying out and instead just do PCA by other software.
Jan Ivanouw posted on Wednesday, October 18, 2017 - 4:48 am