EFA with sampling weight PreviousNext
Mplus Discussion > Exploratory Factor Analysis >
 Anonymous posted on Thursday, January 13, 2005 - 9:26 am
I noticed from the menu that frequency weights are unavailable for EFA, is this correct? Thanks!
 Linda K. Muthen posted on Thursday, January 13, 2005 - 4:24 pm
Yes, that is true.
 ando thornton posted on Tuesday, April 29, 2008 - 4:33 am

I would like to do a two-level EFA of a set of 8 indicators where each indicator is actually an aggregate of a further cluster level factor, and these are therefore recorded with differing precision due to differing sample sizes. I would like to take these differing precisions into account.

An example is EFA of 8 indicators where each is measured on 50 schools each year over a 10 year period. Interest is in both the between and within school factor structures. However, the indicators themselves are an average over measurements of a certain number of students within the schools, but I do not have the indicator data at the student level. However I do have access to the standard errors of the indicators and so would like to use these as inverse-variance weights at the within (ie occasion within school) level.

I have read plenty about sampling weights in Mplus, but I wonder whether inverse variance weights can be used in the WEIGHTS command in MPlus and produce valid chi-squared tests etc? For comparison, Stata calls inverse variance weights "aweights" ("analytical weights"), as distinguished from sampling weights or frequency weights.

 Bengt O. Muthen posted on Tuesday, April 29, 2008 - 7:56 am
While not directly related, you may get ideas from seeing how similar concerns about precision differences is taken care of in the dyadic modeling of

Dagne, G.A., Howe, G.W., Brown, C.H., & Muthén, B. (2002). Hierarchical modeling of sequential behavioral data: An empirical Bayesian approach. Psychological Methods, 7, 262-280.

This paper is available on our web site under Papers, Miscellaneous. Dagne has the Mplus input for maximum-likelihood estimation of this.
 Tihomir Asparouhov posted on Tuesday, April 29, 2008 - 9:54 am
The inverse-variance weights apply only to linear regression and
actually they use residual variances rather than standard errors.

I don't see a viable approach - some corners have to be cut - you
could possibly use constraint= command and use type=complex instead
of type=twolevel but you have to first get the SE info converted to
measurement error info.
 anonymous posted on Wednesday, March 10, 2010 - 5:49 am
I am conducting an EFA within a complex survey sample (clusters and weights). When I attempt to request more than 3 factors, I receive a warning stating that no more than 3 factors are available. Is there a way to examine a model with a greater number of factors?
 Linda K. Muthen posted on Wednesday, March 10, 2010 - 9:32 am
There is a limit to the number of factors that can be extracted from a set of observed variables. The formula is shown in the Topic 1 course handout under EFA. It sounds like this is the issue.
 Fabian Barrera posted on Monday, November 17, 2014 - 4:50 am

I am also conducting EFA within complex servey design (schools/students, weighting factors for both levels)... I am examining a latent dimension based on 5 items collected from schools... some outputs have yielded factor loadings greater than 1 in one of these variables. What could be driving these results?
 Bengt O. Muthen posted on Monday, November 17, 2014 - 7:50 am
See our FAQ

Standardized coefficient greater than 1
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message