Message/Author |
|
|
I have used EFA with geomin rotation on a sample of 319 respondents. I am in a similar situation to many of the people who have posted on this thread. A reviewer has asked me for the percent of variance explained. I have reviewed every post, and I understand that % variance explained is a measure for PCA, not EFA, and that the goal of EFA is to reproduce the correlation matrix. In doing some reading about summarizing and reporting EFA, I found a citation that suggests that the percentage of explained variance can serve as a means of assessing how much the correlation matrix is reproduced. The citation can be found here: Lorenzo-Seva, U. (2013). How to report the percentage of explained common variance in exploratory factor analysis. Technical Report. Department of Psychology, Universitat Rovira i Virgili, Tarragona. Document available at: http://psico.fcep.urv.cat/utilitats/factor/ I have two questions: 1. Is reporting the percentage of explained common variance a reasonable way to summarize a oblique rotation? (In the working paper, Lorenzo-Seva uses Varimax) 2. Is there a way to calculate the percentage of explained common variance in MPlus? Thanks for your help with this! Meredith |
|
|
With an orthogonal rotation, you can sum the squared factors loadings and divide by the number of factor indicators. There is no option for this in Mplus. |
|
Shirley posted on Friday, June 02, 2017 - 3:11 am
|
|
|
Dear Dr. Muthen, I have a survey of 12 items on 3 scales. As a few of the items are positively worded, I fit a CFA with 3 trait factors and 1 method factor (as shown below). The 3-trait-1-method model fit significantly better than the original 3-trait-factor model and had better fit statistics (e.g., RMSEA). As the original 3-trait-factor model provides acceptable fit and is the hypothesis, I am interested to examine the impact of this method factor (similar to the idea of ECV in the previous post). In the CFA context, may I check if the following equation is appropriate? (% of variance in the data the method factor accounts for)=(sum of the squared STDYX loadings on the 3 trait factors from all 12 items)/((sum of the squared STDYX loadings on the 3 trait factors from all 12 items)+(sum of the squared STDYX loadings on the method factor from the 4 cross-loading items))? Thanks! ********************************* VARIABLE: [omitted] CATEGORICAL=q1-q12; CLUSTER=teacher; ANALYSIS: TYPE=COMPLEX; MODEL: f1 by q1* q2-q4; f2 by q5* q6-q8; f3 by q9* q10-q12; Method by q1* q2 q5 q11; f1-Method@1; Method WITH f1-f3@0; |
|
|
I think your numerator should instead be (sum of the squared STDYX loadings on the method factor from the 4 cross-loading items) Otherwise ok. |
|
Shirley posted on Sunday, June 04, 2017 - 8:04 pm
|
|
|
Thanks for spotting the oversight in the equation! |
|
fred posted on Saturday, May 11, 2019 - 8:23 am
|
|
|
Hi, Am I correct to assume that sum of sq loadings divided by n of indicators is not appropriate in Geomin roatated EFA.? Is there then a similar way for the geomin (or other oblique rotations) to get the explained variance? I am familiar with discussion on the issue not being the goal of EFA, still the journal reviewers tend to require the % explained v, I guess because of PCA being the prototype and tradition. |
|
|
I don't think it is appropriate with correlated factors. |
|
Back to top |