
Message/Author 

EFried posted on Thursday, January 31, 2013  6:53 am



We have 9 variables and want to perform EFA/CFA/ESEM with MPLUS. Variables are between 0.00 and 1.00 with an excess of zeroes, we're currently multiplying them with 100 and use COUNT since they are poisson distributed. Unfortunately, all 9 variables always sum to exactly 1 for each person, leading to nonconvergence. Have you dealt with this issue before and would have recommendations as to how to solve this? Thank you 


Are the measures ipsative? 

EFried posted on Monday, February 04, 2013  4:53 am



Not really. We measured distances of people to each other in a social network on a Likert scale (15). The 9 variables are the percentage of types of relations in a social network. These sum to 100%. E.g., if a person has lots of "close social relations", the person has less relations that are not close. I found very little literature on that. The data are fine, but R, MPLUS, and SPSS all run into convergence issues for PCA and EFA/CFA/ESEM, so I assume the problem is that the variables add to 100%. Thank you Linda! 


Without understanding the motivation for factor analysis in this situation, one simple approach is to delete 1 of the 9 variables so they don't sum to 1; it sounds like the 9th variable doesn't carry any extra information. I don't know how relevant this is but, sociometric analysis using crossclassified models is described by Hox in his multilevel book, section 9.2. 

Back to top 

