Bo Fu posted on Monday, October 25, 2010 - 10:16 pm
May I ask about the correlation matrix in EFA? My data contains binary variables, multi-normial variables and continuous variables. I'm wondering, in EFA in MPlus, after I identify the categorical variables' group, how the correlation matrix constructed? by which correlation, (tetrachoric, polyserial, pearson, spearman)
The correlation matrix contains the correlations appropriate for the pair of variables involved, for example, tetrachoric for two binary, Pearson for two continuous, biserial for one binary and one continuous etc.
Bo Fu posted on Tuesday, October 26, 2010 - 1:25 am
Thank you so much!
And, what is used for between categorical and binary variable, categorical and continuous?
I think by categorical you mean ordered polytomous. A polytomous correlation is used for a binary and ordered polytomous variable. A biserial correlation is used for a continuous and ordered polytomous variable.
Bo Fu posted on Tuesday, October 26, 2010 - 4:30 pm
So, different types of correlations are in the correlation matrix, which is for EFA processing. Is that fair, or say appropriate for this analysis? Because, say variables A1, A2, B1, B2 and C1, C2, where
A1, A2: multilevel categorical; B1, B2: binary; C1, C2: continuous for the correlation matrix,
where rAB is for correlation between A and B. so, there is three types of correlations involved: 1) polytomous: rA1A2, rA1B1, rA1B2, rA2B1, rA2B2 2) tetrachoric: rB1B2 3) biserial: rA1C1, rA1C2, rA2C1, rA2C2, rB1C1, rB1C2, rB2C1, rB2C2 4) Pearson: rC1C2
Is it meaningful that one correlation matrix contains different correlation based on different criterion, to used as same scale values in EFA?