

Binary variablezero variance 

Message/Author 

Abe Khademi posted on Saturday, March 14, 2015  5:58 pm



I'm running a qtechnique EFA where the Person x Item matrix is transposed to an Item x Person matrix. The responses are binary (0/1). The problem is that if a person responds all 1's or all 0's to an item, the item gets zero variance and EFA analysis terminates with an error that a variable is not categorical because it is all 1 or all 0. Can we tell Mplus that all variables have two categories and regard the variable with all 1's or all 0's as binary too? In CFA we use $ to show the number of thresholds. How about EFA? What should we do with a variable with no variance? Thanks. 


If you have all zeros or ones in a column of the matrix (so for a certain person), you can't estimate the factor loading. Maybe instead you want to do LCA. People with all zeros may constitute a separate class and same for the ones; 0's and 1's don't hurt. 

Abe Khademi posted on Wednesday, March 18, 2015  6:47 pm



Thanks, Bengt for the explanation and alternative method. 

Back to top 

