

Regression with Variables (loaded on ... 

Message/Author 

MSP posted on Monday, November 16, 2015  11:06 am



Hello! Not sure where to post this, so I just started a new thread (sorry). I was wondering what is the conceptual difference between a simple multiple (logistic) regression, and a regression where my predictors are loaded on latent factors. Does the latter make sense? Y on X1X10; !Y is binary vs. F1F3 by X1X10(*1); Y on X1X10; The output estimates were different (Y on X1X10), but their signs and significance are similar (some became smaller, some became bigger with F1F3). What is the effect of loading X1X10 to F1F3 in X's prediction of Y? Can I say that X1X10's prediction of Y is influenced by F1F3? Any input would be appreciated. Thanks! 


In the model using BY you are imposing a model, that is, putting a restriction on the correlations of the marginal distribution of the 10 x's while the other model allows them to be freely correlated. The model with BY is a bit strange because you don't relate the f1f3 factors to other variables such as Y. 

Back to top 

