

Use of Partial Covariances in SEM (Pa... 

Message/Author 

Lynn Imai posted on Thursday, October 05, 2006  6:47 pm



Dear Prof(s) Muthen, I am trying to use path analysis to test a model. I have one main predictor of theoretical interest (a type of personality variable), leading to a process variable, which then influences an outcome variable. I have 7 other variables I need to control for to see if my main exogenous variable predicts over and beyond the others. My questions are: 1) Is it common practice to use a partialled out covariance matrix to free up degrees of freedom, instead of explicitly modeling all control variables in path analysis and 2) If so, how would you do this using Mplus? All variables are continuous. Thank you in advance for any guidance! 


It is not common practice to analyze a residual matrix in my experience. It is always better to do an anlysis in one step rather than more than one. I would include that control variables in the model as covariates. 

Back to top 

