
Message/Author 


I am attempting to conduct a path analysis in which I have three waves of data. At each wave child aggression (continuous), punitive parenting (continuous), and mother repartnered (binary) are measured. I am also including a host of controls on which the Wave 1 variables are regressed. The final model is one in which all variables become endogenous except for the controls. I ran the models using the WLSMV estimator and the Theta parameterization because Mplus said Delta was unavailable. 1) I chose to use the WLSMV estimator. Is this correct based on my mix of continuous and binary dependent variables? 2) I have the option of running my models with values for child aggression and punitive parenting standardized (i.e. zscored) or running the models using the raw scores. My gut says to run the models using all raw scores. What do you recommend? 3) My output contains unstandardized results with SEs, STDYX results with no SEs, and STD results with no SEs. If I want to report standardized results and if the IV is continuous and the DV is binary, I should look at the STDYX output. If the IV is binary and the DV is binary, then I should look at the STDY output (although this is not in my output). If the IV and the DV are continuous, then I should look at the STDYX results. Is this correct, and what output should I look at for the binary/binary case since STDY is not outputted? 


1. You can use weighted least squares or maximum likelihood. 2. I would use raw scores. 3. For a continuous IV use STDYX. For a binary IV use StdY. 

H. R. posted on Wednesday, March 07, 2012  12:02 am



I have an observed path model with ordered categorical outcomes and mediators, and I use WLSMV with theta parametrization. What is the matrix analyzed in the WLSMV estimation process? I assume it is not a covariance matrix. And how can it be accessed through the MPLUS output? Thanks a lot in advance. 


The sample statistics for model estimation are the thresholds, probit regression coefficients, and residual correlations. To obtain these, use the SAMPSTAT option of the OUTPUT command. 

H. R. posted on Wednesday, March 07, 2012  8:46 pm



Thank you, Linda. 


Dear Linda, In responding to the OP, why do you recommend using the raw scores instead of standardized scores? What if you need to use a composite score from the raw data? Thank you. 


It is risky to do any kind of standardization of your variables when you are comparing across group or time. The person had 3 waves of data. 

Back to top 

