

Covariances estimated that shouldn't ... 

Message/Author 


I have a path analysis model defined as such: (all variables are observed and treated as continuous). MODEL: income on educ; npnews on income educ polint; polint on income educ; natnews on polint educ; pdiscuss on polint npnews; According to the documentation (I have Mplus 5.1, but I get the same results in the Mplus 5.2 demo), the covariance between the residuals for endogenous variables will be fixed to zero by default. Yet one of them, between pdiscuss and natnews, is estimated by Mplus, according to the output. I can fix it to zero using natnews WITH pdiscuss@0 in the model command. It does NOT estimate the covariance between npnews and natnews residuals. This seems to be fixed at zero by default, as the manual says it would be. I just can't figure out why Mplus is freely estimating the residual covariances in the former case but not the latter case. If I specify ANALYSIS: MODEL = NOCOVARIANCES; then it appears that these covariances are being correctly constrained to zero. 


I think I found the answer to my question in a previous post, in which the response was "The Mplus default is to estimate residual covariances for dependent variables that do not influence any other variable." That explains what I see in the output. I overgeneralized information on page 30 of the v5.1 manual, from an example path analysis, which states "The residuals are not correlated as the default." Apparently this applies only to the example provided, in which there is only one variable that isn't modeled to causally affect any other. 


Just a general note. If you obtain parameters you did not specify in the MODEL command, then they are free as the default. If you don't want them, set them to zero. 

Back to top 

