I am running a SEM in AMOS and am having issues with themodel and running a mediation. When I covary error terms and increase model fit, one of the regression weights exceed 1.00. When I try and run the mediation the regression weight is over 1 but insignificant. This doesn't seem correct that these exceed 1. Any suggestions?
Dear Johnson,
I believe you are looking into the Standardized estimates in SEM. If you find the values greater than 1, we call it as Heywood case. For such issue one of the solution is to look into multicollinearity. For treating multi collinearity you can try using stats tool package by James Gaskin.
Yes, I think that is problem, since it implies a correlation of greater than 1.0.
In general, it wise to keep error correlations to a minimum, since it risks "over-fitting" the data. Did your modification indices indicate the value of introducing correlated error?
When we have problem of multicollinearity in data (i.e., two or more variables are highly correlated , >.9), then we get standardize regression weight greater than 1 in SEM. You can check correlation among the variables and remove one variable having very high correlation (>0.9) with other variable.
Thanks for your answers. I decided to covary my mediators which increased model fit, is that an issue. I read some thing that said it was ok for mediators to covary but if the indirect effect is not significant does that mean one of the mediators is needed.
There is no technical reason why you can't allow your mediators to covary, but you should have a clear theoretical justification. You also correct that it will complicate the calculation of the indirect effects.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling