Dear Statalisters,
I'm using Stata 15.1.
I am using the "sem" command for a mediation model with 1 treatment, 4 mediators and 1 outcome, controlling for baseline values.
Both the mediators and the outcome (controlling for baseline values) are measured with error. This implies I have 10 latent variables, each one measured by only one observed variable. The point is: I have information on the standard error of each observed variable, for each observation. How can I use such information? If I well understand, sem has its own weight matrix. I have seen that analytic weights are not allowed, frequency weights are not feasible (they require integer values), I’ve read that importance weights are meant for programmers. The only remaining option would be probability weights, but here weights do not depend on sampling probability. Moreover, I see results differ a lot between importance and probability weights.
Above all, I have no idea how these 10 different standard errors could be combined to build a unique weight variable. Below an image of a simplified version of my model (where I don't include baseline values, so only 5 latent variables are visible, but each of them is affected by its baseline value). My question is: how can I use this information on the standard deviation of the error of each observation of each measured variable?
I'm using Stata 15.1.
I am using the "sem" command for a mediation model with 1 treatment, 4 mediators and 1 outcome, controlling for baseline values.
Both the mediators and the outcome (controlling for baseline values) are measured with error. This implies I have 10 latent variables, each one measured by only one observed variable. The point is: I have information on the standard error of each observed variable, for each observation. How can I use such information? If I well understand, sem has its own weight matrix. I have seen that analytic weights are not allowed, frequency weights are not feasible (they require integer values), I’ve read that importance weights are meant for programmers. The only remaining option would be probability weights, but here weights do not depend on sampling probability. Moreover, I see results differ a lot between importance and probability weights.
Above all, I have no idea how these 10 different standard errors could be combined to build a unique weight variable. Below an image of a simplified version of my model (where I don't include baseline values, so only 5 latent variables are visible, but each of them is affected by its baseline value). My question is: how can I use this information on the standard deviation of the error of each observation of each measured variable?
