standardized and unstandardized estimates different signs

[T Anderson]

I have a rather large SEM mediation model (no latent variables), from which I am calculating the relative importance of the total direct and indirect effects for several endogenous variables. For one of these defined parameters (total direct effects), I've come across something that I do not quite understand how it would be possible: the unstandardized estimate for the defined parameter is negative in sign, while its standardized estimate (std.all) is positive in sign. Perhaps I'm not understanding exactly how the standardized estimates are calculated, but does anyone have any advice or thoughts on how this could be occurring, or am I missing something obvious here? I've tried Googling information and looking through the other threads on how this might happen but have not had much luck so far. Thanks!

[Terrence Jorgensen]

the unstandardized estimate for the defined parameter is negative in sign, while its standardized estimate (std.all) is positive in sign.

Strange.  Could you provide a minimal R script and enough data to reproduce this result?  

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen

[T Anderson]

After much banging my head against the desk, I'm fairly certain I figured this out. There was one variable that was contributing to the derived quantity that had a variance that was much larger than the other variables- when fitting the model, I would get the warning:

lavaan WARNING: some observed variances are (at least) a factor 1000 times larger than others

I had read in other threads on here that this warning "usually" didn't cause any problems. However, after dividing by the variable by 10 to reduce its variance, the unstandardzied and standardized estimates make more sense and have the same sign, so perhaps my model was an example where the warning did matter.

I can still to upload an example if it would be of interest.

[Terrence Jorgensen]

I can still to upload an example if it would be of interest. 

I think it could be, if you can share it.  It can be hard to manufacture problems like this, in order to possibly add a check for them.