mediation with binary outcome and total effects

[Chiara S]

M<-aX

Y<bM+dX

Y+M<-confounder 

indirect:= a*b

direct:=d 

total=?

As far as I understood, a binary/categorical outcome lavaan will use a DWLS estimator; thus, it will assume that the ordinal outcome is a discretization of a latent continuous distribution. Thus, can I interpret the magnitude of the indirect effects derived from the multiplication of a*b? Or do I have a scaling problem, so I should just interpret the sign of the coefficient?

[Terrence Jorgensen]

As far as I understood, a binary/categorical outcome lavaan will use a DWLS estimator; thus, it will assume that the ordinal outcome is a discretization of a latent continuous distribution.

Correct.

 

Thus, can I interpret the magnitude of the indirect effects derived from the multiplication of a*b?

 You can, on the latent scale: https://doi.org/10.1177/0049124113494572

Or do I have a scaling problem, so I should just interpret the sign of the coefficient?

If you are doing this in a multigroup or longitudinal model, then you need to link the scale across groups/occasions in order for coefficients to be on the same scale (and thus comparable) across groups/occasions.  That involves equating at least 2 thresholds.  For binary outcomes, only a single threshold can be equated, so differences in intercepts are confounded with differences in (residual or marginal) variances.  But these parameters are all equated in single-group single-occasion models, which is the situation Breen et al. (2013) discuss.

Terrence D. Jorgensen    (he, him, his)
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
http://www.uva.nl/profile/t.d.jorgensen

[Chiara S]

I'm working with a cross-section of data, but my results should definitely differ between men and women and between rural and urban areas. However, in another paper, they suggest that in the case of multigroup, compare the ratio of indirect/total or direct/indirect: https://academic.oup.com/esr/article-abstract/30/1/107/442812?redirectedFrom=fulltext

Thus, can I interpret the magnitude of the indirect effects derived from the multiplication of a*b?

 You can, on the latent scale: https://doi.org/10.1177/0049124113494572

Does this mean that the scale parameter is implicit in the error variance, which follows a standard normal distribution since it estimates a probit? Thus, I would not need to calculate it explicitly since it is already accounted for in the DWLS estimation process.

Their suggested method estimates the indirect effect, accounting for confounding and rescaling, including the residualized M, assuming that residualized M is orthogonal to X. But this procedure should be done outside Lavaan, correct?

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/8e732db9-4207-4cf4-a957-c0d0716956a9n%40googlegroups.com.