LikeLike

]]>LikeLike

]]>LikeLike

]]>About the effect size of the meditation effect or ‘proportion of mediation’ (Iacobucci, Saldanha, Deng, 2007): it’s indeed not captured by the sumIDE, but it’s the ration between the sumIDE and your total effect. Rather than fitting the whole model with that formula included, you can also compute it in R taking the estimate of ‘sumIDE’ and dividing it by the estimate of ‘total’.

LikeLike

]]>Finally, in your previous comment you suggested that I add the mediation effect size. I’m assuming that means that it’s not being captured in the sumIDE part of the syntax. What can I add to produce the effect size? Does that mean my c’ is NOT the coefficient next to sumIDE in the output?

LikeLike

]]>I tried to reproduce your error by fitting the model you describe above and applying the fit to a dataset after renaming some variables to your names, but I could not reproduce your problem. At the bottom of Defined Parameters, above ‘total’ I have the three contrasts.

Maybe it’s an issue of earlier versions of lavaan? Then a solution could be to write out the whole difference instead of using labels, as in:

contrast1 := a1 * b1 – a2 * b2

instead of:

contrast1 := indirect1 – indirect2.

Does this sort things out?

LikeLike

]]>outmodelTV <-"

DateViolMean ~ c*PopTV52 + b1*SRS8Mean + b2*SABWMn + b3*FSexO10Mn

#mediator models

SRS8Mean ~ a1*PopTV52

SABWMn ~ a2*PopTV52

FSexO10Mn ~ a3*PopTV52

#indirect effects (IDE)

SRS8MeanIDE := a1*b1

SABWMnIDE := a2*b2

FSexO10MnIDE := a3*b3

sumIDE := (a1*b1) + (a2*b2) + (a3*b3)

#contrasts to determine if the IDEs differ

contrast1 := SRS8MeanIDE – SABWMnIDE

contrast2 := SRS8MeanIDE – FSexO10MnIDE

contrast3 := SABWMnIDE – FSexO10MnIDE

#total effect

total := c + (a1*b1) + (a2*b2) + (a3*b3)

#covariates

SRS8Mean ~~ SABWMn

SRS8Mean ~~ FSexO10Mn

SABWMn ~~ FSexO10Mn

"

fit <-sem(outmodelTV, data=mv, se="bootstrap", bootstrap=10000)

#view output

summary(fit, fit.measures=TRUE, standardized=TRUE, rsquare=TRUE)

#bootstrap CIs

bootfitTV <-parameterEstimates(fit, boot.ci.type="bca.simple",level=0.95, ci=TRUE,standardized = FALSE)

bootfitTV

LikeLike

]]>LikeLike

]]>I think your analysis looks sound, but you could add two things:

1 – mediation effect size (sumIDE / total) – see (“A meditation on mediation…”)

2 – if you like to show if a specific indifrect effect through a mediator is larger than another you can contrast the mediators by subtracting the indirect effects from one another (e.g. SRS8MeanIDE – SABWMnIDE; SRS8MeanIDE – FSexO10MnIDE; SABWMnIDE – FSexO10MnIDE). HTH!

LikeLike

]]>And, here is the syntax:

outmodelmovie <-"

DateViolMean ~ c*MovieMonthSum + b1*SRS8Mean + b2*SABWMn + b3*FSexO10Mn

#mediator models

SRS8Mean ~ a1*MovieMonthSum

SABWMn ~ a2*MovieMonthSum

FSexO10Mn ~ a3*MovieMonthSum

#indirect effects (IDE)

SRS8MeanIDE := a1*b1

SABWMnIDE := a2*b2

FSexO10MnIDE := a3*b3

sumIDE := (a1*b1) + (a2*b2) + (a3*b3)

#total effect

total := c + (a1*b1) + (a2*b2) + (a3*b3)

#model correlation between mediators

SRS8Mean ~~ SABWMn

SRS8Mean ~~ FSexO10Mn

SABWMn ~~ FSexO10Mn

"

LikeLike

]]>