Mediation & Bootstrapping with DWLS estimator

[T S]

Hi everyone,

I was performing a SEM analysis in lavaan that included an ordered dependent variable and was looking for a way to conduct bootstrapping for my mediation analysis. After doing some research, I ended up using the following command:

fitboot <- sem(SEMmodel, data=data, se="bootstrap", test="scaled.shifted",
           verbose=TRUE)

The results after bootstrapping look just fine and interpretable. However, I read in some papers that they also used mediation to determine the 95% confidence intervals and I was wondering how I could do that in my case, so I can report it in my paper.

I have two questions now:

1. Due to the ordered variable in the model, the DWLS estimator was used by lavaan. When I implement bootstrapping with the command above, the mediation results differ just slightly from the ones I obtain after estimating the regular model with the DWLS estimator. Question: Is the bootstrapping procedure necessary / recommended when having already calculated indirect effects manually with the DWLS estimator?

2. How can I adjust the command above to make lavaan tell me the 95% confidence interval for the bootstrapping procedure?

Can anybody help me out here?

Thank you very much and kind regards,

Tobias Scholz

[Mauricio Garnier-Villarreal]

Tobias

You can use the function

parameterEstimates( fitboot , boot.ci.type="bca.simple" , ci=T, level=.95)

[T S]

Dear Mauricio,

thank you for the advice, worked perfectly.

Kind regards,

Tobias