mediation analysis, robust method, and bootstrap

WangYH

Hi @EJ @vandenman @elisa

I did a mediation analysis in JASP using robust method. I don't know whether this method calculates confidence interval using boostrap technique. And I don' t understand the difference between robust method and standard method, because both of two methods gave different p value (figures below show the results). I guessed that the difference maybe, is the calculation of standard error.

best wishes,

Yunhong

patc3

Yes the difference is how the standard errors are calculated, and no it's not using bootstrap unless you check the box

WangYH

Thank you @patc3

I also have the question that how do I choose the method between the standard and robust. Moreover, when I check the bootstroop box, it seemly processed under the standard method. If I want a confidence interval using bootstrap under robust method, how do I choose the next step?

best wishes,

Yunhong

patc3

There's no robust vs non robust bootstrap; the choice is between 1) standard SEs, 2) robust SEs, and 3) bootstrap SEs. What bootstrap does is it randomly picks observations from your sample, runs the model, saves the parameter estimates, and repeats this process thousands of times and then calculates the standard deviation of the parameter estimates across all iterations as an estimate of the standard error.

I don't know what guides the choice between standard and robust standard errors--if you have a small sample or variables that are not normally distributed, you could consider using robust (some software like Mplus defaults to robust SEs) or bootstrap SEs.

WangYH

Thank you for your explanation !! @patc3 I got the point. I would also like to inquire if robust estimation is to be used when the predictor, mediator and outcome all do not satisfy a normal distribution, or if it can be used only when the outcome does not satisfy a normal distribution?