Are there alternatives to bootstrapping in mediation analysis (SEM)?

[Anna Warren]

Hello,

I am running a mediation analysis with two parallel mediators which are both latent variables. I am attempting to run 1000 bootstraps on the mediation to get bootstrapped confidence intervals for the direct, indirect and total effects. However, this bootstrapping has already been running for several weeks and looks like it will take several more.

I've read that 1000 bootstraps should be run as a minimum so I'm not sure reducing the number of bootstraps is an option.

I wondered whether there are any alternatives to bootstrapping which may run faster? E.g. is it acceptable to use robust standard errors instead?

Thanks,
Anna

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

If anything takes weeks, I'd first assume that you have a model problem. Either that or you're using a very old machine with very limited RAM. Is your model really taking over an hour to run a single estimation without bootstrap? I don't know if this works in SEM, but in many maximum likelihood estimates you can speed things up by giving the program better start values.

I'm not expert enough in SEM mediation estimates to answer your question about robust se's.