Hi Statalist,

I have a problem about how to use bootstrap test to further detect a mediation process as follows.

My regressions were conducted with GLS (xtgls) because the panel data is of unbalanced time series. The sample size is smaller than 400. There is one IV, one DV, one mediator, one moderator, and several controlled variables with year and firm fixed effects. The regressions include: H(1) IV has an inverted U shape relationship with DV (coefficient c). H(2) IV positively relates with the mediator (coefficient a). H(3) The mediator has an inverted U shape relationship with DV (coefficient b), while IV is held constant (coefficient c'). H(4) A moderator negatively impacts on the positive relationship between IV and mediator.

Now the GLS regressions have generated outcomes that support H(2), H(3), and H(4). According to literature of mediating effect in recent years, the above results can also be regarded as a mediating process.




I want to use bootstrap test to see whether the indirect effect (a*b) of the mediation process is significant according to bias corrected percentile bootstrap CI method (i.e., whether the 95% confidence intervals in the bootstrap test has 0) and then to detect the sampling distribution of the mediated effect with Bootstrap tests. How to solve these problems?

Many thanks!

Elaine