Dear Linda or Bengt, we ran tests of mediation, I reported confidence intervals and signficance levels of the indirect effects. Journal is asking that we use MacKinnon et al.(2002) joint significance test. Is the MPlus significance of the indirect effect and the standardized indirect effect a joint significance test in the alpha times beta mediation model? Thank you in advance for your response.
Thank you Bengt, the following response was provided by Dave MacKinnon, I was asking on whether the default indirect tests or the asymetrical bootstrapped standard errors were equivalent to the joint significance test referred to in their 2002 paper: "The default Mplus indirect tests are based on the first order multivariate delta standard error (Sobel test). That assumes that the ratio of indirect effect to standard error has a normal distribution but it doesn't. That test has lower power and less accurate Type I error rates (MacKinnon et al., 2002 2004). Mplus also has bootstrap indirect effects which do accomodate the nonnormal distribution of the product and do have higher power and nominal Type I error rates. You could also use the distribution of the product itself, available online at my website with the PRODCLIN program. Testing mediation with the distribution of the product is almost identical to the joint significance test but you get confidence intervals which are helpful. The joint significance test just tests each link in the mediation chain--a reasonable test but it does not provide confidence intervals. The 2002 reference is the good one for the joint significance test. These tests are described in Chapter 3s and 4 (Chapter 11 for bootstrapping) in my book on mediation analysis."
Hi Linda, I was given the following syntax for MacKinnon's joint significance test using the MPlus model constraint command, but I want to make sure this is accurate (I notice if I switch the order of alpha and beta in the "new" command, I get different significance levels):
x = IV; y = DV; m = mediator
y ON m (alpha); m ON x (beta);
Model Constraint: new (newmed); newmed = alpha beta;