Hello, I have data with three treatment categories dummy coded D1 and D2, a binary mediator (M2), and a continuous outcome (Y2). There is a pretest for the mediator (M1) and outcome (Y1). If I run these equations: Y2 = D1 D2 Y1 M2 ; M2 = D1 D2; with the logit link and use 8.68 from your book I get the results I get by using IND once for D1 and once for D2. If I change the second equation to M2 = D1 D2 M1; and include the coefficient for M1 in the application of 8.68, I do not get the results I get from using IND. Questions: Is 8.68 valid with a pretest for the mediator? Is IND valid with a pretest for the mediator? Thanks, Jamie
I think you want to center M1 when you include that as a covariate (i.e. control variable "C" in the book notation) and do the effects at M1 = 0. Equation (8.68) doesn't cover C. See how I discuss it for the vaccination example in Section 8.18 starting on page 323 of our Regression and Mediation book.