

Mediation with a nominal control vari... 

Message/Author 

Inma posted on Thursday, May 14, 2015  7:39 am



My model is: X(continuous)>M(continuous)>Y(continuous) And my control variable is nominal: I built 2 (C1)dummy variables My questions are: 1. Would it be appropriate to use Maximun Likelihood Method despite the dummy variables (C1 and C2)? 2. Are the following instructions ok? MODEL: Y on M C1 C2; M on X C1 C2; MODEL INDIRECT: Y IND M X; 3. I have not specify anywhere that C1 and C2 are dummy variables. Should I? 4. My guess is that statistics for model estimation would be covariance matrix. Right? 


1. Yes, ML assumes normality only for the DV residuals. 2. Yes, or just Y IND X; 3 No, you only declare DV status. 4. Raw data is better. 

Inma posted on Friday, May 15, 2015  8:11 am



Thanks a lot! 1) If I want to check if the model fits equally well in the three groups (instead of controling "company" using dummies): Would it be ok: USEVARIABLES = X M Y company; grouping are company (1 = cI 2 = cII 3 = cIII); ANALYSIS: TYPE = GENERAL; ESTIMATOR = ML; ITERATIONS = 10000; CONVERGENCE = 0.00005; MODEL: Y ON X; Y ON M; M ON X; MODEL INDIRECT: Y IND X; MODEL cI: Y ON X (c1); Y ON M (b1); M ON X (a1); MODEL cII: Y ON X (c2); Y ON M (b2); M ON X (a2); MODEL cIII: Y ON X (c3); Y ON M (b3); M ON X (a3); MODEL constraint: new (f g h); f=a1*b1; g=a2*b2; h=a3*b3; 0=fg; 0=fh; 0=gh; OUTPUT: MOD STAND; 2) If I do this and the fit indexes (RMSEA, ...) are ok: Might I conclude that the model fits equally well in the three samples? 


1. Looks ok. 2. Yes. 

Back to top 

