

Mediation. Single item measure 

Message/Author 

Inma posted on Friday, November 21, 2014  5:45 am



I want to run a mediation model where the mediating variable is a singleitem measure. X (2 dimensions)> Z (singleitem measure)> Y (2 items) żIS IT CORRECT TO DO THAT? ANALYSIS: TYPE = GENERAL; ESTIMATOR = ML; ITERATIONS = 10000; CONVERGENCE = 0.00005; MODEL: X BY x1 x2; Y BY y1 y2; Y ON Z X; Z ON X; MODEL INDIRECT: Y IND Z X; 


That's ok. 

Inma posted on Friday, November 21, 2014  8:21 am



 I have also run the "reversed causal model" as a competing model where: The mediating variable act as an outcome The outcome is tested as mediating  Chi sq, RMSEA, CFI, SRMR, AIC and BIC were exactly the same! The only difference were the estimate for each parameter 1. Is that possible? 2. Should I choose then between these competing models based only on theoretical arguments? 


1. The model test concerns restrictions on the covariance matrix, which are the same in your two scenarios. Statistics can't always tell you what's right. 2. Yes. 

Back to top 

