I have a path model with three exogenous variables (A, B, and C) directly influencing variable D, which then influences E. (So D is my mediator between the exogenous variables and E.) The model has moderately good fit. If I remove A, B, or C, the RMSEA increases by .02. How do I interpret this?
Hi Stephanie,
So the full model has effects
a -> d
b -> d d -> e
c -> d
and the restricted model has only
d -> e.
Do you do this by completely removing the variables a to c from analysis OR do you fix their effect to zero? If the latter, then this would tell you, if the effects not important in the model. In the former case you are assuming that the means and variances of a-c are zero.
Without knowing anything else about the model and data (e.g. do the models have the same sample size?), it would seem to suggest that slightly less variability is left unexplained with the larger model.
Regards,
Timo
Thanks, Timo! I'm actually looking at the restricted model of having only one exogenous variable removed. All models have equal sample size.
I was afraid my finding alluded to some systematic error in the model. I've been sent a private message that says this finding just means that the model fits better with all three exogenous variables with no systematic error implied. I feel better!
Hello Stephanie,
it appears that models with low df and small N tend to inflate RMSEA values.
RMSEA = sqrt (chi^2 - df) / sqrt (df * (N-1)) , thus removing a path from the model will alter the df, and consequently, the RMSEA.
See:
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1, 130-149.
good luck,
Angel
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling