Need help: reg EQS and Goodness of Fit tests

#1
Hi! Hope you guys could help me out here... I'm doing a path analysis on EQS, and the goodness of fit test results i got are the ff - i'm not sure if these are valid:
Chi-square = .001, p.97 (df = 1)
CFI = 1.00
RMSEA = 0.00

I'm not sure if this could be called a good fit, a perfect fit, or if this is at all valid as I've never encountered this before.
 
#2
I have the feeling that the model is saturated, hence, the perfect fit. Most of the time, in path analysis if you estimate all the possible paths, the models gets just-identified, or saturated, giving you perfect CFI, TLI, RMSEA SRMR. Chi-square is not affected but is highly sensitive to sample size. You should liberate one parameter to have real fit indicators.
 
#3
thanks for the reply, really appreciate it! but how can I then loosen my parameters? I don't know how this is done. I actually only have 3 variables. The model basically looks like this: Var1 --> Var2 --> Var3

:-S
 
Last edited:
#4
mmm... a path mediation model with 3 variables will be always saturated... you should consider adding another variable to the model, or move to SEM (Latent variables). What's your sample size??
 
#6
mmm... the sample size is hardly enough but you should try anyway. :) May be you can help me with my question!! is about 5 or 6 treads below yours, and is about power analysis XD
 
#7
looks like i'm stuck with a saturated model. I've tried SEM and adding other variables, and other permutations with my data but none appears to work. but thanks so much for the help! really appreciate it :) would you also have any idea on how to report it?

sorry i could't be of help to you regarding power analysis - i'm not familiar with it :-S good luck though!
 
#8
well... saturated model estimation is not wrong, so can report the betas, and the fit is not "wrong" either is just "saturated", meaning that can perfectly replicate the real variance-co variance matrix of the data. it's not a sin, so just report it the way it is.