Hello.
I have tried fitting a path analysis model in AMOS and Lavann. In some model fit indices the results are very similar, but in AIC and BIC the results are very distant. Why can this be?
LAVVAN
Test statistic 4,485
Degrees of freedom 2
P-value (Chi-square) 0,106
AMOS
Test statistic 4,466
Degrees of freedom 2
P-value (Chi-square) 0,107
LAVAAN
Comparative Fit Index (CFI) 0,992
AMOS
Comparative Fit Index (CFI) 0,995
LAVAAN
RMSEA 0,072
AMOS
RMSEA 0,072
LAVAAN
Akaike (AIC) 3743,234
AMOS
Akaike (AIC) 42,466
LAVANN
Bayesian (BIC) 3788,536
AMOS
Bayesian (BIC) 108,677
AIC and Bic aren't goodness of fit statistics but rather statistics to compare different prediction models so that one can decide which prediction model is better under what circumstances. If you are using OLS AIC is often called Mallows Cp statistic. See the attachment for more details.. Best wishes David Booth
Different programs for structural equation modeling (SEM) use different formulas for computing chi-square, AIC, BIC, and other fit statistics. For chi-square, some programs use N in the denominator, whereas others use N - 1. In larger samples, this usually makes only a small and not practically meaningful difference. I'm assuming this probably explains the very slight differences in chi-square values in your example.
For AIC and BIC, the differences in the computations (formulas) used by different programs are larger as shown by your example. The rank order of models (i.e., for which model AIC or BIC are smaller vs. larger when comparing different models) should be the same though regardless of the computational approach.
In any case, I would check the program manuals. They should show you which formulas are used by either program to compute chi-square, AIC, and BIC.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA