|
|
Choosing the best fitted model |
|
Message/Author |
|
Oana Lup posted on Sunday, June 26, 2011 - 12:16 pm
|
|
|
Hello, I have a question about choosing the best fitted model. i am testing for causal relationships between two variables. i am estimating cross-lagged and synchronous models with unidirectional and reciprocal effects. i have six models. when i use chi-2 test, nested models are preferred. if i look at other measures of model fit though (AIC, BIC, adjusted BIC, RMSEA), one of the larger model is preferred. how is this possible and how can i choose. thanks very much, Oana |
|
|
The only way to statistically test two nested models is a chi-square difference test. Other measures cannot provide statistical tests just comparisons. |
|
Oana Lup posted on Tuesday, June 28, 2011 - 1:24 pm
|
|
|
thanks very much. my question is which of these models give a better description of the relationships between my variables. can i then use AIC, BIC, adjusted BIC, and RMSEA to compare the models and decide? On the other hand, I still don't understand how comes model 2 fits significantly better than model 1 when i do a chi-square test difference but when i compare them based on AIC, BIC, and the rest, model 1 has a better fit than model 2. thanks, Oana |
|
|
Fit statistics do not always agree. I would probably rely on a chi-square difference test when it is available. |
|
Back to top |
|
|