|
|
BIC vs. Sample-size adjusted BIC for ... |
|
Message/Author |
|
cheng hong posted on Sunday, February 21, 2016 - 12:48 pm
|
|
|
I'm comparing a model with only main effects with a model with both main effects and two interactions. Since mPlus doesn't give me model fit information for interaction model except AIC and BIC, I compare these two models based on AIC and BIC. But my problem is: for AIC, the moderation model fits better; for BIC, it fits worse. However, the result of sample-size adjusted BIC is consistent with AIC, indicating the model with interactions is better. In this case, can I go by the SA BIC? My sample size is 482 and there are significant interactions. |
|
|
You can use compare the models using a likelihood-ratio chi-2 difference test. All you need are the loglikelihoods. |
|
cheng hong posted on Tuesday, February 23, 2016 - 6:45 pm
|
|
|
Thank you so much for your response!! I tried to use likelihood-ratio, but the problem is I couldn't find scaling correction factor for my model with only main effects. Below are my results, can you take a look at it? Model with both main effects and interactions: Number of Free Parameters 138 Loglikelihood H0 Value= -24946.434 H0 Scaling Correction Factor = 1.721 for MLR Information Criteria Akaike (AIC)= 50168.868 Bayesian (BIC)= 50745.424 Sample-Size Adjusted BIC= 50307.424 (n* = (n + 2) / 24) Model with only main effects: Number of Free Parameters = 132 Loglikelihood H0 Value=-24957.359 H1 Value=-23840.497 Information Criteria Akaike (AIC)= 50178.718 Bayesian (BIC)= 50730.207 Sample-Size Adjusted BIC=50311.250 (n* = (n + 2) / 24) Thank you so much!!! |
|
|
Please send the two outputs to Support along with your license number. |
|
Back to top |
|
|