AIC and BIC values

#1
Hi all,
I am running two models which both fit well. Thus, I am looking at the AIC and BIC values. I ran my analyses through Mplus and my AIC and BIC for Model 1 are 5297.428 and 5346.977 respectively. They appear to be higher for Model 2 at 5386.379 and 5435.929. Would I report each of these values in a paper (they seem quite large compared to what I have seen?) or would I report the difference (approx 80ish)? Furthermore, based on these values and existing theories, are these values different enough to be able to say that Model 1 is a stronger and more parsimonious fit compared to Model 2?

Thanks in advance!
 

noetsi

Fortran must die
#2
I think you would simply say which model had the lower scores, I doubt many will care what the specific values are. There is no clear rule as far as I know on how much a difference between AIC really matters. Note choosing based purely on AIC might not be ideal depending on what type of model you are running. Including variables that are not significant simply because the model has a lower AIC if you do this is at least questionable so you might want to explain why you did this.

With time series the norm is to use AIC (or AICC for small samples) and a hold out data set.
 

noetsi

Fortran must die
#4
I have never been able to understand what a small sample is (the common response is "it depends"). My guess is 201 is not a small sample (given that by 30 z and t distributions are close and the central limit theorem kicks in, plus what samples I have seen in journals) but that is a wild guess.