Model comparison with AIC based on different sample size

[peter a]

Hi all,

Let's assume I have two models Q1 and Q2:

Q1 and Q2 have different number of covariates 

Since one of the variable in model Q2 has some missing values, the sample size of Q2 is lower than sample size of Q1

missing is about 25%

Is it still admissible to use AIC to compare models based on different sample sizes?

[Giulio Flore]

Hi,

It depends. The AIC just measures how well the model fit the data. A model that uses more data (e.g. more information) is bound, by and large, to do always  better than a model which is applied to data with a lower information content (such as, missing data). 

The Bayesian Information Criterion might be more suitable as it adds a penalization in terms of the available observations. Thus a model that uses more data is equalised in some way in comparison to a model that uses less data.  That said, BIC is reliable if you have observations well in excess of the model parameters and you are not dealing with high dimensional problems.

There is also another criterion, the Deviance Information Criterion, that generalizes the AIC and BIC to hierarchical models.  

To sum up, I would think that BIC is a better benchmark to judge the relative merits of the two models.

Regards,

Giulio

--
--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.