Hello, I ran AIC for some competing models I created. I get df and an AIC score from the AIC procedure. Can I use the models with the lowest AIC scores from this procedure to choose my 'best' models? If not, what else do I need to do (and know) and how can I do it in R to chose the 'best' models? Thank you kindly, Michael [[alternative HTML version deleted]]
Michael Just <mgjust <at> gmail.com> writes:> I ran AIC for some competing models I created. I get df and an AIC score > from the AIC procedure. Can I use the models with the lowest AIC scores from > this procedure to choose my 'best' models?Depends. You told us nothing. Have you tried to search the list for AIC?>If not, what else do I need to do > (and know) and how can I do it in R to chose the 'best' models?You could try stepAIC in package MASS. It is quite good in rejecting illegal cases, such as REML based lme. However, also check the field you are working in, and read other messages on this list on the subject. Model selection might be acceptable in sociology, but it is seen critical in formal medical studies. Dieter
Choosing the model with minimum AIC is just one consideration that might be used. If you look at books such as McQuarrie and Tsai (1998), Regression and Time Series Model Selection, World Scientific, you will find about 450 pages dealing mainly with the use of this and similar criteria to select appropriate models. You could also look at Konishi and Kitagawa (2008), Information Criteria and Statistical Modeling, Springer. You will then realise that there is no simple answer to your question. John Frain 2008/10/14 Michael Just <mgjust at gmail.com>:> Hello, > I ran AIC for some competing models I created. I get df and an AIC score > from the AIC procedure. Can I use the models with the lowest AIC scores from > this procedure to choose my 'best' models? If not, what else do I need to do > (and know) and how can I do it in R to chose the 'best' models? > > Thank you kindly, > Michael > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. >-- John C Frain Trinity College Dublin Dublin 2 Ireland www.tcd.ie/Economics/staff/frainj/home.html mailto:frainj at tcd.ie mailto:frainj at gmail.com
John, Thank you for those references. Cheers, Michael On Tue, Oct 14, 2008 at 8:27 AM, John C Frain <frainj@gmail.com> wrote:> Choosing the model with minimum AIC is just one consideration that > might be used. If you look at books such as McQuarrie and Tsai > (1998), Regression and Time Series Model Selection, World Scientific, > you will find about 450 pages dealing mainly with the use of this and > similar criteria to select appropriate models. You could also look > at Konishi and Kitagawa (2008), Information Criteria and Statistical > Modeling, Springer. You will then realise that there is no simple > answer to your question. > > John Frain > > 2008/10/14 Michael Just <mgjust@gmail.com>: > > Hello, > > I ran AIC for some competing models I created. I get df and an AIC score > > from the AIC procedure. Can I use the models with the lowest AIC scores > from > > this procedure to choose my 'best' models? If not, what else do I need to > do > > (and know) and how can I do it in R to chose the 'best' models? > > > > Thank you kindly, > > Michael > > > > [[alternative HTML version deleted]] > > > > ______________________________________________ > > R-help@r-project.org mailing list > > https://stat.ethz.ch/mailman/listinfo/r-help > > PLEASE do read the posting guide > http://www.R-project.org/posting-guide.html > > and provide commented, minimal, self-contained, reproducible code. > > > > > > -- > John C Frain > Trinity College Dublin > Dublin 2 > Ireland > www.tcd.ie/Economics/staff/frainj/home.html > mailto:frainj@tcd.ie > mailto:frainj@gmail.com >[[alternative HTML version deleted]]