Nobody seems to have suggested the lasso approach which could
(or could not) be usefull. Search CRAN for "lasso", there are
multipler packages.
Kjetil
On Tue, Jun 15, 2010 at 10:26 AM, David Wooff <[log in to unmask]> wrote:
> I would read Chapter 15 of Draper & Smith, Applied Regression Analysis (3rd
> ed), New York: Wiley (1998). Having done that, I'd go back and re-read
> Chapter 11. Next, I'd read the contrary views expressed in Chapter 4.3 of
> Harrell, Regression Modelling Strategies, New York:Springer (2001) [which,
> inter alia, expresses doubts about the bootstrapping suggested by Dorothy].
> Having read those, you should be in a position to judge for yourself what
> might be appropriate.
>
> If your number of candidate variables is large in relation to your number of
> observations, you are pretty much walking blindfold in a minefield.
>
> Cheers,
>
> David
>
> On 15/06/2010 14:57, Dorothy Middleton wrote:
>>
>> Some of the variables are spurious. Strongly urge you to use the
>> bootstrap to assess which variables are genuine. Or, for that matter, to
>> use a decision tree approach in the first place.
>>
>> -------- Original Message --------
>> Subject: [SPAM] Variable selection in multiple regression
>> From: Noori Akhtar-Danesh<[log in to unmask]>
>> Date: Mon, June 14, 2010 1:20 pm
>> To: [log in to unmask]
>>
>> Dear List members,
>>
>> I would appreciate if I could be directed to a reference (journal
>> article or book) for the following approach of variable selection in
>> multiple regression. This approach seems to be quite common in some area
>> of research but I have not seen any reference for it.
>>
>>
>>
>> Approach: When there are many explanatory variables, first each
>> explanatory variable is regressed individually against the outcome
>> (dependent) variable. Then, for each variable, if the p-value is, say
>> <=.20, it is chosen to be included in a multiple regression model (this
>> stage may be called the screening phase). Next, these selected variables
>> are used in a multiple regression approach to come up with a final model
>> (the multiple regression can be, I guess, conducted using a backward,
>> forward, or stepwise approach).
>>
>> Many thanks in advance,
>>
>> Noori
>>
>> ============================
>> Noori Akhtar-Danesh, PhD
>> Faculty of Health Sciences,
>> McMaster University,
>> 1200 Main St. West, Room 3N28B
>> Hamilton, ON L8N 3Z5,CANADA
>> Tel: 905-525-9140 Ext. 22297& 22725
>> Fax: 905-521-8834
>> http://www.fhs.mcmaster.ca/ceb/faculty_member_akhtar-danesh.htm
>> =============================
>>
>> You may leave the list at any time by sending the command
>>
>> SIGNOFF allstat
>>
>> to [log in to unmask], leaving the subject line blank.
>>
>> You may leave the list at any time by sending the command
>>
>> SIGNOFF allstat
>>
>> to [log in to unmask], leaving the subject line blank.
>
> --
> David Wooff,
> Director, Statistics and Mathematics Consultancy Unit,
> & Senior Lecturer in Statistics, University of Durham.
> Department of Mathematical Sciences, Science Laboratories,
> South Road, Durham, DH1 3LE, UK. email: [log in to unmask]
> Tel. 0191 334 3121, Fax 0191 334 3051.
> Web: http://maths.dur.ac.uk/stats/people/daw/daw.html
>
> You may leave the list at any time by sending the command
>
> SIGNOFF allstat
>
> to [log in to unmask], leaving the subject line blank.
>
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|