[R] Re : PCA and automatic determination of the number of components

[Stéphane Dray]

ade4 has the 'testdim' function which implements a recent method for 
estimating the number of dimension for PCA on correlation matrix. Paper 
describing the approach is available at 
http://pbil.univ-lyon1.fr/members/dray/articles/SD805.php




William Revelle wrote:
>
> At 12:08 PM +0000 4/20/09, Jari Oksanen wrote:
>> justin bem <justin_bem <at> yahoo.fr> writes:
>>
>>>
>>>  See ade4 or mva package.
>>>   Justin BEM
>>>  BP 1917 Yaoundé
>>>
>> I guess the problem was not to find PCA (which is easy to find), but
>>  finding an automatic method of selecting ("determining" sounds like
>> that selection would be correct in some objective sense) numbers of
>> components to be retained. I thin neither ade4 nor mva give much support
>> here (in particular the latter which does not exist any more).
>>
>> The usual place to look at is multivariate task view:
>>
>> http://cran.r-project.org/web/views/Multivariate.html
>>
>> Under the heading "Projection methods" and there under
>> "Principal components" the taskview mentions packages
>> nFactors and paran that help in selecting the number
>> of components to retain.
>>
>> Are these Task Views really so invisible in R that people don't find
>> them? Usually they are the first place to look at when you need
>> something you don't have. In statistics, I mean. If they are invisible,
>> could they be made more visible?
>>
>> Cheers, Jari Oksanen
>>
>>>  ________________________________
>>>  De : nikolay12 <nikolay12 <at> gmail.com>
>>>  À : r-help <at> r-project.org
>>>  Envoyé le : Lundi, 20 Avril 2009, 4h37mn 41s
>>>  Objet : [R] PCA and automatic determination of the number of 
>>> components
>>>
>>>  Hi all,
>>>
>>>  I have relatively small dataset on which I would like to perform a 
>>> PCA. I am
>>>  interested about a package that would also combine a method for 
>>> determining
>>>  the number of components (I know there are plenty of approaches to 
>>> this
>>>  problem). Any suggestions about a package/function?
>>>
>>>  thanks,
>>>
>>>  Nick
>>
>> ___
>
> Henry Kaiser once commented that the "Solving the number of factors 
> problem is easy, I do it everyday before breakfast. But knowing the 
> right solution is harder"
>
> The psych package includes a number of ways to determine the number of 
> components.  Parallel analysis (comparing your solution to random 
> ones), Minimum Absolute Partial correlations, Very Simple Structure 
> are three of the better ways.  Try functions fa.parallel and VSS.
>
> Bill
>
>
>
>

-- 
Stéphane DRAY (dray at biomserv.univ-lyon1.fr )
Laboratoire BBE-CNRS-UMR-5558, Univ. C. Bernard - Lyon I
43, Bd du 11 Novembre 1918, 69622 Villeurbanne Cedex, France
Tel: 33 4 72 43 27 57       Fax: 33 4 72 43 13 88
http://pbil.univ-lyon1.fr/members/dray/