ventilation

[Agnese Vardanega]

I've found this group, and hope that my questions will be of use to
the others.

A propos de ventilation, in the ACM procedure, I have two questions:

1) is it possible to exclude (ventiler) NA values, since they are
treated as independent columns? I have many filtered answers, so that
certain questions applies only to certain cases. Of course, they are
all the same cases, and ... the inertia is enormously overestimated. I
cannot use average values, of course, and the frequency is too high to
use the "level.ventil" option. I have tried to work on partial
analyses, but levels were not enough (I got an error message).

2) by the way, to exclude some other categories, I would use the
"ventil.level", but: what number should I input? a percentage? How
does it work with weighted data? where can I see the result of the
ventilation, i.e., the categories actually "ventilée"? (in the ACM
output, I cannot even find the poids effectifs)

Many thanks

Agnese

[François Husson]

Dear Agnese,

Missing values can be handled using the missMDA package that we have developed. This package is a complement to FactoMineR.
You can use the function imputeMCA for your problem with missing values in MCA.
A small website is dedicated to this package. It shows how to use the function: http://math.agrocampus-ouest.fr/infoglueDeliverLive/developpement/missMDA
It is better to use this function than to ventilate because the objective is really to skip the missing values.

If you use ventil.level, you have to give a percentage.
Best,
F. Husson

[Agnese Vardanega]

I thought missMDA was conceived to substitute a missing value with an
estimated value ... I will try,

Thanks

Agnese

On Nov 20, 8:46 pm, François Husson <francois.hus...@agrocampus-

ouest.fr> wrote:
> Dear Agnese,
>
> Missing values can be handled using the missMDA package that we have
> developed. This package is a complement to FactoMineR.
> You can use the function imputeMCA for your problem with missing values in
> MCA.
> A small website is dedicated to this package. It shows how to use the

> function:http://math.agrocampus-ouest.fr/infoglueDeliverLive/developpement/mis...

[François Husson]

In fact, the objective of missMDA is to perform PCA, MCA, or MFA skipping the missing values. To do so, the missing values are imputed in a way that they do not contribute to the construction of the dimensions. And the principal component methods is then performed on the completed dataset. But this imputation is just an intermediate step.

FH

[Agnese Vardanega]

Great!

I have to say that it is not very clear, neither in the help files,
nor in the website. Probably it is a problem of mine, depending on my
poor English and on my terrible French :) ... Can you suggest a paper
or a journal article about missMDA I can cite in my research paper?
About the treatment, not necessarily the package, I mean.

Another question. What about the results of ventilation (for the other
categories)? can I find these results somewhere in the res.mca? My
problem is that rows are weighted, and I would like to know what
happens to categories, one by one. Instead I cannot find the output
with "effectifs", and, in the var output, all categories seem to
contribute to the axes, whatever percentage I use (I've tried up to
50! :))) ).

Thanks again!
PS: when using (evidently too) many supplementary variables, dynGraph
gives me an error such as "the object is not a matrix". It seems a
bug.

On Nov 21, 8:29 am, François Husson <francois.hus...@agrocampus-

[François Husson]

You can find 2 references :

- Josse J, Pagès J & Husson F. (2011). Multiple imputation in principal component analysis. Advances in Data Analysis and Classification. 5, pp231-246.

- Josse, J., Chavent M, Liquet B Husson, F., & Pagès (2011). Handling Missing Values with Regularized Iterative Multiple Correspondence Analysis. To be published in Journal of Clcassification.

You cannot find the results for the ventilated categories. This output is not yet available.

For dynGraph, you should ask your question to S. Le: sebast...@agrocampus-ouest.fr

[Agnese Vardanega]

> You cannot find the results for the ventilated categories. This output is
> not yet available.

OK. What about the effectifs? More important: do excluded categories
appear in the $var$contrib output, also if they didn't actually
contribute? If not, as I suppose, I quite simply cannot use the
option ... :(

Thank you again!

[Agnese Vardanega]

Error:

Errore in svd(X) : infinite or missing values in 'x'
Inoltre: Warning messages:
1: In sqrt((nrow(don) * ncol(don))/apply(tab.disj.comp, 2, sum)) :
Si è prodotto un NaN
2: In sqrt(M) : Si è prodotto un NaN

:(

[Agnese Vardanega]

In a new analysis: I cannot exclude a category with level.ventil, using the following:

level.ventil = 2 (2%)

level.ventil = 2.0 (2%)

level.ventil = 11 (< or = 11)

level.ventil = 12 (< 12)

Would you pls. post an example of a working command-line?

Thanks

Agnese

[Sébastien Lê]

Hi Agnese,

level.ventil should take its value between 0 and 1.

Best,

S. Lê

Le 10 août 12 à 12:47, Agnese Vardanega a écrit :

_____________________________

Sébastien Lê

Associate Professor

Applied Mathematics Laboratory

Agrocampus Ouest, Rennes, France

Tel: +33 (0)2 23 48 58 81

Fax: +33 (0)2 23 48 58 71

Website: http://sebastien.ledien.free.fr

[Agnese Vardanega]

Thank you! as a proportion, thus, not as a percentage ... (now I have another problem, but I will be opening a new thread)

[Leu]

Hi

The use of missMDA is for me interesting too, because I've got al lot of missing (filtered, not plausible and no answer) values . I'm using FactoMineR and I tried to use the package and the lines from the references (tutorial and example), but I didn't succeed. Do I have to start in the R Console or can I write it in the Rcmdr too? And why can't "np" not be found?

Kind regards

Leu