- hello mail list

I have a categorical variable that describe rock type and I'm

interesting in quantify the relation and discriminant capacity of

continuous variables (grade elements) over each category.

The problem is that I have no normality over the domain and the classes,

the variance and covariances are not the same, and the categories are

not independent, that limit for example the Discriminant analysis, some

alternative (example logistic regression )?

a) what to use as statistical method?

b) is possible to use discriminant analysis?

Regards

Adrian Martínez - Re : Use of discriminant analysis: You might try quadratic

disciminant analysis, wherein differences among variances and

covariances are taken into account. You don't describe how far the

class distributions are from being normally distributed - but in some

cases data transformations (e.g. applying the log transformation to

skewed data) will make the data normally distributed or almost so.

In any case, it's a good idea to get a measure of performance -

splitting data with class identifications into training (for

computing discriminants) and test (for computing accuracy of

discriminat anaoysis) subsets if you have sufficient data, or

jacknifing in the case of fewer data.

>hello mail list

--

>

>I have a categorical variable that describe rock type and I'm

>interesting in quantify the relation and discriminant capacity of

>continuous variables (grade elements) over each category.

>

>The problem is that I have no normality over the domain and the classes,

>the variance and covariances are not the same, and the categories are

>not independent, that limit for example the Discriminant analysis, some

>alternative (example logistic regression )?

>

>a) what to use as statistical method?

>b) is possible to use discriminant analysis?

>

>Regards

>Adrian Martínez

>

>

>

>* By using the ai-geostats mailing list you agree to follow its rules

>( see http://www.ai-geostats.org/help_ai-geostats.htm )

>

>* To unsubscribe to ai-geostats, send the following in the subject

>or in the body (plain text format) of an email message to

>sympa@...

>

>Signoff ai-geostats

***************************************

Chris Hlavka

NASA/Ames Research Center 242-4

Moffett Field, CA 94035-1000

(650)604-3328 FAX 604-4680

Christine.A.Hlavka@...

*************************************** - Hi, Martinez,

It seems to me that you are analysing coded categorical variables (eg. soft

= 1, coarse = 2, hard = 3 etc.). I also assume that your coding is not

reflecting any inherent characteristics of your rocks or your objects of

examination.

To be able to answer your question I would like to know the nature of your

response variable (usually called regressand or predictand etc.). Especially

I would like to know if it is a continuous variable or a categorical

variable. If it is a categorical variable, how many levels do you have for

it? Bionomial or multinomial? Ordered or nomial?

If you have bionomial dependent variable you can use logistic regression or

binary probit. Even though their distribution assumptions are different, the

final conclusion will be the same. If you have ordered variable, you can use

ologit (ordered logit) or oprobit (ordered probit). If you have multi-level

nominal variable, you can use multinomial regression analysis.

You can use discriminat analysis, but I would rather go for logist or probit

methods as they are more advanced and any way will answer your quesions.

If you need more help, pleas contact me.

I hope this helps a little bit.

Cheers

Mahdi

> hello mail list

--

>

> I have a categorical variable that describe rock type and I'm

> interesting in quantify the relation and discriminant capacity of

> continuous variables (grade elements) over each category.

>

> The problem is that I have no normality over the domain and the classes,

> the variance and covariances are not the same, and the categories are

> not independent, that limit for example the Discriminant analysis, some

> alternative (example logistic regression )?

>

> a) what to use as statistical method?

> b) is possible to use discriminant analysis?

>

> Regards

> Adrian Martínez

>

>

>

>

-----------------------------------

Mahdi Osman (PhD)

E-mail: m_osm@...

-----------------------------------

DSL Komplett von GMX +++ Supergünstig und stressfrei einsteigen!

AKTION "Kein Einrichtungspreis" nutzen: http://www.gmx.net/de/go/dsl - Hi all:

Sorry to disturb with a different question, I was wondering if anyone could

point out some good references on multilevel modeling within a spatial

framework, besides the work of the folks creators of Mlwin.

Thanks in advance,

Gaston Pezzuchi

Mensaje citado por MARTINEZ VARGAS Adrian <martinez@...>:

> hello mail list

-------------------------------------------------

>

> I have a categorical variable that describe rock type and I'm

> interesting in quantify the relation and discriminant capacity of

> continuous variables (grade elements) over each category.

>

> The problem is that I have no normality over the domain and the classes,

> the variance and covariances are not the same, and the categories are

> not independent, that limit for example the Discriminant analysis, some

> alternative (example logistic regression )?

>

> a) what to use as statistical method?

> b) is possible to use discriminant analysis?

>

> Regards

> Adrian Martínez

>

>

>

>

Internet Way - Webmail - http://www.way.com.ar