correspondence analysis and logistic regression

[butasbutauskas]

Hi,

I know that if you want to reduce variable size you have to use for instance principal component analysis. In addition to this, this method could eliminate multicollinearity problem in regression analysis.


In my research I have a lot of variables (about 30) of categorical data , which I want to reduce (therefore I am thinking to use correspondence analysis), and then use these results to predict categorical outcome (i.e. through logistic regression).

The questions would be:
1. Is it possible to get scores from correspondence analysis (similar as in Principal Component analysis) and to use it in logistic regression?
2. Is Correspondence analysis could eliminate multicollinearity problem?

Thanks in advance:)

Regard

[Art Kendall]

CatPCA and CatReg in SPSS might fit your situation.

Are your variables strictly nominal? or dichotomous, or merely ordered,
or equal-appearing?
Are there subsets of the independent variables designed to measure
pretty much the same thing? E.g., are they attitude questions?


If you describe your data in more detail list members would be better
able to give their reactions.

Art Kendall
Social Research Consultants



On 2/2/2011 5:02 AM, butasbutauskas wrote:

> Hi,
>
> I know that if you want to reduce variable size you have to use for instance
> principal component analysis. In addition to this, this method could
> eliminate multicollinearity problem in regression analysis.
>
>
> In my research I have a lot of variables (about 30) of categorical data,
> which I want to reduce, and then use these results to predict categorical
> outcome (i.e. through logistic regression).
>
> The questions would be:
> 1. Is it possible to get scores from correspondence analysis (similar as in
> Principal Component analysis) and to use it in logistic regression?
> 2. Is Correspondence analysis could eliminate multicollinearity problem?
>
> Thanks in advance:)
>
> Regard
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/correspondence-analysis-and-logistic-regression-tp3367586p3367586.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[butasbutauskas]

OK:), little details about my data:

I am studying archaeological artifacts from two big cemeteries, that is from archaeological sites I have about 40 types of artifacts in each of cemetery (knifes, axes, various ornaments, work tools and so on). Therefore, variables are nominal. I would like to count the number of artifacts in my analysis. On the other hand, it would also be useful to make analysis with only presence/absence (dichotomous) of particular artifact.

I want to make general complex of archaeological artifacts (to reduce 40 type of artifacts into smaller sets, which will strongly correlate with each other).

Next, I would like to use logistic regression in order to find out the probability that particular artifact complex (that should be calculated from the above mentioned operation) is characteristic to particular age (or social) class.

Could you suggest, what could be optimal scaling technique (that could reduce huge amount of artifact type number) for this task?

[Bruce Weaver]

Regarding the use of principal components analysis (PCA) to reduce the number of variables or deal with multicollinearity, you may find this article interesting:

   http://www.questia.com/googleScholar.qst?docId=5001333495

You can get it via JSTOR if you have institutional access.

HTH.

butasbutauskas wrote

Hi,

I know that if you want to reduce variable size you have to use for instance principal component analysis. In addition to this, this method could eliminate multicollinearity problem in regression analysis.


In my research I have a lot of variables (about 30) of categorical data , which I want to reduce (therefore I am thinking to use correspondence analysis), and then use these results to predict categorical outcome (i.e. through logistic regression).

The questions would be:
1. Is it possible to get scores from correspondence analysis (similar as in Principal Component analysis) and to use it in logistic regression?
2. Is Correspondence analysis could eliminate multicollinearity problem?

Thanks in advance:)

Regard

[Jon K Peck]

Another approach would be to use Partial Least Squares.  It has the advantage over principal components of taking the dependent variable into account when picking the components.  It is available in SPSS Statistics via the PLS extension command available from the SPSS Community.  Installation instructions are in the download.  Installation is a bit of a pain, so be sure to read them.

HTH,

Jon Peck
Senior Software Engineer, IBM
[hidden email]
312-651-3435



From:        Bruce Weaver <[hidden email]>
To:        [hidden email]
Date:        02/02/2011 10:16 AM
Subject:        Re: [SPSSX-L] correspondence analysis and logistic regression
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Regarding the use of principal components analysis (PCA) to reduce the number
of variables or deal with multicollinearity, you may find this article
interesting:

  http://www.questia.com/googleScholar.qst?docId=5001333495

You can get it via JSTOR if you have institutional access.

HTH.



butasbutauskas wrote:
>
> Hi,
>
> I know that if you want to reduce variable size you have to use for
> instance principal component analysis. In addition to this, this method
> could eliminate multicollinearity problem in regression analysis.
>
>
> In my research I have a lot of variables (about 30) of categorical data ,
> which I want to reduce (therefore I am thinking to use correspondence
> analysis), and then use these results to predict categorical outcome (i.e.
> through logistic regression).
>
> The questions would be:
> 1. Is it possible to get scores from correspondence analysis (similar as
> in Principal Component analysis) and to use it in logistic regression?
> 2. Is Correspondence analysis could eliminate multicollinearity problem?
>
> Thanks in advance:)
>
> Regard
>


-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/correspondence-analysis-and-logistic-regression-tp3367586p3368120.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD