AW: [SPSSX-L] PLS/PLSR

[Heiko Klemm]

Thanks for the explanation. Actually I´m trying to predict the uppermost instance in a hierarchical model. But I have trouble doing this prediction because of multicollinearity and a lot of missings in the data. As a side effect i´m trying to forecast all other instances in my model as well.

The users of my model should then be in the position to estimate the outcome of the prediction (in terms of the variable that is at the top of the hierarchy) regarding a let`s say 10% increase of the mean on a specific variable.

If you or any of the listers have a better approach, I would be glad for assistance

Kind regards,

Heiko Klemm
MRSC

Deutsche Post World Net
Business  Consulting GmbH
Market Research Service Center
Tulpenfeld 2
 53113 Bonn
Germany

T. [phone]
mailto:[hidden email]


-----Ursprüngliche Nachricht-----
Von: Peck, Jon [mailto:[hidden email]]
Gesendet: Montag, 7. August 2006 15:01
An: Klemm, Heiko BC/MRSC
Betreff: RE: [SPSSX-L] PLS/PLSR

I am not a PLS expert, by any means, but the difference between PLS-1 and PLS-2 is that PLS-2 results in one set of scores and one set of eigenvalues while in PLS-1, you get separate scores and eigenvalues for each Y variable.  PLS-1 can take significantly longer to calculate, but it will tend to give more accurate predictions.  For a single variable, then, there is actually no difference.

I am glad that you are enjoying Python.  It seems to be addictive.

-Jon

-----Original Message-----
From: [hidden email] [mailto:[hidden email]]
Sent: Montag, 7. August 2006 07:35
To: Peck, Jon
Subject: AW: [SPSSX-L] PLS/PLSR

Dear Jon,

I`m not quite sure since I don´t have a great deal of experience with pls, but to put it this way, i think pls-1 should bet be appropriate procedure to make a prediction for on single variable (please correct me if i`m wrong!!!).

Thanx again for explaining python to us in munich, I really enjoy exploring it (especially now, where it is running on the german systems ;-)



Heiko Klemm
MRSC

Deutsche Post World Net
Business  Consulting GmbH
Market Research Service Center
Tulpenfeld 2
 53113 Bonn
Germany

T. [phone]
mailto:[hidden email]


-----Ursprüngliche Nachricht-----
Von: Peck, Jon [mailto:[hidden email]]
Gesendet: Montag, 7. August 2006 13:40
An: Klemm, Heiko BC/MRSC
Betreff: RE: [SPSSX-L] PLS/PLSR

It is interesting that you should ask about this.  I am working right now on PLS using Python and scipy.

Are you interested in PLS-1 or PLS-2?

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Heiko Klemm
Sent: Montag, 7. August 2006 03:26
To: [hidden email]
Subject: [SPSSX-L] PLS/PLSR

Dear Listers,



I wonder if there is a way to compute "partial least square regression"
(pls / plsr) in spss, or if anybody knows of another spss procedure that
allows to develop prediction models if there is multicollinearity in the
data.



Kind regards



Heiko Klemm

MRSC



Deutsche Post World Net

Business  Consulting GmbH

Market Research Service Center

Tulpenfeld 2

 53113 Bonn

Germany



T. [phone]

mailto:[hidden email] <mailto:[hidden email]>