regression help

[Kizzy M Parks]

Hello:

Using regression, I need to analyze continuous and categorical IVs, but my
DV is continuous.  Any suggestions on the method I should use?  I have a
sample of over 400,000.


Kizzy M. Parks, ABD
Florida Tech

[Noble, Lyndsay]

Kizzy,
I'm not sure what you mean by "method".  My initial advice would be to use
regular, plain, old, regression with your categorical variables dummy coded
(or effect coded, same thing).  That is, if your categorical variable has n
levels, you would create n-1 variables with values of 0 and 1 (the nth level
of the initial variable would be indicated by having zeros in all n-1 new
variables).  Then, I would enter all of the variables into the regression
model and go from there.

If what you mean by "method" is stepwise, forward, backward, or some other
selection algorithm, I have no opinion.  Each has its pros and cons and I
will defer to the rest of the list.

HTH,
Lyndsay



-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Kizzy M Parks
Sent: Tuesday, July 10, 2007 8:55 AM
To: [hidden email]
Subject: regression help

Hello:

Using regression, I need to analyze continuous and categorical IVs, but my
DV is continuous.  Any suggestions on the method I should use?  I have a
sample of over 400,000.


Kizzy M. Parks, ABD
Florida Tech

[Mark A Davenport MADAVENP]

No, no, NO! No stepwise.  There is little reason to let the computer organize your variables.  You should dictate how and when the variables are loaded into the equation based on prior findings, lit review, prevailing theory, logic, temporal contiguity, almost ANYTHING but stepwise selection. See

 

Thompson, B. Stepwise regression and stepwise disciminant analysis  not apply here: A guideline editorial

Educational and Psychological Measurement, Vol. 55, No. 4, 525-534 (1995)
 

***************************************************************************************************************************************************************
Mark A. Davenport Ph.D.
Senior Research Analyst
Office of Institutional Research
The University of North Carolina at Greensboro
336.256.0395
[hidden email]

'An approximate answer to the right question is worth a good deal more than an exact answer to an approximate question.' --a paraphrase of J. W. Tukey (1962)

-----"SPSSX(r) Discussion" <[hidden email]> wrote: -----

To: [hidden email]
From: "Noble, Lyndsay" <[hidden email]>
Sent by: "SPSSX(r) Discussion" <[hidden email]>
Date: 07/10/2007 12:20PM
Subject: Re: regression help

Kizzy,
I'm not sure what you mean by "method".  My initial advice would be to use
regular, plain, old, regression with your categorical variables dummy coded
(or effect coded, same thing).  That is, if your categorical variable has n
levels, you would create n-1 variables with values of 0 and 1 (the nth level
of the initial variable would be indicated by having zeros in all n-1 new
variables).  Then, I would enter all of the variables into the regression
model and go from there.

If what you mean by "method" is stepwise, forward, backward, or some other
selection algorithm, I have no opinion.  Each has its pros and cons and I
will defer to the rest of the list.

HTH,
Lyndsay



-----Original Message-----
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of
Kizzy M Parks
Sent: Tuesday, July 10, 2007 8:55 AM
To: [hidden email]
Subject: regression help

Hello:

Using regression, I need to analyze continuous and categorical IVs, but my
DV is continuous.  Any suggestions on the method I should use?  I have a
sample of over 400,000.


Kizzy M. Parks, ABD
Florida Tech