Statistical Question

[Brandon Long-2]

This listserv is a great resource, so I was hoping someone might be able
to help me get some clarification on a certain statistical question.  If
you have risk-adjusted a sample of data by certain factors using
logistic regression, is it incorrect to further stratify these results
by some sort of grouping variable?  I do not believe it would be
statistically correct but am having a hard time justifying this.  Am I
wrong?  Any information would be helpful.
 
Thank you!

[Albert-Jan Roskam]

Hi Brandon,

If you want to stratify the odds of e.g. having asthma
by educational level you could specify your education
variable as a categorical variable in your regression
model. You will see the odds relative to one reference
group, which is dependent on the way you sorted your
data.

Cheers!
Albert-Jan

--- Brandon Long <[hidden email]> wrote:

> This listserv is a great resource, so I was hoping
> someone might be able
> to help me get some clarification on a certain
> statistical question.  If
> you have risk-adjusted a sample of data by certain
> factors using
> logistic regression, is it incorrect to further
> stratify these results
> by some sort of grouping variable?  I do not believe
> it would be
> statistically correct but am having a hard time
> justifying this.  Am I
> wrong?  Any information would be helpful.
>
> Thank you!
>

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

[Hector Maletta]

More exactly, the odds are computed for each cateogyr relative to one
reference category which is chosen by default (usually the last one) but
there are several other contrasts to choose from.

Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de
Albert-jan Roskam
Enviado el: Tuesday, August 08, 2006 5:51 PM
Para: [hidden email]
Asunto: Re: Statistical Question

Hi Brandon,

If you want to stratify the odds of e.g. having asthma
by educational level you could specify your education
variable as a categorical variable in your regression
model. You will see the odds relative to one reference
group, which is dependent on the way you sorted your
data.

Cheers!
Albert-Jan

--- Brandon Long <[hidden email]> wrote:

> This listserv is a great resource, so I was hoping
> someone might be able
> to help me get some clarification on a certain
> statistical question.  If
> you have risk-adjusted a sample of data by certain
> factors using
> logistic regression, is it incorrect to further
> stratify these results
> by some sort of grouping variable?  I do not believe
> it would be
> statistically correct but am having a hard time
> justifying this.  Am I
> wrong?  Any information would be helpful.
>
> Thank you!
>

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

[Jason McNellis]

Great question.  I can't really speak to the statistical aspect--I'm
suspicious it would be violating an assumption regarding structure in the
data.  I can tell you that in a business setting logistic regression is
sometimes applied to a grouping variable (like customer segments).  The way
we do this is build a separate model for each segment.  This is done when we
have prior knowledge that suggests different groups will respond
differently.

Jason





-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Brandon Long
Sent: Tuesday, August 08, 2006 3:35 PM
To: [hidden email]
Subject: Statistical Question

This listserv is a great resource, so I was hoping someone might be able
to help me get some clarification on a certain statistical question.  If
you have risk-adjusted a sample of data by certain factors using
logistic regression, is it incorrect to further stratify these results
by some sort of grouping variable?  I do not believe it would be
statistically correct but am having a hard time justifying this.  Am I
wrong?  Any information would be helpful.

Thank you!