Stepwise binary logistic Regression

[E. Bernardo]

Do you recommned stepwise binary logistic regression in choosing a parsimonious model that can be formed from a number of continuous and categorical predictors.  

 

Thank you.

Eins

---

Fast, Ad-free, Unlimited Storage - Yahoo! Mail allows you to have it all.

[SR Millis-3]

Absolutely not!

As Harrell discusses:

Here are some of the problems with stepwise variable selection.

   1. It yields R-squared values that are badly biased to be high.
   2. The F and chi-squared tests quoted next to each variable on the printout do not have the claimed distribution.
   3. The method yields confidence intervals for effects and predicted values that are falsely narrow (See Altman and Anderson, 1989, Statistics in Medicine).
   4. It yields p-values that do not have the proper meaning, and the proper correction for them is a difficult problem.
   5. It gives biased regression coefficients that need shrinkage (the coefficients for remaining variables are too large; see Tibshirani, 1996).
   6. It has severe problems in the presence of collinearity.
   7. It is based on methods (e.g., F tests for nested models) that were intended to be used to test prespecified hypotheses.
   8. Increasing the sample size doesn't help very much (see Derksen and Keselman, 1992).
   9. It allows us to not think about the problem.
  10. It uses a lot of paper.

“All possible subsets” regression solves none of these problems.


I recommend Bayesian model averaging.

Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Fri, 6/12/09, Eins Bernardo <[hidden email]> wrote:

> From: Eins Bernardo <[hidden email]>
> Subject: Stepwise binary logistic Regression
> To: [hidden email]
> Date: Friday, June 12, 2009, 8:32 AM
> Do you recommned� stepwise
> binary logistic regression in choosing a parsimonious model
> that can be formed from a number of continuous and
> categorical predictors.� �
> �
> Thank you.
> Eins
>
>       Fast, Ad-free, Unlimited Storage -   Yahoo! Mail allows you to
> have it all.

=====================
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

[E. Bernardo]

Thank you, Scott. 

Nice! Very informative.  

I have a follow-up question: Can we do Bayesian model averaging in SPSS?

 

Eins

--- On Fri, 6/12/09, SR Millis <[hidden email]> wrote:

From: SR Millis <[hidden email]>
Subject: Re: Stepwise binary logistic Regression
To: "Eins Bernardo" <[hidden email]>, "SPSS" <[hidden email]>
Date: Friday, 12 June, 2009, 12:43 PM

Absolutely not!

As Harrell discusses:

Here are some of the problems with stepwise variable selection.

   1. It yields R-squared values that are badly biased to be high.
   2. The F and chi-squared tests quoted next to each variable on the printout do not have the claimed distribution.
   3. The method yields confidence intervals for effects and predicted values that are falsely narrow (See Altman and Anderson, 1989, Statistics in Medicine).
   4. It yields p-values that do not have the proper meaning, and the proper correction for them is a difficult problem.
   5. It gives biased regression coefficients that need shrinkage (the coefficients for remaining variables are too large; see Tibshirani, 1996).
   6. It has severe problems in the presence of collinearity.
   7. It is based on methods (e.g., F tests for nested models) that were intended to be used to test prespecified hypotheses.
   8. Increasing the sample size doesn't help very much (see Derksen and Keselman, 1992).
   9. It allows us to not think about the problem.
  10. It uses a lot of paper.

“All possible subsets” regression solves none of these problems.


I recommend Bayesian model averaging.

Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  smillis@...
Tel: [phone]
Fax: [phone]


--- On Fri, 6/12/09, Eins Bernardo <einsbernardo@...> wrote:

> From: Eins Bernardo <einsbernardo@...>
> Subject: Stepwise binary logistic Regression
> To: SPSSX-L@...
> Date: Friday, June 12, 2009, 8:32 AM
> Do you recommned stepwise
> binary logistic regression in choosing a parsimonious model
> that can be formed from a number of continuous and
> categorical predictors.  
>  
> Thank you.
> Eins
>
>       Fast, Ad-free, Unlimited Storage -   Yahoo! Mail allows you to
> have it all.

---

Fast, Ad-free, Unlimited Storage - Yahoo! Mail allows you to have it all.

---

Get your preferred Email name!
Now you can @ymail.com and @rocketmail.com.

[SR Millis-3]

Unfortunately, I'm not aware of any macros/programs to perform Bayesian model averaging in SPSS.  However, it can be done easily in R---the free, open source statistical software:

http://www.research.att.com/~volinsky/bma.html




Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Fri, 6/12/09, Eins Bernardo <[hidden email]> wrote:

> From: Eins Bernardo <[hidden email]>
> Subject: Re: Stepwise binary logistic Regression
> To: "SPSS" <[hidden email]>, "SR Millis" <[hidden email]>
> Date: Friday, June 12, 2009, 10:04 AM
> Thank you, Scott.�
> Nice! Very informative.� �
> I have a follow-up question: Can we do Bayesian model
> averaging in SPSS?
> �
> Eins
>
> --- On Fri, 6/12/09, SR Millis
> <[hidden email]> wrote:
>
>
> From: SR Millis <[hidden email]>
> Subject: Re: Stepwise binary logistic Regression
> To: "Eins Bernardo"
> <[hidden email]>, "SPSS"
> <[hidden email]>
> Date: Friday, 12 June, 2009, 12:43 PM
>
>
>
> Absolutely not!
>
> As Harrell discusses:
>
> Here are some of the problems with stepwise variable
> selection.
>
> � � � 1. It yields R-squared values that are
> badly biased to be high.
> � � � 2. The F and chi-squared tests quoted
> next to each variable on the printout do not have the
> claimed distribution.
> � � � 3. The method yields confidence intervals
> for effects and predicted values that are falsely narrow
> (See Altman and Anderson, 1989, Statistics in Medicine).
> � � � 4. It yields p-values that do not have
> the proper meaning, and the proper correction for them is a
> difficult problem.
> � � � 5. It gives biased regression
> coefficients that need shrinkage (the coefficients for
> remaining variables are too large; see Tibshirani, 1996).
> � � � 6. It has severe problems in the presence
> of collinearity.
> � � � 7. It is based on
>  methods (e.g., F tests for nested models) that were
> intended to be used to test prespecified hypotheses.
> � � � 8. Increasing the sample size doesn't
> help very much (see Derksen and Keselman, 1992).
> � � � 9. It allows us to not think about the
> problem.
> �  10. It uses a lot of paper.
>
> “All possible subsets” regression solves none of these
> problems.
>
>
> I recommend Bayesian model averaging.
>
> Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
> Professor & Director of Research
> Dept of Physical Medicine & Rehabilitation
> Dept of Emergency Medicine
> Wayne State University School of Medicine
> 261 Mack Blvd
> Detroit, MI 48201
> Email:�  [hidden email]
> Tel: [phone]
> Fax: [phone]
>
>
> --- On Fri, 6/12/09, Eins Bernardo <[hidden email]..ph>
> wrote:
>
> > From: Eins Bernardo <[hidden email]>
> > Subject: Stepwise binary logistic Regression
> > To: [hidden email]
> > Date: Friday, June 12, 2009, 8:32 AM
> > Do you recommned� stepwise
> > binary logistic regression in choosing a parsimonious
> model
> > that can be formed from a number of continuous and
> > categorical predictors.� �
> > �
> > Thank you.
> > Eins
> >
> >�  �  � � � Fast, Ad-free,
> Unlimited Storage
>  -� � � Yahoo! Mail allows you to
> > have it all.
>
>
>       Fast, Ad-free, Unlimited Storage -   Yahoo! Mail allows you to
> have it all.

=====================
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

[Hector Maletta]

There is a book that may be useful: Jim Alberts, Bayesian Computation with R
(Springer, 2007).

Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SR
Millis
Sent: 12 June 2009 13:32
To: [hidden email]
Subject: Re: Stepwise binary logistic Regression

Unfortunately, I'm not aware of any macros/programs to perform Bayesian
model averaging in SPSS.  However, it can be done easily in R---the free,
open source statistical software:

http://www.research.att.com/~volinsky/bma.html




Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
Professor & Director of Research
Dept of Physical Medicine & Rehabilitation
Dept of Emergency Medicine
Wayne State University School of Medicine
261 Mack Blvd
Detroit, MI 48201
Email:  [hidden email]
Tel: [phone]
Fax: [phone]


--- On Fri, 6/12/09, Eins Bernardo <[hidden email]> wrote:

> From: Eins Bernardo <[hidden email]>
> Subject: Re: Stepwise binary logistic Regression
> To: "SPSS" <[hidden email]>, "SR Millis" <[hidden email]>
> Date: Friday, June 12, 2009, 10:04 AM
> Thank you, Scott.
> Nice! Very informative.
> I have a follow-up question: Can we do Bayesian model
> averaging in SPSS?
>
> Eins
>
> --- On Fri, 6/12/09, SR Millis
> <[hidden email]> wrote:
>
>
> From: SR Millis <[hidden email]>
> Subject: Re: Stepwise binary logistic Regression
> To: "Eins Bernardo"
> <[hidden email]>, "SPSS"
> <[hidden email]>
> Date: Friday, 12 June, 2009, 12:43 PM
>
>
>
> Absolutely not!
>
> As Harrell discusses:
>
> Here are some of the problems with stepwise variable
> selection.
>
>    1. It yields R-squared values that are
> badly biased to be high.
>    2. The F and chi-squared tests quoted
> next to each variable on the printout do not have the
> claimed distribution.
>    3. The method yields confidence intervals
> for effects and predicted values that are falsely narrow
> (See Altman and Anderson, 1989, Statistics in Medicine).
>    4. It yields p-values that do not have
> the proper meaning, and the proper correction for them is a
> difficult problem.
>    5. It gives biased regression
> coefficients that need shrinkage (the coefficients for
> remaining variables are too large; see Tibshirani, 1996).
>    6. It has severe problems in the presence
> of collinearity.
>    7. It is based on
>  methods (e.g., F tests for nested models) that were
> intended to be used to test prespecified hypotheses.
>    8. Increasing the sample size doesn't
> help very much (see Derksen and Keselman, 1992).
>    9. It allows us to not think about the
> problem.
>   10. It uses a lot of paper.
>
> "All possible subsets" regression solves none of these
> problems.
>
>
> I recommend Bayesian model averaging.
>
> Scott R Millis, PhD, ABPP (CN,CL,RP), CStat, CSci
> Professor & Director of Research
> Dept of Physical Medicine & Rehabilitation
> Dept of Emergency Medicine
> Wayne State University School of Medicine
> 261 Mack Blvd
> Detroit, MI 48201
> Email:  [hidden email]
> Tel: [phone]
> Fax: [phone]
>
>
> --- On Fri, 6/12/09, Eins Bernardo <[hidden email]..ph>
> wrote:
>
> > From: Eins Bernardo <[hidden email]>
> > Subject: Stepwise binary logistic Regression
> > To: [hidden email]
> > Date: Friday, June 12, 2009, 8:32 AM
> > Do you recommned stepwise
> > binary logistic regression in choosing a parsimonious
> model
> > that can be formed from a number of continuous and
> > categorical predictors.
> >
> > Thank you.
> > Eins
> >
> >       Fast, Ad-free,
> Unlimited Storage
>  -   Yahoo! Mail allows you to
> > have it all.
>
>
>       Fast, Ad-free, Unlimited Storage -   Yahoo! Mail allows you to
> have it all.

=====================
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