Using a segmented linear regression to find a transition point

[Moshe Marko]

Hi,

I would like to run a piecewise regression and a logistic regression models
where the goal is to find the transition point where the association between
the dependent and the independent variables change (i.e. adding a change
point c) - any ideas for the correct syntax?

Thanks

Moshe

Moshe Marko, PT, DPT, MHS, OCS, CSCS
Assistant Professor
Department of Physical Therapy Education
College of Health Professions
SUNY Upstate Medical University
Room 2232  Silverman Hall
750 Adams Street
Syracuse, NY 13210-1834
315 464 6577
FAX [phone]
[hidden email]

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

[Bruce Weaver]

Moshe Marko wrote

Hi,

I would like to run a piecewise regression and a logistic regression models
where the goal is to find the transition point where the association between
the dependent and the independent variables change (i.e. adding a change
point c) - any ideas for the correct syntax?

Thanks

Moshe

This should help for the piecewise linear regression model:

  http://www.spsstools.net/Syntax/RegressionRepeatedMeasure/PiecewiseRegression.txt

For the case where the knot points are known, the same methods described there could be applied to logistic regression, I should think.  But I'm not aware of a method for using the data to estimate the knot points for logistic regression.  I guess I'd start by seeing if I could cajole NLR into running an ordinary binary logistic regression model first.  If that can be done, then you should be able to get it to determine the knot points for you too.

HTH.

[Jon K Peck]

Take a look at the CATREG command.  You can tell it to discretize a predictor into a specified number of groups according to various criteria, and it will find the optimal placement for the segments.

HTH,
Jon Peck
SPSS, an IBM Company
[hidden email]
312-651-3435

From:	Moshe Marko <[hidden email]>
To:	[hidden email]
Date:	09/22/2010 05:38 AM
Subject:	[SPSSX-L] Using a segmented linear regression to find a              transition point
Sent by:	"SPSSX(r) Discussion" <[hidden email]>

---

Hi,

I would like to run a piecewise regression and a logistic regression models
where the goal is to find the transition point where the association between
the dependent and the independent variables change (i.e. adding a change
point c) - any ideas for the correct syntax?

Thanks

Moshe

Moshe Marko, PT, DPT, MHS, OCS, CSCS
Assistant Professor
Department of Physical Therapy Education
College of Health Professions
SUNY Upstate Medical University
Room 2232  Silverman Hall
750 Adams Street
Syracuse, NY 13210-1834
315 464 6577
FAX [phone]
[hidden email]

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

[Bruce Weaver]

Jon K Peck wrote

Take a look at the CATREG command.  You can tell it to discretize a
predictor into a specified number of groups according to various criteria,
and it will find the optimal placement for the segments.

HTH,
Jon Peck
SPSS, an IBM Company
[e-mail]
312-651-3435

Hi Jon.  I don't have access to CATREG, but I have taken a look at the tutorial in the Help.  I may be misunderstanding, but I don't think discretization is what the OP is after here.  It would be one of the various spline options, wouldn't it?  

Thanks,
Bruce