Logistic regression, please help

[kangaroo]

I am having a nightmare and unfortunately my supervisors can not figure this out either!
 
I have an equation from logistic regression, it predicts treatment outcome from psych variables, of which the equation includes an interaction variable (2 variables combined)-anxietyscore*controloverpain score
 
I need to produce a worked example and use new patients scores within the model to predict outcome, we can’t find out what happens to the baseline scores to make the interaction.
 
In other words, I need to demonstrate what happens to the 2 baseline scores within the interaction, within the log regression equation.
 
We assumed it was variable*variable, multiply anxiety score by control over pain score, but it appears not.
 
Please can you help? I am due to submit my PhD and really can’t get to the bottom of this.
Best Wishes
Liz

[Bruce Weaver]

Have you ever looked at James Jaccard's nice monograph on interactions in logistic regression?  If not, you should.

 http://books.google.com/books/about/Interaction_effects_in_logistic_regressi.html?id=tj1Wn5u9gSMC

Another pointer:  After you've read and understood Jaccard's book, you might consider running your model via GENLIN (with a logit link and binomial error distribution) rather than via LOGISTIC REGRESSION.  With GENLIN, you can use EMMEANS sub-commands to obtain fitted values (of the log-odds) at selected combinations of the interacting variables.  These can then be plotted to illustrate very nicely the nature of the interaction (in the way Jaccard describes).

HTH.

kangaroo wrote

I am having a nightmare and unfortunately my supervisors can not figure this out either!
 
I have an equation from logistic regression, it predicts treatment outcome from psych variables, of which the equation includes an interaction variable (2 variables combined)-anxietyscore*controloverpain score
 
I need to produce a worked example and use new patients scores within the model to predict outcome, we can’t find out what happens to the baseline scores to make the interaction.
 
In other words, I need to demonstrate what happens to the 2 baseline scores within the interaction, within the log regression equation.
 
We assumed it was variable*variable, multiply anxiety score by control over pain score, but it appears not.
 
Please can you help? I am due to submit my PhD and really can’t get to the bottom of this.
Best Wishes
Liz

[Hector Maletta]

I do not understand exactly what is the problem. Your model, apparently, must include the two scores separately, and then another variable which is the product of the two (three predictors in total: "anxietyscore", "controloverpain", and "anxietyscore x controloverpain"), apart from any other predictors that you have in the equation. Your database must have one column for the anxiety score, and another column for control-over-pain. LOGISTIC REGRESSION creates the interaction variable by itself, but if you want you may also create a new variable yourself:
COMPUTE INTERACTIONVAR= ANXIETYSCORE *controloverpain.
And then include the three variables in the equation: controloverpain, ANXIETYSCORE, and INTERACTIONVAR.
In any case, you would have the separate scores in the dataset, and the three coefficients in the outcome. What is the problem?

Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de kangaroo
Enviado el: Wednesday, July 20, 2011 18:02
Para: [hidden email]
Asunto: Logistic regression, please help

I am having a nightmare and unfortunately my supervisors can not figure this
out either!

I have an equation from logistic regression, it predicts treatment outcome
from psych variables, of which the equation includes an interaction variable
(2 variables combined)-anxietyscore*controloverpain score

I need to produce a worked example and use new patients scores within the
model to predict outcome, we can’t find out what happens to the baseline
scores to make the interaction.

In other words, I need to demonstrate what happens to the 2 baseline scores
within the interaction, within the log regression equation.

We assumed it was variable*variable, multiply anxiety score by control over
pain score, but it appears not.

Please can you help? I am due to submit my PhD and really can’t get to the
bottom of this.
Best Wishes
Liz

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Logistic-regression-please-help-tp4617275p4617275.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
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
-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 10.0.1390 / Virus Database: 1518/3776 - Release Date: 07/20/11

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

[kangaroo]

Following on- my problem is that now I have the predicative equation from the log regression- I need to use it with new patients- but as it includes an interaction, I don't know what calculation to do to include the interaction?

Thanks so much for any help
Sent using BlackBerry® from Orange

-----Original Message-----
From: "Hector Maletta [via SPSSX Discussion]" <[hidden email]>
Date: Thu, 21 Jul 2011 04:43:06
To: <[hidden email]>
Subject: Re: Logistic regression, please help

        I do not understand exactly what is the problem. Your model, apparently, must include the two scores separately, and then another variable which is the product of the two (three predictors in total: "anxietyscore", "controloverpain", and "anxietyscore x controloverpain"), apart from any other predictors that you have in the equation. Your database must have one column for the anxiety score, and another column for control-over-pain. LOGISTIC REGRESSION creates the interaction variable by itself, but if you want you may also create a new variable yourself: COMPUTE INTERACTIONVAR= ANXIETYSCORE *controloverpain. And then include the three variables in the equation: controloverpain, ANXIETYSCORE, and INTERACTIONVAR. In any case, you would have the separate scores in the dataset, and the three coefficients in the outcome. What is the problem? Hector -----Mensaje original----- De: SPSSX(r) Discussion [mailto:[hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=0> ] En nombre de kangaroo Enviado el: Wednesday, July 20, 2011 18:02 Para: [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=1> Asunto: Logistic regression, please help I am having a nightmare and unfortunately my supervisors can not figure this out either! I have an equation from logistic regression, it predicts treatment outcome from psych variables, of which the equation includes an interaction variable (2 variables combined)-anxietyscore*controloverpain score I need to produce a worked example and use new patients scores within the model to predict outcome, we can't find out what happens to the baseline scores to make the interaction. In other words, I need to demonstrate what happens to the 2 baseline scores within the interaction, within the log regression equation. We assumed it was variable*variable, multiply anxiety score by control over pain score, but it appears not. Please can you help? I am due to submit my PhD and really can't get to the bottom of this. Best Wishes Liz -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Logistic-regression-please-help-tp4617275p4617275.htmlSent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=2>  (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 ----- No virus found in this message. Checked by AVG - www.avg.com Version: 10.0.1390 / Virus Database: 1518/3776 - Release Date: 07/20/11 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=3>  (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
----------------
 
 
If you reply to this email, your message will be added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Logistic-regression-please-help-tp4617275p4618151.html 
  To unsubscribe from Logistic regression, please help, click here < .

[Bruce Weaver]

Include the product of the interacting variables.  

But here's another option.  If you have the raw data that generated the equation, append rows of data for the new cases to that data file, and fill in the values of the explanatory variables for those new cases, but leave the outcome variable as missing.  Create variable to flag the new cases so you can select them easily later.  Now re-run your logistic regression model, and save the predicted probabilities to the file.  They will be saved for all cases in the file, including the new ones that were not used to generate the model.  If you want the odds or log-odds rather than the predicted probability, use COMPUTE commands to get them.

HTH.

kangaroo wrote

Following on- my problem is that now I have the predicative equation from the log regression- I need to use it with new patients- but as it includes an interaction, I don't know what calculation to do to include the interaction?

Thanks so much for any help
Sent using BlackBerry® from Orange

-----Original Message-----
From: "Hector Maletta [via SPSSX Discussion]" <[hidden email]>
Date: Thu, 21 Jul 2011 04:43:06
To: <[hidden email]>
Subject: Re: Logistic regression, please help

        I do not understand exactly what is the problem. Your model, apparently, must include the two scores separately, and then another variable which is the product of the two (three predictors in total: "anxietyscore", "controloverpain", and "anxietyscore x controloverpain"), apart from any other predictors that you have in the equation. Your database must have one column for the anxiety score, and another column for control-over-pain. LOGISTIC REGRESSION creates the interaction variable by itself, but if you want you may also create a new variable yourself: COMPUTE INTERACTIONVAR= ANXIETYSCORE *controloverpain. And then include the three variables in the equation: controloverpain, ANXIETYSCORE, and INTERACTIONVAR. In any case, you would have the separate scores in the dataset, and the three coefficients in the outcome. What is the problem? Hector -----Mensaje original----- De: SPSSX(r) Discussion [mailto:[hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=0> ] En nombre de kangaroo Enviado el: Wednesday, July 20, 2011 18:02 Para: [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=1> Asunto: Logistic regression, please help I am having a nightmare and unfortunately my supervisors can not figure this out either! I have an equation from logistic regression, it predicts treatment outcome from psych variables, of which the equation includes an interaction variable (2 variables combined)-anxietyscore*controloverpain score I need to produce a worked example and use new patients scores within the model to predict outcome, we can't find out what happens to the baseline scores to make the interaction. In other words, I need to demonstrate what happens to the 2 baseline scores within the interaction, within the log regression equation. We assumed it was variable*variable, multiply anxiety score by control over pain score, but it appears not. Please can you help? I am due to submit my PhD and really can't get to the bottom of this. Best Wishes Liz -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Logistic-regression-please-help-tp4617275p4617275.htmlSent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=2>  (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 ----- No virus found in this message. Checked by AVG - www.avg.com Version: 10.0.1390 / Virus Database: 1518/3776 - Release Date: 07/20/11 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] </user/SendEmail.jtp?type=node&node=4618151&i=3>  (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
----------------
 
 
If you reply to this email, your message will be added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Logistic-regression-please-help-tp4617275p4618151.html 
  To unsubscribe from Logistic regression, please help, click here <http://spssx-discussion.1045642.n5.nabble.com/template/NamlServlet.jtp?macro=unsubscribe_by_code&amp;node=4617275&amp;code=bGl6emVzcGFya2VzQGhvdG1haWwuY29tfDQ2MTcyNzV8MTE1MDQwNjI0Mg==> .