inflated exp(b) in logistic regression
Locked
 
|
Hi,
I'm running a logistic regression (n=870) with 8 IVs most of which are continuous. For one of my continuous IVs I'm getting an inflated exp(b) of 86.73 which seems wrong. The B is 4.4, se = 2.2 and wald = 3.9 which all seem okay. I did box-tidwell to check linearity of logit assumption and am not violating this assumption. Any ideas why this is going on? On a related note, my constant has a wald of 59.6- is that okay? seems high |
Re: inflated exp(b) in logistic regression
|
|
Have you checked the collinearity diagnostics?
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: 313-993-8085 Fax: 313-966-7682 --- On Sat, 5/30/09, delliott <[hidden email]> wrote: > From: delliott <[hidden email]> > Subject: inflated exp(b) in logistic regression > To: [hidden email] > Date: Saturday, May 30, 2009, 11:33 PM > Hi, > I'm running a logistic regression (n=870) with 8 IVs most > of which are > continuous. For one of my continuous IVs I'm getting an > inflated exp(b) of > 86.73 which seems wrong. The B is 4.4, se = 2.2 and > wald = 3.9 which all > seem okay. I did box-tidwell to check linearity of > logit assumption and am > not violating this assumption. Any ideas why this is > going on? > > On a related note, my constant has a wald of 59.6- is that > okay? seems high > -- > View this message in context: http://www.nabble.com/inflated-exp%28b%29-in-logistic-regression-tp23799527p23799527.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 > ===================== 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 |
|
|
In reply to this post by delliott
Need a Holiday? Win a $10,000 Holiday of your choice. Enter now.. |
Re: inflated exp(b) in logistic regression
|
|
In reply to this post by delliott
exp(b) is, of course, the odds ratio. I think that we need more information about the model, eg, the predictors/covariates, the response variable, and the confidence interval for the variable having the large OR.
In general, I like to perform a range of diagnostics, including examining the degree of collinearity among the covariates, ie, seeing if any of the condition indexes exceed 20-30 and whether 2 or more of the variance decomposition proportions are >.50 on a given condition index. Scott Millis --- On Sun, 5/31/09, Jims More <[hidden email]> wrote: > From: Jims More <[hidden email]> > Subject: Re: inflated exp(b) in logistic regression > To: "SR Millis" <[hidden email]> > Date: Sunday, May 31, 2009, 7:48 AM > Hi delliott, Scott, etc, > > I am interested of this issue. I apologize for > interferring. What is the range of exp(b) to be > acceptable? > > Jims > > --- On Sun, 31/5/09, SR Millis > <[hidden email]> wrote: > > > From: SR Millis <[hidden email]> > Subject: Re: inflated exp(b) in logistic regression > To: [hidden email] > Received: Sunday, 31 May, 2009, 4:13 AM > > > Have you checked the collinearity > diagnostics? > > 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: 313-993-8085 > Fax: 313-966-7682 > > > --- On Sat, 5/30/09, delliott <[hidden email]> > wrote: > > > From: delliott <[hidden email]> > > Subject: inflated exp(b) in logistic > regression > > To: [hidden email] > > Date: Saturday, May 30, 2009, 11:33 PM > > Hi, > > I'm running a logistic regression (n=870) with 8 > IVs most > > of which are > > continuous. For one of my continuous IVs I'm > getting an > > inflated exp(b) of > > 86.73 which seems wrong. The B is 4.4, se = 2.2 > and > > wald = 3.9 which all > > seem okay. I did box-tidwell to check linearity > of > > logit assumption and am > > not violating this assumption. Any ideas why > this is > > going on? > > > > On a related note, my constant has a wald of 59.6- is > that > > okay? seems high > > -- > > View this message in context: http://www.nabble.com/inflated-exp%28b%29-in-logistic-regression-tp23799527p23799527.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 > > > > ===================== > 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 > > > > > > > Need a Holiday? Win a $10,000 Holiday of your choice. Enter now.. ===================== 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 |
Re: inflated exp(b) in logistic regression
|
|
In reply to this post by delliott
If not apparent in my previous posting, I should have added that the range of the odds ratio is 0 to infinity. How big is big is dependent on the scaling of the variable and other model characteristics.
Scott Millis --- On Sun, 5/31/09, SR Millis <[hidden email]> wrote: > From: SR Millis <[hidden email]> > Subject: Re: inflated exp(b) in logistic regression > To: [hidden email] > Date: Sunday, May 31, 2009, 1:35 PM > exp(b) is, of course, the odds > ratio. I think that we need more information about the > model, eg, the predictors/covariates, the response variable, > and the confidence interval for the variable having the > large OR. > > In general, I like to perform a range of diagnostics, > including examining the degree of collinearity among the > covariates, ie, seeing if any of the condition indexes > exceed 20-30 and whether 2 or more of the variance > decomposition proportions are >.50 on a given condition > index. > > Scott Millis > > > > --- On Sun, 5/31/09, Jims More <[hidden email]> > wrote: > > > From: Jims More <[hidden email]> > > Subject: Re: inflated exp(b) in logistic regression > > To: "SR Millis" <[hidden email]> > > Date: Sunday, May 31, 2009, 7:48 AM > > Hi delliott, Scott, etc, > > > > I am interested of this issue. I apologize for > > interferring. What is the range of exp(b) to be > > acceptable? > > > > Jims > > > > --- On Sun, 31/5/09, SR Millis > > <[hidden email]> > wrote: > > > > > > From: SR Millis <[hidden email]> > > Subject: Re: inflated exp(b) in logistic regression > > To: [hidden email] > > Received: Sunday, 31 May, 2009, 4:13 AM > > > > > > Have you checked the collinearity > > diagnostics? > > > > 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: 313-993-8085 > > Fax: 313-966-7682 > > > > > > --- On Sat, 5/30/09, delliott <[hidden email]> > > wrote: > > > > > From: delliott <[hidden email]> > > > Subject: inflated exp(b) in logistic > > regression > > > To: [hidden email] > > > Date: Saturday, May 30, 2009, 11:33 PM > > > Hi, > > > I'm running a logistic regression (n=870) with 8 > > IVs most > > > of which are > > > continuous. For one of my continuous IVs I'm > > getting an > > > inflated exp(b) of > > > 86.73 which seems wrong. The B is 4.4, se = > 2.2 > > and > > > wald = 3.9 which all > > > seem okay. I did box-tidwell to check > linearity > > of > > > logit assumption and am > > > not violating this assumption. Any ideas > why > > this is > > > going on? > > > > > > On a related note, my constant has a wald of > 59.6- is > > that > > > okay? seems high > > > -- > > > View this message in context: http://www.nabble.com/inflated-exp%28b%29-in-logistic-regression-tp23799527p23799527.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 > > > > > > > ===================== > > 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 > > > > > > > > > > > > > > Need a Holiday? Win a $10,000 Holiday of your > choice. Enter now.. > > ===================== > 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 |
|
|
In reply to this post by SR Millis-3
Hi,
I did perform some diagnotics and may have a collinearity issue (someone else suggested this might be the issue as well). When I ran collinearity diagnostics I found a correlation of .6 with another variable- high, but not overly high. I'm not sure what you mean by condition indices exceeding 20-30 and whether 2 or more of the variance decomposition proportions are greater than .5. Can you explain? The model is examining the persistence of students in higher education so the DV is continued enrollment or dropout. I have 8 IVs: hs gpa- continuous college selectivity- categorical gender- dichotomous parent's education - categorical academic integration- continuous, social integration- continous academic efficacy- continuous social efficacy- continuous The last variable is the one with the high odds ratio and the correlation of .6 is with academic efficacy, however when I run the model without academic efficacy I still get an unusually high odds ratio. Thanks for your help :)
|
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |