Interpreting logistic regression models

[Bianco, Joseph]

Hi Listers:

 

I’m looking to build a model of the ways in which coping style, social support, and depression influence adherence to medication. Coping, social support, and depression are continuous. Adherence is a dichotomous DV (0=nonadherent, 1 = adherent). This was chosen b/c the meds in question require consistently high levels of adherence (>95% of the time), so the idea of examining degrees of adherence isn’t as relevant.

 

All variables are significantly correlated (in the expected directions).

 

In any event, I don’t have access to Lisrel or Amos, so I’m using logistic regressions. I am not sure how to interpret the results. I don’t’ understand why some variables lose significance when paired with others in the model, since the regressions do not seem to indicate suppression or mediation (at least in my limited understanding of them).

 

Regression 1:  Logistic Regression, forced entry:  I entered Coping, Support, and Depression together (block 1) on Adherence. Results:  only Coping is significant (Wald 5.21, p =.022).  Support p = .07, Dep = p .93.

 

Regression 2: Logistic Regression, with Coping and Support entered in block 1. Adherence is still the DV. Results: Coping (Wald = 5.87, p = .015) and Support (4.22, p = .04) are both significant once Depression is no longer in the equation.

 

Regression 3:  Logistic regression. Adherence is the DV and Depression alone is the IV. Depression is significant. Wald = 4.50, p =.034.

 

Regression 4:  Multiple regression with Depression (continuous variable) as the DV and Coping and Support as predictors.   Results: Rsq = .41 F(3,161) = 54.7, p <.001. Both support and coping are significant at the <.000 levels.

 

 

Can anyone help me understand why depression is significant alone, but not when paired with support and coping? I guess I don’t understand why all 3 predictors aren’t significant in Regression 1. If depression weren’t a predictor of adherence, I’d assume it’s a suppressor.

 

Any help with this (admittedly low-level) question would be greatly appreciated.

Thanks

Joe

 

 

 

Joseph A. Bianco, Ph.D.

Assistant Professor, Research

Department of Geriatric Medicine/Gerontology

Ohio University College of Osteopathic Medicine

Athens, OH 45701

 

[mpirritano]

The best references to help explain general logistic and logistic regression with interactions

 

‘Interaction Effects in Logistic Regression’ James Jaccard

 

It’s a little sage paper.

 

Matthew Pirritano, Ph.D.

Research Analyst IV

Medical Services Initiative (MSI)

Orange County Health Care Agency

[phone]

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Bianco, Joseph
Sent: Monday, August 24, 2009 10:44 AM
To: [hidden email]
Subject: Interpreting logistic regression models

 

Hi Listers:

 

I’m looking to build a model of the ways in which coping style, social support, and depression influence adherence to medication. Coping, social support, and depression are continuous. Adherence is a dichotomous DV (0=nonadherent, 1 = adherent). This was chosen b/c the meds in question require consistently high levels of adherence (>95% of the time), so the idea of examining degrees of adherence isn’t as relevant.

 

All variables are significantly correlated (in the expected directions).

 

In any event, I don’t have access to Lisrel or Amos, so I’m using logistic regressions. I am not sure how to interpret the results. I don’t’ understand why some variables lose significance when paired with others in the model, since the regressions do not seem to indicate suppression or mediation (at least in my limited understanding of them).

 

Regression 1:  Logistic Regression, forced entry:  I entered Coping, Support, and Depression together (block 1) on Adherence. Results:  only Coping is significant (Wald 5.21, p =.022).  Support p = .07, Dep = p .93.

 

Regression 2: Logistic Regression, with Coping and Support entered in block 1. Adherence is still the DV. Results: Coping (Wald = 5.87, p = .015) and Support (4.22, p = .04) are both significant once Depression is no longer in the equation.

 

Regression 3:  Logistic regression. Adherence is the DV and Depression alone is the IV. Depression is significant. Wald = 4.50, p =.034.

 

Regression 4:  Multiple regression with Depression (continuous variable) as the DV and Coping and Support as predictors.   Results: Rsq = .41 F(3,161) = 54.7, p <.001. Both support and coping are significant at the <.000 levels.

 

 

Can anyone help me understand why depression is significant alone, but not when paired with support and coping? I guess I don’t understand why all 3 predictors aren’t significant in Regression 1. If depression weren’t a predictor of adherence, I’d assume it’s a suppressor.

 

Any help with this (admittedly low-level) question would be greatly appreciated.

Thanks

Joe

 

 

 

Joseph A. Bianco, Ph.D.

Assistant Professor, Research

Department of Geriatric Medicine/Gerontology

Ohio University College of Osteopathic Medicine

Athens, OH 45701

 

[SR Millis-3]

I suspect that there may be high collinearity among coping style, social support, and depression.  I'd suggest that you run collinearity diagnostics and examine the condition indexes and their associated variance decomposition proportions.

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 Mon, 8/24/09, Bianco, Joseph <[hidden email]> wrote:

> From: Bianco, Joseph <[hidden email]>
> Subject: Interpreting logistic regression models
> To: [hidden email]
> Date: Monday, August 24, 2009, 1:44 PM

> Hi Listers:
>
>  �
>
> I’m looking to build a model of
> the ways in which
> coping style, social support, and depression influence
> adherence to medication.
> Coping, social support, and depression are continuous.
> Adherence is a
> dichotomous DV (0=nonadherent, 1 = adherent). This was
> chosen b/c the meds in
> question require consistently high levels of adherence
> (>95% of the time),
> so the idea of examining degrees of adherence isn’t
> as relevant.
>
>  �
>
> All variables are significantly
> correlated (in the expected
> directions).
>
>  �
>
> In any event, I don’t have
> access to Lisrel or Amos,
> so I’m using logistic regressions. I am not sure how
> to interpret the
> results. I don’t’ understand why some variables
> lose significance
> when paired with others in the model, since the regressions
> do not seem to
> indicate suppression or mediation (at least in my limited
> understanding of
> them).
>
>  �
>
> Regression 1:�  Logistic
> Regression, forced entry:�
> I entered Coping, Support, and Depression together (block
> 1) on Adherence. Results:�
> only Coping is significant (Wald 5.21, p =.022).�
> Support p = .07, Dep = p
> .93.
>
>  �
>
> Regression 2: Logistic Regression,
> with Coping and Support
> entered in block 1. Adherence is still the DV. Results:
> Coping (Wald = 5.87, p
> = .015) and Support (4.22, p = .04) are both significant
> once Depression is no
> longer in the equation.
>
>  �
>
> Regression 3:�  Logistic
> regression. Adherence is the DV
> and Depression alone is the IV. Depression is significant.
> Wald = 4.50, p
> =.034.
>
>  �
>
> Regression 4:�  Multiple
> regression with Depression
> (continuous variable) as the DV and Coping and Support as
> predictors. � � Results:
> Rsq = .41 F(3,161) = 54.7, p <.001. Both support and
> coping are significant
> at the <.000 levels.
>
>  �
>
>  �
>
> Can anyone help me understand why
> depression is significant
> alone, but not when paired with support and coping? I guess
> I don’t
> understand why all 3 predictors aren’t significant in
> Regression 1. If
> depression weren’t a predictor of adherence,
> I’d assume it’s
> a suppressor.
>
>  �
>
> Any help with this (admittedly
> low-level) question would be
> greatly appreciated.
>
> Thanks
>
> Joe
>
>  �
>
>  �
>
>  �
>
> Joseph A. Bianco, Ph.D.
>
> Assistant Professor, Research
>
> Department of Geriatric
> Medicine/Gerontology
>
> Ohio University College of Osteopathic
> Medicine
>
> Athens, OH 45701
>
>  �
>
>
>
>
>
>
>

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