Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Logistic Regression Results Question: Large OR, Sm...

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 29, 2017 7:55 AM
(539 views)

Hello,

Having some trouble interpreting data from a logistic regression. Attempting to determine if various drugs were associated with outcome in a case-control study. Using logistic regression because I want to control for two co-variates (age and site of disease onset). For one drug, the nominal logistic spits out a huge OR, small Chisq, and no CIs (see photo). When I run the model with each co-variate individually, I do not have this problem.

Does anyone have any idea what could have gone wrong here? I have not had the same issue with the other drugs I have analyzed.

Thanks,

Dan

5 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 29, 2017 8:26 AM
(530 views)

There is a convergence problem with the estimation. JMP reports that most of the estimates are *unstable*.

It might be the 'separation problem' at work. Try changing the personality from **Nominal Logistic** to **Generalized Linear Model**, select **Binomial** for the distribution, accept the default **Logit **link function, and select both of the options, including the **Firth** adjustment. This GLM is essentially the same as the logistic regression. If that method corrects the problem, then it is likely due to separation.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 30, 2017 10:03 AM
(497 views)

Hi Mark,

Thank you for your assistance! Using the GLM does resolve the issue of unstable parameter estimates.

A little confused about where to go from here. Is it still possible to compute odds ratios for the independent variable and covariates using this model? If so, how is this done?

Appreciate the assistance!

Best,

Dan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 31, 2017 2:36 AM
(475 views)

Odds ratios are not available using the GLM like they are using the Nominal Logistic platform. You can save the model as a prediction formula and then use this formula to compute odds and finally odds ratios.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 31, 2017 11:35 AM
(459 views)

Thanks again, Mark! I was able to save the model as a prediction formula (Save Columns>Prediction Formula) but was unable to figure out where to go from here to compute odds and an OR. Could you clarify this step for me?

Eventually, I'll need to compute CIs and a p-value for my OR as well. Please let me know if there are additional steps I'll need to carry out for this part of the analysis!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Mar 31, 2017 1:00 PM
(449 views)

I can give you some direction but that is all. You now have the formula for the probability. Generally the odds are simply the Pr( outcome ) / Pr( not outcome ). The odds ratio is then the odds( under condition A ) / odds( under condition B ). Select **Help** > **JMP Help**, then search 'odds ratio.' You should find one of the top results is **Logistic Regression Models** > **Statistical Details** > **Odds Ratios** for more information.

I can't help you with the confidence interval or p-values for these results. You will have to consult a logistic regression reference such as:

- Agresti, Alan (2013)
*Categorical Data Analysis*, Third Edition, John Wiley & Sons. - Hosmer, David W., Stanley Lemeshow, and Rodney X. Sturdivant (2013)
*Applied Logistic Regression*, Third Edition, John Wiley & Sons.

or some other source.

Learn it once, use it forever!