turn on suggestions

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

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Logistic Regression: Confidence interval of predic...

Topic Options

- 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
- Permalink
- Email to a Friend
- Report Inappropriate Content

03-02-2013 01:46 PM

I've been doing a logistic regression of two variables. Dose [1,5,10,15] and response [binomial list of how many died after being given a specific treatment dose]. In PROC LOGISTIC, you can ask for confidence intervals with the l= and u= statements in the output. This results in a logistic regression model of what percentage of individuals you can expect to to die after being given a specific doseage. The output will give the confidence intervals for predicted mortality at doses 1,5,10,and 15. I'm using the events / trails syntax, so I'm using a statement like this:

proc logistic data =data plots=effect plots=ROC ;

model dead/trtsize =dose ;

output out=mortalitymeasures p=LT l=lower95 u=upper95;

run;

However, what if I'm interested in confidence intervals between the treatment doses? Specifically I'm interested in 50% mortality, which occurs between doses 5 and 10 at 6.5. I'm rusty on my logistic regression, but is it even statistically feasible to try to calculate a confidence interval for mortality at 6.5 since there were no observations at this dose, and hence no sample size to generate a confidence interval? Seems like an issue of interpolation. If it's actually appropriate look for confidence intervals for predictions between treatments, how would I go about this for something like PROC LOGISTIC or LIFEREG?

Accepted Solutions

Solution

03-03-2013
01:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Reeza

03-03-2013 01:24 PM

All Replies

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to hanson4022

03-02-2013 02:11 PM

See the example in the user guide on how to score a dataset.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to hanson4022

03-03-2013 12:22 PM

Does including in the original logistic regression a "dummy" observation with a value of DOSE=6.5 but missing values for DEAD and TRTSIZE yield a predicted value and its 95% confidence interval for that DOSE?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to 1zmm

03-03-2013 12:43 PM

Missing values are excluded from the fitting of the data and the model wouldn't be able fit because there are no dead or trtsize values. Typically, when you fit for a specific variable the others are set to the average value of the observed data.

Solution

03-03-2013
01:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Reeza

03-03-2013 01:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to 1zmm

03-03-2013 04:59 PM

You are correct. I forgot that dead/trtsize were the dependent variables.

Message was edited by: Reeza

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Reeza

03-03-2013 06:08 PM

Dead and Trtsize are the DEPENDENT variables whose predicted value PROC LOGISTIC estimates. If any of the independent variables were missing from an observation, PROC LOGISTIC could NOT estimate its predicted value.