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
- /
- Used Bootsrapping to Create Model, Now Need to App...

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
- Highlight
- Email to a Friend
- Report Inappropriate Content

08-19-2010 03:06 PM

Hi SAS Users,

I used a bootstrapping technique to create a thousand logistic regression models using 80% of my dataset. I am using the median point estimates of the beta coffecients for each variable and level (for categorical variables), thus my 1000 logreg models result in a single set of point estimates. I also computed the standard deviation for each of these point estimates.

What I want to do now is use these point estimates and std. deviations to setup proc logistic to actually use my model to assign predicted probability of the outcome for each observation in my test set. For example:

proc logistic inmodel=lib.model;

score data=lib.testset out=lib.predictions;

run;

Here's the problem:

I do not know how to actually create the lib.model dataset. Currently, I have run a single logistic regression model using the same model structure as my bootstrap-created model, used the output table from the simple model as a base, then manually changed the beta coefficients for each variable/level. Of course, this is (a) manual and therefore I don't like it and (b) presumably incorrect. This method assigns each observation a new predicted probability, but of course I don't know if it's correct since I don't know if my editing of the model specification dataset is correct.

Ultimately, what I'd like is to know how to go from having a set of beta coefficients and standard deviations of my choosing to having SAS use that data to create a logistic regression model that can be fed data from a test dataset and assign predicted probabilities to each observation.

Thank you for reading and please advise!

-A.

I used a bootstrapping technique to create a thousand logistic regression models using 80% of my dataset. I am using the median point estimates of the beta coffecients for each variable and level (for categorical variables), thus my 1000 logreg models result in a single set of point estimates. I also computed the standard deviation for each of these point estimates.

What I want to do now is use these point estimates and std. deviations to setup proc logistic to actually use my model to assign predicted probability of the outcome for each observation in my test set. For example:

proc logistic inmodel=lib.model;

score data=lib.testset out=lib.predictions;

run;

Here's the problem:

I do not know how to actually create the lib.model dataset. Currently, I have run a single logistic regression model using the same model structure as my bootstrap-created model, used the output table from the simple model as a base, then manually changed the beta coefficients for each variable/level. Of course, this is (a) manual and therefore I don't like it and (b) presumably incorrect. This method assigns each observation a new predicted probability, but of course I don't know if it's correct since I don't know if my editing of the model specification dataset is correct.

Ultimately, what I'd like is to know how to go from having a set of beta coefficients and standard deviations of my choosing to having SAS use that data to create a logistic regression model that can be fed data from a test dataset and assign predicted probabilities to each observation.

Thank you for reading and please advise!

-A.