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
- /
- Using Classification Error Costs in Building Logis...

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

03-02-2017 06:17 AM

When you use a Binary Logistic Regression Model to estimate the probability of an event you can then set a threshold value

for this probability so that if P > K then classify the case as Event = 1 otherwise classify the case as Event = 0

There are four possible decisions :

Classify the case as Event = 0 when it is in fact Event = 0

Classify the case as Event = 0 when it is in fact Event = 1 TYPE 2 ERROR [ Cost of False Negative Error = X ]

Classify the case as Event = 1 when it is in fact Event = 0 TYPE 1 ERROR [ Cost of False Positive Error = Y ]

Classify the case as Event = 1 when it is in fact Event = 1

Question:

If I can estimate the values of the Costs of making either a Type 1 Error or a Type 2 Error before I start to build the Logistic Regression Model, how can I include this information in the Proc Logistic Code in order to optimise my model with respect to

the Error Costs.

Without this inclusion, the default is to assume that the costs are all the same, which in my project is clearly not the case.

I would appreciate it if you could provide guidance w.r.t. SAS Procedures, Documentation , SAS Code as well as Published Papers / Case Studies.

Thanks

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

Posted in reply to JonDickens1607

03-02-2017 11:19 PM

It looks like you want do some Bayes analysis.

Check PEVENT= option of MODEL in PROC LOGISTIC.

or you could check PROC MCMC .