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
- /
- Multiple Discriminant Analysis: SAS

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

02-23-2017 04:47 AM - edited 02-26-2017 03:37 PM

Hello,

I have a non metric DV (ordinal) and both metric and non metric Independent variables and need to classify /predict group membership.

I know that PROC DISCRIM uses only continuous variables in the VAR Statement. How can I run a multiple discriminant analysis in SAS

Any help would be more than appreciated

Regards,

MS

Accepted Solutions

Solution

02-26-2017
03:19 PM

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

02-23-2017 06:23 AM

I don't know what you mean by " multiple discriminant analysis in SAS"

For predict model, most used is

1) proc logistic

2) proc hpsplit --- decision tree

Discriminant is very low powerful, and only can apply to continuous variables.

All Replies

Solution

02-26-2017
03:19 PM

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

02-23-2017 06:23 AM

I don't know what you mean by " multiple discriminant analysis in SAS"

For predict model, most used is

1) proc logistic

2) proc hpsplit --- decision tree

Discriminant is very low powerful, and only can apply to continuous variables.

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

02-23-2017 09:36 AM

Well, linear discriminant analysis requires continuous independent variables, however multiple discriminant analysis works with a metric dependent variable and non metric independent variables.

I'm looking at predicting group membership.

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

02-23-2017 11:18 AM - edited 02-23-2017 11:22 AM

mszommer wrote:

Well, linear discriminant analysis requires continuous independent variables, however multiple discriminant analysis works with a metric dependent variable and non metric independent variables.

I think you have it backwards. Multiple discriminant analysis would have several class dependent variables, and metric independent variables. However, SAS PROC DISCRIM does not perform Multiple discriminant analysis, it only works on a single dependent variable.

If you really have multiple dependent class variables, you could combine them into a single class variable encompassing all of the multiple class variables, or perhaps something like PROC PLS will work (or maybe it won't, I haven't really tried).

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

02-23-2017 11:43 AM - edited 02-25-2017 03:50 PM

Hello Paige,

I have just one dependent variable (rating of delivery: ordinal) and varied non metric (nominal, ordinal) and metric (discrete) independent variables.

I wish to classify the respondents based on their delivery rating. What test would you suggest?

Regards

MS

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

02-23-2017 12:01 PM - edited 02-23-2017 01:18 PM

mszommer wrote:

Hello Paige,

I have just one dependent variable (rating of delivery: ordinal) and varied non metric (nominal, ordinal) and metric (discrete) independent variables.

This contradicts everything you have previously stated ... now you have one dependent variable that is ordinal. Is this the correct statement of the problem now?

I wish to classify the respondents based on their delivery rating. What test would you suggest?

Did you mean to say you want to classify the respondents based upon their independent variables?

If your answer to both questions above is YES, then I think PROC DISCRIM is fine, or logistic regression would also be a tool that can be used in this situation.

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

02-23-2017 03:41 PM

mszommer wrote:*I have just one dependent variable (rating of delivery: ordinal) and varied non metric (nominal, ordinal) and metric (discrete) independent variables.*

Given the nature of your predictors, I think you should at least consider @Ksharp's suggestion to use (a recent version of) HPSPLIT. A decision tree model is easier to interpret than discriminant or logistic models and relies on fewer assumptions about the predictors.

If you have access to JMP, try the Partition platform to develop your decision tree. It's a lot easier to use.

PG

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

02-25-2017 03:50 PM

Thank you all for your response. I really appreciate it.

I did run PROC HPSPLIT, however, I have difficulties interpreting the results. Could you kindly help with it?

Also, I do not have access to JMP.

Regards

MS

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

02-25-2017 05:37 PM

I only have access to version 13.1 of SAS/STAT. From what I can read, proc HPSPLIT is much friendlier in version 14.1. I suggest you ask for help on HPSPLIT as a new topic. Good luck!

PG