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
- /
- SAS Programming
- /
- General Programming
- /
- F-test to compare two models

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

03-19-2014 12:29 PM

Hi guys,

I have two models that i need to compare by conducting a F-test but got no clue how

y=b0+b1x1+b2x2

y=b0+b1x1+b2x2+b3x3

I did google and came across this :

proc reg data = mydata ;

model y = x1 x2 x3 ;

test x3 = 0;

run ;

Is this something that i need to do,anyone please?

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

03-19-2014 12:53 PM

I would suggest to take some basic statistical training to learn why the **test** statement is redundant and how to interpret the analysis of variance of your model to determine if variable **x3** is useful to predict **y**.

PG

PG

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

03-19-2014 01:42 PM

my second model is actually more complicated than it shows here in my post. For example lets say it is b0+b1x1+b2X2+....+b6x6

and i know how to test the null hypothesis based on the global F test regardless of whether the individual variables (x3-x6) contribute to the model or not but are satisfactory F and p results for the second model enough to chose it over the 1st model?

so there is no way i can compare these models and based on F and p determine if the complete model is good to go or not without running them separately ?

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

03-20-2014 07:52 AM

A likelihood ratio test seems in order here. Fit the full and reduced models in PROC MIXED, and get the difference in -2 log likelihood values. This will be distributed as a chi square with degrees of freedom equal to the number of parameters "deleted" by reducing the model.

Steve Denham