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
- /
- probing interactions

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

08-25-2016 10:20 PM

Hello,

I am trying to probe a three way interaction between predictor variables. I used PROC mixed to estimate a three level HLM. I tried to use the lsestimate statement, but it only works if the variables are in the class statement. How would I go about probing a three way interaction with variables listed only in the model statement?

Thanks!

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

08-26-2016 06:02 AM

Is there a technical meaning of "probing interactions?" Or do you just means that you want to understand the interactions?

If it is helpful, SAS provides various "slice" and "at" options that enable you to graphically display the predicted values at certain values of the explanatory variables. See the article "Use the EFFECTPLOT statement to visualize regression models in SAS."

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

08-30-2016 02:10 PM

Sometimes it is better to put variables into a CLASS statement when it may be the case that nonlinearities are hiding in there someplace. For instance, suppose you had 3 levels of a treatment, plus a control, and for further fun, let's suppose that the treatment is a fertilizer and you are measuring plant growth. Now suppose the response, in some units, are

Control 1

Low rate 1.5

Med rate 4

High rate 2

If you fit this as a continuous, linear response, you are going to miss the fact that the high dose actually brings down the response. You can try quadratic and cubic functions, but it is just a lot easier to be semi-parametric, and look at this as a categorical variable. Then with the LSMESTIMATE statement, you could fit polynomial coefficients, if that is what you are interested in. It seems to me to be a lot easier to go that way.

Steve Denham