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
- /
- Forecasting
- /
- Help with Proc Surveyreg and Interaction terms

- 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

02-09-2016 04:43 PM

Hi Everyone,

I am using proc surveyreg on an unbalanced panel data. Here is my model:

Y = X1 + X2 + X3 + C + X1*X2 + X1*X3 + X2*X3 + X1*X2*X3 (no intercept)

where.

X1 and X2 are dummy variables

X3 is Continous Variable

C are Control Variables

If I create percentile based on X3 (Low. medium, & high), and I want to find the interaction between X1, X2, & X3, how would I do that? More specifically, do I still include the continous variable X3 and then have dummy variables for percentile values, such LowX3, MedX3, and HighX3? What about the interaction term?

Thanks for your suggestion,

The rock

Accepted Solutions

Solution

02-10-2016
03:40 PM

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

Posted in reply to therock

02-09-2016 07:42 PM

The one i mentioned would allow different X3 slopes for different X3 percentile values. Which one is better? I would start with the model you proposed and add additional complexity such as the terms included in the model i mentioned to see if they add any significant value in terms of goodness-of-fit or validation using a holdout data. Typically, however, the simplier the model, the better.

Cheers

All Replies

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

Posted in reply to therock

02-09-2016 05:16 PM

I think you should include X3 in addition to the percentile dummies. You can test X3 to see if it is significant after inlcuding the percentile dummies. Also interations are possible predictors to be considered if data shows different slop of X3 for each percentile value.

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

Posted in reply to alexchien

02-09-2016 05:23 PM

So how would you actually model it?

Thanks!

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

Posted in reply to therock

02-09-2016 07:09 PM

you would need 2 dummy variables for the percenticle values: LowX3 and MedX3 (or whichever 2 you pick). X3 with high percentile values can be represented by setting LowX3 = MedX3 = 0.

original model

X1 + X2 + X3 + C + X1*X2 + X1*X3 + X2*X3 + X1*X2*X3

New model

X1 + X2 + X3 + C + X1*X2 + X1*X3 + X2*X3 + X1*X2*X3

+ LowX3 + MedX3

+ X1*LowX3*X3 + X1*MedX3*X3 + X2*LowX3*X3 + X2*MedX3*X3

+ X1*X2*X3*LowX3 + X1*X2*X3*MedX3

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

Posted in reply to alexchien

02-09-2016 07:15 PM

Thanks for quick reply. A question:

What would the difference be between the one you mentioned and this one:

New model

X1 + X2 + X3 + C + X1*X2 + LowX3 + MedX3

+ X1*LowX3 + X1*MedX3 + X2*LowX3 + X2*MedX3

+ X1*X2*LowX3 + X1*X2*MedX3

Which one is better?

Thanks so much!

Solution

02-10-2016
03:40 PM

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

Posted in reply to therock

02-09-2016 07:42 PM

The one i mentioned would allow different X3 slopes for different X3 percentile values. Which one is better? I would start with the model you proposed and add additional complexity such as the terms included in the model i mentioned to see if they add any significant value in terms of goodness-of-fit or validation using a holdout data. Typically, however, the simplier the model, the better.

Cheers