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
- /
- Creating a meaninful composite outcome

- 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

04-06-2017 09:11 PM

I have some exam data for some middle school students, with 2 continuous outcomes i am interested in looking at

(1) exam score

(2) time it takes to complete the exam

Is it possible to combine these to 1 meaningful outcome variable (they have different scales and better is higher for score and lower for the time)? I want to compare the effect of a predictor and also look at differences in this association by county.

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

Posted in reply to LucyB

04-07-2017 08:04 AM

LucyB wrote:

(1) exam score

(2) time it takes to complete the exam

Is it possible to combine these to 1 meaningful outcome variable (they have different scales and better is higher for score and lower for the time)? I want to compare the effect of a predictor and also look at differences in this association by county.

You will probably get many suggestions. My suggestion is to use a multivariate statistical method, such as MANOVA, which will form a linear combination (or perhaps two linear combinations) of your 2 continuous outcomes; or Partial Least Squares Regression (which also will form one or two linear combinations of your 2 continuous outcomes).

I'm also thinking that none of this is necessary in the case where you have only two response variables and two predictor. In the case where you have so few variables, it's easy to plot the variables against one another and multivariate methods are really overkill unless the two response variables are highly correlated.

--

Paige Miller

Paige Miller