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
- /
- Ratio variables in regression analysis

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

03-30-2012 09:25 AM

Hi all,

Should we be using ratio variable as predictors in regression analysis? If yes, how the estimate should be interpreted?

Thanks

Accepted Solutions

Solution

a month ago

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

03-30-2012 09:46 AM

Many predictors are natural ratios and are interpreted like any other continuous measure. The place that gets tricky is if the ratio is bounded and the data are near the bound. For instance, often a % is bounded at 0 and 100. If the data are in the middle, the usual normal theory works fine (An example of that is the "ejection fraction" for the heart. It can officially be between 0 and 100%, but 60% is "normal" and it is rarely observed outside 10-90%, so we just use the standard analysis approach). If the data are near the edge, then you probably need to explore some variance stabilizing transformation (see any good regression reference).

Doc Muhlbaier

Duke

All Replies

Solution

a month ago

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

03-30-2012 09:46 AM

Many predictors are natural ratios and are interpreted like any other continuous measure. The place that gets tricky is if the ratio is bounded and the data are near the bound. For instance, often a % is bounded at 0 and 100. If the data are in the middle, the usual normal theory works fine (An example of that is the "ejection fraction" for the heart. It can officially be between 0 and 100%, but 60% is "normal" and it is rarely observed outside 10-90%, so we just use the standard analysis approach). If the data are near the edge, then you probably need to explore some variance stabilizing transformation (see any good regression reference).

Doc Muhlbaier

Duke

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

04-02-2012 07:52 AM

Doc's advice is excellent. My concern with ratio variables is that they are often bounded below by zero, but unbounded above. One way to address this is to separate the numerator and denominator in the predictor by taking the logs of both and including those as predictors. The correlation between the two predictors should (note SHOULD) cover the situation. It is even more important for ratio variables as dependent variables.

Steve Denham