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
- /
- Back-transformation where dependent variable is lo...

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

07-11-2017 11:28 PM

I have a lognormal response variable (depressive score - heavily right skewed), and I run GLIMMIX to assess the assocation between the transformed (DIST=lognormal) score and distance to roadway. I read from a couple previous posts from Mr. Steve Denham that we can get the estimates on the original scale using the forumlas (see image) from the GLIMMIX MODEL statement, DIST options. I have tried it but the back-transformed "original scae" estimates, and the confidence intervals don't make sense to me. If they are back in the original scale, the CI for row 2 that is not statistically signficiant should contain 0, but it is not the case. It looks more like a ratio than mean change. If my math is wrong, please let me know where is the problem. if it is correct, how can I interpret the back-transformed values? I am looking for the interpretation similar to something like "Xx change in the depressive score per 50 meter closer to the main road".

For example, I have the estimate, StdErr, and p-value as the following:

1) -0.0085 0.002541 0.0008

2) -0.00307 0.002115 0.1463

The corresponding back-transformed estimate (Ey) variance (VarY), lower and upper CI should be:

1) 0.99154 6.35E-06 0.98661 0.99648

2) 0.99693 4.45E-0.6 0.99280 1.00107

Thank you very much.