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
- /
- Analyzing zero-inflated near-normally distributed ...

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
- Permalink
- Email to a Friend
- Report Inappropriate Content

12-30-2014 10:55 PM

Hi all,

I have an outcome variable measuring change of numbers of unhealthy days across two time points. The data is highly zero-inflated with both negative and positive numbers . Does anyone know how to transform the data to make it normally distributed? Ideally I want to use linear regression model but there might be other models which deal better with non-parametric data like this.

Please feel free to throw in any ideas!

Thanks a lot!

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

Posted in reply to cindyforest7

12-31-2014 12:02 AM

You should look at the FMM procedure. Start with the introductory zero-inflation example:

PG

PG

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

Posted in reply to cindyforest7

12-31-2014 01:03 PM

Is the response you are looking at the difference between timepoint A and timepoint B (these might be intervals of equal length)? If so, rather than looking at differences, could you treat the values as repeated measures, and then look at the difference in the least squares means.

The reason I suggest this is because I think an observation of 3 days for A and 3 days for B resulting in a zero is fundamentally different from an observation of 90 days for A and 90 days for B, which also results in a zero.

Steve Denham

Message was edited by: Steve Denham