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
- /
- is my data normal or poisson? proc mixed or glimmi...

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

05-20-2011 07:38 PM

My outcome is percent per month. Conceptually should I model this as count data or continuous data? or do I graph it and decide based on the shape of the graph? Thanks.

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

05-21-2011 01:13 AM

Whether you view your response as continuous depends on measurement properties of both the numerator and the denominator. If the denominator is the number of days in the month (which I am assuming it is), AND if the numerator is NOT limited to just a few days per month (at any point in the range - that is, whether the few days are near zero or near 15 or near 30), then you could probably treat the response as continuous. However, if either the numerator or the denominator has a restricted range, then treating the response as continuous is probably not reasonable.

How to model the response is another question entirely. You don't indicate whether the relative frequency is near 0 or near 1. If near zero, then you could probably model the response as a count with distribution Poisson or negative binomial and treat log(number of days in the month) as an offset variable. If the relative frequency is closer to 1 and you have a limited response range, then Poisson and negative binomial distributions are probably not reasonable.

In order to advise to any greater degree, we need to know more about the measurement characteristics of your data.

How to model the response is another question entirely. You don't indicate whether the relative frequency is near 0 or near 1. If near zero, then you could probably model the response as a count with distribution Poisson or negative binomial and treat log(number of days in the month) as an offset variable. If the relative frequency is closer to 1 and you have a limited response range, then Poisson and negative binomial distributions are probably not reasonable.

In order to advise to any greater degree, we need to know more about the measurement characteristics of your data.

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

05-21-2011 11:30 PM

The numerator is number of successes. The denominator is number of trials. I have this percent recorded every month for 3 years for 10 subjects.

In my preliminary data:

Outcome #1: ranges from 0% to 67% with mean of 33%.

Outcome #2: ranges from 0 to 33% with mean of 10%. Message was edited by: proctice

In my preliminary data:

Outcome #1: ranges from 0% to 67% with mean of 33%.

Outcome #2: ranges from 0 to 33% with mean of 10%. Message was edited by: proctice