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
- /
- Glimmix procedure using a weibull distribution

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

12-12-2016 04:02 PM

I would like to analyze a multi-location complete randomized block design where resposne variable is days to maturity (of a crop). After running a proc severity the weibull distribution was selected. My questions are :

Is there any information about using the weibull distribution in the glimmix procedure, if so which is the appropriate link function?

Do I need a fully saturated model with weibull distribution as in a binomial distribution?

Accepted Solutions

Solution

12-12-2016
04:47 PM

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

12-12-2016 04:33 PM

I don't think a univariate histogram/analysis of the response variable tells you very much. The DIST= option in GENMOD and GLIMMIX specify the (conditional) error distribution, which might be very different from the distribution of the response. See the pictures in the article "Error distributions and exponential regression models."

To answer your implied question, the GLIMMIX procedure does not support a Weibull distribution function for the errors. It sounds like you are treating the response variable as a continuous variable. If you think you have long-tailed errors, you might look at the gamma or lognormal distributions. (The gamma is sometimes used for waiting times.) Look at the goodness-of-fit statistics and the diagnostic residual plots to guide you in choosing a good model.

All Replies

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

12-12-2016 04:09 PM

You say "after running PROC SEVERITY, a Weibull distribution was selected." Did you use PROC SEVERITY to run a fixed-model regression, or did you just analyze the response variable in isolation (a univariate model)?

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

12-12-2016 04:12 PM

In Isolation (Univariate model)

Solution

12-12-2016
04:47 PM

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

12-12-2016 04:33 PM

I don't think a univariate histogram/analysis of the response variable tells you very much. The DIST= option in GENMOD and GLIMMIX specify the (conditional) error distribution, which might be very different from the distribution of the response. See the pictures in the article "Error distributions and exponential regression models."

To answer your implied question, the GLIMMIX procedure does not support a Weibull distribution function for the errors. It sounds like you are treating the response variable as a continuous variable. If you think you have long-tailed errors, you might look at the gamma or lognormal distributions. (The gamma is sometimes used for waiting times.) Look at the goodness-of-fit statistics and the diagnostic residual plots to guide you in choosing a good model.

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

12-12-2016 04:16 PM

Thank you for taking the time to help Rick.

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

12-12-2016 04:52 PM

Once again thank you Rick, for pointing me in the right direction. I will take a look at your sugestions and explore both the lognormal and gamma distribution.