Turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Re: Why is there no power estimate for the last category of categorical variable...

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

Highlighted

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Apr 17, 2019 5:29 PM
(3324 views)

In Evaluate Design > Power Analysis, why is there no power estimate for the last category of categorical variables? For example lets lay you have a categorical variable X with three categories. The resulting power analysis for an experiment will provide a Power for X 1 and X 2 but not X 3. Why is that and how can I find out what it is?

1 ACCEPTED SOLUTION

Accepted Solutions

Highlighted

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Created:
Apr 18, 2019 6:39 AM
| Last Modified: Apr 18, 2019 6:44 AM
(3291 views)
| Posted in reply to message from LEP 04-17-2019

JMP uses the *effect parameterization* for categorical predictors in the linear predictor. That parameterization means for a k-level predictor, only k-1 parameters are estimated. All k parameters must sum to zero, so the first k-1 estimates are free but the last parameter is fixed to be -1 times the sum of the first k-1 parameter estimates.

Since there is not estimate, there is no test. If there is no test, then there is no meaning to power.

This simple case with a single categorical factor with 3 levels illustrates your situation and what I explained. Note that JMP also reports the power for the whole term as well as for individual parameters that make up a term. I find this useful because I think in terms of effects, not levels.

Learn it once, use it forever!

2 REPLIES 2

Highlighted

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Created:
Apr 18, 2019 6:39 AM
| Last Modified: Apr 18, 2019 6:44 AM
(3292 views)
| Posted in reply to message from LEP 04-17-2019

JMP uses the *effect parameterization* for categorical predictors in the linear predictor. That parameterization means for a k-level predictor, only k-1 parameters are estimated. All k parameters must sum to zero, so the first k-1 estimates are free but the last parameter is fixed to be -1 times the sum of the first k-1 parameter estimates.

Since there is not estimate, there is no test. If there is no test, then there is no meaning to power.

This simple case with a single categorical factor with 3 levels illustrates your situation and what I explained. Note that JMP also reports the power for the whole term as well as for individual parameters that make up a term. I find this useful because I think in terms of effects, not levels.

Learn it once, use it forever!

Highlighted
##

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Re: Why is there no power estimate for the last category of categorical variables?

Thanks Mark, makes sense.

For future readers, a good explanation of categorical parameterization (effect coding) that puts Mark's answer in context is here: https://www.jmp.com/en_us/statistics-knowledge-portal/what-is-multiple-regression/mlr-with-categoric...

Article Labels

There are no labels assigned to this post.