Turn on suggestions

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

Showing results for

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Effect size in GLIMMIX

Options

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

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Posted 05-29-2018 11:05 AM
(3091 views)

Dear Madam, dear Sir,

I would like to estimate effect size in PROC GLIMMIX (in SAS 9.4)

I am seeking for an association between a continuous dependent variable (normal distribution) and a series of independent predictors.

Predictors consist in 1 class variable (sex) and 6 continuous variables (age, education, etc.).

Some predictors are correlated with one another but not redundant.

How could I get something which ressemble an effect size estimation (r^squared, r, Cohen's, etc.) for each of the (significant) predictors (there are at least 2 significant predictors)?

We tried to standardize all variables prior to computing the model. We do get (beta) estimates that are in what seem to be the same scale. Can this be seen as some sort of effect size measure?

Thanks in advance for your help.

Regards,

Gilles

7 REPLIES 7

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

For categorical predictor variables, the effect size can be obtained by the LSMEANS statement.

For continuous predictor variables, the effect size (which is actually the slope) can be obtained by using the SOLUTION option in the MODEL statement.

You ask for R-squared and similar measures, but these are not "effect size" measures in any reasonable usage of the term "effect size"; furthermore GLIMMIX does not compute R-squared. If you really want R-squared, then PROC GLM could be used if it fits the requirements of the problem.

--

Paige Miller

Paige Miller

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Hi,

I'm trying to get effect size values for my mixed model using PROC GLIMMIX. I used SOLUTION in the MODEL statement as suggested but I'm not sure how to interpret the output of this. Could someone please help?

Thanks,

Dave

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

@djw23 wrote:

Hi,

I'm trying to get effect size values for my mixed model using PROC GLIMMIX. I used SOLUTION in the MODEL statement as suggested but I'm not sure how to interpret the output of this. Could someone please help?

I believe I explained this in my earlier post in this thread. If that doesn't help, please be more specific, show us the output, and then state exactly what you don't understand in the output.

--

Paige Miller

Paige Miller

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Hi there,

So some background to start.

I have an dataset with several fixed effects and I'm measuring all 2 and 3-way interactions. As I have a random effect too, I'm using proc glimmix to analyse my data in a generalised linear mixed model. I also have a gamma distribution with a log link.

What I wish to do is calculate effect size for each of my fixed effects and interactions. I've tried adding the oddsratio function into my code, but this is the message i get:

NOTE: Odds ratios are computed only for the logit, cumulative logit, or the generalized logit link function.

When I use the SOLUTION function that you suggested, what I wish to know is if the effect size corresponds to the estimate column and how I should interpret it? Also, why are there zeros next to certain levels of my fixed effects?

Thanks in advance

David

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Also, why are there zeros next to certain levels of my fixed effects?

Let's suppose you have a variable that has two levels, male and female, and your analysis is of the heights (in inches) of the subjects. Then the results COULD BE that males have an effect of +5 and that females have an effect of –5. Or, the same information could be represented as males +10 and females 0; or it could be represented as males 0 and females –10. These are all the same representations of the results, there is no unique representation. So, SAS chooses one of these, using the rule that the level which is last alphabetically gets a zero.

This is why I recommend you really really really really *really really* *really really*** really really** want to look at the LSMEANS, which doesn't have this problem, and are much easier to interpret.

what I wish to know is if the effect size corresponds to the estimate column and how I should interpret it?

If the variable is categorical, don't interpret it. Use LSMEANS and interpret those. If the variable is continuous, then the effects are the slopes estimated in the model.

--

Paige Miller

Paige Miller

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

Hi,

What do you look at in the LSMEANS output to get the effect size?

Where do you find the slope for the continuous predictor in the Solution output?

Thank you for your help.

Cedric

- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content

@cedric5 wrote:

Hi,

What do you look at in the LSMEANS output to get the effect size?

Where do you find the slope for the continuous predictor in the Solution output?

Thank you for your help.

Cedric

Please start a new thread with your question, as it doesn't belong in this thread. In your new thread, please be sure to show us the output you are looking at.

--

Paige Miller

Paige Miller

Build your skills. Make connections. Enjoy creative freedom. Maybe change the world. **Registration is now open through August 30th**. Visit the SAS Hackathon homepage.

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.