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
- /
- SAS Programming
- /
- General Programming
- /
- Specifying EDF option for longitudnal studies in ...

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

06-06-2013 11:46 AM

Dear All,

I have a quick question, while using SAS proc mianalyze for a longitudinal data with 442 clusters each of size two,what should I specify in the option ( EDF=? ),which I believe requests the adjusted degrees of freedom be used in the analysis?

**proc mianalyze parms(classvar=level)=gmparms edf=?;**

Thanks

Tasneem

Accepted Solutions

Solution

06-06-2013
02:21 PM

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

Posted in reply to TZaihra

06-06-2013 02:21 PM

Is there an error degrees of freedom output anywhere in the output you got these estimates from? I am starting think perhaps not, that everything is tested with a Wald chi-squared, which does not depend on the error degrees of freedom. Anyway, with an exchangeable error structure for your GEE, you only have 442 subjects, so that is the number I would plug into your formula. I would use 14 (number of parameters fit plus one [intercept doesn't count], ignoring scale), and obtain edf=424.

However, if I fit this in GLIMMIX, I would specify ddfm=KR2, and then use the value that comes up for denominator degrees of freedom (smallest value) minus 4.

Steve Denham

All Replies

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

Posted in reply to TZaihra

06-06-2013 01:42 PM

Hi Tasneem,

The following formula is a good start: edf= Error degrees of freedom expected if dataset were complete, using the proposed model minus Number of values imputed. You can get the first number by running a side calculation on the first imputation dataset.

Steve Denham

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

Posted in reply to SteveDenham

06-06-2013 02:09 PM

Hi Steve,

Thank you for your response. As always, it's helpful but just to make sure I understand your formula correctly I was wondering if below is what you meant:

For the longitudinal data with GEE after imputing (at two time points)with total sample size (442*2=884), I have the below model and I imputed ACT, AQLQ, ASES, BMQ thus the

**EDF = 442(or884?)-15(# of parameters estimated**, **should I count "scale"..since this is calculated by default)-4(the # of variables for which I imputed)**

Also, a point to note is that in my study there are no missing values at t1 only at time t2 there are missing values..

I would really appreciate any comments and suggestions from you as well as the community.

Thanks

Tasneem

Obs Imp Parameter Estimate StdErr

1 1 Intercept 0.0229 0.8838

2 1 Gender -0.3055 0.1828

4 1 Age Group -0.2936 0.3878

5 1 Age Group 0.5481 0.1816

7 1 SES -1.0081 0.5467

8 1 SES 0.4542 0.1854

10 1 Smoker -0.7517 0.8320

12 1 Comorb1 -0.6721 0.2127

14 1 Comorb2 0.0690 0.2645

16 1 ACT -0.1079 0.0254

17 1 AQLQ -0.1824 0.0973

18 1 BMQ 0.0226 0.0154

19 1 ASES 0.1280 0.1229

20 1 ACT*ASES -0.0390 0.0251

21 1 Scale 1.0000 0.0000

Solution

06-06-2013
02:21 PM

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

Posted in reply to TZaihra

06-06-2013 02:21 PM

Is there an error degrees of freedom output anywhere in the output you got these estimates from? I am starting think perhaps not, that everything is tested with a Wald chi-squared, which does not depend on the error degrees of freedom. Anyway, with an exchangeable error structure for your GEE, you only have 442 subjects, so that is the number I would plug into your formula. I would use 14 (number of parameters fit plus one [intercept doesn't count], ignoring scale), and obtain edf=424.

However, if I fit this in GLIMMIX, I would specify ddfm=KR2, and then use the value that comes up for denominator degrees of freedom (smallest value) minus 4.

Steve Denham

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

Posted in reply to SteveDenham

06-06-2013 03:30 PM

Thanks a lot Steve that makes much sense