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
- /
- Proc mixed

- 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

09-11-2013 03:57 AM

Hey,

One newbie asks for help building a right model for my longitudinal data analysis. I cant find help in my faculty, so you're my last hope

It's my first time using SAS and building multilevel model, so my understanding is very limited.

So here is my situation:

1. There are 3 years longitudinal data, kids were tested every year same time for 3 years (0,12,24).

2. There are 4 groups: kids were grouped according to their BMI: underweight -1 (20 boys), normal weight -2 (50 boys), overweight-3 (20 boys), obese-4 (23 boys). They stayed in the same BMI group for all 3 years of study.

3. My data is already converted from wide to long.

4. My predictors (fixed) should be: biological_age, biological_age2, height, weight, lean_mass, fat_mass, physical activity.

5. Biological age, and biological age2 should be added both as fixed and random effect.

6. Do you think I should center the study start to 12 month?

7. My purpose is build group specific multilevel model for 3 skeletal sites: BMD_T, BMD_FN, BMD_LS.

8. I want to analyze the relationships between PA(physical activity) and bone parameters (BMD_T, BMD_FN, BMD_LS) in boys from all 4 groups (different body composition).

9. Basically I want group-specific model for BMD_T, BMD_FN, BMD_LS controlling for biological age, height, weight and testing for the effect of PA(physical activity).

I have some more predictors in the list and some more ideas, but first I would like to learn the basics, according to my situation. I am planning to use stepwise procedure.

So, please if someone will find some time for me to write the model and then some time to answer my few questions about that, I would appreciate it a lot! I think if I understand the way to build model and what means all these commands, I will be able to go on myself and then help others

Thank you!

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

09-24-2013 01:51 PM

First of all, give up on the stepwise. See this site for the many drawbacks.

Second, regarding point 5, why do you want to consider age and age2 (I assume that this is age*age) as both random and fixed. Not that it is wrong, but be sure of what you are doing here.

Third, don't bother centering model start, especially with only three timepoints that are equally spaced. However, I would really, really encourage centering the covariates..

Fourth, what kind of variable is PA? Ordinal, continuous, nominal? It will make a difference.

It looks like BMD_xx will be a response variable, time a repeated fixed effect, a boatload of continuous covariates, and PA a fixed categorical from over here. I would strongly suggest getting a copy of SAS for Mixed Models, 2nd ed. by Littell et al. and really looking over the chapters on analysis of covariance and repeated measures. Try fitting less complete models to your data until you are comfortable with going after all of the effects. And, oh yeah, don't go down the stepwise road. It's not very good for linear models and is absolutely death (IMO) in mixed models.

Steve Denham

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

09-26-2013 07:07 AM

In addition to Steve's points, all of which I agree with.

Why is BMI categorized? Binning continuous variables causes problems. See Harrell.

Have you corrected/noted the problems with BMI? Some of these are summarized in my post on Yahoo (very nontechnical) but there are additional problems when using it with children as good values change with age.

Your predictor values seem highly likely to be colinear. You can check this with the COLLIN option in PROC REG (since colinearity is a function only of the predictors, the incorrectness of REG doesn't matter).