BookmarkSubscribeRSS Feed
Calcite | Level 5

PROC Mixed newbie HELP!

I am just starting out with PROC MIXED and unable to find help at my university, so hoping someone here can help!

I am setting up a model where my outcome (weight) is measured 3 times (baseline, 12m, 18m) there is so much variability with the time weight was actually measured (baseline-68months) that I am using a continuous "time since intervention" variable.

1. Do I need to center my time variable? why?

2. Is there a general rule of thumb on when to use random slope v random intercept v both random slope and intercept? When I run it with random intercept + slope the model shows no interaction between gender*time, while the random intercept only model does—any thoughts on why this might be?

My base model is:

Weight= time since intervention

and my model w covariates

weight= time since intervention +gender + age

I was going to include a time*time var since we believe people lose more weight earlier in the study- does this make sense?

When I DON’T center time this time*time var is significant vs not signif. when I do center time- any idea what this means?

i want to test time*gender to see if the rate of weight loss differs by gender and will add that in the next model. - does this make sense? does the time variable in this case need to be time*time*gender?

Any help or thoughts in general re mixed models is appreciated.

2 REPLIES 2
Super User

Re: PROC Mixed newbie HELP!

Example of the raw data. Best as data step code pasted into a text box opened on the forum with the </> icon above the message window. Only need variables used in the model if that reduces the burden.

Example of what "centering" means to you or as applied to the raw data.

The entire Proc code.

Calcite | Level 5

Re: PROC Mixed newbie HELP!

Here is an example of the data (there are more ppts, this is just the first 25 rows)

Time= time in months from intervention.

When I speak of centering, I would grand mean center the variable as follows (subtract the mean of TIME from TIME for each row):

Time_Cent= TIME- Time_MEAN;:

</>

 ID WEIGHT AGE GENDER TIME 25003 122.041 52 2 0 25003 122.002 52 2 5 25003 121.883 52 2 12 25001 121.762 56 3 0 25001 121.729 56 3 8 25001 121.691 56 3 12 25004 122.122 42 2 0 25004 122.051 42 2 7 25004 122.023 42 2 13 25009 121.79 66 1 0 25009 121.763 66 1 7 25009 . 66 1 . 25015 121.904 40 2 0 25015 . 40 2 . 25015 121.777 40 2 56 25026 121.909 56 2 0 25026 121.869 56 2 6 25026 121.843 56 2 12 25027 . 36 2 0 25027 . 36 2 . 25027 . 36 2 . 25048 121.619 48 2 0 25048 121.573 48 2 6 25048 121.568 48 2 12 25059 . 57 1 0

</>

Discussion stats
• 2 replies
• 159 views
• 0 likes
• 2 in conversation