How do I set up this mixed model in PROC MIXED?

[CATENA]

Hi: I have a data set that I would like to analyze with PROC MIXED. I have used this procedure before, but never with a model with as many levels as this one. I am hoping you can tell me how to set it up (SAS 9.3).

My data consist of enamel formation rate measurments taken sequentially at points along the tooth in three different regions (top third, middle third, and bottom third).

Each tooth is a canine tooth from a single individual, so each "subject" is represented by one canine tooth.

Sequential measurements along a tooth are chronological: measurements at the top represent the rate of enamel formation at an earlier period than measurements at the bottom.

Each subject is either a male or female and is also either a chimpanzee, gorilla, orangutan, or human.

Here is what I am interested in knowing:

1.What is the effect of subject, tooth region (top, middle, bottom thirds), sex (M/F), and species (chimp/gorilla/orangutan/human) on these measurements when each of the other factors is accounted for?

2, How do these measurments change from the top of the tooth to the bottom of the tooth in the teeth of different individuals, sexes and species?

Thanks for any help pointing me in the right direction.

Best, Catena

[SteveDenham]

Hi Cateria,

I think you are well on your way to a good analysis.  I have one question regarding the sequential nature of the data--are there successive measurements at each of the three points on the tooth over time, or is all of the time dependency captured by location on the tooth with a single date of measurement?  Given the answer to that, it should be (relatively) painless to shape your data and write a PROC MIXED or GLIMMIX script that will give you what you want.

Steve Denham

[CATENA]

Many thanks Steve:

All of the time-dependency is captured by location on the tooth. Hope this helps and thanks in advance! Best, Catena

[CATENA]

Hi Steve: One correction to the problem as I explained it to you. It turns out that the measurements were taken "opportunistically" within each of the three regions of the tooth. In other words, they are repeated measures trying to capture the same value -- the value of enamel formation within a particular tooth region (top, middle, and bottom).

I would still like to know how the values change per region, and an analyisis over the tree regions does preserve information about time as the top region forms before the middle region which forms before the bottom region. So, it would be neat to examine the lines for slope differences by sex and species, though I realize we are only looking at three time periods each represented by a particular tooth region. 

Thanks once again, Catena

[SteveDenham]

Sorry to be so late in responding.  I would try the following:

proc mixed data=yourdata;
class sex species location subject_id;
model enamel = sex|species|location / solution;
repeated location / type=un subject=subject_id;
lsmeans sex|specie|location;
run;

From here, I would start to look at whether I needed to fit separate repeated effects by sex or by species, or by both.  This can be accomplished by using the group= option in the repeated statement.  If you start down this road, you may want to shift to PROC GLIMMIX, which offers better ways of testing for covariance parameters through the use of the COVTEST statement.

Steve Denham

[CATENA]

Many thanks Steve! I will start with this and get back to you if I have any questions. I very much appreciate your thought and time. Cheers, Catena