A mixed model for trend analysis

Posted on 19 December 2007


I am working on my first application of PROC MIXED on actual data. I offered to help a professor with trend analysis for a longitudinal study on utterances of infants/toddlers from different linguistic and geographic groups. The n for each group is different and there is attrition at each step. I asked her and her assistant to send me the data on a vertical format and I have run a first unconditional means model using “age in months” as wave. I am yet to analyse the output and interpret it. I know already I had a significant effect of age (slope) and a non-significant effect of the intercept (I didn’t center age or set the lowest age as zero, so I think that is expected).

I will post the model and the SAS script of this first model in case I refer anyone to this site for help. In the meantime I am trying to figure out how to deal with the group variable (remember the participants come from different linguistic and/or geographical groups). Mainly, should I create a 3-level mixed model or should I treat “group” as a predictor in the 2-level model? I will get a consulting appointment in the new year to check I am on the right track or if the data set is even suitable for this approach.

Here is the SAS code, I’ll come back to it later to explain it.

proc mixed data=utterances covtest;
class id;
model  utt= age /solution ddfm=bw;
random intercept age/subject=id type=un;