Tao Yang posted on Monday, April 20, 2015 - 8:15 am
Dear Dr. Muthen
I had about 70 individuals responding to a short survey each day over 10 days. I'd like to use a person-level binary variable (x) to predict day-level continuous outcome variable (y). Multilevel variance decomposition suggested that the between-person level variance of y was not significant. So I reverted to a single-level model using x to predict y and taking into accounted non-independence of observations. I used TYPE=COMPLEX and CLUSTER = personid (i.e., individual id). I did not have a sampling weight variable, assuming equal probability of selection for each observation. The output included the statement:
Variables with special functions
Cluster variable PERSONID Weight variable _WEIGHT
My questions are:
1) Do you think my specification was the right approach?
2) What does "_WEIGHT" mean in the output? How does it influence the results?