I am trying to run a multilevel model with GLLAMM. The data consist of many
(i.e., 50 - 100) repeated measures on participants plus baseline
demographic and trait measures. From what I have read about these analyses,
some advocate to context center and some advocate grand mean centering the
variables and then including (or not depending on theory) a level-2 mean of
the level-1 centered variables. In my analyses, I am frequently getting a
correlation between the random intercept and slope of -1. I have several
questions.
1. I am wondering if I am doing something wrong or if the -1 correlation
signifies that the intercept and slope terms are redundant. That is, should
I just be running a random intercept model. I have run a series of analyses
to look at this. I notice that the correlation is always high (e.g., .90).
It becomes -1 if either I add additional level 2 predictors or if I use the
context centered repeated measures predictor.
2. On a related note, I notice that there is a nocor option to set this
correlation to zero and am wondering when it is appropriate to do so.
3. If I have gender as a level 2 predictor do I need to use this "s" option?
If so how?
s(eqname) specifies that the log of the standard deviation (or coefficient
of variation) at level 1 for normally (or gamma) distributed responses
should be given by the linear predictor defined by eqname. This is
necessary if the level-1 variance is heteroscedastic. For example, if
dummy variables for groups are used, different variances are estimated
for different groups.
4. If using the gamma family with the canonical link function. Is
interpretation of the signs of the slope coefficients opposite to the
direction of the relationship. That is, given it is a reciprocal link, does
a negative coefficient actually signify a positive relationship between the
variables? Is it reasonable to use an identity link instead?
5. Finally, being new to this type of analysis, I was wondering if anyone
could comment on the relative strengths and weaknesses of using GLLAMM
versus a program such as HLM.
Here is an example of my commands, varx and vary are the repeated measures
the remaining variables are level 2 predictors:
Gen cons=1
Eq cons: cons
Eq slope: varx
gllamm vary varx varz varw ,i(id) family(gamma) nrf(2) eqs(cons slope)
adapt
Any assistance will be most appreciated.
Jeffrey Simons
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
