Hi Iain,
Are there any resources you would suggest explaining MLM? The stuff I've
found online hasn't quite done it for me.
I understand that MLM involves accounting for variance at different
levels, but I'm not sure I quite understand how it looks.
Does the following seem right for a three-level model of normally
distributed outcomes?
Y ~ N(a_jk+BX_ijk,sigma_y2)
a ~ N(gamma_o0 + u_oj,sigma_a2)
B ~ N(gamma_r0 + u_rj, sigma_B2)
Thanks,
Jason
Iain Osgood wrote:
Hi Nicola and David,
This is a great question. I have seen a couple ways to succinctly
represent hierarchical models while staying true to the idea of
representing the stochastic and systematic parts of our model
(although of course here, what was once the systematic part know has
stochastic elements so things become a little jumbled). For one type
of representation, see the Zelig manual:
http://gking.harvard.edu/zelig/docs/ls.mixed.pdf
The other thing I have seen involves essentially merging the
systematic component into the stochastic component, for the
distribution of Y; the write out the stochastic component with
systematic component merged in for random parameters; then write out
the stochastic component with merged systematic component for the next
random parameters. No matter how you slice, that's a lot of latex.
By the way, a lot of these models are in Zelig for those who want to
try out some of these models.
Iain
On Wed, Mar 24, 2010 at 2:18 PM, Nicola Bretscher
<nicolabretscher at
gmail.com <mailto:nicolabretscher at gmail.com>> wrote:
Dear all,
How do you specify the systematic and stochastic components of a
model in Gary's notation if you are using a
multilevel/hierarchical/mixed effects model?
The model we want to use has a three level structure, comprising
both fixed and random effects. Our independent variables are
linear ability, quadratic ability and school-average ability and
dependent variable is maths self-concept.
In standard notation our model is defined as follows:
Y_ijk = alpha_jk + beta_rjk*X_rijk + epsilon_ijk
where
alpha_jk = gamma_00 + u_0j + v_0jk (intercept)
beta_rjk = gamma_r0 + u_rj + v_rjk (slope)
gamma's represent the fixed component of the intercept and slope.
u's represent the random component of the intercept and slope at
the second level. u~N(0, SIGMA_u)
v's represent the random component of the intercept and slope at
the third level. v~N(0, SIGMA_v)
epsilon is the error in the model as a whole. epsilon~N(0, SIGMA_e)
Note SIGMA is a covariance matrix.
Many thanks,
Nicola and David
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
<mailto:gov2001-l at lists.fas.harvard.edu>
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
------------------------------------------------------------------------
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l