Hi All,
I am using Stata Version 11. I am a beginner with Stata, and one with
Multilevel Modeling. My data are police arrests, which may not be
independent. For example, police officers respond to calls for service
together, they make arrests together, and they usually make more than
one arrest within a police department. This situation reflects the
fact that arrests within each police department may be more similar to
each other and not independent. We can go one-step higher within a
state and say that arrests from police departments serving the same
community population level may also be more similar to each other than
arrests from police departments serving different community population
levels. Multilevel modeling would be the appropriate statistical
technique to employ if this is true. So below is output for my DV
(police force), and for an Intercept-Only Model across level 2 units
(police departments). The way I see this output is (1) small
variability between police departments, (2) larger variability within
police departments, and (3) rho = .03 (seems trivial). So there
appears no meaningful average difference on the DV among police
departments, and I may analyze the data at level 1 (arrest cases)
using simple multiple regression. Are there some additional analyses
you would suggest before I make this leap? I have 21 predictors. What
would be the syntax using "xtmixed" or "xtreg" for testing a random
intercept and random slopes (do the DV-IV relationships vary across
pds) model? Thank you.
Best,
Frank
xtsum pforce, i(pd)
Variable | Mean Std. Dev. Min Max |
Observations
-----------------+--------------------------------------------
+----------------
pforce overall | 3.430618 .7753135 1.69 8.94 | N
= 3300
between | .1435753 3.136604 3.668141 | n
= 16
within | .7584913 1.523233 8.702478 | T-bar
= 206.25
xtmixed pforce || pd:, variance
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -3793.0922
Iteration 1: log restricted-likelihood = -3793.0922
Computing standard errors:
Mixed-effects REML regression Number of obs
= 3300
Group variable: pd Number of groups
= 16
Obs per group: min
= 22
avg
= 206.2
max
= 696
Wald chi2(0)
= .
Log restricted-likelihood = -3793.0922 Prob > chi2
= .
------------------------------------------------------------------------------
pforce | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------
+----------------------------------------------------------------
_cons | 3.365488 .0392125 85.83 0.000 3.288633
3.442343
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf.
Interval]
-----------------------------
+------------------------------------------------
pd: Identity |
var(_cons) | .0190313 .0081169 .
0082495 .0439046
-----------------------------
+------------------------------------------------
var(Residual) | .5776606 .0142495 .
5503966 .6062752
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) = 104.96 Prob >= chibar2
= 0.0000
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
