Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: interpreting xtmixed results


From   sjsamuels@gmail.com
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: interpreting xtmixed results
Date   Wed, 26 Aug 2009 14:43:09 -0400

Sorry that I'm late in responding to this:

You asked about multi-level models vs. survey analysis.


"3.  I am using xtmixed to analyze a 3-level model (child student
district).  However, the variance estimates at levels 2 and 3 are
almost identical.  I would assume this means that I don't need a
multilevel model and would be safe in using svyset."


 Equality of variances at different levels has nothing to do with
which levels are needed in a model.  If both variances are non-zero,
this is definite evidence  FOR  retaining both  levels.  Consult a
multi-level text to gain a better understanding or talk to a local
faculty statistician.

Whether you should use the survey analysis  or xtmixed depends on your
questions of interest.  In particular, if it is important to know to
quantify the random  variation between units at each level, for
example, as a way of judging explanatory power of fixed effects at
each level, then you should apply the hierarchical model.  Also, the
mixed model can provide information about random coefficients, not
just random intercepts. As to why -svy: reg-   was giving you smaller
standard errors: you might have unusual data; or perhaps you set up
the models incorrectly.   It's impossible to tell becaused you did not
show us your commands and output (read the FAQ!). .   The survey
analysis can answer some questions robustly, but not all.  Without
knowing your study questions, I cannot advise you further.  I would
present both approaches.

Of course, you can have both multi-level models and survey
design-based standard errors. In Stata,  the -gllamm- command ("findit
gllamm") will fit multi-level models with probability weights and
clusters.  If your school has the HLM software package, the analyses
might be e  If you wish to estimate the variances on each level, you
probably scale the weights.  See references in:
http://www.stata.com/statalist/archive/2009-07/msg00766.html

-Steve

-- 
Steven Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
845-246-0774
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index