Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: In this particular case: should I prefer clustering or a random-effects model


From   Steven Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: In this particular case: should I prefer clustering or a random-effects model
Date   Thu, 7 Jul 2011 23:31:45 -0500

Well, I see that I was too rigid about the definition of "random sample", which can apply to observations generated from a  theoretical distribution.
In a cluster-randomized design, the "random"ness is generated by the treatment assignment process. Although sample size calculations are frequently done under a parametric model, such a model is not a requirement for cluster-based standard errors or fixed-effects inference.
  

Steve
sjsamuels@gmail.com

 Again, there is no necessary connection to a theoretical distribution.  

Andrea Bennet

"But just from a theoretical point of view, I thought that a random effects model would be preferred because then I would treat the effects of "classID" as a random sample of the effects of all the classes in the full population."

There is no theoretical justification for your preference, which is apparently based on the coincidence that "random" appears in the term "random effects" and "random sampling".  "Random effects" are assumed to be generated from a  parametric distribution (e.g. Gaussian). "Random samples" arise from a designed sampling process.  There is no theoretical relation between the two.

Steve


Steven J. Samuels
Consultant in Statistics
18 Cantine's Island
Saugerties, NY 12477 USA
Voice: 845-246-0774
Fax:   206-202-4783 
sjsamuels@gmail.com



Thanks for the link and your help!

I indeed do have a cluster-randomized design (treatment intervention was on the class level).

With respect to FE. I have included fixed-effects dummies for each class. This results in dropped variables (classID dummies) and renders the treatment intervention to be insignificant. Performing a standard regression with <reg score treatment controls, cluster(classID)> is fine. Performing <xtreg score treatment controls, i(classID) mle/re> is fine too while <xtreg ... , i(classID) fe> results in dropped independent variables (which measure differences on the class level).

But just from a theoretical point of view, I thought that a random effects model would be preferred because then I would treat the effects of "classID" as a random sample of the effects of all the classes in the full population.

Best regards!

Andrea


On Jul 7, 2011, at 14:37 , Austin Nichols wrote:

> f in fact you have a cluster-randomized design, you should have
> calculated power (required sample size, minimum detectable effect
> size, etc.) in advance assuming the analysis design (pooled, FE,
> multilevel hierarchical model, etc.) to be used once data is
> collected, using e.g.
> http://www.urban.org/publications/1001394.html
> or your own custom simulations, so you should not be designing the
> analysis after the data has been collected!


*
*   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/


*
*   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   |   Site index