Daniel wrote:
>I have a question about the ``cluster'' option (common to many
>commands, but I am invoking it from the simple ``reg'' command).
>In a model where every individual i is also a part of a group j, and
>the dependent variable is determined by characteristics not only of the
>individual i but also of the group j, it seems like a regression ought
>to attempt to fit the model
>y_i = x_i b + w_j g + e_i + u_j
>where x_i are individual characteristics, w_j are group
>characteristics, e_i is an individual error term, and u_j is a group
>error term. Can anyone help me contrast an error component model like
>this (fitted either through f-GLS or ML) with what ``cluster'' does? I
>have already read William Rogers' article ``Regression standard errors
>in clustered samples'', but am still unclear.
Both f-GLS and ML procedures involve estimation of the correlation structure
of the data (i.e. elements of the covariance matrix of the data) and
regression function parameters. The option -cluster- only adjusts standard
errors and degrees of freedom for the hypothesis tests when data is
clustered. The estimates of the coefficients remain unchanged.
The section 23.14 Obtaining robust variance estimates, p.270 in the User's
Guide discusses the variance estimator used when -cluster- option is
specified.
*
* 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/
