st: Re: -areg- question

[Kit Baum <baum@bc.edu>]

<>
This is misguided. If you have a dataset of children, identified also  
by their mothers, a family with one child will have one value of 1 and  
N-1 values of 0 in that mother's dummy variable. That variable still  
has variance over the N observations, so it will not be dropped from  
the regression. However that OBSERVATION is essentially dropped from  
the regression. The FE estimator allows each unit (mother) to have her  
own constant term, and when she only has one observation, that means  
that the value of that mother's coefficient will be chosen to make  
that observation's residual precisely zero. What that means is that if  
you drop all single-child families, you will change the number of  
observations, but those observations are not contributing to the  
explanatory power of the model anyway. However, changing N will affect  
small-sample standard errors, t statistics, etc. This is the problem  
of 'singleton dummies'.

Conceptually there is no meaningful difference between running a  
regression with one dummy per mother (regardless of number of  
children) with -regress-, using -areg- to absorb the mother_id  
variable, or xtsetting the data by mother_id and running -xtreg, fe-.

Kit

On Sep 18, 2009, at 2:33 AM, Sue wrote:

> with
> fixed effects using the variable "mother_rc", there are
> obervations(children) with only one count per mother_rc, that is
> mothers with only one child. Once I include the mother fixed effects,
> those with only one child should be dropped from the regression since
> with only one child, there is no variation within the mother.


Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                              An Introduction to Stata Programming   
|   http://www.stata-press.com/books/isp.html
   An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html

*
*   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/