Minimum within variation for a FE model

[Steven Spivack]

Hi,

I am attempting to run a FE model with data from 2013-2015. My main IV is a continuous variable for the percentage of a hospital's patients that are Medicare (0-100%). My unit of analysis is the hospital. However, I am concerned that I do not have enough within variation in order to use a FE model for this question. Here is the xtsum output:

Variable | Mean Std. Dev. Min Max | Observations
-----------------+--------------------------------------------+----------------
MCRPCT overall | .5192264 .2027019 .0017073 1 | N = 11451
between | .1966962 .0018414 1 | n = 5842
within | .0530289 .1233522 .9151007 | T-bar = 1.96012

As you can see the within variation is minimal. However, because I am using a percentage that is limited between values of 0-100 I am curious if this limited amount of within variation is less of an issue. Any thoughts would be greatly appreciated.

Steven

[Carlo Lazzaro]

Steven:
welcome to the list.
As per FAQ, please post what you typed and what Stata gave you back. Thanks.
In the meantime: did you run the -hausman- specification test to investigate whether -fe- or -re- is good for your panel data analysis?

[Steven Spivack]

Thank you Carlo and sorry for not including the full code. Right now I am just doing some preliminary investigation into my main IV. Here is my code/output:

xtset ID year
panel variable: ID (unbalanced)
time variable: year, 2013 to 2014
delta: 1 unit

xtsum MCRPCT

Variable | Mean Std. Dev. Min Max | Observations
-----------------+--------------------------------------------+----------------
MCRPCT overall | .5192264 .2027019 .0017073 1 | N = 11451
between | .1966962 .0018414 1 | n = 5842
within | .0530289 .1233522 .9151007 | T-bar = 1.96012

I have not run a Hausman yet because I do not have all the data necessary to construct the full model. Of course, when I do obtain these data I will compare FE to RE and pooled OLS models. But my current concern is that the small level of within variation will result in astronomical SEs for the estimate of my main independent variable. I anticipate my final sample will include around 4,000 hospitals so while it is not a small dataset it is by not means so large that I do not have to worry about the impact of small within variation on my SEs.

[Carlo Lazzaro]

Steven:
as you stated, the usual drawback of -fe- specification is it that it gasps on limited within variation.
My advice is to collect the remaining data before taking a decision based also on the outcome of the -hausman- test.
As an aside, FAQ reminds that the best way to report what you typed and what Stata gave you back is via CODE delimiters (that wipes out any formatting issues). Thanks.