[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Large dummy variable set in LSDV regressions
I have a couple of questions on Least Squares Dummy Variable (LSDV)
My dataset has product, country, and year dimensions. Each observation
is on a particular product, from a certain country, for each year from
1988 to 2003. There are about 2000 products and 180 countries. So, the
dataset is pretty large about 5.4 million observations.
The first question is on model specification. Is there a difference
between having three-way error components (e.g. product, country, and
year separately) and having two-way error components with two of the
three components merged into a single dimension (e.g. product-country
and year, or product and country-year)? Just from some casual
regressions I've run, these produce different results.
The second question is on Stata estimation of two- or three-way error
component models with large dummy variable sets. The areg command and
absorb option helps out with one of the components, but does anybody
have any tips for how to handle the other components? In my dataset, if
I have two error components and they are product-country and year, areg
combined with 15 year dummies works. If the error components are
product and country-year, I run into memory problems.
Any help would be much appreciated!
* For searches and help try: