Re: st: Clustering by school year

[David Kantor <kantor.d@att.net>]

At 10:03 PM 10/23/2010, Jose A wrote:
> Would clustering by school year be as simple as generating a 
> variable school_year = school identifier * year, and then using this 
> new varialbe as the cluster?

Just from a practical standpoint, this could work, provided that 
school_identifier is numeric (preferably an integer).
But you also need to assure that the values you get will constitute a 
one-to-one mapping of school_identifier and year to the resulting number.
That is, there should be no distinct pairs of school_identifier and 
year that map to the same value.
Say that you have school_identifiers 200 and 201, and your years are 
2000 and 2010. You would have,
2000 * 201 = 402000
2010 * 200 = 402000
-- thus, a many-to-one mapping.

You need to inspect your set of years and school_identifiers to see 
if something like this would happen.

If this situation arises, then you need some other scheme. You could 
extract the unique pairs of school_identifier and year that occur in 
the data. (Or, if you need to be more general, obtain the sets of 
years and school_identifier separately; form the cross-product; see 
-help cross-.) With this set, -gen long clusterid = _n-; save it, and 
later merge your analysis file to this file.

HTH
--David

----
P.S., there is another numeric-based solution: either,
 k1 * school_identifier + year
or
 k2 * year + school_identifier

where k1 or k2 are strategically chosen constants:
 k1 > max(year)
 k2 > max(school_identifier)

One possibility is
 10000 * school_identifier + year
----

*
*   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/