Re: st: RE: Data Manipulation Question

[Phil Schumm <pschumm@uchicago.edu>]

On Sep 20, 2013, at 2:37 PM, Robert Picard <picard@netbox.com> wrote:
> This is a tricky problem if individuals can have more than one job per quarter.


That depends on whether job spells occur serially or whether they can overlap (IIRC, the OP was unclear on this detail).  If there is no overlap, and assuming that there is sufficient information to uniquely order each job by date, e.g.,

    quarter    date   individual  employer
    -------  -------  ----------  --------
      212    01jan13       1          A
      212    15feb13       1          B
      213    01apr13       1          B
               etc.

then for the purposes of identifying changes in employer IDs, it may be adequate simply to ignore the intermediate observations and retain only those at the beginning (or end) of each quarter.  You'll miss any job changes that don't span at least two quarters (e.g., A -> B -> C within a single quarter), but that might not be an important limitation.  This same approach would easily extend to the situation in which you can't order the observations within quarter (i.e., you don't have the date variable above), by simply ordering the employers within individual/quarter so that they match up with the end of the preceding quarter or the beginning of the next.

If there is overlap between job spells, then the problem is no longer well defined.  For example:

    quarter  individual  employer
    -------  ----------  --------
      212        1          A
      212        1          B
      212        1          C
      213        1          A
      213        1          D

Now, does this represent a change from B -> D, from C -> D, or no change at all (i.e., the individual's primary employer is A, he or she dropped secondary jobs B and C, and then at some later point, picked up a new secondary job with D)?  This level of detail may not matter much for the limited task at hand (i.e., identifying changes in employer IDs), but any implementation will require making a decision about how to handle situations like this.  Certainly, however, you wouldn't want to count B -> A or C -> A as job changes from 212 to 213 in the example above.


-- Phil


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/