KIM A. WEEDEN 
RECODING OCCUPATIONS

One of the, um, occupational hazards of studying occupations is that you spend a significant amount of time and effort trying to reconcile different versions of administrative occupation schemes. In order to prevent other researchers from reinventing the wheel, I've provided Stata code that implements two sets of translations: 2000 SOC into 1990 SOC, and 1980 SOC into 1970 SOC. (The 1980-1970 Stata "do" file also translates 1990 into 1970, but this is relatively trivial.)

1. The 2000-to-1990 translation files are here (in a zip file, just to keep them all together).
2. The 1980-to-1970 translation files are here.

Both translations differ from other available algorithms (e.g., those provided by the BLS) in that they do not assign a single early-year SOC code (e.g., 1990) to each later-year SOC code (e.g. 2000). Instead, if a 2000 SOC code was constructed from ten 1990 SOC codes, my algorithm will assign all ten of the 1990 codes.

The trick is to create a new data set in which each person in the original data set contributes Y records, where Y equals the number of 1990 SOC codes that contributed to his or her 2000 SOC code. Each record then receives a (sex-specific) weight that equals the proportion of people in the 2000 code that would have been coded into that record's 1990 code. The weights are, in turn, based on double-coded data collected by the Bureau of Labor Statistics.

It's not as complicated as it sounds. Trust me, or, better still, read the "readme" file that accompanies each set of Stata files.

If you use these files or the translation algorithms they implement, I would appreciate the courtesy of an acknowledgement or citation. See the readme.txt files for details.

DISCLAIMER! Although I have made every effort to provide files that are free of errors, I strongly recommend that you check my work as if it was your own. I'm not responsible for any erroneous findings that you might generate, blah blah blah. If you should happen to find any errors in these files, future users and I would very much appreciate a quick e-mail.

It goes without saying that you should think carefully about the implications of using this recoding procedure in the context of your own research. For example, if you are studying trends in occupational sex segregation, an algorithm that assigns to 1990s data sex-specific weights constructed from 1979 data will, most likely, understate the extent of desegregation.

Finally, these files are provided as a free service to the social scientific research community. They are not to be sold or used for commercial purposes.