I am using IPUMS CPS to collect data on Union membership, per state, per year, per certain occupations.
In doing my data analysis I am getting some peculiar results. For example, I am looking at per state per year per occupation, and for one state I am getting 40% reporting “member of a labor union” in one year, and 0% in every other year.
The variables I am using are:
Union (with earnings weight)
Occ1950
Statefip
Year
I am thinking that it might be something with the occupation code? I am collecting data from 1990-2013.
Does anyone have any ideas of what might be happening?
When I performed a quick crosstabulation, I was not able to find any states with zero union members from the March samples of the year range you mentioned. I think you are correct that the 0’s issue is related to the OCC1950 variable. Because OCC1950 has so many categories, and the sample size of the earner study in any given CPS sample is fairly small (in some year-state combinations only 100 people even asked the UNION question), it is unlikely that every occupation will be represented in every state. One approach would be to group occupations into larger categories to insure representation in each state in each year. If occupational specificity is more important, you could group states as well. If you are seeing a trend in the weighted data that seems unrealistic, try performing the analysis on the unweighted data. It may be that the 40% union membership of an occupation in one state is only 2 out of 5 actual respondents.