Following up on another question...


I would like to follow up on a question asked in august 2015, this was the response:

It is not possible to directly extract the Merged Outgoing Rotation Groupsample; however, you can keep only the ORG cases after downloading your IPUMS data extract. Outgoing Rotation Group members are those respondents that meet all of the following criteria: AGE=15+,EMPSTAT=10-12, CLASSWKR=20-28, and MISH=4 or 8. Researchers should use the earnings weight variable EARNWT when analyzing the outgoing rotation group/earner study.

Expanding the number of basic monthly samples and variables is one of the major projects for this year. Unfortunately, it is not clear at the moment how quickly IPUMS-CPS will be able to make the 1976-1988 basic monthlysamples available. In the meantime, the source data are available throughNBER.

Hope this helps.

My question is this. If I follow this and only using those respondents that fit the variables listed about in MISH 4 & 8 will I have duplicate individuals? If so, does EARNWT take care of duplicates? Or should I drop the duplicates from my analysis?

Thank you in advance.


Following the selection methods outlined in the response you quoted, you will (usually) end up with two observations of each person. One observation at MISH 4 and one at MISH 8. I say “usually” because there are a number of people who are not observed at all 8 months in sample. If you further limit your population to just MISH 4 or MISH 8 you will ensure that you have no duplicates. If you are pooling data across samples to create estimates, then I recommend restricting your sample to one of the two MISH outgoing rotation groups (4 or 8). If your estimates are meant to represent the month from which those outgoing rotation groups were drawn, then you do not need to worry about duplicates as there will be no duplicates within a single month sample. Basically, if you are showing a trend over time, for example age distribution from May, June, and July 2010, then EARNWT correctly weights the samples from those months up to the population total. However, if you want an estimate of the age distribution from the pooled population using May-July 2010, you will end up with a population total of three-times the true population, so you will need to divide the weights by 3. If you expanded that pool to May, June, and July of 2011 (when MISH 4 people will then be in MISH 8) you will end up with two observations of most of the population, potentially introducing bias.

I hope this helps.


ah, I see I forgot to mention that I am indeed trying to do a time trend but over years. From 1995 to 2015. So I would use one sample from each year or just MISH=4 & 8 from the outgoing rotation.

Sorry, I am very new at this.


For each individual year, a respondent will be included in the MORG group only once. As a result, you do not need to account for duplicates in the time trend you describe. The decision whether to use one sample per year or to pool the twelve MORG samples for an entire year is up to you. Keep in mind that if you pool the MORG samples, you will need to divide EARNWT by twelve to get the true population.

Hope this helps.