I’ve taken the advice in the answer to “Why lots of NIUs when using PAIDHOUR?”, and done what I wanted to do in the first place: Isolating full-time, non-hourly-paid wage&salary workers (for 2014) and then analyzing that group by INCWAGE and OCC. But I’m still struggling with how to interpret the results. I used EARNWT, as advised, and came up with a total of 32.3 million non-hourly (and 33.7 million hourly, for a total of 66.2 million FT W&S). But the numbers still seem low. I know from BLS data that there were ~139 million W&S workers last year, and even if I subtract all 26+ million PT workers there were still more than 100 million FT W&S workers. But I’m only getting about three-fifths of them. How can I translate the ORG-based numbers (and, more critically, my subgroups of interest) to the full universe – if indeed I can? Should I simply multiply everything by 1.667? Or use only percentages rather than counts? Or simply contextualize my results by saying that such-and-such a subgroup may be an undercount?

# Need guidance in interpreting outgoing-rotation data

Note that the 2014 ASEC has a split-design structure with two separate sets of income-related questions. IPUMS-CPS currently provides only the 5/8 of the sample who received income questions similar to those that have previously been on the CPS. Data from the other 3/8 of the sample (approximately 30,000 households) with the redesigned income questions should be available later this summer.

In the meantime, it appears that the publicly-available CPS data does not currently adjust the earner study weight values to account for the missing 3/8 sample. Unfortunately, we do not have an official recommendation for adjusting the EARNWT variable. You could choose to simply multiply your 2014 ORG numbers by 1.6 to approximate the missing 3/8ths of the sample. This appears to produce numbers closely in line with prior months. Alternatively, you could consider investigating the weight variables further. Specifically, the 3/8 sample was randomly chosen from respondents with MISH=2-8, while all MISH=1 respondents are included in the 5/8 sample. As you will notice, to account for this the values of WTSUPP (which have been adjusted for the missing 3/8 sample) are smaller, on average, for March 2014 respondents with MISH=1. You might be able to use the relationship of weight values between MISH groups to more accurately adjust the EARNWT values.

Hope this helps.