Codes and Frequencies
Month in sample - 1962-2023
In 2016, 2018, and 2020, the original data from the Census Bureau inverts the month-in-sample values such that those records who should be in MIS 1 are coded as MIS == 5, and those in MIS 5 are coded as MIS == 1, who should be in MIS 2 are coded as MIS == 6 and those min MIS 6 are coded as MIS == 2, etc. The harmonized variable MISH corrects this inversion.
Unicon Corporation made the following notes about this variable:
Rotation groups 4 and 8 are not in the March Income sample in 1962. Variables in columns P133-P172 do not contain valid data for these two groups. (per Census docs) Comparing these values to the Feb-Mar match variable (match) it appears that the month-in-sample value for 1962 is actually the MIS value for February, not March. MIS groups 4 and 8 show up as non Feb-Mar matches. If the March MIS value were listed here, it would show MIS groups 1 and 5 as non Feb-Mar matches. Rotation groups 4 and 8 are not in the 1964-1965 files. In the 2001S and 2002 files, the original mis values were not the values used in the creation of sample weights. The Census provided patch files with the corrected mis values for these two years. As part of the January 2003 changes in the weighting scheme of the CPS, the Census Bureau changed the way it weights rotation groups. Instead of preserving the eight groups at the state level, the Bureau now pairs rotation groups as follows: 1 & 5 2 & 6 3 & 7 4 & 8 As a result, if one compares distribution across the eight rotation groups (using H-MIS) by state in the 2002 and preceding March ASEC files with the same distribution in 2003 and 2004, significantly more apparent variation is seen in the distribution in 2003 and 2004. This apparent variation disappears when the eight rotation groups are collapsed into the four paired groups. New state-by-state calculations are shown in the Appendix AA. The new pairing of the groups was part of the overall redesign of CPS weighting. The pairing allowed the Census Bureau to improve weighting routines by increasing cell sizes, thus yielding better results when controlling for detailed demographic distributions at the state level.
- All households.
|1962 – 2000||-||-||X||-||-||-||-||-||-||-||-||-|
|2003 – 2023||-||-||X||-||-||-||-||-||-||-||-||-|