I am unable to match observations in June and July 2026; did CPSIDP change?
There are two primary reasons that persons or households would not link from June 2026 to July 2026. The first is that they do not appear in the July 2026 sample. This could be because they were not surveyed in July 2026 because of the CPS rotating panel design (described briefly below). This could also be because they did not respond to the July 2026 survey. The second reason that persons or households would not link from June 2026 to July 2026 is recent changes to CPSID(P).
Based on your extract history, I suspect you are trying to link person records from an extract containing only July 2026 BMS data (submitted shortly after the release of these data) to an extract of June 2026 BMS records (submitted shortly after their release in June). There was a change to CPSID(P) values in between the release of the June and the July 2026 data. You don’t note how many cases you cannot link. As a point of reference, I just downloaded an extract that includes both the June and July 2026 data and about 68% of the records have a CPSDIP values that is observed twice. Only 75% of June BMS person records are eligible to be linked to the July BMS; this is an upper bound of records that might link because of non-response. Based on your description and extract history, I expect that submitting a new data extract that includes both June and July 2026 (or resubmitting your June 2026 extract to get a new file that has the latest version of the data), you will be able to link records successfully across samples.
Below is the July 13, 2026 entry from the IPUMS CPS revision history that describes the change (this change was also briefly noted in the IPUMS email newsletter sent on July 27, 2026).
Edited samples.
- In September 2025, the Census Bureau re-released revised BMS data files from January 2024 through July 2025 that contained a fix to the variable HRHHID2. See the Census Bureau explanation of the issue in the CPS Basic Monthly Footnotes (Item 24). IPUMS has now processed and released these revisions. HRHHID2 is a household identifier variable. Because person records are grouped into households based on sharing a household ID, the errors in HRHHID2 affected how household records were formed, and how person records were placed within households. Compared to the previous version of the data containing the error in HRHHID2, the revised data include fewer household records and the same number of person records. Each affected BMS sample now contains between 98 to 187 fewer household records. See the IPUMS CPS Sample Size page for more details. Users should note that most household variables within the affected samples will have different frequencies due to the decrease in total households. Similarly, person-level variables that rely on household-level variables may also have different frequencies (e.g. family interrelationship pointers).
Revisions to the Census Bureau identifiers (i.e., HRHHID2) affect IPUMS linking keys (i.e., CPSID(P) values), and have downstream implications on CPSID(P) in subsequent samples (i.e., beyond those re-released by the Census Bureau). As you likely know, the CPS is a short, interrupted panel that follows a 4-8-4 rotation pattern: dwelling units are sampled for four consecutive months, rotate into an eight-month break, are interviewed for four consecutive months, and then rotate out of the CPS. This means that persons in those households may be interviewed for the same four months in two adjacent calendar years. Therefore, persons whose HRHHID2 values were revised in the June or July 2025 BMS will have revised CPSID(P) values in June/July 2025 as well as June/July 2026.
I assume you are linking the July 2026 BMS to the version of the June 2026 BMS that predates the HRHHID2 and CPSID(P) revisions, which is why you are not able to match or link records as expected. If this were the case, the records you are not able across these two files should be persons in the second half of the CPS panel (i.e., IPUMS CPS MISH values of 5-8). You should be able to successfully link persons in the first half of the rotation pattern (i.e., MISH 1-4) as their HRHHID2 (and by extension their CPSID(P) values) should remain unchanged.
Follow up with questions or if I have misunderstood the issue you are encountering.