Hello forum,
I am calculating the official poverty rate using the ACS data and I found a discrepancy between the calculated estimates and the published Census ACS table when using CBPOVERTY in a particular way.
According to the guidance given on the IPUMS variable page, CBPOVERTY == 0 should be excluded from the analysis as those observations are documented as missing, however if you restrict a sample under this methodology (CBPOVERTY > 0) the estimated poverty rate is too low. After some digging through the IPUMS forum I found a post from 2025 outlining the appropriate restrictions for using the CBPOVERTY variable to match published data, specifically to include GQ == 5. When I changed the restrictions from CBPOVERTY > 0 to GQ == c(0:2, 5) I was able to get the data to converge to published estimates, including (and in particular interest to me) by state. Here are my national results below:
| Year | Published poverty rate | Estimated poverty rate: CBPOVERTY > 0 | Estimated poverty rate: !GQ %in% c(3, 4) |
|---|---|---|---|
| 2024 | 12.1% | 10.3% | 12.1% |
| 2023 | 12.5% | 10.6% | 12.5% |
Note that throughout my analysis I am using an indicator variable for “poverty_1x” which is defined as 1 if CBPOVERTY < 100 and 0 otherwise to produced “Estimated poverty rate.”
I am concerned that the data that does converge to published data (right-most column) still includes households (but not group quarters) with CBPOVERTY == 0 due to how poverty_1x is defined, which diverges from the guidance in the IPUMS forum post linked above. To be clear, if I edit poverty_1x to be defined as 1 if 0 < CBPOVERTY & CBPOVERTY < 100 per the forum guidance the estimates follow the same trend as the middle column (CBPOVERTY > 0).
I’m wondering if anyone at IPUMS has any insight into how this variable is constructed and if these restrictions and approach/methodology seem appropriate.