Questions about implementing RELSHIPP and POVERTY variable logics in Census Data

USA
1

Hello,

We are working on recreating several IPUMS-derived variables directly in PUMS as preparation for a project using restricted ACS data. We want to ensure that our replication procedures match IPUMS logic as closely as possible, particularly for variables that rely on household relationships and poverty thresholds. I have two main questions:

  1. Household relationships used to create linked variables from MOM_LOC and POP_LOC:
    It is our understanding that the ACS PUMS RELSHIPP variable doesn’t distinguish between several relationship types needed for second and third-level parent links (parental value rule = 2 or 3, e.g., Aunt/Uncle, Cousin, Sibling-in-Law, Other Relatives n.s.). How does IPUMS derive or infer these relationship categories from the raw ACS data?

  2. We attempted to replicate POVERTY values using the adjustment factors from the IPUMS documentation link, but observed significant error distributions compared to IPUMS-provided values. The sample we tested this with was the 2023 1-Year data, limiting our pilot analyses to the state of Massachusetts, so we used the 0.407 multiplier. Can you:
    Confirm the exact formula and rounding procedures used in computing POVERTY?
    Clarify whether you use alternative or updated adjustment factors beyond those published in the documentation?

    Python Code to compute POVERTY_HEAD using the IPUMS provided logic is attached here as well

2

Thank you for your patience in awaiting a response. IPUMS creates the family interrelationship pointer variables, including MOMLOC and POPLOC, using algorithms that take into account information from multiple variables. These variables include age, sex, marital status, relationship to household head, and the values of these variables for other household members. Our algorithm uses a household search method, meaning household members are examined in turn and tested to determine whether they likely have a given relationship with each other household member. The code we use for creating these is stored in a format that is ideal for our custom data harmonization software, but it is cumbersome to use outside of the IPUMS infrastructure. I would recommend reading this paper by Gorsuch & Williams on how IPUMS created the family interrelationship pointer variables as a starting point. Feel free to follow up by email (ipums@umn.edu) if you have specific questions or if certain pieces of the code would be particularly helpful to you.

Code review is outside of the scope of IPUMS User Support. When attempting to replicate the IPUMS variable POVERTY, be sure to use the IPUMS family definitions. Families are identified by the variable FAMUNIT, which is based on the IPUMS-created family interrelationship variables. Families are identified differently by the Census Bureau versus IPUMS (note that the variable CBPOVERTY reports the poverty status assigned by Census Bureau in the original PUMS data). I assume you are referring to the adjustment factors used by IPUMS to convert total family income (FTOTINC) to 1989 dollars, found in table 2 on this page. Note that these are not state-specific—they are sample specific and do not vary by state within a sample. If you could provide a link to the estimate from IPUMS that you are unable to replicate, that may help me better assist you. I am not aware of any IPUMS publications that estimate state-level poverty rates using 2023 ACS data.