ASEC Family Weights and Family-Based Inferences

Some variables in the IPUMS CPS are family-based while others are household based, even if the variable assigned at the household level. For example, the CPS contains both measures of household income and measures of family income. However, regardless of the family a household member belongs to, every individual in a household is assigned the family income of the householder.

As far as I can tell, there is no family identifier or family sampling weight in the IPUMS CPS except for the ASEC (ASECFWT). Instead, there are individual or household weights. This can make inference about distributions of family variables quite awkward. For example, using the household weights leads to the calculation of the distribution of family income across households. This is seems like an odd quantity: what we’d like to summarize is the distribution of family income, where each family is weighted equally. However, we instead have unequal weighting of families in our inferences about the distribution according to the number of families in a household.

  • Is this interpretation of inferences about families based on household weighting correct?
  • Is there a reason the family weights are only available in the ASEC?
  • The documentation notes that “ASECFWT indicates the householder or reference person weight and is to be used for family-level analyses of ASEC data.” It’s not clear to me why the householder/reference person is used here and why this would not then equal the household weight.
  • Is this post Family weight in CPS out-of-date and IPUMS-CPS has now added the family weight variable?

I’ll answer your questions in order:

  1. Yes this is the correct interpretation. I would note that the number of multi-family households is relatively small, and a common practice is to calculate statistics for single-family households only.

  2. The family weights are currently available as unharmonized variables for the basic monthly samples in IPUMS CPS, called UH_FAMWGT_1 and UH_FAMWGT_2. However these are currently only available up through 2009. IPUMS CPS is planning to add harmonized versions of these weights eventually, but there is currently no timeline for doing so.
    These weights are also available for basic monthly samples (for all years) in the original public use microdata files available at NBER. You can merge the variable (called PWFMWGT) from the NBER files into your IPUMS extract using the process described here.

  3. You’re right that the documentation for this variable is not completely correct, and the family weight is not always equal to the person weight of the reference person. There are adjustments to the weights made to ensure the number of married men equals the number of married women. Unfortunately there is no simple way to calculate the family weights using the person weights. We’ll update this documentation in a future release.

  4. The post you linked to is not entirely correct. See answers above.

Update: the latest IPUMS CPS release now includes the unharmonized family weights for basic monthly samples through Jan 2020. The variable is called UH_FAMWGT_2.

1 Like