Family versus Person level weights in CPS ASEC

Hello, I am using 2000-2007 IPUMS CPS ASEC data and I am trying to calculate federal/state income tax as well as tax credits using NBER TAXSIM program. Thus, my unit of analysis would be tax units and I plan to use the FILESTAT variable to identify the tax filing status. Given this, I have a few questiosn about weights.

  1. Should I use family-level weights (ASECFWT) or person-level weights (ASECWT)?
  2. When I apply either family or person level weights using Stata, is it correct that I should use the following code? [pweight = variable name].
  3. The replicate weights seem to be availble from 2005, and since I am using data in 2000-2004, I don’t think I should use the replicate weights. Is my understanding correct?

I’d highly appreciate your help.

  1. Neither ASECFWT nor ASECWT are designed to work with tax units and will need to be modified to run the type of analysis you mention. ASECFWT will work only in cases when the family is equivalent to the tax unit. However, since a family includes all members related to the household head, a family may also have multiple tax units within it. This report details how the census bureau attempts to identify tax units for its model. For example, a family consisting of a married couple and an adult child may have two tax units if the adult child files single rather than being listed as a dependent. You will therefore need to drop families with multiple tax units in order to run your analysis with ASECFWT. This should hopefully not pose too many issues since the majority of families will form a single unit, but I recommend keeping track of how many are dropped and what their characteristics are. Once you drop these observations, you will also want to restrict your analysis using ASECFWT to one individual per family/tax unit so as to not count each unit multiple times.

  2. Yes, you should assign them as probability weights.

  3. Yes, replicate weights are only available beginning from the 2005 ASEC sample. If you aren’t using replicate weights, please be aware that your standard errors will be less accurate. You might be interested in a set of papers by Davern et al. (2006, 2007) which propose a method to improve estimation of standard errors in ASEC data without replicate weights.

    I do want to point your attention towards the tax variables already available in the ASEC on IPUMS CPS since these are calculated by the CB in a manner similar to TAXSIM.

Thank you so much for your helpful responses!