Estimating children ever born in linked historical census data when CHBORN is unavailable

USA
1

Hello,

I am working with linked historical census data and would like to estimate the total number of children ever born for men and women aged 40 and older.

I realize that this is not directly observed for all persons/years. My question is whether there is any recommended or reasonably defensible way to approximate this quantity when a direct fertility variable is unavailable.

One possible strategy I have considered is linking an individual across censuses and then taking the maximum number of co-resident own children ever observed in that person’s household across waves. But I am not sure this is appropriate, for several reasons:

  1. Older children may already have left the parental household by the next census.

  2. If births are widely spaced, siblings may never appear together in the same household at any one census.

  3. Children themselves can sometimes be linked across censuses, but not all children link successfully. When a child observed in two censuses cannot be linked, it becomes difficult to know whether this is the same child observed twice or two different children.

  4. If one child disappears between censuses, it is hard to distinguish between leaving home, death, and linkage failure.

So my main question is: when CHBORN is not available, is there any best practice in IPUMS linked historical census data for approximating completed fertility or total children ever born? For example, have people used a defensible lower-bound/upper-bound approach, an own-child-style reconstruction, or some other method?

Thank you very much.

2

You already seem to have a good grasp of some of the challenges of reconstructing a birth history using linked census data. The basic idea you have has indeed been used by other researchers. Your specific questions are beyond the scope of what IPUMS User Support can assist with, since we do not offer such tailored analytical guidance. That said, I would suggest reading the following paper, which uses linked census data to reconstruct birth histories (and child mortality histories) using 19th century linked census data: Wealth and Child Mortality in the Nineteenth-Century United States: Evidence from Three Panels of American Couples, 1850-1880 by Hacker, Dribe, and Helgertz. This paper also describes using inverse probability weighting, a standard method for adjusting for the unequal probability of being linked from one census dataset to another. You can likely find other relevant papers by forward and/or backward searching this one, or searching for demography research that involves estimating fertility using historical U.S. census data.

Note that using the IPUMS Multigenerational Longitudinal Panel, you can be fairly certain that linked individuals are the same person. The IPUMS team that constructed the MLP used restricted use census data that includes names and other identifying information to make the links. You can read more about the linking method on the linked page.