# Replicate family structure estimates

**URL:** <https://forum.ipums.org/t/replicate-family-structure-estimates/3753>\
**Category:** CPS\
**Created:** [September 14, 2020, 7:32pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753 "2020-09-14T19:32:27Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![Emily\_Maxfield](https://avatars.discourse-cdn.com/v4/letter/e/7ea924/32.png) [@Emily\_Maxfield](https://forum.ipums.org/u/Emily_Maxfield)\
**Post date:** [September 14, 2020, 7:32pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/1 "2020-09-14T19:32:27Z")

</div>

We are trying to replicate family structure estimates found in table C3 at [America’s Families and Living Arrangements: 2019](https://www.census.gov/data/tables/2019/demo/families/cps-2019.html).

With the 2019 CPS data, we created three family structure variables: two parents, one parent and no parent. Then we created a subpop of those who completed the ASEC, and excluded group quarters and entries that are either the reference person or the spouse (as indicated below the table). Our estimates for total N match when we restrict the age to under 3, and the published estimate is within the confidence interval for two parent and one parent. However, we are underestimating the no parent number.

Does anyone have any insight into how to replicate the published family structure estimates? Thank you!

---

<div class="post-metadata">

**Author:** ![Johannes\_Fleck](https://avatars.discourse-cdn.com/v4/letter/j/50afbb/32.png) [@Johannes\_Fleck](https://forum.ipums.org/u/Johannes_Fleck)\
**Post date:** [September 15, 2020, 7:10am UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/2 "2020-09-15T07:10:22Z")

</div>

Just to clarify: Are you using the weight variables?

---

<div class="post-metadata">

**Author:** ![Emily\_Maxfield](https://avatars.discourse-cdn.com/v4/letter/e/7ea924/32.png) [@Emily\_Maxfield](https://forum.ipums.org/u/Emily_Maxfield)\
**Post date:** [September 15, 2020, 1:25pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/3 "2020-09-15T13:25:14Z")

</div>

Yes, we are using asecwt because the published estimates use ASEC data only.

---

<div class="post-metadata">

**Author:** ![Matthew\_Bombyk](https://avatars.discourse-cdn.com/v4/letter/m/34f0e0/32.png) [@Matthew\_Bombyk](https://forum.ipums.org/u/Matthew_Bombyk)\
**Post date:** [September 16, 2020, 8:40pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/4 "2020-09-16T20:40:56Z")

</div>

I noticed this note in table C2 for children living with neither parent: “Excludes children who are also in a subfamily.”

The numbers in table C2 line up by age with the numbers in table C3, so I assume this note should also apply to table C3. I’m not entirely sure how to interpret that, but it might help get you closer. The IPUMS variable [FTYPE](https://cps.ipums.org/cps-action/variables/FTYPE#codes_section) identifies individuals in subfamilies.

---

<div class="post-metadata">

**Author:** ![Emily\_Maxfield](https://avatars.discourse-cdn.com/v4/letter/e/7ea924/32.png) [@Emily\_Maxfield](https://forum.ipums.org/u/Emily_Maxfield)\
**Post date:** [October 15, 2020, 8:39pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/5 "2020-10-15T20:39:27Z")

</div>

I have an update to this question - we have been also been trying to replicate some census estimates using a similar process with IPUMS ACS data, and similarly finding that we are not able to replicate population estimates. Could this be due to differences between census and IPUMS?

---

<div class="post-metadata">

**Author:** ![Matthew\_Bombyk](https://avatars.discourse-cdn.com/v4/letter/m/34f0e0/32.png) [@Matthew\_Bombyk](https://forum.ipums.org/u/Matthew_Bombyk)\
**Post date:** [October 19, 2020, 6:19pm UTC](https://forum.ipums.org/t/replicate-family-structure-estimates/3753/6 "2020-10-19T18:19:34Z")

</div>

It is possible that the discrepancy is due to differences between the internal ACS data and the public-use sample, which are documented [here](https://ask.census.gov/prweb/PRServletCustom/YACFBFye-rFIz_FoGtyvDRUGg1Uzu5Mn*/!STANDARD?pyActivity=pyMobileSnapStart&ArticleID=KCP-2951). Because of these differences, we generally do not expect estimates from the ACS microdata to exactly replicate the official Census Bureau publications. But the numbers should be close, so if you are seeing large discrepancies, there is probably something else going on. Note that this applies to public use microdata (PUMS) files obtained either direct from the Census Bureau or from IPUMS USA, which are derived from the Census Bureau PUMS files.