Hello IPUMS-ATUS users,
Using ATUS 2003-2019 I am trying to identify time diary respondents who are in same-sex partnerships with spsex.
Going so I find a very small proportion of respondents (around 0.5%) that are in same-sex partnerships out of all coupled respondents (have a spouse or unmarried partner in RELATE).
Is it possible that part of the reason for this has to do with CBS data edits to RELATE? It notes that:
“Starting in 2010, edits were updated to permit same-sex unmarried couples. Same-sex married couples were coded as same-sex unmarried couples. Starting in 2020, edits were updated to permit same-sex married couples.”
Am I to understand from this that only married same-sex couples can be identified prior to 2010? That is, is it true that cohabiting same-sex partners were not linked as spouses at all in years 2003-2009? And if so, is there any information on how they were coded in the relate variable in these years?
Your reply will be very much appreciated and thank you for your time!
I see two parts to your question: the first is about understanding SPSEX and changes to RELATE, the second is about the representativeness of ATUS estimates of same-sex couples. I will answer each in turn.
First, SPSEX reports the sex of the respondent’s unmarried partner or spouse (as reported in the RELATE variable). Accordingly, any edits to RELATE implemented by the Bureau of Labor Statistics (BLS) (the original ATUS data providers) will be reflected in SPSEX. The note on the IPUMS ATUS comparability tab for RELATE reflects the level of detail available in this BLS document. Looking at the data, I see cases of same-sex unmarried partners in all years, and a very small number (6 or fewer per year) of same-sex married partners prior to 2020. I have a query out to BLS for more clarification on the edits and will update this thread with any additional information I receive.
Second, to determine if ATUS estimates are reasonable, I suggest comparing to other datasets and reviewing these two Demography articles about identifying same-sex couples from ATUS with information from the CPS (Flood & Genadek 2021, Prickett et al. 2016).
I did a quick comparison of ATUS to the NHIS data. Below are the number of same-sex couples (married and unmarried) for 1997-2021 using IPUMS NHIS data. The numbers are pretty close considering the small sample sizes, particularly in the later years (using 2015/Obergefell v. Hodges as a reference for “later” here). I will also note there are survey design changes beginning with the 2019 NHIS that might be a better comparison to the random selection of the ATUS respondent.
When comparing these numbers to other sources, bear in mind that these capture coresident couples (i.e., they do not capture the LGB population or same-sex couples who do not live together) and that SPSEX is only available for respondents (e.g., it won’t identify a same-sex couple in a household with a 16-year-old ATUS respondent who has same-sex parents). Finally, while these data don’t cover your full study period of interest, you may be interested in NCHAT data.
Thank you very much for your informing and detailed reply! it is not taken for granted.
As for the first part of your reply, it seem that the BLS document provides, more or less, the same phrasing as the IPUMS ATUS RELATE description regarding the 2010 and the 2020 changes. Therefore, I thank you for putting out a query and will be very interested to hear their answer. I think that the fact that there is quite a substantial increase in same-sex partnerships after 2009 (excluding 2011) is perhaps telling.
As for estimates presented in the table, can you perhaps share the definition (or code) you used for “same-sex couple” in the ATUS data? (did you use the Prickett et al. (2016) or the Flood & Genadek 2021 method?) The estimates I get are quite a but lower than yours in most years (using my working sample, which does exclude respondents with partners, but I don’t think this can explain the differences).
Thanks again for all your help!
You are correct that the IPUMS ATUS data doesn’t reflect additional detail beyond what is in the BLS documentation. I will report back when I learn more about this. My Stata code for generating a same-sex partner flag and weighted tab of couples by year is below. I am wondering if we are using a different denominator; SPSEX is only available for respondents, so I am not including other married/coupled household members.
*look at sex distribution for householders and unmarried partners;
*not including any RELATE restrictions because SPSEX is only available for respondents;
tab sex spsex;
*SPRULE values of 21 should capture spouses and their unmarried partners only;
bys year: tab sex spsex if sprule == 21;
*SPRULE values of 11 capture direct relationships where there is only 1 potential spouse or partner. Combined with SPSEX only being available for respondents, this should capture married persons.
bys year: tab sex spsex if sprule == 11;
gen samesex = .;
replace samesex = 0 if sex != spsex & spsex != 99;
replace samesex = 1 if sex == spsex & spsex != 99;
*generate weight for weighted estimates--skipping replicate weights for now;
gen new_wt = .;
replace new_wt = wt06/365 if year <= 2019 | year >= 2021;
replace new_wt = wt20/312 if year == 2020;
svy: tab year samesex, row;
tab year samesex;
Thank you very much for your time and for sharing the code!
The differences between our estimations are due to weighting (originally, I wasn’t trying to estimate the same-sex proportion in the population but rather just looking at the actual number of same-sex cases in the data to make sure I can analyze this group so I didn’t use weights), and my selection of the working sample (dropping some respondents with spouses from the data).
I hope you hear back from BLS soon