SPLOC in ATUS seems not so reliable?

SPLOC is a constructed variable to link the person’s spouse. And if the person does not have a spouse, SPLOC =0. (https://www.atusdata.org/atus-action/variables/SPLOC#description_section)

I’m using SPLOC to link the spouse of ATUS respondents and non-respondents. However, when I created a contingency table of SPLOC and marital status(MARST), I find something looks not right.

As the tables shows, SPLOC gives nearly half of the “Married - spouse present” group value 0. And in some cases in the marital status groups that a person should not have a spouse, SPLOC returns the valid line number. Although the marital status information is collected from the final CPS interview, which takes place 2-5 months before the ATUS interview, it would be too abnormal to conclude that nearly half of the couples get divorced in such a short period.

Meanwhile, another spouse pointer variable ASPOUSE from CPS seems more reliable. There are no cases that tell different story in ASPOUSE variable and MARST variable. (Due to the new user limit, I cannot upload this second image showing the table.)


Here is the contingency table that shows the reliability of ASPOUSE.

In general, discrepancies between SPLOC and ASPOUSE are expected, since SPLOC assigns spouse links based on information in MARST, RELATE, AGE, and SEX and does not use information about self-reported spouses. More specifically, there seem to be a variety of different reasons that explain cases with SPLOC==0 “no spouse in household” and MARST==01 “married, spouse present” in your observation here.

First, and most importantly, the algorithm used to generate SPLOC can only be used on household members that are present in the ATUS sample. Therefore, anyone who is a “non-respondent” will have SPLOC==0. This is why you see such a large share of your “respondents and non-respondents” sample with SPLOC==0 even though they have MARST==01. In the “respondents and household members” samples, I find that about 4% of the samples have SPLOC==0 and MARST==01.

Second, these 4% of cases, can be explained by a variety of reasons. Such as a) ASPOUSE is just missing. See, for example, CASEID==20030101031400 in the 2003 ATUS sample. This explains roughly a third of the cases I identify with SPLOC==0 and MARST==01 in the “respondents and household members” samples. b) ASPOUSE identifies a line number (via LINENO_CPS8) that is not included in the ATUS household. See, for example, CPSID==20070101070727 in the 2007 ATUS sample. The individual who has SPLOC==0 and MARST==01 has ASPOUSE==4, but there is no value of 4 in LINENO_CPS8. c) There is insufficient or inconsistent information in RELATE. This prevents the algorithm used to generate SPLOC to assign spouse location within the household. See, for example, CASEID==20070112060764 in the 2007 ATUS sample. The two individuals who self-report being each other’s spouses (and have SPLOC==0 and MARST==01), have RELATE values indicating “brother/sister” (to the household head) and “Don’t know.”

Thanks for the explanation, Jeff!

Now it’s pretty clear. It seems ASPOUSE is more suitable for nonresponse analysis.