This user update is described on our Web site (www.census.gov) as:
Technical Note on Same-Sex Unmarried Partner Data From the 1990 and 2000 Censuses
The release of data in the SF 1 files from the 2000 census has brought with it a number of analyses documenting change that has occurred since the last census was conducted in 1990. While many of the variables and processes between the two censuses are comparable, some are not, and direct comparison of some estimates may lead to misleading conclusions. This note discusses one such topic, that of "unmarried partners," and advises that for some analyses - those involving unmarried same-sex partners - direct comparison of the 1990 and 2000 estimates is not substantively valid.
The household relationship item in both the 1990 and the 2000 censuses offered many ways of identifying how other people in the household were related to the householder (the person in whose name the house is owned or rented). Categories included ""spouse"", child or other relative of the householder, housemate/roommate, roomer/boarder, and unmarried partner. In all circumstances, the respondent was asked to choose the category that best represented how other members of the household were related to the householder.
In both censuses, the "spouse" and "unmarried partner" response categories were defined and asked the same way. However, there were important differences in data processing that mean that some of the data are not comparable, limiting the usefulness of comparisons of the number of same-sex unmarried partners between these two censuses.
In both censuses, if a person was identified as the "spouse" of the householder and was the same sex as the householder, the "spouse" response was flagged for further review and allocation, that is, assignment of a value other than that originally reported, based on other data on the form. In 1990, the edit and allocation procedures did not allow same-sex "spouse" combinations to occur, thus resulting in the allocation of one of these two items in order to achieve editing consistency among the responses.
Processing steps were changed for Census 2000 for households that contained same-sex "spouses." If the person with the "spouse" category was the same sex as the householder and if neither person had their sex previously allocated, a relationship response of "spouse" was allocated as an "unmarried partner" response. Since marital status was no longer on the short form, its given value could not be considered (or modified) in this allocation procedure as it had been in 1990.
Data allocation is a standard statistical practice that is followed by most data collection agencies. Data on the relationship item (as other items) were subject to allocation in the census, as they are in virtually all Census Bureau surveys. In 1990, the marital status item was available on the 100 percent (short) form and aided in both the evaluation of the consistency of responses between the householder and the "spouse," and in the subsequent allocation procedure. The 1990 procedure allocated responses via a statistical model that distributed allocated responses from answers given by respondents in a proximate geographic area. This procedure used key demographic data from the census form, including marital status, as stratifying factors to provide a reasonable distribution of allocated responses. This procedure, while ensuring that no same-sex "spouse" response could be subsequently allocated, produced a set of allocated responses that could have included an "unmarried partner" response as well as any other response that was consistent with the age/sex/marital status profile of the respondent. This would include being allocated as a sibling or a relative, for example, or if the age differences were far enough apart (15 or more years), even a parent or child of the householder.
Three principal factors affected our decision to take this approach for Census 2000.
1. Same-sex "spouse" responses were flagged as invalid to comply with the 1996 Federal Defense of Marriage Act (H.R. 3396) passed by the 104th Congress. This act instructs all federal agencies only to recognize opposite-sex marriages for the purposes of enacting any agency programs. In order for Census Bureau data to be consistent with this act and the data requirements of other federal agencies, same-sex "spouse" responses were invalidated. The legislation defines marriage and "spouse" as follows:
"In determining the meaning of any Act of Congress, or of any ruling, regulation or interpretation of the various administrative Bureau's and agencies of the United States, the word marriage means only a legal union between one man and one woman as husband and wife, and the word "spouse" refers only to a person of the opposite sex who is a husband or wife."
In order for the Census Bureau to be consistent with this act and the data requirements of other federal agencies, same-sex "spouse" responses were invalidated.
2. The second issue was statistical in nature. The principal basis of any good statistical allocation routine rests on the selection of the stratifying or input factors to provide a good statistical model. Without marital status data on the 100 percent form in Census 2000, the allocation routine would be relatively weak. Since many partners are roughly the same age, a statistical routine without marital status as one of its factors would have likely resulted in an overestimate of adult siblings or relatives, as the majority of people living in households are relatives, and this is the population from which we would draw our allocated responses. Additionally, if the same-sex partners were more than 15 years difference in age, the statistical routine would have likely allocated the invalidated "spouse" response as either a "child" or "parent" of the householder, as these types of relatives predominate in households in this age range of differences. This was an unacceptable outcome, as it would actually destroy the intent of the original "spouse" response, which clearly indicated a nonparental type of relationship. It should be noted that the "spouse" response on the form is assumed to be deliberate - not accidental - as it was the first response category on the question and was not placed between other possible response categories that may have been meant to be marked, such as housemates or roomers.
3. The third factor took into consideration that couples in long term same-sex relationships may consider themselves as "married partners" and thus respond as such on the census form. In addition, at the time of writing the editing program for Census 2000, there were several challenges in the courts concerning the legality of same-sex marriages. Clearly, we could not ignore the fact that same-sex spouse responses were going to be recorded during Census 2000. In light of these social and legal aspects - and the lack of a key variable in the statistical allocation routine (marital status) - the assignment of same-sex "married" couples to the same-sex "unmarried partner" category was the procedure chosen for the editing process. We were adverse to a randomized allocation of these responses after people had clearly marked a close relationship preference on the census form.
As a result of these changes in the processing routine, estimates of same-sex unmarried partners are not comparable between the 1990 and 2000 census. We believe 2000 census estimates of this category are better estimates than those produced in 1990. It should also be noted that estimates of opposite-sex unmarried partners, however, were not affected by these editing procedures and changes and are comparable between the two censuses. For further information on this topic, please contact the Fertility and Family Statistics Branch on 301-457-2416.
July 2001
