Direct and Indirect Measures of Cohabitation in Census Bureau Data: Who's Just Sharing Rent?

Jason Fields, U.S. Census Bureau
Martin T. O'Connell, U.S. Census Bureau

Direct and Indirect Measures of Cohabitation in Census Bureau Data: Who's Just Sharing Rent? When are cohabiting couples really unmarried partners and when are they sharing rent? Using data from Census 2000, the Current Population Survey (CPS), and the Survey of Income and Program Participation (SIPP) we address this question. The Census Bureau has a long tradition of measuring social trends with data designed to do something else. In the late 1960's and early 1970's Demographers at the Census Bureau recognized that their was an emerging trend, couples increasingly were living together outside of marriage. The Bureau researchers, using their best assumptions at the time created an indirect measure that has served reasonably well for nearly 30 years - the POSSLQ household was born. POSSLQ - Persons of the Opposite Sex Sharing Living Quarters - A POSSLQ household is defined as a household with only two members over age 15, they must be non-related, non-married, opposite-sex adults. There can be any number of household members under age 15. Now, and really for some time, this definition has left researchers suffering with unacceptable ambiguity. There are very few sanctions preventing premarital cohabitation any more, child rearing is still another issue. Now, compared with 10 or 20 years ago, it is more acceptable for people, regardless of gender, to share living arrangements for economic reasons, as housing costs have continued to climb. In many circumstances it is more acceptable and more common for men and women to live together outside of a romantic relationship. This last issue is really at the heart of the problem. How do you measure the difference between a living arrangement of a romantic or intimate nature from one that is based on convenience, economics, or more basically, not of a romantic or intimate nature? And to even further increase the challenge, how does one do this with data that was never designed to do so, and which contains no direct questions about the nature of the relationship? Background Cohabitation and its measurement by Federal agencies and surveys has continued to be a significant area of investigation (Casper and Cohen, 2000), and an important topic included in the Counting Couples Workshop held in December, 2001, to identify areas for improvement in demographic data collected by Federal agencies /1. Since 1995, the Census Bureau has included "Unmarried Partner" as a distinct tabulation category in the relationship to reference person variable. This same category was included in both the decennial censuses of 1990 and 2000. This item can also be compared with the indirect measure of cohabitation (POSSLQ) developed in the 1970's at the Census Bureau which was to measure households with "Persons of the Opposite Sex Sharing Living Quarters." This section will examine how well these direct and indirect estimates match up with each other. Data and Methods The data for this analysis comes from three sources. First, we use data from the March Current Population Survey (CPS) for the years from 1990 to 2002. This data provides continuous cross-sectional data to use for trend analysis. It also provides a significant amount of labor force, economic, and demographic data to use in show the characteristics of cohabiting couples. Second, Decennial Census data is used for 1990 and 2000. The Decennial data is useful because it has the size to provide data for relatively rare events like cohabitation, and even though it is more limited in terms of variables, it still provides some useful data to characterize cohabiting couples. For this analysis we us data primarily from the short and long form of Census 2000. Finally, data from the 1996 Survey of Income and Program Participation (SIPP) is included in the analysis. The SIPP data provides extensive detail about the household and family economy, as well as including marital and fertility histories, and detailed household relationships in wave 2. The SIPP also provides the only source of data at the Census Bureau that currently records person number and type identifiers for co-resident mothers and fathers of all respondents. This allows the identification of cohabiting parents who are both the biological parents of a given child in the household. The CPS sample is a nationally representative sample of households, and for the years included in this analysis surveys approximately 50,000 households each month. For 2001 and 2002 the CPS sample was increased to produce estimates for the State Children's Health Insurance Programs. Census 2000 is the most recent Decennial Census. This data collection is primarily a self-administered mail-out/back instrument. The Census Bureau attempts to enumerate each resident in the United States every 10 years. There are two primary data collection forms used in this process. The first is the Census Short Form, which includes basic information for each member of the household. This data primarily collected for redistricting and apportionment. Additionally for 1/6 of the population a longer instrument is used, the Census Long Form, that collects a number of additional characteristics of the housing, the household economy, and other legislatively mandated items. For the detailed portion of this analysis, data from both the Long and Short forms will be used. The SIPP is a longitudinal panel survey that re-interviews respondents every 4 months for the life of the panel. In the case of the 1996 SIPP panel the survey was conducted for 12 waves, or 4 years. The SIPP also collects detailed information on a variety of topics in supplemental data collections called topical modules. The 1996 panel included topical modules in the second wave of data collection that collected detailed marital and fertility histories, detailed household relationships, as well as education and migration histories. The initial sample for the SIPP survey was about 37,000 households. Variables A POSSLQ household is defined as a household with only two people 15 and over, who must be unmarried, not related to each other, and of the opposite gender. There can be any number of people under age 15 in a POSSLQ household. Also, the second adult in a POSSLQ household can report any "unrelated" relationship to the householder, including unmarried partner, roommate, housemate, border and other. Unmarried-partner households are identified in these data collections as a relationship to the reference person. This means that it only identifies cohabiting couples where one of the partners is the householder. This is likely not a significant limitation, though it does mean that the total universe of cohabiting couples is not measured. Data from the SIPP can add these people back into the universe using data collected in the detailed household relationship module, additionally cohabiting couples with children are more easily identified in the SIPP, and more accurately described. Another component of the universe of all cohabiting couples that is not included in this analysis, but bears mention, are same-sex cohabiting couples. Data from the CPS and from the SIPP do not reasonably measure this population due to its size, Census 2000 is the only source of data for this population. Variables from Census 2000 One of the fundamental questions that we are trying to answer is the degree to which households identified as POSSLQ households are simply households sharing rent. To this end, for households meeting the POSSLQ definition, we will compare those who identify as unmarried partners versus other POSSLQ's by the number of bedrooms in their housing unit. This useful long-form data item will help to separate the sharing rent component of POSSLQ from the romantic/intimate component. Single-earner POSSLQ's are compared with dual-earners in each of the data sources. Additional characteristics that are included in the analysis are the marital status of the householder and partner, their education, their income, their race and ethnicity (including mixed race couples), housing characteristics, and geographic characteristics. Results In Table 1, the trend in the number of POSSLQ households is shown from 1990 to 2002 from the CPS. This estimate is further decomposed into those households with and without children under age 15. The comparable numbers from the CPS for direct measurement of the number of unmarried-partner households are shown from 1995 to 2002, and similarly shown by the presence of children under 15 years of age for comparison purposes with the POSSLQ estimates (This table differs from that on the internet in Table UC-1 /2, which shows the presence of children under 18 in the household.) In addition, Table 1 shows the estimate of unmarried partner households using census data for 1990 and 2000, and POSSLQ estimates for Census 2000. Table 1. Trends in POSSLQ and unmarried partner data from the Current Population Survey and from the Decennial Census Cohabitation has certainly been increasing during the 1990s, based on both estimates using the POSSLQ definition and using the unmarried partner estimate. Current Population Survey estimates of POSSLQ households increased from 2.9 million in 1990 to 3.7 million in 1995 to 4.7 million in 2000. Unmarried-partner estimates from the CPS, although at a lower level, also increased about 1 million from 2.6 million households in 1995 to 3.8 million households in 2000. The proportion of POSSLQ households with children under 15 in CPS data has stayed between 31 percent and 36 percent over the 1990s while the comparable proportion using the unmarried-partner definition was about 40 percent since 1995. The total estimated number of POSSLQ households in the CPS in 2000 (4.7 million) is also very close to the number of unmarried-partner households using Census 2000 data (X.X million). Estimates of the number of POSSLQ households based on Census 2000 data show a close agreement between these two different measures (POSSLQs and unmarried-partner households). While the CPS estimates of POSSLQ and unmarried-partner households in 2000 differ by about 1.1 million households, Census 2000 data show very little difference between these two differently defined populations. A further comparison can be made by looking at the specific households that report they are unmarried partner households, and compare their indirect identification as a POSSLQ household. In Table 2, we do this for the March 2000 CPS comparing the agreement between the direct and indirect measures for all households, and then separately for those with and without children under 15. Table 2. Agreement between POSSLQ and unmarried partners: March 2000 CPS and Census 2000 On average, only 67 percent of the POSSLQ households in the CPS were also classified as unmarried-partner households. What this means is that of the 4.7 million households in the CPS that were identified as cohabiting POSSLQ households, only 67 percent (3.2 million) actually reported that they were unmarried-partner households. Clearly, one-third are either misclassified based on these criteria or are reluctant to call themselves unmarried-partners. Preliminary data from Census 2000 indicate that a slightly more consistent reporting between the two differently defined household types. In households without children, these percentages decline slightly for CPS to 65 percent and also show a similar decline of a few percentage points for data. When children are present, the fit improves to 72 percent for the CPS and also improves for Census 2000. Obviously, the presence of children may promote more consistency in the replies and may indicate a more permanent "family" structure in the household that would elicit an "unmarried-partner" response. Figure 1. Venn-Diagram of CPS and Census 2000 universes. Table 3. SIPP data improve - who is sharing rent? Table 4. CPS data improvements - economics? Table 5. Census 2000 - how many bedrooms? Table 6. Logistic analysis of who's who, by data source. Results Forthcomming.