ArticlePDF Available

The rise of cohabitation in the United States: New historical estimates

The Rise of Cohabitation in the United States: New Historical Estimates
Catherine Fitch, Ron Goeken and Steven Ruggles
Minnesota Population Center
University of Minnesota
March 2005
Prepared for presentation at the annual meeting of the Population Association of America,
Philadelphia, March 31-April 2 2005.
Contact Catherine Fitch at
The rise of cohabitation in the United States in the late twentieth century is an important
component of the dramatic changes in marriage, family formation and childbearing (e.g.,
Bumpass, Sweet and Cherlin 1991; Bumpass and Lu 2000; Casper and Bianchi 2002; Manning
1995). This increase, first noted in the 1970s, was initially inferred from household composition
because few data sources collected direct information on couples “living together” (Glick and
Norton 1977, Glick and Spanier 1980). Research on cohabitation exploded in the 1980s as the
trend accelerated and when longitudinal data sources, such as the National Survey of Families
and Households, provided nationally representative datasets designed for studying cohabitation.
Despite a wealth of new information on cohabitating couples—including the implications
of cohabitation on family formation, relationship stability, childbearing and child well-being—
there are few estimates of cohabitation that assess the dramatic change over time (e.g., Bumpass
and Sweet 1989, Bumpass and Lu 2000). Studies that examine cross-sectional differentials, such
as race, educational attainment and place of residence, are even rarer.
This paper improves on previous attempts to infer cohabitation from the decennial
census. The 1990 and 2000 censuses included specific responses for “unmarried partner” in the
relationship question; previous censuses classified these individuals in broader
“partner/roommate” or “partner/friend” categories. Our goal is to infer as best we can which
individuals in the censuses of 1960 though 1980 would have described themselves as opposite-
sex unmarried partners if that option had been available on the census. We do this by first
developing rules to identify households that are likely contain an unmarried partner, and then by
applying a regression model to refine these measures.
POSSLQ and Adjusted POSSLQ
The acronym POSSLQ—“Persons (or Partners) of Opposite Sex Sharing Living
Quarters”—was coined by Census Bureau staff in the late 1970s. POSSLQ households—termed
“Unmarried Couple Households” by the Census Bureau—are composed of two unrelated adults
of the opposite sex (one of whom is the householder) who share a housing unit with or without
the presence of children under 15 years old. According to this definition, unmarried couple
households may contain only two adults (Casper and Cohen 2000). Scholars use POSSLQ (or
unmarried couple) households to estimate the number of cohabiting couples. The number of
unmarried couple households from the decennial censuses of 1960 to 1980 serves as one of the
few measures of cohabitation in that period (e.g., Smock and Manning 2004).
These widely-cited census statistics on the number of POSSLQ households are flawed.
The numbers cited for 1960 and 1970 derive from the census volumes on “Persons by Family
Characteristics” (U.S. Census Bureau 1964: Table 15; 1973: Table 11), and refer not to
unmarried-couple households but to the total number of persons residing with primary
individuals of the opposite sex.
This measure misses many actual POSSLQ households because
it does not allow for the presence of children of the household head who are under 15 years old.
Much more problematic, however, the measure is not a count of couples, but rather of all
individuals residing with a head of the opposite sex. Thus, for example, a household containing
an elderly female head with four male lodgers has been interpreted as four separate POSSLQ
households. These errors are easily corrected using census microdata, and the corrections have a
significant impact on the trend over time.
Primary individuals are persons residing in households with no family members and with or without non-family
Figure 1 compares the published unmarried-couple statistics (U.S. Census Bureau 2004)
with new estimates from the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al.
2004). The published numbers for 1960 and 1970 are significantly overstated compared with the
IPUMS estimates, and those for 1980, 1990 and 2000 are slightly understated. The latter
discrepancy probably arises from a difference in the source; the IPUMS numbers are tabulated
from the decennial census and the published numbers for 1970 through 2000 come from the
Current Population Survey. Using a consistent measure with a consistent source suggests that the
increase in cohabitation between 1960 and 2000 was more than 60% greater than previously
Figure 1. Comparison of published and tabulated estimates
of unmarried couple (POSSLQ) households
1960 1970 1980 1990 2000
Census year
Number of couples (000s)
Published statistics
Tabulated from IPUMS
Several authors have suggested refinements of the POSSLQ measure (Chevan 1966;
Hatch 1995; Moffit, Reville, and Winkler 1998). The most influential of these changes was
implemented by Casper and Cohen (2000), who broadened the definition to allow other adults in
the household. In particular, Casper and Cohen’s “Adjusted POSSLQ” measure permits any
number of adults related to the householder and any adult children in unrelated subfamilies.
Casper and Cohen correctly noted that the traditional measure excludes cohabitors who have
children aged 15 or older, and the adjustment was designed to capture these cases. Table 1
compares the rules for POSSLQ and Adjusted POSSLQ.
Table 1. Rules for designating POSSLQ households
a. Household must have a householder aged 15+
b. Household must include one other person aged 15+ who is unrelated and of the opposite sex as
the householder
c. Household cannot include any other persons aged 15+
ADJUSTED POSSLQ (Casper and Cohen 2000)
a. Household must have a householder aged 15+
b. Household must include one other person aged 15+ who is unrelated, not a foster child, and of
the opposite sex as the householder
c. Household cannot include any other persons aged 15+, except for relatives of the reference
person and persons listed as a child in an unrelated subfamily
Figure 2 compares POSSLQ and Adjusted POSSLQ for the period 1960 to 1980 and
compares both POSSLQ measures with opposite sex “Unmarried Partners” in 1990 and 2000.
The relationship to householder question on the census forms in 1990 and 2000 included an
explicit category for unmarried partners (see Figure 3), instead of the vaguer categories of
The census, unlike the Current Population Survey (CPS) did not classify unrelated subfamilies after 1960. To
implement Adjusted POSSLQ in the census, we therefore inferred unrelated subfamily status from the IPUMS
family interrelationship variables (Ruggles 1995). Since the CPS subfamily variables are problematic, this
implementation of Adjusted POSSLQ is probably more accurate than the CPS-based version (Ruggles and Brower
partner, friend, or roommate that were enumerated in previous census years. As shown in Figure
2, the Adjusted POSSLQ measure yields substantially higher estimates than either POSSLQ or
unmarried partners in all census years.
Figure 2. Comparison of number of cohabiting households according to POSSLQ,
Adjusted POSSLQ, and Unmarried Partner measures
1960 1970 1980 1990 2000
Census year
Number of couples (000s)
Adjusted POSSLQ
Umarried Partner
Both the POSSLQ and the Adjusted POSSLQ measures capture many persons who were
not actually cohabiting, especially in the earlier census years. Between 30 and 40 percent of
households identified by these measures in 1990 and 2000 do not include an unmarried partner.
Moreover, when we examine individual census records in earlier census years, most of the cases
do not look like cohabiting couples. Table 2 shows some typical examples, drawn from the
beginning of the 1960 census sample. Most of the POSSLQ households in this period appear to
be typical boarding or domestic service arrangements, and we have little reason to suspect that
they were actually cohabitors.
Figure 3. Census 2000 inquiry on relationship to householder
Table 2. Examples of POSSLQ households in 1960
Relationship Age Sex Race Marital Status
Head 51 Female White Widowed
Child 11 Male White Never married
Boarder 32 Male White Divorced
Head 30 Female White Married, spouse absent
Child 12 Female White Never married
Child 11 Female White Never married
Child 1 Female White Never married
Boarder 42 Male White Widowed
Head 44 Male Black Widowed
Child 13 Female Black Never married
Child 12 Female Black Never married
Child 11 Male Black Never married
Child 8 Male Black Never married
Child 8 Female Black Never married
Child 7 Female Black Never married
Child 5 Female Black Never married
Child 3 Female Black Never married
Child 3 Male Black Never married
Employee 53 Female Black Widowed
Head 45 Female White Divorced
Employee 46 Male Black Separated
Potential Unmarried Partners
Our goal is to infer unmarried partner status in census years before it was a specific
census category. We do not attempt to uncover cohabitation that would not have been revealed
by the census in 1990 or 2000. In essence, we seek to estimate how many people acknowledged
cohabitation in each census; thus, we focus on self-identified cohabitation. Self-identified
cohabitation is analogous to the census concept of self-identified race. Like racial categories,
categories of relationships between couples, such as cohabitation and marriage, are socially
constructed. Just as the race question does not attempt to measure genetic heritage, we do not
attempt to estimate the number of legally unmarried coresident persons with a sexual
relationship. Clearly, estimation of sexual relations is impossible with census data for any period;
acknowledged cohabitation is the salient measure for study of historical change in living
Other sources, such as the National Survey of Households and Families (NSFH) and
National Survey of Family Growth (NSFG), ask multiple questions on cohabitation and uncover
higher numbers of cohabitors than are self-identified in the census (Casper and Cohen 2000).
Some of this difference may occur because cohabitors intentionally or unintentionally fail to
identify themselves as unmarried partners in the census. In other cases, people may cohabit
informally but still maintain two residences, and so are not enumerated in the same household by
the census. In the NSFH, for example, persons are counted as cohabitors if they “stay in the
household half the time or more,” regardless whether or not they also have a separate household.
Moreover, the census only identifies persons who are partners of a householder, whereas surveys
can identify persons cohabiting with other household members; the NSFH data suggest that the
census thereby misses about 3 percent of cohabitors (Casper and Cohen 2000).
Both POSSLQ and Adjusted POSSLQ identify many households without unmarried
partners, and exclude many households that do have an unmarried partner. To better identify
households likely to include an unmarried partner, we developed a new measure, which we term
potential partner. We have upper- and lower-bound definitions of potential partner, termed
maximum potential partners and minimum potential partners respectively. These measures
impose restrictions on age and marital status that do not appear in the POSSLQ measures, but
they impose no restrictions on the number of adults in the household. The minimum potential
partner measure also imposes restrictions on the relationship to householder. The definitions for
these measures appear in Table 3.
The definition of potential partners differs slightly from the POSSLQ rules. We restricted
potential partners to persons aged 17 or more. Unmarried partners under 17 are extremely rare;
exclusion of persons aged 15 or 16 eliminates a trivial percentage of cohabiting couples while
greatly reducing the number of cases falsely identified as potential partners. We also eliminated
currently married persons as potential unmarried partners. The 2000 census microdata do not
include any married persons listed as unmarried partners, presumably because of post-
enumeration editing. Because married persons were apparently ineligible to be unmarried
partners in 2000, we made them ineligible in earlier years as well.
Table 3. Rules for Identifying Potential Unmarried Partners
a. Must be age 17 or older and opposite sex as the householder
b. Must be first unrelated person listed in the household
c. Neither householder nor partner can be currently married
a. Must be age 17 or older and opposite sex as the householder
b. Must be first unrelated person listed in the household
c. Must be listed as partner, roommate, or friend
d. Neither householder nor partner can be currently married
For the lower-bound minimum potential partner measure, we also restricted the
relationship categories for potential partners. POSSLQ and Adjusted POSSLQ allow any
unrelated person to be a partner, as long as they are at least fifteen and of the opposite sex as the
householder. We consider it unlikely that a true unmarried partner prior to 1990 would have been
enumerated as an employee, boarder, or lodger. We therefore restricted the minimum potential
partners to the categories of partner, friend, housemate, or roommate (see the Appendix for
details on the census relationship question in each census year).
If cohabitors in the period from 1960 through 1980 frequently identified themselves as
lodgers or domestic servants, the minimum potential partner measure would understate the
cohabitation in that period. If, however, we are correct in thinking that few people listed as
lodger or employee in 1960, 1970, or 1980 would have identified themselves as unmarried
partner if that response had been offered on the census form, then the minimum potential partner
measure is justified.
In one respect, we broadened the POSSLQ definition. A substantial number of unmarried
partners reside in households that contain multiple adults, and the prohibition of such households
may significantly bias the characteristics of cohabitors. We therefore have no restriction on the
number of related or unrelated adults in the household. We did, however, impose a rule that the
potential unrelated partner must be the first-listed person in the household who is unrelated to the
householder. Virtually all unmarried partners in 2000 (99.4 percent) were in fact the first
unrelated person listed. In the few cases with preceding unrelated individuals in the household,
we suspect that these unmarried partners are not actually unmarried partners of the householder,
but rather unmarried partners of another unrelated adult. Moreover, by imposing this rule we
limit each household to a single potential partner, which simplifies analysis.
Figure 4 compares the potential partner measures with the POSSLQ measures and with
the unmarried partner variable. Of the four measures, the original POSSLQ comes closest to
matching the number of unmarried partners in 1990 and 2000, but this apparent reliability is
deceptive. Table 4 shows the number of false positives and false negatives for each of the four
measures. False positives are households identified by each measure that do not include an
unmarried partner; false negatives are households with unmarried partners that are not identified
by each measure. The original POSSLQ fares well only because the high percentage of false
positives is canceled out by the high percentage of false negatives. The Adjusted POSSLQ
reduces the false negatives significantly, but at the price of a high rate of false positives. Our
maximum potential partner misses fewer unmarried partners than the POSSLQ measures, and
our minimum potential partner measure also substantially reduces the number of falsely
identified partners.
Figure 4. Comparison of POSSLQ, potential Partners, and unmarried partners,
1960 1970 1980 1990 2000
Census year
Number of couples (000s)
Adjusted POSSLQ
Maximum potential partners
Minimum potential partners
Umarried Partner
Table 4. Error rates for alternate inferred measures of cohabitation
a. False b. False c. Total error
positive negative (a+b)
POSSLQ 25.8 22.7 48.5
Adjusted POSSLQ 28.6 8.5 37.1
Maximum Potential Partner 28.9 5.8 34.7
Minimum Potential Partner 20.7 5.8 26.5
POSSLQ 23.4 20.0 43.4
Adjusted POSSLQ 25.9 6.2 32.1
Maximum Potential Partner 25.8 0.7 26.5
Minimum Potential Partner 16.2 0.7 16.9
Note: false positive is the percent of households identified by the measure that do
not include an unmarried partner; false negative is the percent of households with
unmarried partners not identified by measure.
Our goal for the potential partner measures was to make the definition as narrow as
possible without discarding a significant number of unmarried partners. In 1990, however, our
measures fail to identify 5.8 percent of the opposite-sex unmarried partners listed in the census
microdata. Two-thirds of these cases were excluded based on the marital status rule we imposed
for consistency with Census 2000. As noted, Census 2000 editing procedures did not permit
persons who were currently married to be listed as unmarried partners. Some of these cases, no
doubt, actually represent cohabitors, but the Census 2000 editing rule may be sound: many of
these married cohabitors probably represent coding errors. Excluding false negative cases that
were edited in Census 2000, the 1990 potential partner measures would yield just 1.9 percent
false negatives.
The remaining false negatives among potential partners in 1990 occurred because the unmarried partner
was not the first-listed unrelated person in the household or because the head or partner was under age 17. In many
of these cases, however, we suspect that the relationship may be miscoded.
Extending these measures backwards in time highlights the inappropriateness of inferring
cohabitation for persons listed as boarders, lodgers, and employees. Figure 5 shows the
percentage of all couple households (married and unmarried couple households) that are
estimated to include unmarried couples according to each of the four methods, for the entire
period 1880 through 2000. Figure 6 shows the same statistics for the period from 1880 to 1960.
According to most of the measures, the 1960 census represented the low point of cohabitation,
and cohabitation was much more common in the early-twentieth century. This is not because
cohabitation actually declined over the first half of the twentieth century. Rather, these measures
reflect that boarding, lodging, domestic service, and farm labor were much more common before
1950, and often involved residence with unrelated persons of the opposite sex (Goeken 1999).
In the period before 1940, we have access to the actual words used by enumerators to
describe living arrangements. We have recorded 28 cases listed explicitly as concubine or
mistress, several hundred companions, and several thousand partners and friends (most of whom
were apparently business partners). It may have occasionally happened that an enumerator would
record lodger or servant for a person who acknowledged cohabitation, but it was probably rare.
In all periods, the number of cohabitors erroneously enumerated as boarders or employees must
have represented a tiny minority of total persons enumerated in these categories.
Figure 5. Unmarried couples as a percentage of all couple households, 1880-2000
1880 1910 1910 1920 1940 1950 1960 1970 1980 1990 2000
Census year
Percent of all couple households
Adjusted POSSLQ
Maximum potential partners
Minimum potential partners
Figure 6. Unmarried couples as a percentage of all couple households, 1880-1960
1880 1910 1910 1920 1940 1950 1960
Census year
Percent of all couple households
Adjusted POSSLQ
Maximum potential partners
Minimum potential partners
Predicting potential partners
The potential partner measures include virtually all unmarried partners as they were
defined in Census 2000, but they also include a substantial number of households that did not
contain unmarried partners. To refine the measures further, we used binary logistic regression to
impute a set of unmarried partners probabilistically. Our strategy was to predict unmarried
partner status for potential partners in 1990 and 2000, and then to calculate the predicted
probability of being an unmarried partner for all potential partners from 1960 to 2000.
Table 5 describes the variables used in the models. We ran four models, separating male
and female potential partners in each potential partner universe (maximum and minimum
potential partners). The dependent variable in each case is a dichotomous variable indicating
whether or not a potential partner was an unmarried partner. We selected the independent
variables for their predictive power, and the relationships differ from those cited in the literature
on cohabitation. Most studies contrast cohabitors and married couples (Smock and Manning
2004); our analysis seeks to identify the characteristics that distinguish roommates or other
unrelated persons from unmarried partners. We include census year in the model, since the
overall proportion of potential partners who were actually unmarried partners increased slightly
between 1990 and 2000. We tested for interactions between census year and the other
independent variables but did not find any significant relationships.
The regression results are shown in Table 6. The strongest predictors of unmarried
partner status were age difference between householder and potential partner, number of adults
in household, presence of own children, presence of children under five, home ownership, age,
and census year.
Table 5. Independent variables for regression analyses
Variable short name Variable description
Year Census year (coded 100 for 2000, 99 for 1990, 98 for 1980 and so on)
Region Region (dummy variables: Northeast, Midwest, South and West)
Metro Households residing in a metropolitan area, as defined in each census
year (coded 1=yes and 0=no)
Owns Housing unit is owned (coded 1=yes and 0=no)
Num of Adults Number of other adults in the household in addition to householder and
potential partner and excluding adult children
Div-sep Potential partner is divorced or separated (coded 1=yes and 0=no)
HH div-sep Householder divorced or separated (coded 1=yes and 0=no)
Child Potential partner's own child(ren) present (coded 1=yes and 0=no)
HH child Householder's own child(ren) present (coded 1=yes and 0=no)
HH child5 Householder's own child(ren) under age 5 present (coded 1=yes and
Age Potential partner age groups (dummy variables; see regression)
Age diff Age difference between the householder and potential partner (dummy
variables; see regression)
School Potential partner in school (coded 1=yes and 0=no)
HH school Householder in school (coded 1=yes and 0=no)
Education Potential partner's educational attainment (dummy variables: has not
finished high school, completed high school, some college, and 4+ years
of college)
Educational diff. Difference in educational attainment levels between householder and
HH income Householder's income in 1000s--adjusted to 2000$; topcode of 150,000
(1960 topcode in 2000$); all negative values coded 0
Income diff. Difference in householder's and potential partner's income (measured as
Unemployed Potential partner is unemployed (coded 1=yes and 0=no)
HH unemp. Householder is unemployed (coded 1=yes and 0=no)
Not in labforce Potential partner is not in the labor force (coded 1=yes and 0=no)
HH not in labforce Householder is not in the labor force (coded 1=yes and 0=no)
When we use the coefficients in Table 6 to estimate the predicted number of unmarried
partners in earlier census years, the results suggest that previous estimates of cohabitation before
1990 may be dramatically overstated. Figure 7 compares our predicted number of unmarried
partners in each census year with the number of POSSLQ households. In 1980, our maximum
estimate of unmarried partners is two thirds of the Adjusted POSSLQ measure, and in 1960 and
1970 our maximum estimate is less than half that obtained from Adjusted POSSLQ. Our
Table 6. Binary logistic regressions of unmarried partner status:
potential partners, 1990-2000
Minimum potential partners Maximium potential partners
Male partners Female partners Male partners Female partner
ariables Exp(B) Exp(B) Exp(B) Exp(B)
Year 1.399 *** 1.335 ***
1.197 *** 1.114 ***
Midwest 0.942 0.944
0.927 *** 0.944 ***
South 0.816 *** 0.848 ***
0.822 *** 0.854 ***
West 0.737 *** 0.772 *** 0.785 *** 0.816 ***
Metro 0.836 *** 0.836 *** 0.809 *** 0.825 ***
Ownership 1.338 *** 1.344 *** 1.136 *** 1.153 ***
Num. of adults 0.626 *** 0.659 *** 0.582 *** 0.623 ***
Div-sep 1.057 1.023 1.057 *** 0.976 ***
HH div-sep 1.079 * 1.107 **
1.084 *** 1.134 ***
Child 1.499 0.895 1.066 0.830 ***
HH child 1.646 *** 1.975 *** 1.233 *** 1.787 ***
HH child5 1.519 *** 1.714 *** 1.586 *** 1.970 ***
Chid*div-sep 0.946 1.402 ** 1.118 ** 1.429 ***
HHchild*HHdiv-sep 0.769 *** 0.669 *** 0.784 *** 0.545 ***
20-24 1.270 * 0.883
1.284 *** 0.957 ***
25-29 1.344 ** 0.934 1.370 *** 1.031 ***
30-34 1.257 * 0.866 1.271 *** 0.977 ***
35-39 1.249 * 0.843 * 1.208 *** 0.908 ***
40-44 1.104 0.843 * 1.036 *** 0.875 ***
45-49 0.989 0.790 **
0.847 *** 0.791 ***
50-59 0.873 0.583 *** 0.647 *** 0.580 ***
60-69 0.664 ** 0.469 *** 0.462 *** 0.408 ***
70-79 0.575 *** 0.384 *** 0.372 *** 0.347 ***
80+ 0.469 *** 0.323 *** 0.285 *** 0.260 ***
Age diff.
less than -35 0.873 0.417
0.428 *** 0.268 ***
-34 thru -30 0.761 0.095 0.731 *** 0.045 ***
-29 thru -25 0.904 0.362 ** 0.839 *** 0.264 ***
-24 thru -20 0.787 0.498 ** 0.920 *** 0.386 ***
-19 thru -15 0.970 0.753 1.005 0.655 ***
-14 thru -10 1.178 * 0.816 * 1.215 *** 0.878 ***
-9 thru -5 1.072 0.913 1.079 *** 0.959 ***
4 thru -2 1.042 0.939
1.032 *** 0.960 ***
-1 thru 1
2 thru 4 0.906 * 1.135 **
0.890 *** 1.105 ***
5 thru 9 0.737 *** 1.018 0.694 *** 1.001
10 thru 14 0.580 *** 0.851 ** 0.489 *** 0.818 ***
15 thru 19 0.390 *** 0.722 *** 0.224 *** 0.622 ***
20 thru 24 0.189 *** 0.483 ***
0.074 *** 0.348 ***
25 thru 29 0.110 *** 0.452 *** 0.033 *** 0.255 ***
30 thru 24 0.082 *** 0.282 *** 0.022 *** 0.135 ***
35 + 0.082 *** 0.299 *** 0.012 *** 0.066 ***
School 0.764 *** 0.820 *** 0.760 *** 0.805 ***
HH school 0.767 *** 0.688 *** 0.792 *** 0.726 ***
Education ***
Some high school 1.623 *** 1.479 ***
1.308 *** 1.203 ***
Completed high school 1.384 *** 1.282 *** 1.184 *** 1.131 ***
Some college 1.224 *** 1.258 *** 1.142 *** 1.167 ***
4+ years of college
Educational diff. 0.960 * 0.908 ***
0.971 *** 0.929 ***
HH income 1.007 *** 1.006 *** 1.006 *** 1.004 ***
Income diff. 0.997 *** 0.998 *
0.997 *** 0.999 ***
Unemployed 1.105 0.951 1.010 * 0.895 ***
HH unemployed 0.967 0.857 * 0.955 *** 0.845 ***
Not in labforce 0.886 ** 0.965 0.897 *** 0.785 ***
HH not in labforce 0.934 0.814 *** 0.777 *** 0.916 ***
Constant 0.000 *** 0.000 *** 0.000 *** 0.000 ***
Figure 7. Predicted partners compared with POSSLQ measures, 1960-2000
1960 1970 1980 1990 2000
Census year
Number of couples (000s)
Adjusted POSSLQ
Predicted partners-maximum
Predicted partners-minimum
Figure 8. Predicted partners compared with POSSLQ measures, 1960-2000
(log scale)
1960 1970 1980 1990 2000
Census year
Number of couples (000s)
Adjusted POSSLQ
Predicted partners-maximum
Predicted partners-minimum
minimum estimates—which we consider more realistic—are so small that they are difficult to
see on the graph in the early census years.
Figure 8 presents the same data as Figure 7 on a log scale, to allow easier comparison.
Our minimum measure suggests that just 8,000 people in 1960 would have listed themselves as
unmarried partners had the category been available, 78,000 in 1970, and just over a million in
1980. Many more couples in 1960 must have been living together without being legally married,
but because of the stigma associated with cohabitation in that period, these couples may have
reported themselves as husband and wife. In other cases, respondents may simply have neglected
to acknowledge the presence of an unmarried partner.
Our goal, however, is to estimate self-
identified unmarried partners. The low number of cohabitors in 1960 estimated by the model
may indeed approximate the number of persons who identified themselves as unmarried couples.
This analysis expands the chronological scope of research on cohabitation in the United
States. We hope that these estimates will provide a baseline for understanding one of the most
profound changes in household formation of the late twentieth century. These estimates can also
provide further insight into other demographic phenomena, such as rising marriage ages and the
uncoupling of nuptiality and fertility. Our approach moves beyond the simple set of rules used in
both versions of the POSSLQ measure; instead, our analysis of the 1990 and 2000 census years
examines which characteristics distinguish unmarried partners from roommates (or other groups
of unrelated persons). Instead of a simple dichotomy, the predicted probability approach
provides a continuous measure and new analytic possibilities.
We consider it more likely that cohabiting opposite-sex couples in 1960 were reported as spouses or omitted than
that they were reported as employees or boarders. The more inclusive measures of cohabitation, such as POSSLQ,
cannot account for such misreporting any more than does our minimum potential partner measure.
Previous estimates of pre-1990 cohabitation have understated the magnitude and pace of
change during the period from 1960 to 1990. The published estimates for 1960 and 1970
overstate the number of households that meet the POSSLQ definition. More important, the
POSSLQ and Adjusted POSSLQ definitions greatly overstate cohabitation before 1990. Even if
it is justified to count relationship types such as boarders and domestic servants as cohabitors, the
POSSLQ measures probably overstate cohabitation in 1960 and 1970 by at least a factor of two.
We favor a more conservative approach that focuses on self-identified cohabitation. By this
standard, the traditional measures overstate cohabitation before 1980 by an order of magnitude.
These revisions make a significant difference. Table 7 summarizes the estimates
presented in Figures 1, 2, 4, and 7. The widely-accepted published POSSLQ measures suggest an
increase in cohabitation of just over 10-fold between 1960 and 2000. Our predicted partner
measure, by contrast, implies an increase between 28-fold (under the upper-bound universe) and
576-fold (under the lower-bound universe). The timing of change also varies by estimation
method. According to the more inclusive methods, there were already several hundred thousand
cohabitors by 1960 but there was only a modest rise in cohabitation during the 1960s. Under our
lower-bound estimates, however, there was very cohabitation in 1960, and there was a dramatic
increase (in percentage terms) during the following decade.
Our methods improve the potential for individual-level analysis of cohabitation in the
pre-1990 censuses. The POSSLQ measures are of limited use for this purpose, since they
introduce substantial biases with respect to household size, education, race, and presence of
children (Baughman, Dickert-Conlin, and Houser 2002). Our preliminary analysis suggests that
these problems are dramatically reduced for predicted partners, using either the lower-bound or
upper-bound universe. In future work, we plan to use these measures to assess changes in the
characteristics of cohabitors during the 1960s and 1970s.
Table 7. Comparison of cohabitation estimates, 1960-2000
Percent change
1960 1970 1980 1990 2000
POSSLQ, published 439 523 1,589 2,856 4,746 1,081 166
POSSLQ, corrected 280 430 1,759 3,232 4,861 1,736 150
Adjusted POSSLQ 399 575 2,077 3,975 5,900 1,479 148
Potential partners-maximum 437 485 2,127 4,110 6,219 1,423 151
Potential partners-minimum 17 128 1,412 3,686 5,505 32,382 149
Predicted partners-maximum 164 230 1,428 2,923 4,615 2,814 158
Predicted partners-minimum 8 78 1,042 2,923 4,613 57,663 158
Opposite sex unmarried partners 3,102 4,646 150
Unmarried partners, restricted*
2,982 4,646
*Opposite sex unmarried partners under Census 2000 editing rules (both partners currently married).
We have not yet addressed the issue of same-sex unmarried partners. We plan to apply
similar techniques to analyze this population, but we are not optimistic about our prospects for
successfully identifying same-sex cohabitors before 1980. We suspect that it will be difficult to
distinguish same-sex unmarried partners from roommates and business partners. Nevertheless,
because of the importance of the topic and the paucity of alternative sources, it is worth
investigating the feasibility of historical analysis.
Works Cited
Baughman, Reagan, Stacy Dickert-Conlin, and Scott Houser. 2002. “How Well Can We Track
Cohabitation Using the SIPP? A Consideration of Direct and Inferred Measures,”
Demography 39(3):455-465.
Bumpass, Larry L., James A. Sweet, and Andrew Cherlin. 1991. “The Role of Cohabitation in
Declining rates of Marriage,” Journal of Marriage and the Family 53: 913-27.
Bumpass, Larry L. and Hsien-Hen Lu. 2000. “Trends in Cohabitation and Implications for
Children’s Family Context in the United States,” Population Studies 54: 29-41.
Bumpass, Larry L. and James A. Sweet. 1989. “National Estimates of Cohabitation,”
Demography 26: 615-25.
Casper, Lynne M. and Philip N. Cohen. 2000. “How Does POSSLQ Measure Up? Historical
Estimates of Cohabitation” Demography 37(2): 237-245.
Casper, Lynne M. and Suzanne M. Bianchi. 2002. Continuity and Change in the American
Family. Thousand Oaks, CA: Sage.
Chevan, Albert. 1996. “As Cheaply as One: Cohabitation in the Older Population.” Journal of
Marriage and the Family 58:656-667
Glick, Paul C. and Arthur .J. Norton. 1977. “Marrying, Divorcing and Living Together in the
U.S. Today,” Population Bulletin 32(1):4-34.
Glick, P.C. and G.B. Spanier. 1980. “Married and Unmarried Cohabitation in the United States,”
Journal of Marriage and Family 42: 19-30.
Goeken, Ron. 1999. Unmarried Adults and Residential Autonomy: Living Arrangements in the
United States, 1880-1990. Dissertation: University of Minnesota.
Hatch, Rebecca Gronvold. 1995. Aging and Cohabitation. New York:Garland.
Manning, Wendy D. 1995. “Cohabitation, Marriage and Entry into Motherhood,” Journal of
Marriage and the Family 57: 191-2000.
Moffitt, Robert A., Robert Reville, and Anne E. Winkler. 1998. “Beyond Single Mothers:
Cohabitation and Marriage in the AFDC Program.” Demography 35(3):259-278.
Ruggles, Steven. 1995. “Family Interrelationships,” Historical Methods 28: 52-58.
Ruggles, Steven and Susan Brower. 2003. “The Measurement of Family and Household
Composition in the United States, 1850-2000.” Population and Development Review
29(1): 73-101.
Ruggles, Steven, Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken, Patricia
Kelly Hall, Miriam King, and Chad Ronnander. Integrated Public Use Microdata Series:
Version 3.0 [Machine-readable database]. Minneapolis, MN: Minnesota Population
Center [producer and distributor], 2004.
Smock, Pamela J. and Wendy Manning. 2004. “Living Together Unmarried in the United States:
Demographic Perspectives and Implications for Family Policy.” Law and Policy 26: 87-
U.S. Census Bureau. 2004. Families and Living Arrangements, Table UC-1.
U.S. Bureau of the Census. 1964. 1960 Census of Population PC(2)-4B, “Persons by Family
Characteristics”. Washington, D.C.: U.S. Government Printing Office.
U.S. Bureau of the Census. 1973. 1970 Census of Population PC(2)-4B, “Persons by Family
Characteristics”. Washington, D.C.: U.S. Government Printing Office.
Census Questions on Unrelated Household Members
Relationship information in the 1960 census
Write names in this order
Head of household on first line
Wife of head
Unmarried children, oldest first
Married children and their families
Other relatives
Others not related to head of household If "Other not related to
head," also
give exact relationship, for example, partner, maid, etc.
What is the relationship of each person to the head of this
household? (For example, wife, son, daughter, grandson, mother-
in-law, lodger, lodger's wife)
Relationship information in the 1970 census
O Head of household
O Wife of head
O Son or daughter of head
O Other relative of head - Print exact relationship
O Roomer, boarder, lodger
O Patient or inmate
O Other not related to head - Print exact relationship
If "Other not related to head," also
give exact relationship, for example, partner,
maid, etc.
If two or more unrelated people live together and share the rent, mark the first
one you list Head of household. Mark the rest Other not related to head
and print "partner" in the space. A stepchild or legally adopted child of the
head should be marked Son or daughter."
Relationship information in the 1980 census
If not related to person in column 1:
O Roomer, boarder
O Partner, roommate
O Paid employee
O Other nonrelative _______________
"Fill a circle to show how each person is related to the person in
column 1. A stepchild or legally adopted child of the person in
column 1 should be marked Son/daughter. Foster children or wards
living in the household should be marked Roomer, boarder."
1990-2000 “unmarried partner” category
Relationship to householder question:
How is this person related to person 1?
If not related to person 1:
O Roomer, boarder
O Housemate, roommate
O Unmarried partner
O Foster child
O Other nonrelative
... This limits the ability to observe the precise relationships of taxpayers who reassign dependents. However, using a variant of the Persons of Opposite Sex Sharing Living Quarters (POSSLQ) method, which is commonly applied to infer relationship status when it is unavailable, we are able to obtain broad estimates of relationship statuses in the tax data (see Casper andCohen, 2000, andFitch, Goeken, andRuggles, 2005, for overviews of this approach and see Dokko, Li, and Hayes, 2015, for a modification of this approach that incorporates age bands into the relationship status imputation). We consider three relationship statuses in the data: independent cohabiting couples, multigenerational households, and roommates/other living situations. ...
... This limits the ability to observe the precise relationships of taxpayers who reassign dependents. However, using a variant of the Persons of Opposite Sex Sharing Living Quarters (POSSLQ) method, which is commonly applied to infer relationship status when it is unavailable, we are able to obtain broad estimates of relationship statuses in the tax data (see Casper andCohen, 2000, andFitch, Goeken, andRuggles, 2005, for overviews of this approach and see Dokko, Li, and Hayes, 2015, for a modification of this approach that incorporates age bands into the relationship status imputation). We consider three relationship statuses in the data: independent cohabiting couples, multigenerational households, and roommates/other living situations. ...
Using a panel of household level tax data, we estimate the degree to which dependents are "reassigned" between tax units within households, and how these reassignments affect combined tax liabilities. Reassigning dependents reduces combined tax liabilities on average, suggesting some household level coordination. Additionally, when EITC benefits expanded in 2009, reassignments increasingly involved adding a third child to tax returns to claim these new benefits. However, the subgroup reassigning towards three child tax units actually increased total household tax liabilities, suggesting that some taxpayers may prioritize minimizing their own tax burden or focus on particularly salient aspects of tax policy.
... Cohabitation rates are increasing not only among young people. They are rapidly growing also among the oldest cohorts (Fitch, Goeken, & Ruggles, 2005;Chevan, 1996). For example, in the United States between 1980 and 1990 for unmarried people under 40 the cohabitation rate doubled and for those over 60 it even tripled. ...
Full-text available
In recent decades, the issue of life quality and its determinants has been increasingly discussed in the scientific literature. One of the key determinants of subjective well-being is the fact of being in a relationship (formal or not) as well as its quality. The results of analyzes for other countries show that people who have a partner have a higher subjective well-being than people who do not live with their spouse, while divorced people have the lowest (and the highest level of depression). In addition to being in a relationship, its quality is also important: people more satisfied with marriage declare greater well-being than those less satisfied. The aim of the study is to analyze quality of life (expressed in terms of well-being and depression) among older people in Poland, putting emphasis on their individual (e.g. age, sex, health status, level of education) and household (financial situation) characteristics. Importantly, the analysis takes into account the quality of the marriage approximated by the satisfaction with marriage declared by both spouses separately. For the purposes of this study I use data of ‘Social Diagnosis’ study carried out in 2015 in Poland. The final sample was limited to partnered older people aged 65+. Basic individual socio-demographic and economic as well as household characteristics were controlled for. The results show that older partnered men in Poland have higher subjective well-being than older partnered women. Also, older partnered men report significantly higher levels of marital satisfaction than older partnered women do. Moreover, it should be underlined that higher levels of marital satisfaction among older spouses may be beneficial for their own quality of life expressed by well-being and depression. Thus, better opinion about marriage may increase subjective well-being and decrease depression level. However, this relationship is different for older partnered men and women. Positive relationship between marital satisfaction and well-being is stronger for partnered women when subjective well-being is taken into account and for men when well-being is understood as depression level.
... This has implications for the computation of household income as contributions provided by other members of the household are not included in the analysis. 9 Unmarried couples were not identified in 1973, as such status was not recorded in the CPS, but estimates suggest that the prevalence of unmarried couples was limited to about only 1% of all couples (Fitch et al., 2005). income shares (according to WID.World data and reported in Figure 1). ...
Full-text available
We analyze the association between spouses' incomes using a rank-rank specification that takes non-linearities along both spouses' income distribution into account. We also document that the relationships between income and labor force participation and income and couple formation are non-linear. Using simulations, we then analyze how changes in spouses' rank-dependence structure, labor force participation and couple formation contribute to the upsurge of income inequality in the U.S between 1973 and 2013. We find that an increased tendency towards positive sorting contributed substantially to the rise in inequality among dual-earner couples, but contributed little to overall inequality across households. When considering all households, the factor accounting most for the increased inequality during this period is an increased tendency for individual men and women to remain single. JEL codes: J12, J22, D31
Trends in life expectancy and marriage patterns work together to determine expected lifetime years married. In 1880, adult life expectancy was short and marriages were more likely to end by death than divorce. Since then, although there have been substantial life expectancy gains in adulthood, marriage has been increasingly delayed or forgone and cohabitation and divorce are far more prevalent. Whether adults today can expect to spend more or fewer years married than in the past depends on the relative magnitude of changes in mortality and marriage. We estimate trends in men's expected lifetime years married (and in other marital statuses) from 1880 to 2019 and by bachelor's degree (BA) status from 1960 to 2019. Our results show a rise in men's expected lifetime years married between 1880 and the Baby Boom era and a subsequent fall. There are large and growing differences by BA status. Men with a BA have had high and relatively stable expected lifetime years married since 1960. For men without a BA, expected lifetime years married has plummeted to lows not seen among men since 1880. Cohabitation accounts for a substantial fraction, although not all, of these declines. Our results demonstrate how increasing inequality in both life expectancy and marriage patterns combine to amplify educational differences in lifetime experiences of coresidential partnerships.
We use a simple structural matching model with unobserved heterogeneity to produce counterfactual marriage patterns, and thus quantify the contribution of changes in marital patterns in rising income inequality. We propose an algorithm that allows us to fix the degree of assortative mating without changing the level of marital gains and hence isolate the intensive and extensive margins (i.e. isolate changes in assortative mating from changes in marriage rates). We apply this approach to US data from 1962 to 2017, and show that marital patterns can explain about a quarter of the rise in income inequality, the intensive margin contributing 7%, the extensive margin the remaining 93%. Our algorithm also allows us to show that the extensive margin is itself driven for three-fifths by a change in the total number of singles and for two-fifths by a change in the distribution of types among singles (in particular low-educated women).
Cohabitation is one of the fastest growing family forms in the United States. It is widespread and continues to increase but has not been consistently measured across surveys. It is important to track the quality of data on cohabitation because it has implications for research on the correlates and consequences of cohabitation for adults and children. Recent rounds of the Current Population Survey (CPS), National Longitudinal Survey of Adolescent to Adult Health (Add Health), National Longitudinal Survey of Youth (NLSY-97), and National Survey of Family Growth (NSFG) provide an opportunity to contrast estimates of cohabitation status and experience using nationally representative data sets and assess the quality of data on cohabitation in these data sets. Results demonstrated that the surveys provide similar estimates of current cohabitation status, except the CPS resulted in lower estimates. In terms of cohabitation experience (i.e., having ever cohabited), Add Health produced higher estimates, whereas both the NSFG and NLSY-97 produced lower estimates. We documented a strong education gradient across all surveys, with lower levels of current cohabitation and cohabitating experience and with increases in educational attainment. Racial/ethnic differences in cohabitation were inconsistent across surveys. We discuss aspects of sampling and measurement that potentially explain differences in estimates. Our findings have implications not only for survey design but also for the interpretation of results based on these four national surveys.
This study uses Interdependence Theory, specifically cognitive interdependence and the investment model of commitment, to further understand the impact of marital intent in cohabiting versus dating relationships. Contrary to the hypothesis posed, results revealed that individuals in cohabiting relationships and dating relationships experience similar levels of interdependence. However, people who report an intent to marry their partner, whether dating or cohabiting, have higher degrees of centrality of relationship, commitment, satisfaction, investments, and a lower level of perceived relationship alternatives than those who did not report marital intent. The results of this study suggest that marital intent may work similarly in dating relationships and cohabiting relationships, and that Interdependence Theory has utility for understanding why marital intent makes a difference in relational stability.
The decline of marriage over the past half century ranks among the most profound demographic transformations in American demographic history. This chapter puts recent change into historical context by providing new estimates of long-run trends in marriage. I then describe change in the family economy and explore the impact of economic changes on marriage behavior. I conclude with a discussion of cultural and structural explanations for change and their implications for the future.
This article proposes explanations for the transformation of American families over the past two centuries. I describe the impact on families of the rise of male wage labor beginning in the nineteenth century and the rise of female wage labor in the twentieth century. I then examine the effects of decline in wage labor opportunities for young men and women during the past four decades. I present new estimates of a precipitous decline in the relative income of young men and assess its implications for the decline for marriage. Finally, I discuss explanations for the deterioration of economic opportunity and speculate on the impact of technological change on the future of work and families.
Full-text available
We investigate the extent and implications of cohabitation and marriage among U.S. welfare recipients. An analysis of four data sets (the Current Population Survey, the National Survey of Families and Households, the Panel Study of Income Dynamics, and the National Longitudinal Survey of Youth) shows significant numbers of cohabitors among recipients of AFDC. An even more surprising finding is the large number of married women on welfare. We also report the results of a telephone survey of state AFDC agencies conducted to determine state rules governing cohabitation and marriage. The survey results indicate that, in a number of respects, AFDC rules encourage cohabitation. Finally, we conduct an analysis of the impact of AFDC rules on cohabitation, marriage, and single motherhood and find weak evidence in support of incentives to cohabit.
Prior research has neither explicitly compared the entry into motherhood of cohabiting with that of married women nor examined the impact of cohabitation on marital fertility in the United States. Subsamples of 2,056 women in first unions and 1,763 married women from the National Survey of Families and Households are used to address those questions. Entry into motherhood occurs more often and sooner in marriage than in cohabitation. Yet the transition from cohabitation to marriage does not appear to be influenced by desires to begin bearing children. Once nonpregnant cohabiters marry, the timing of the marital first birth is similar to that of women who never cohabited. Cohabitation accelerates the timing of marital first births only among White women who were pregnant when they married Instead, the impact of cohabitation on marital first birth timing operates partly via duration of time spent coresiding (in marriage and cohabitation).
Sharp declines in both first marriage rates and rates of remarriage have been largely offset by increasing cohabitation. The increase in the proportion of unmarried young people should not be interpreted as an increase in "singlehood" as traditionally regarded: young people are setting up housekeeping with partners of the opposite sex at almost as early an age as they did before marriage rates declined. The characteristics of cohabiting couples are documented here, including the role of the least educated in leading this trend, and the presence of children with 40% of the couples. While most cohabitors expect to marry their partner, there is a substantial proportion who disagree about marriage, and a high proportion are concerned about the stability of their relationship. Thus the picture that is emerging is that cohabitation is very much a family status, but one in which levels of certainty about the relationship are lower than in marriage.
This study examines the prevalence of cohabitation, observes trends in cohabitation between 1960 and 1990, and investigates the conditions leading older persons to cohabit. An indirect strategy is used to measure cohabitation as a result of a validation study of approaches to its measurement. The trend analysis with Public Use Microdata Samples finds 2.4% of unmarried persons age 60 and older cohabiting in 1990, up from virtually 0% in 1960. By 1990 there were 407,000 elderly cohabitors, compared with 9,600 in 1960. Cohorts with considerably higher levels of cohabitation will shortly enter old age. Variables measuring individual characteristics, economic motivations, and the social context are used to predict cohabitation.
This paper documents increasing cohabitation in the United States, and the implications of this trend for the family lives of children. The stability of marriage-like relationships (including marriage and cohabitation) has decreased despite a constant divorce rate. Children increasingly live in cohabiting families either as a result of being born to cohabiting parents or of their mother s entry into a cohabiting union. The proportion of births to unmarried women born into cohabiting families increased from 29 to 39 per cent in the period 1980-84 to 1990-94, accounting for almost all of the increase in unmarried childbearing. As a consequence, about two-fifths of all children spend some time in a cohabiting family, and the greater instability of families begun by cohabitation means that children are also more likely to experience family disruption. Estimates from multi-state life tables indicate the extent to which the family lives of children are spent increasingly in cohabiting families and decreasingly in married families.
Never-married and formerly married adults living with unrelated adults of the opposite sex are contrasted with married couples living together. Social and economic correlates of these living arrangements and joint characteristics of the partners are described on the basis of national data from the Census Bureau's Current Population Surveys. An estimated 1.8 percent of all couples living together in 1975, and 2.3 percent in 1978, were unmarried. The ways in which the married and unmarried individuals differ is discussed. Finally, the two partners in married and unmarried couples are compared with respect to race, age, educational attainment, employment characteristics, and marital history.
This paper synthesizes research on the demographic correlates and consequences of unmarried, heterosexual cohabitation in the United States. First, we place cohabitation in the context of recent demographic trends in union formation and dissolution. Second, we consider the implications of cohabitation for child well-being. Third, we review population subgroup variation in the role of cohabitation in family patterns, focusing on social class and race and ethnicity. Finally, we discuss how and why unmarried cohabitation is implicated in recent dialogues about family policy.
Thesis (Ph. D.)--University of Minnesota, 1999. Includes bibliographical references.