The Role of Cohabitation in Family Formation: The United States in Comparative Perspective

Abstract

The prevalence of nonmarital cohabitation is steadily increasing in the United States. In evaluating the contribution of this new living arrangement to family formation, analysts have relied primarily on comparisons between individuals who cohabit and those who do not. We complement this line of inquiry by comparing the United States and 16 industrialized nations. We first identify six conceptually distinct ideal types of cohabitation with respect to family formation. We then propose empirical indicators to distinguish between the different ideal types, and estimate the values of these indicators for each of the 17 nations. Our findings indicate that although a number of countries fit an empirical pattern corresponding to one ideal type, cohabitation in the United States is more difficult to characterize.

Normative attitudes on family formation have been changing rapidly in the United States since at least the 1960s (Pagnini & Rindfuss, 1993; Thornton, 1989). As shown by Axinn and Thornton (1993), these attitudinal trends have both contributed to and resulted from concomitant declines in the prevalence of a traditional family formation sequence in which adults first get married, then live together, and finally have children. The deviation from this sequence that has received the most attention in the United States is undoubtedly childbearing before marriage. The increase in the proportion of births to unmarried mothers—from 4.0% in 1950 to 33.0% in 1999 (Ventura & Bachrach, 2000)—is indeed one of the most impressive trends, and a large literature now documents the effects on children of growing up with a single parent (e.g., McLanahan & Sandefur, 1994).

Until recently, living together before marriage generated much less public attention, although family scholars have debated whether to interpret unmarried cohabitation as a prelude to marriage—that is, a simple inversion in the timing of two events (marrying and cohabiting)—or as an alternative to marriage—that is, a decision not to marry. This scholarly debate might have remained just that, as long as unmarried cohabitation was not publicly perceived as a childbearing institution. In fact, Rindfuss and VandenHeuvel (1990) found that unmarried cohabiting couples in the United States exhibited much of the same characteristic behaviors as single dating people, and suggested that cohabitation was an alternative to being single rather than to marriage.

Although cohabiting couples postpone childbearing longer than married couples do (Manning, 1995), cohabitation is nevertheless becoming a significant feature of the modern reproductive landscape. In the early 1990s, births to cohabiting mothers represented more than 10% of all births, and nearly two fifths of out-of-wedlock births (Bumpass & Lu, 2000). More recent data from the Fragile Families and Child Wellbeing Study suggest that nearly half of all out-of-wedlock births are to cohabiting mothers (J. O. Teitler, personal communication, April 30, 2003). The presence of children has brought cohabitation into public discourse, leading concerned policymakers to propose incentives to incite unmarried cohabitating partners to marry. Although we know relatively little about the consequences for children of parental cohabitation (Smock, 2000)—with the possible exception of economic circumstances (Manning & Lichter, 1996; Morrison & Ritualo, 2000)—cohabitation is generally less stable than marriage and thus presents higher risks for children to experience parental separation. Cohabitation is also spreading as a postmarital institution. The proportion of births occurring to cohabiting mothers nearly doubles to 20% of births among previously divorced mothers (Brown, 2000), and children are more likely to experience cohabitation in a stepfamily than with their two biological parents. Accounting for types of parental cohabitation, Graefe and Lichter (1999) found that one fourth of all children live with a cohabiting parent at some point during childhood, whereas Bumpass and Lu (2000), using a different data set, estimated the same proportion at nearly two fifths.

To understand the role of cohabitation in the landscape of family formation in the contemporary United States, several studies compare the characteristics of individuals forming different types of partnerships (e.g., Blackwell & Lichter, 2000; Brown & Booth, 1996; Manning, 1993; Rindfuss & VandenHeuvel, 1990). Cross-sectional studies of individual characteristics cannot provide insight into the contextual factors that affect all individuals, however, and may have contributed to the departure from the sequence of marriage, then coresidence, and finally, childbearing. Looking backward, it is tempting to link this departure to contemporaneous cultural, social, and economic changes in the United States. Even a cursory review of the comparable trends in other Western nations that have experienced many of the same changes, however, renders such an interpretation of recent trends in the United States unconvincing. Comparative European research (e.g., Kuijsten, 1996) has found little evidence of convergence on several indicators of the “second demographic transition” (van de Kaa, 1987), and cohabitation is no exception (Kiernan, 2001; Prinz, 1995).

In this article, therefore, we attempt to contribute to the literature that compares individuals in different living arrangements in the contemporary United States by comparing U.S. patterns of cohabitation formation and fertility among women, and exposure to cohabitation among children, with those in 16 other nations. Our goal is to understand cohabitation in the United States by situating it in a larger context of other Western industrialized societies.

Our analysis proceeds in several steps. In the next section, we identify six conceptually distinct ideal-typical ways in which couples experience unmarried cohabitation, derived from the sociological and demographic family literature. We next adduce a set of empirical indicators, as well as predictions about the relative magnitudes of those indicators, that should yield sufficient cross-national variation to enable identifying the presence of these ideal types in different countries. A key factor in our scheme is the fertility behavior of cohabiting adults, because the more cohabitation enters “the arena for reproduction,” the more it is likely to be a substitute for marriage (Smock, 2000, p. 10). Using data from 17 nationally representative Family and Fertility Surveys, we then employ single and multiple-decrement life table analysis to estimate the incidence of, duration in, and routes of exit from cohabitation for female respondents and the children of childbearing female respondents. Comparing these indicators with the patterns corresponding to each ideal type, we draw conclusions about the presence or absence of a dominant orientation toward cohabitation in the different countries.

Theoretical Background

Early research on cohabitation was framed by the prelude to marriage versus alternative to marriage dichotomy. Rindfuss and VandenHeuvel (1990) conceptualized cohabitation instead as an alternative to being single. Casper and Bianchi (2002) proposed four cohabitation types, essentially introducing one more distinction within the prelude to marriage type: (a) alternative to marriage, (b) precursor to marriage, (c) trial marriage, and (d) coresidential dating. In this typology, the two main dimensions are the expectation to marry (present in b, absent in a and d, and undecided in c), and the expected duration of the relationship (long in a and b, short in d, and undecided in c). Several researchers (e.g., Manning, 1993; Raley, 2001; Smock, 2000) have also emphasized differences in the arena of reproduction, showing, for instance, how a pregnancy may affect the future of a relationship, especially in the trial-marriage type of cohabitation. Kiernan (2001) recently introduced the idea that in the final stage of development, “cohabitation and marriage become indistinguishable with children being born and reared in both” (p. 3), which can be seen as a further distinction within the alternative to marriage type. Integrating these critical distinctions, we identify and describe six ideal-typical ways in which cohabitation contributes to family formation in Western societies based on actual behavior (duration of cohabitation and childbearing) rather than on expectations at the start of cohabitation. More specifically, we define these ideal types by considering unmarried couples’ decisions (a) to live together, (b) to have children together, and (c) to stay together.

In any given society at any given time, we expect all combinations of decisions corresponding to each of these ideal types to be found; however, we also expect cross-national variation in the prevalence of these different combinations. Research on cohabitation in the United States has tended to focus on documenting the differences in individual characteristics that explain why some couples pursue or do not pursue particular sequences of partnership formation, marriage, childbearing, and childrearing (e.g., Bumpass & Lu, 2000; Manning, 1993). Kiernan (2000) compared individual differences in the likelihood to marry before ever cohabiting in 14 European countries. Although these comparisons reveal modest education and religion gradients, subgroup differences within countries are small compared with international differences. In other words, although within countries, the more educated and those never attending church are less likely to marry without prior cohabitation, members of these subgroups in Spain or Italy are still much more likely to marry without prior cohabitation than the least educated and most religious subgroups in Sweden or Finland.

Cross-national differences in underlying individual characteristics therefore contribute to differences in national-level prevalence, but differences at the national level likely depend much more on the cultural norms, expectations, attitudes, and institutional supports that families encounter in each society. As long as marriage continues to be a socially approved arrangement for family formation, we should find institutional and cultural rewards for living together, having children, and staying together within marriage in all societies. Reher (1998), for instance, documented persistent cultural differences in the subjective contours of family life across Western Europe. In addition, Western countries feature different welfare policies (Esping-Andersen, 1990) that have different impacts on families in general (Gauthier, 1996), and families with children in particular (Bradshaw, Ditch, Holmes, & Whiteford, 1993). Individual characteristics may explain who might be more or less likely to forgo these rewards and benefits and form a family outside wedlock, but the nature of these rewards and benefits likely affects the overall proportion of couples who decide to form a partnership outside marriage. Selection into unmarried cohabitation may exist in all countries, but the strength of this selection should vary and be revealed in part by the prevalence of decisions to bear children and to remain unmarried for a substantial length of time.

Ideal Types of Cohabitation

Marginal

In countries where unmarried cohabitation continues to be culturally frowned upon and institutionally penalized, cohabitation will attract only a small minority of couples. In these countries, the incidence and duration of adulthood cohabitation should be low, and children’s exposure to and duration in cohabitation should be even lower.

Prelude to marriage

Family formation may be initiated by unmarried cohabitation as a “testing” ground for a relationship. Couples may feel a greater need for this premarital experience when they observe high rates of divorce. Choices to cohabit first may also depend on access to affordable housing, or to reliable contraception to postpone childbearing. If norms or institutions continue to be unsupportive of unmarried couples with children, however, couples would be expected either to end the relationship or to marry before children are born. This implies that the average duration of an episode of unmarried cohabitation should be fairly short, cohabitation should frequently transition into marriage, and children’s exposure to cohabitation should remain relatively low.

Stage in the marriage process

This ideal type is closely related to the previous one. Although Casper and Bianchi (2002) distinguished between trial marriage and precursor to marriage based on the motivation of individuals entering cohabitation with respect to marriage, we distinguish this type from the previous one, based on the actual timing of marriage and childbearing. We see cohabitation as a stage in the marriage process rather than a prelude to marriage when cohabiting couples who decide to have a child do not feel strongly about the precise order and timing of childbearing and marriage. Birth followed by marriage should become more frequent when there continue to be institutional incentives to raise children within marriage, but when cultural sanctions against out-of-wedlock childbearing have eroded. In this scenario, couples may increasingly experience competing opportunities that they could pursue by briefly postponing marriage, as long as it is understood that they intend to marry eventually. When this behavior becomes more prevalent, we should observe a slightly longer average duration than in the strictly ordered prelude to marriage, and substantially more childhood exposure to cohabitation. The average duration of children’s exposure should be relatively short, however, with the assumption being that once children are born, cohabiting adults fairly quickly formalize their relationships as marriages.

Alternative to single

Following Rindfuss and VandenHeuvel (1990), we include this ideal type to capture cohabiting partners who want to postpone forming a family, and instead prefer cohabiting rather than living separately during courtship. As with the prelude to marriage ideal type, increased access to affordable housing and to efficient contraception should increase the prevalence of this behavior when children are still expected to be born within marriage. The behavior should also be more frequent when young adults consider themselves “too young” to seriously consider marriage, and hence enter cohabitation with no immediate intention to marry. Because the commitment of these cohabiters is more like a dating relationship than a marital one, the presence of this form of cohabitation should depress the average duration of cohabitation episodes and increase the proportion of these episodes ending in separation.

Alternative to marriage

The alternative to marriage corresponds to the choice to cohabit instead of marrying, but simultaneously to form a family as a married couple would. As in the marginal ideal type, we expect population heterogeneity to be manifest, and individual characteristics to determine which couples choose to form a family outside rather than within marriage. We also expect that a greater cultural approval of and better institutional support for children raised entirely out of wedlock would reduce the strength of selectivity, however, and lead to a greater proportion of couples opting for cohabitation as compared with the previous types. When these conditions are met, cohabitation should become more prevalent and should last longer, on average, than in the marginal ideal type, and should transition into marriage less frequently than in the prelude to marriage or stage in the marriage process ideal types. Children should also be exposed to cohabitation more frequently, and should remain longer with cohabiting parents, when the prevalence of this type of cohabitation increases.

Indistinguishable from marriage

Following Kiernan (2001), we introduce one more ideal type that shares many similarities with the alternative to marriage type, in the sense that in both, unmarried couples form families as married couples do. In this ideal type, however, couples are not driven to an alternative to marriage by their own characteristics and attitudes toward marriage, but in fact are indifferent to marrying because of the general acceptability of unmarried cohabitation, and institutional supports for parents that essentially ignore marital status. We expect the incidence of cohabitation among adults and exposure to cohabitation among children to be even higher than in the alternative to marriage type because unmarried cohabitation is less constrained by individual characteristics. We also expect the duration of cohabitation episodes with and without children to be slightly shorter, and a larger proportion of cohabitation episodes to transition into marriage as compared with the alternative to marriage type. This is because cohabitation is not viewed as antithetical to, or something to be done instead of marriage; thus, couples may become more pragmatic in their decision to marry.

Indicators and Empirical Predictions

The foregoing discussion suggests that cross-national variation in the incidence of, duration in, and route of exit from cohabitation for both adults and children should provide insights into the different ways that couples approach unmarried cohabitation as a component to their family-formation strategies. Thus, our key indicators for adults are the percentage of women expected to experience at least one cohabitation between ages 15 and 44, the expected duration of such cohabitations, and the route of exit, either marriage or separation. For children, our indicators are the percentage of children expected to experience a parental cohabitation from birth to age 16, the expected duration of exposure to a parental cohabitation, and the route of exit, either the marriage or separation of the child’s parents. We describe these measures and their estimation in greater detail below. The measures are indicators of the central tendency of the distribution across women, children, or episodes of cohabitation. As noted above, however, when the prevalence of one type of cohabitation increases, the average indicators are expected to behave in a corresponding manner.

Table 1 summarizes the ideal types identified above and lists the empirical predictions for the relative magnitudes of the indicators corresponding to the significant presence of any of these types. Table 1 illustrates in particular the importance of considering both women’s and children’s experiences. We see that according to several of these ideal types, cohabitation could be fundamentally altering the way that couples are partnering, but may have little implication for the family structures in which children are born and raised (Timberlake & Heuveline, in press).

Table 1.

Ideal-Typical Roles of Cohabitation, Descriptions of Types, and Empirical Indicators and Predictions

			Empirical Indicators and Predictions
			Adults			Children
Role		Description	Incidence of Own Cohabitation	Median Duration	% Ending in Marriage	Exposure to Parental Cohabitation	Median Duration	% Ending in Marriage
A.	Marginal	Cohabitation is not prevalent and is likely  discouraged by public attitudes and policies.	Lower	Shorter	Higher	Lower	Shorter	Higher
B.	Prelude  to marriage	Exists as a prereproductive phase for adults.  Unions tend to be brief and nonreproductive,  but end in marriage.	Higher	Shorter	Higher	Lower	Shorter	Higher
C.	Stage in  marriage process	Exists as a transitory phase in reproduction.  Unions tend to be longer, and children are  more likely to be born into a cohabitation than  in (B), but with short duration of exposure.	Higher	Shorter	Higher	Higher	Shorter	Higher
D.	Alternative  to single	Cohabitation primarily for brief,  nonreproductive unions that end in separation  instead of marriage.	Higher	Shorter	Lower	Lower	Shorter	Lower
E.	Alternative  to marriage	Is a discrete family component. Adulthood  cohabitation prevalent, and for longer duration  than in (C). Low proportion leading to  marriage, more exposure to cohabitation during  childhood than in (C), and for longer duration.	Higher	Longer	Lower	Higher	Longer	Lower
F.	Indistinguishable  from marriage	Little social distinction between cohabitation  and marriage. Children more likely than in (E)  to experience the marriage of parents because  cohabitation not seen as an alternative to  marriage.	Higher	Longer	Lower	Higher	Longer	Higher

Method

Data: The Family and Fertility Surveys

The Family and Fertility Surveys comprise an international sample survey program focusing on fertility and family change in the member countries of the United Nations Economic Commission for Europe. The list of participants includes over 20 European countries, of which we analyze 14, as well as Canada, New Zealand, and the United States. The program did not fund ad hoc data collection, so participating countries typically included survey modules within their regular survey-taking activities. For example, in Canada, survey questionnaires appeared in cycles 4 and 5 of the General Social Survey, whereas they were included in the 1994 Annual Employment Surveys in France, and in cycle 5 of the National Survey of Family Growth in the United States (see the Appendix).

Most important for the present analyses, the data provide histories of all births and partnerships for female respondents. The partnership histories provide the dates of coresidence (beginning and end) for up to nine cohabiting relationships, if and how the partnership ended, and the date of marriage, if applicable. For women, we kept track of transitions to and from cohabiting status directly from these histories. We reconstruct children’s exposure to cohabitation by combining the partnership and fertility histories of the female respondents. For each natural-born child, we combined these two histories to create an early life course record of living arrangements with the mother and her partners from birth to exact age 16. As long as the child was living with the mother, we knew whether she also lived with a partner and, if so, whether the couple was married.

The Family and Fertility Surveys data represent an unparalleled source of information about differences in family-formation trends across these Western nations, but fitting the survey into existing national data collection programs introduced idiosyncrasies into the sampling designs. As shown in the Appendix, respondents’ age ranges vary across countries (upper limits from 40 to 60 years), as do the years of data collection (1989–1997). Further differences in survey design, data quality and comparability, and reasons for excluding certain countries have been described elsewhere (Heuveline, Timberlake, & Furstenberg, 2003; Kveder, 2002; Macura & Klijzing, 1992). Although overall, the surveys’ data vary in content, they are fairly standardized in terms of the information on births and partnerships that we use for the present analyses. Dates of birth (of self and own children) and marriage are among the most accurately reported items in retrospective surveys, especially by women (Poulain, Riandey, & Firdion, 1991). Retrospective reports on the incidence and timing of cohabitation might be less reliable, so some early and short partnerships might have gone unreported (Casper & Cohen, 2000; Murphy, 2000). This could tend to bias the estimates of the overall incidence of cohabitation downward, and the estimates of the median duration in cohabiting spells upward. These biases are likely to be extremely small, however, because as explained below, we limited our observations to a period shortly before the interview.

Analyses

Our primary analytic tools in this article are single- and multiple-decrement period life tables, which we use to estimate the incidence of, duration in, and route of exit from cohabiting relationships for the female survey respondents and their natural-born children. Life tables are a general class of demographic models that describe the transition over time of a cohort of individuals from one life state to another (Preston, Heuveline, & Guillot, 2001).

For adult women, our state of interest is to live in an unmarried, cohabiting partnership between ages 15 and 44. We had to settle for a younger upper age limit in several countries because of sampling age restrictions at the time of the survey (age 40 in Germany, age 42 in Belgium and Hungary). In addition, because the surveys were fielded in different years, we attempted to make the reference period more comparable by observing rates several years before the survey in some countries. Variations in the lower age limit are a lesser concern. When the lower age limit is greater than age 15—say, age 18—we only observe period rates from age 18 on, but we also observe cohort rates before age 18 from the retrospective histories. We then use the reference period rates to estimate the experience of a synthetic cohort from age 18 on, and complete it with the experience of the actual cohorts turning 18 during the reference period.

For children, the state of interest is to be under exact age 16 and living with a mother who is unmarried and cohabiting. Retrospective data on children reported by reproductive-age mothers are known to be subject to selectivity biases with respect to maternal age at birth (Rindfuss, Palmore, & Bumpass, 1982). The risk of bias is present here because younger mothers are less likely to be married at the time of birth (Morgan & Rindfuss, 1999). Little bias would be expected in countries where the adult age range for the period of observation reached age 44, because children up to exact age 16 would include children born to mothers up to the age of 28. Selectivity problems could be more severe in the countries with younger age limits, so we also estimated child life tables with an upper age limit of 12 instead of 16. The two sets of tables proved to be sufficiently comparable, so we only present here the tables ending at age 16.

For adults and children, we estimated two types of life tables to study entries into and exits from cohabitation. For adult women, the single-decrement “entry” life tables start at age 15 and follow, ideally up to exact age 45, the “risk” of forming an unmarried cohabitation. To allow us to distinguish between first cohabitations being formed before or after any marriage, two “at-risk” states are considered among women who have never cohabited: never married and ever married. For children, the entry life tables start at birth and follow, up to exact age 16, the probability of experiencing a maternal cohabitation.

From the entry tables, we estimate the percentage of adult women expected to live in an unmarried cohabitation by exact age 45 (or younger when required by sampling restrictions) before or after any marriage, and the percentage of children expected to live with an unmarried, cohabiting mother at birth, by exact age 1, and by exact age 16. For both adults and children, we present from the exit tables the median duration in the relationship, that is, how long adults and children are expected to live in premarital (adults) or parental (children) cohabitation before half of these partnerships end, and the percentage that by then would have ended in marriage or separation. The reason for using median duration for the adult tables rather than the more conventional life expectancy indicator is that the latter depend on how we would “close” the life table (Preston et al., 2001, p. 48). Because our observations are based on retrospective histories, we in effect remove the risk of death for adult women. This is not particularly problematic because that risk is extremely low at the ages under consideration, but it does allow for the possibility that a partnership “survives” forever, unless we impose arbitrary rules to end partnerships at very long durations. For children, the use of median duration also limits the analyses to younger ages and reduces the potential selectivity bias due to maternal age at the time of birth.

The “exit” tables are multiple-decrement life tables that account for the different ways that the observation of a spell or episode of interest may end, and in particular, distinguish between the observations ended by an event that effectively terminates the spell (referred to as an event of interest) and those ended by an event that removes the possibility of further observing a spell that is still unfinished (referred to as censoring; Preston et al., 2001, pp. 80–86). These life tables can then be reestimated under the assumption that spells whose observation was ended by a censoring event at a given duration would have had the same likelihood to survive the events of interest as those that can be observed to survive past that duration. Adult women reaching the maximum observable age for the period while in an unmarried cohabitation are treated so, for instance. In contrast, the entry life tables are simply closed at the upper age limit because events are not observed past that age.

For both adults and children, the temporal dimension of the exit life tables is the duration of the cohabiting relationship rather than the age of either the adult or the child. For adults, the table starts either when a premarital cohabitation partnership is formed, or at exact age 15 for never-married women who are already cohabiting at age 15. For children, the table starts either at birth for children born to a cohabiting mother, or at the time that the cohabitation begins for children born to mothers living alone who form a partnership before the child’s first birthday. Although these data do not allow a direct test of the new partner’s relationship to the child, we assume that the partner is most likely the child’s father and refer to cohabitations started at birth or before age 1 as parental cohabitations.

Taken together, the entry and exit life tables depict what would happen to a synthetic cohort exposed at each duration or age to the estimated duration- or age-specific transition probabilities of a reference period. To avoid excessive year-to-year variations, we selected a 3-year reference period in each country. Because these surveys were fielded in different years with different sampling age restrictions, imposing exactly the same age range and the same reference period would have wasted valuable information in countries that had more recent data on a more extended age range. Our choice of reference period and age range across countries, shown in the Appendix, reflects several compromises between scope, selectivity (on maternal age), timeliness, and comparability. Although our objective was to estimate adult rates from ages 15 to 44 during a 3-year period in the early 1990s, the actual reference period thus varies across countries. With the exception of those of Finland (1984–1986) and Sweden (1987–1989), the reference period of all countries include 1 to 3 years in the early 1990s (1990–1995). Whereas all values presented below are therefore estimates based on the probabilities observed in a somewhat variable reference period, we sometimes describe those estimates by resorting to the shorthand expression at early 1990s rates.

Results

Adulthood Cohabitation

We first examine the incidence of, duration in, and route of exit from cohabitation for adult women. Figure 1 shows the life table estimates of the percentage of 15-year-old female cohorts who enter a cohabiting partnership at least once by exact age 45, by prior marital status—that is, never or ever married. A comparison of these latter two estimates, depicted in the different colored sections of the bars, suggests that at early 1990s rates, the overwhelming majority of women who experience a first cohabitation do so before their first marriage. This confirms a recent transformation in the nature of cohabitation, because cohabitation following marital breakdown was likely the principal form of cohabitation through the 1950s and 1960s (Kiernan, 2001, p. 2).

Figure 1. Expected Probability (%) of Experiencing at Least One Adulthood Cohabitation by Age 45,^a by Previous Marital Status: Female Respondents in Family and Fertility Surveys.

Note: Countries sorted in descending order by total percentage expected to cohabit. Estimates derived from single decrement life tables.

^aAge 36 in Germany, 37 in Belgium and Hungary, 38 in Sweden, 40 in the Czech Republic and the United States, 41 in Slovenia, and 43 in Italy and Switzerland.

The total percentage of adults forming a pre- or postmarital cohabitation is likely reaching record levels, although high cross-national variation remains. By imposing the same upper age limit across the board, we found that the different upper age limits do not contribute much to this cross-national variation. This is because most women between the exact ages of 15 and 45 experience cohabitation before marriage, and because most women who marry between those ages are married by their late 30s. At early 1990s rates, this percentage ranges from 4.4% in Poland to 83.6% in France. Between these extremes, quantifying exactly which percentages are high or low is somewhat arbitrary. The lower end of the distribution exhibits a clear cutoff point, however, with three countries at 15% or less (Spain, Italy, and Poland), while the next country (Hungary) is at 24%. Cohabitation therefore continues to contribute relatively little to family formation in these three countries.

Table 2 shows the median estimated duration of cohabitation spells involving never-married women (i.e., not just the first cohabiting partnership, but all cohabiting partnerships before first marriage)—that is, the number of years it would take, given the transition probabilities observed during the reference period, for 50% of these cohabitation spells to end with the couple either separating or transitioning into marriage. Table 2 also breaks down the route of exit for these first 50%. The estimation has to be restricted to the countries that provide enough cohabitation-years of observation during the reference period. Setting a lower limit at 500 cohabitation-years, 11 countries are left and there is again substantial variation in both duration and route of exit. Three countries exhibit median durations of 3 years or more (France, 4.28 years; Sweden, 3.44 years; and Canada, 3.32 years), whereas the median duration of premarital cohabitation spells is distinctly shorter in the United States (1.17 years) than in other countries.

Table 2.

Expected Duration of and Route of Exit From Adulthood Cohabitation: Never-Married Women Ages 15 to 44^a

Country	Cohabitation Years	Median Durationb	Route of Exitc
			Marriage	Separation
France	1,490	4.28	46.3	53.7
Sweden	1,799	3.44	61.2	38.8
Canada	1,310	3.32	36.1	63.9
Slovenia	707	2.90	78.1	21.9
Austria	1,218	2.69	76.8	23.2
Belgium	504	2.39	76.2	23.8
Finland	1,045	2.38	80.7	19.3
Germany	1,470	2.24	53.3	46.7
New Zealand	801	2.18	39.5	60.5
Switzerland	1,367	1.78	75.6	24.4
United States	2,171	1.17	48.0	52.0
Country  median		2.39	61.2	38.8

Note: Table sorted in descending order by expected median duration, excluding countries with fewer than 500 observed cohabitation-years during the reference period. Estimates derived from multiple-decrement life tables, adjusted for the risk of reaching the upper age limit before the end of a cohabitation spell (see Preston et al., 2001, pp. 80–84).

^aActual age ranges vary across countries; see Appendix.

^bDuration by which half of the premarital cohabitation spells have ended in either marriage or separation.

^cFigures relate to the 50% of the cohabitation spells that are ended by the median duration.

When most cohabitation spells are initiated as a prelude to marriage or as a stage in the marriage process, the median duration of these spells should be relatively short, and a fairly high proportion of them should transition into marriage. We find that for 5 of these 11 countries, the percentage of cohabitation spells transitioning into marriage is indeed above three fourths, with a high of 80.7% in Finland. The percentage is lowest in Canada (36.1%), and relatively low as well in the other two countries with long median duration (France, 46.3%; Sweden, 61.2%). Two other countries with relatively low median durations also exhibit percentages ending in marriage under 50%, however: New Zealand (39.5%) and the United States (48.0%). In contrast to Canada, France, and Sweden, where low likelihood of marriage, combined with relatively long duration, suggests stable cohabitation regardless of marriage, the combination of low likelihood of marriage and short durations in New Zealand and in the United States suggests unstable cohabitation unless they are converted into marriages.

Childhood Cohabitation

We now turn to children’s experience of cohabitation. Figure 2 presents life table estimates of the percentage of early 1990s birth cohorts expected to experience a maternal cohabitation at least once by exact age 16. These estimated percentages are further broken down into three parts: those born to cohabiting parents, those born to single mothers who then transition into a parental cohabitation by age 1, and those whose first experience of maternal cohabitation is between exact ages 1 and 16, which we treat as stepcohabitations. Again, we observe wide variation in childhood experiences of maternal cohabitation, ranging from 4.7% in Poland to 53.5% in Sweden.

Figure 2. Expected Probability (%) of Exposure to at Least One Maternal Cohabitation by Age 16, By Route of Exposure: Children of Family and Fertility Surveys Female Respondents.

Note: Countries sorted in descending order by total percentage expected to experience maternal cohabitation by age 16. Estimates derived from single-decrement life tables.

Not surprisingly, at the low end of the distribution of children’s exposure we find the three countries with low adulthood exposure, but three other countries (Hungary, Belgium, and Switzerland) also exhibit childhood proportions below 15% (Figure 2). For Switzerland, the contrast between adulthood and childhood percentages is remarkable, with 59.9% of women and only 9.6% of children experiencing maternal cohabitation. According to our estimates based on early 1990s rates, women in Switzerland are thus six times more likely to experience cohabitation than children during the ages under consideration.

At the high end of the distribution, Sweden is an outlier with more than half of a birth cohort expected to experience maternal cohabitation during childhood, followed by France at near 40%. More surprisingly, the United States, which is about average on the adult-cohabitation scale, is next, with about one third of a birth cohort expected to experience maternal cohabitation by exact age 16. Our estimate is higher than Graefe and Lichter’s (1999) one-fourth estimate based on a different survey, the 1997 National Longitudinal Survey of Youth, but it is slightly lower than Bumpass and Lu’s (2000) two-fifths estimate using the U.S. Family and Fertility Survey and a broader reference period (1990–1994). In any event, the relatively high exposure to maternal cohabitation in the United States is due in part to the higher exposure to maternal cohabitation between exact ages 1 and 16, which we infer to be with stepparents rather than with biological parents. The proportion of children experiencing their first maternal cohabitation between exact ages 1 and 16 is the largest in the United States (more than one fifth of a birth cohort), and accounts for more than one half of children’s first exposure to maternal cohabitation. By contrast, the proportion is about 5% in Sweden, only accounting for about one tenth of children’s first exposure to maternal cohabitation. In part, children in the United States are more at risk of experiencing maternal cohabitation between age 1 and age 16 because they are more likely than children in other countries to be living with a single mother at some point during their childhood, most often as a result of parental separation (Heuveline et al., 2003).

Table 3 presents children’s median durations in and routes of exit from parental cohabitations (begun either at birth or by age 1) for the nine countries that had a sufficient number of observed cohabitation-years to enable reliable estimation. As for adults, median durations for children are shortest in the United States and longest in France, ranging among the remaining countries from 2.11 years to 9.48 years (compared with 1.17 years to 4.28 years for adults). Although children are less likely to experience cohabitation than adult women, the median duration of a childhood parental cohabitation is generally longer than the median adult premarital cohabitation. We find the four countries with percentages of parental cohabitation ending in marriage under 50% to be those with percentages of premarital cohabitation ending in marriage also under 50%: Canada, France, New Zealand, and the United States. Perhaps more surprisingly, the highest proportion of parental cohabitations ending in marriage at the median duration among the remaining nine countries is found in Sweden (81.6%).

Table 3.

Expected Duration of Exposure to and Route of Exit From Parental Cohabitation: Children Age 0 to 16

Country	Cohabitation Years	Median Durationa	Route of Exitb
			Marriage	Separation
France	1,583	9.48	23.8	76.2
Slovenia	591	6.11	76.9	23.1
Canada	724	4.17	30.5	69.5
Finland	444	3.48	69.7	30.3
New Zealand	660	3.03	34.5	65.5
Austria	606	2.82	76.0	24.0
Sweden	2,401	2.71	81.6	18.4
Germany	1,085	2.62	63.3	36.7
United States	1,630	2.11	48.2	51.8
Country  median		3.03	63.3	36.7

Note: Table sorted in descending order by expected median duration, excluding countries with fewer than 500 observed cohabitation-years during the reference period. Estimates derived from multiple-decrement life tables (see Preston et al., 2001, pp. 80–84).

^aDuration by which half of the parental cohabitation spells have ended in either marriage or separation.

^bFigures relate to the first half of the parental cohabitation that are ended by the median duration.

Summary

Table 4 summarizes national indicators and compares them to the ideal types presented in Table 1. Several countries exhibit indicators that do not conform entirely to one of the patterns. These countries are classified in Table 4 on the basis of the preponderance of fit between the observed values and the patterns. We mark cells that clearly do not conform to that pattern with a boldface. First, we find a few countries where cohabitation remains relatively rare and continues to play a marginal role in the landscape of family formation. We estimate that at early 1990s rates, fewer than 15% of women will experience cohabitation in Italy, Poland, and Spain. As a result, childhood experience of cohabitation is even rarer, which generates an insufficient sample of cohabitation spells to estimate other indicators reliably, such as median duration or percentage transitioning into marriage.

Table 4.

Family and Fertility Surveys Countries by Ideal-Typical Role of Cohabitation

			Empirical Indicators
			Adulthood Premarital Cohabitation			Childhood Parental Cohabitation
Role		Country	Incidence	Median Duration	Proportion Ending in Marriage	Incidence	Median Duration	Proportion Ending in Marriage
A.	Marginal	Italy	9.4	na		4.7	na
		Poland	3.6	na		2.6	na
		Spain	14.7	na		5.5	na
B.	Prelude to marriage	Belgium	27.5	2.39	76.2	4.5	naa
		Czech Republic	28.2	na		10.4	naa
		Hungary	23.2	na		7.1	naa
		Switzerland	59.4	1.78	75.6	5.0	naa
C.	Stage in marriage  process	Austria	58.3	2.69	76.8	18.0	2.82	76.0
		Finland	75.4	2.38	80.7	13.1	4.08 b	74.1
		Germany	53.0	2.24	53.3	12.7	2.62	63.3
		Latvia	47.0	naa		12.4	naa
		Slovenia	43.2	2.90	78.1	15.9	6.11	76.9
D.	Alternative to single	New Zealand	46.6	2.18	39.5	14.6	3.03	34.5
		United States	71.9	1.17	48.0	20.1	2.11	48.2
E.	Alternative to marriage	Canada	55.2	3.32	36.1	17.4	4.17	30.5
		France	83.3	4.28	46.3	25.9	9.48	23.8
F.	Indistinguishable  from marriage	Sweden	66.7	3.44	61.2	48.0	2.71	81.6
Median values			47.0	2.39	61.2	12.7	3.03	63.3
Cutoff values (higher/lower,  longer/shorter; see Table 1)			15.0	3.00	50.0	12.0	4.00	50.0

^a“na” indicates nonavailable indicators in countries where we observed fewer than 500 cohabitation-years during the reference period.

^bBoldface denotes indicators whose respective position relative to the corresponding cutoff value is contrary to the expectation for that type (see Table 1).

At the other end of the distribution, we estimate that 83.6% of adult women will experience cohabitation between ages 15 and 45 in France, and 53.5% of children will experience maternal cohabitation by age 16 in Sweden. In both countries, the median duration of an adult premarital cohabitation is relatively long (3.44 years in Sweden, 4.28 years in France), and the proportion of these premarital cohabitations ending in marriage is relatively low (61.2% in Sweden, 46.3% in France). Yet indicators for these countries differ when we estimate the likely trajectories of parental cohabitation. Although fewer children are expected to be born to cohabiting parents in France, half of them can expect to remain with unmarried cohabiting parents for at least 9.48 years, and only 23.8% of those who left parental cohabitation did so through marriage rather than through separation. In Sweden, the median duration is shorter, 2.71 years, but more important, the proportion exiting through parental marriage is much larger: 81.6%. Sweden thus appears to fit Kiernan’s (2001) description of the end point in the emergence of cohabitation as a family-building institution, when cohabitation eventually evolves to be almost indistinguishable from marriage. Children are born to cohabiting parents nearly as frequently as to married parents, and when they are, their parents are quite likely to marry later on.

Children born to cohabiting parents in France can expect to remain with unmarried cohabiting parents much longer than in Sweden, and are much less likely to see them marry. A substantial minority of French parents thus seem to choose long-term cohabitation as an alternative to marriage. Although his study focuses on adult cohabitation, Toulemon’s (1997) title also accurately describes childhood experience of parental cohabitation in asserting that “Cohabitation Is Here to Stay” in France. This seems to also be the case in Canada, where parental cohabitations are also fairly long (median duration 4.17 years), with only 30% ending in marriage. The incidence of both adulthood and childhood cohabitation is significantly lower than in either France or Sweden, however. The intermediate values for Canada likely result from the sharp difference in propensity to cohabit between Québec and the other provinces (Lapierre-Adamcyk, Le Bourdais, & Marcil-Gratton, 1999).

In the majority of countries studied here, most unmarried couples appear to enter cohabitation with the intention of marrying. Most cohabitation spells are relatively short and frequently result in marriage. For those resulting in marriage and childbearing, the normative sequence of marriage first, then birth is respected to a variable extent across countries. On one hand, cohabiting parents in Switzerland seem to conform to this timing, with 59.9% of adult women experiencing cohabitation, 75.6% of cohabitating partnerships resulting in marriage, and only 5.0% of a birth cohort experiencing a parental cohabitation by age 1. This characterization of family formation in Switzerland is consistent with previous studies (e.g., Charton & Wanner, 2001).

On the other hand, Austria may best represent the stage in the marriage process ideal type. Compared with Switzerland, roughly the same percentage of women experience unmarried cohabitation (58.8% vs. 59.9%), and roughly the same percentage of these cohabiting partnerships are expected to result in marriage (76.8% vs. 75.6%). The percentage of children experiencing parental cohabitation by age 1 is substantially higher in Austria, however (18.0% vs. 5.0%), suggesting that some of the cohabiting couples who eventually marry do so after rather than before their first birth. This particularity of the Austrian family system has also been described in previous studies (e.g., Prinz, 1995). Most couples in the remaining countries also seem to cohabit as a more or less timed transition to marriage; the indicators for these countries are intermediate between the Swiss and the Austrian values, with the notable exceptions of New Zealand and the United States.

Cohabitation is not negligible in the United States; about one half of adult women are expected to experience cohabitation at least once between ages 15 and 45, and one third of children are expected to experience maternal cohabitation by age 16, but the median duration of the cohabitation spells was the shortest of all of the countries reviewed here for adults (1.17 years) and for children (2.11 years). Couples who cohabit as a durable alternative to marriage are clearly a minority. However, couples who enter cohabitation as a prelude to marriage or even a stage in the marriage process do not clearly dominate either, because about as many cohabitation spells end in separation as in marriage for adults and for children. Rindfuss and VandenHeuvel (1990) described this cohabitation pattern of short duration and relatively high separation frequency as alternative to single. An indicator perhaps less consistent with this characterization of cohabitation in the United States is the nonnegligible percentage of children experiencing parental cohabitation. As noted above, this is largely due to the contribution of stepcohabitation, which likely results from the higher incidence of divorce in the United States than in European countries, and subsequent partnership formation out of marriage. Another American characteristic is the substantial percentage of children born to single noncohabiting mothers (Heuveline et al., 2003). Recent data from the Fragile Families project show that nearly two thirds of these mothers are romantically involved with the father, who is not living with the mother but visits more or less regularly (J. O. Teitler, personal communication, April 30, 2003). We may hence retain the alternative to single characterization of cohabitation in the United States, as long as we take into account the evolution of being single as a state that involves dating, sexual intercourse, and frequently, childbearing.

Discussion

As noted by Smock (2000), it is difficult to understand cohabitation by comparing it with other living arrangements, and to marriage in particular, because marriage itself is a moving target rather than a stable reference point. It is thus debatable whether the different contours of cohabitation found here to separate the United States and New Zealand from other countries reveal a lesser tendency of cohabiting partners in these two countries to behave as married partners rather than as single adults, or a fading of traditional characteristics of marriage, such as childbearing and expected durability. One interpretation of the relatively short median duration and high separation frequency of cohabitation spells in the United States and New Zealand is that cohabitation resembles being single, yet being single does not preclude childbearing in these two countries. An alternative interpretation would point to the differences in the reference institution, because marriage is not the same across countries and appears particularly vulnerable to disruption in these two countries. Life table estimates from the same data suggest that the relative risk of experiencing parental disruption for children born to cohabiting versus married parents is not higher in the United States or New Zealand than in other countries (Heuveline et al., 2003). Thus, cohabitation in these two countries could hence be as much a stage in the marriage process as it is in other countries, a process that exhibits a higher failure rate in these countries regardless of how it is started.

It is noteworthy that the three non-European countries are more difficult to characterize than their European counterparts, perhaps due to their internal diversity. As mentioned above, regional differences are sharp in Canada, and both New Zealand and the United States are ethnically diverse countries, with well-documented ethnic differences in family behavior in the United States (e.g., Bumpass & Lu, 2000; Manning, 1993). More generally, within-country heterogeneity is clearly important and not captured by the summary indicators presented here. It would be problematic if one interpreted these indicators as applying to whole populations rather than to use them for what they are: monodimensional indicators of the central tendency of a whole distribution. In the above analyses, our argument is simply that when a behavior becomes prevalent enough (e.g., adulthood cohabitation, childbearing while cohabiting before marriage, remaining together but unmarried following a birth), average indicators can detect it even if it remains a minority behavior. The presence of partners who cohabit as an alternative to marriage in France is a case in point. Although we see evidence of this behavior from the fact that after 9.48 years, half of the children whose parents were cohabiting at birth or by age 1 were still together and unmarried, these children only constitute 25.9% of a birth cohort; hence, less than 13% of a birth cohort can actually expect to live 9.48 years or more with cohabiting parents.

We are therefore not claiming that the patterns of family formation for which we found evidence in one country or another apply to the whole population of these countries, nor do we claim that these differences represent permanent national characteristics. The snapshot of cross-national differences in cohabitation presented here likely catches many of these countries in the midst of substantial transition. Because the age limit of the survey respondents is typically around age 45, we cannot go backward in time without encountering some well-known selectivity biases related to maternal age at birth (Rindfuss et al., 1982). Even with these limitations, the Family and Fertility Surveys data provide clear evidence of fast-paced changes (Heuveline et al., 2003). It is possible that the differences found here relate more to differences in the timing of a common transition from the traditional Western family system to a new family regime rather than to stable differences in family formation across countries. From these data, however, there is arguably more evidence of path dependency (e.g., postnatal marriage in Austria) than clear signs of convergence.

In any event, we believe that the single-period differences presented here should be attributed partly to cross-national variation in the underlying distributions of individual traits related to the propensity to cohabit, marry, and raise children out of wedlock, and also to the contextual factors that affect these propensities for all or part of a population. As mentioned above, earlier work (e.g., Kiernan, 2000) has shown that within-country differences influenced by individual traits (e.g., education) are small compared with between-country differences. Although the identification of the macrolevel factors responsible for these differences is beyond the scope of this article, the rise of cohabitation has generated different institutional responses across the countries considered here. For example, the current U.S. administration has launched a “marriage initiative” to facilitate the transition from cohabitation to marriage. In contrast, the French government recently instituted the Pacte Civil de Solidarité as an alternative living arrangement, removing some of the requirements of marriage while providing social protections more comparable with that of married couples (Martin & Théry, 2001). Although we have learned a great deal about differential propensities to cohabit from individual-level survey data and cross-sectional analyses (see Smock 2000, for a review), there remains a great need for complementary comparative research on the effects of contextual-level factors on these propensities.

APPENDIX.

Family and Fertility Survey Names, Dates, Respondent Upper Age Limits, and Sample Sizes

Country	Family and Fertility Survey Name	Survey Dates		Survey Age Range	N		Reference Period	Period Age Range
		From	To		Women	Children
Austria	Austrian Family and Fertility Survey	12/95	4/96	20–55	4,535	7,039	1990–92	18–45
Belgiumb	Nationale Enquête Gezinsontwikkeling,  V (NEGO V)	12/91	12/92	20–42	3,231	3,771	1988–90	20–37
Canada	Family History Survey	1/95	12/95	15–55	4,050	5,452	1991–93	15–45
Czech Republic	Czech Family and Fertility Survey	11/97	12/97	15–45	1,730	2,363	1993–95	15–40
Finland	Population Survey	8/89	1/90	22–53	4,140	6,220	1984–86	20–45
France	Annual Employment Survey	1/94	5/94	20–51	2,936	4,525	1989–91	18–45
Germany	Familienbildung Und  Kinderwunsch in Deutschland	7/92	7/92	20–40	5,990	5,923	1989–91	20–36
Hungary	European Fertility and Family Survey	12/92	12/93	18–42	3,590	4,933	1989–91	18–37
Italy	National Sample Survey on Fertility  Control and Expectation	11/95	1/96	20–50	4,794	5,495	1990–92	18–43
Latvia	Latvian Family and Fertility Survey	9/95	10/95	18–50	2,688	3,902	1991–93	17–45
New Zealand	Women, Work, Family, and Education  Survey	9/95	11/95	20–60	2,901	5,847	1990–92	18–45
Poland	Fertility and Family in Poland, 1991	12/91	12/91	18–50	4,210	7,016	1988–90	18–45
Slovenia	Fertility Behavior of the Inhabitants of  Slovenia	11/94	11/95	15–46	2,788	4,005	1991–93	15–41
Spain	Spanish Fertility and Family Survey	11/94	11/95	18–50	4,021	5,065	1991–93	18–45
Sweden	Family Study, 1992	10/92	5/93	22–45	3,272	4,686	1987–89	20–38
Switzerland	Swiss Family Microcensus	10/94	6/95	20–50	3,876	5,440	1989–91	18–43
United States	National Survey of Family Growth,  Cycle 5	1/95	10/95	15–45	10,711	14,824	1991–93	15–40

^aSurvey names taken from United Nations Economic Commission for Europe (2003).

^bThe Belgium survey covers only the Flanders region.