© 2006 Royal Statistical Society0964–1998/06/169757
J. R. Statist. Soc. A (2006)
169, Part 4, pp. 757–779
The formation and outcomes of cohabiting and
marital partnerships in early adulthood: the role
of previous partnership experience
University of Bristol, UK
and Constantinos Kallis and Heather Joshi
Institute of Education, London, UK
[Received August 2005. Final revision January 2006]
Summary. We estimate a joint model of the formation and dissolution of cohabiting and marital
unions among British women who were born in 1970.The focus of the analysis is the effect of
previous cohabitation and marriage on subsequent partnership transitions.We use a multilevel
simultaneous equations event history model to allow for residual correlation between the haz-
ards of moving from an unpartnered state into cohabitation or marriage, converting a cohabiting
union into marriage and dissolution of either form of union.A simultaneous modelling approach
allows for the joint determination of these transitions, which may otherwise bias estimates of
the effects of previous partnership outcomes on later transitions.
Keywords: Multiple states; Partnership dissolution; Partnership formation; Repeated events;
Simultaneous equations model
Most studies of the formation and outcomes of co-residential partnerships in Britain con-
sider only the first union, typically investigating the timing of entry into marriage and marital
breakdown. The focus on the first marriage is consistent with the traditional route to family
formation whereby an individual left the parental home to marry before having children. There
of union formation and childbearing (Berrington, 2003). A notable change is the sharp increase
their definition of a partnership to include non-marital co-residential unions, and to study the
partnership behaviour is an increase in the rate of dissolution, leading to multiple partnerships
per individual (Ferri and Smith, 2003; Murphy and Wang, 1999). Nevertheless, studies that
consider the formation and dissolution of second- and higher order partnerships are rare.
In this paper, we study repartnering behaviour, and in particular how it is shaped by past
partnership events. Specifically, we examine the relationship between previous cohabitation or
marriage and the timing of the formation and dissolution of subsequent partnerships. We ask
Address for correspondence: Fiona Steele, Centre for Multilevel Modelling, Graduate School of Education,
University of Bristol, 35 Berkeley Square, Bristol, BS8 1JA, UK.
758 F. Steele, C. Kallis and H. Joshi
risk of marital dissolution. There is some evidence for first marriages in Britain between 1970
and 1989 that those which were preceded by premarital cohabitation were more likely to dis-
solve (Haskey (1999), using data from the 1989 General Household Survey). There was, how-
ever, no control for other characteristics, observed or otherwise. In a study of first marriages
among the 1958 birth cohort which adjusts for the effects of a range of background charac-
teristics, Berrington and Diamond (1999) found that previous cohabitation, with either the
marital or a previous partner, is associated with a higher risk of dissolution. Haskey (1992) also
cited evidence from the USA, Canada and Sweden that showed a positive association between
cohabitation and the breakdown of a subsequent marriage.
Following the break-up of a marriage, most divorcees (especially women) cohabit with their
next partner (Haskey, 1999). We explore whether the experience of divorce acts as a deterrent
to converting these cohabitations into another marriage. Finally, we consider the relationship
between previous marriage or cohabitation and subsequent partnership dissolution. Are later
partnerships at a higher risk of dissolution because the possibility of break-up has already been
experienced, or does the previous experience, on the contrary, discourage repeating it? Previous
research on divorce in Britain tends to support the former hypothesis. Haskey (1983, 1996) has
found that marriages in which a partner is remarrying after divorce carry an above-average risk
of subsequent divorce.
A major difficulty in determining the effects of previous events on the timing of later events
invariant characteristics that affect the probability of event occurrence across the observation
period, leading to correlation in the durations between successive events in the same process.
For example, the positive association that is observed between previous marital breakdown
and the rate of dissolution of remarriages may be due to differences between individuals who
marry only once and those who have divorced; the latter group may contain individuals whose
unobserved characteristics put them at high risk of separation from any partner. Another form
of selection arises when the unobserved risk factors for events in one process are correlated with
those for events in a different but related process. A study in the USA found that this type of
selection explained an apparent positive effect of premarital cohabitation on the risk of marital
breakdown (Lillard et al., 1995). Women who chose to cohabit with their partner before mar-
riage differed from those who married directly in unobserved ways which were related to their
repeated entries into and exits from partnerships.
Using data from the 1970 British cohort, we study the formation and outcomes of all part-
nerships that were formed by women before age 30 years. In these years of their early adulthood
nearly all women formed co-residential partnerships, but they differed from previous genera-
ships to be cohabitations rather than marriage (Ferri and Smith, 2003). These partnerships
were also being formed in an era when divorce was becoming a normal occurrence which may
have encouraged a more relaxed attitude to the importance of entering formal marriage, while
parallel developments in education and employment meant that these young women need be
less dependent on finding a husband than their predecessors had been. We focus on women for
ease of comparison with our previous research on partnership transitions among the 1958 and
1970 birth cohorts (Steele et al., 2005, 2006), and because men’s later age at first partnership
results in shorter exposure to multiple partnerships.
Our study builds on previous research in several ways. First, whereas other studies that have
jointly modelled partnership formation and dissolution (Aassve et al., 2006; Goldstein et al.,
Formation and Outcomes of Cohabiting and Marital Partnerships 759
Second, we extend the work of Steele et al. (2005) who jointly modelled transitions out of mar-
riage and cohabitation to incorporate transitions into both forms of partnership. Third, we
extend the simultaneous equations model of Lillard et al. (1995) to examine the effect on the
risk of marital dissolution of premarital cohabitation not only with the current partner but also
with any previous partner.
the need for a joint modelling approach
Assessing the effect of previous partnership events on subsequent transitions:
In an analysis of multiple partnership transitions, and in particular estimation of the effects
of previous partnership outcomes on subsequent transitions, it is important to consider the
potential for observed and unobserved heterogeneity and the endogeneity of past partnership
outcomes with respect to later behaviour. One covariate which may or may not be observed
which links cohabitation and marital break-up is religious belief and whether the marriage
was a religious ceremony (Murphy, 1983). Other relevant personality traits are inherently less
observable. To illustrate the implications of unobserved heterogeneity, suppose that we are
interested in the effect of divorce on the stability of a later marriage. There are likely to be unob-
served individual characteristics, which are constant over time, that influence an individual’s
risk of dissolution in any marriage that they form. The presence of unobserved time invariant
risk factors leads to unobserved heterogeneity between individuals. If such heterogeneity is not
taken into account, the effect of previous marriage on subsequent marital dissolution cannot be
interpreted in a causal way; rather the risk of dissolution among the previously married is likely
them at increased risk. To take an example, suppose that the true effect of previous marriage
is zero, i.e. the risk of dissolution is actually the same for first and second marriages. If there
is selection of higher risk individuals into remarriage and this is ignored in the analysis, it will
appear that second marriages carry a higher risk of dissolution than first marriages.
By the same argument, failure to account for unobserved heterogeneity is likely to lead to a
biased estimate of the effect of previous cohabitation on the risk of dissolution of later cohab-
iting partnerships. Aassve et al. (2006) and Lillard et al. (1995) highlighted the need to control
for unobserved heterogeneity when considering repeated events. Using data from the British
Household Panel Survey, Aassve et al. (2006) found that an apparently strong increase in the
risk of separation with the number of previous partners vanished after adjusting for the effects
of unobserved individual-specific factors. In the USA, Lillard et al. (1995) also found that the
increased rate of dissolution that was observed among previously married women disappeared
after accounting for unobserved heterogeneity.
These examples concern the effect of a previous event on the probability that an event of
the same type occurs again. More generally, we may be interested in the effect of any previous
partnership event (e.g. the breakdown of a marriage or cohabitation, and premarital cohabi-
tation) on later transitions, which include both the dissolution and the formation of marital
and cohabiting partnerships. To examine the effect of the break-up of a marriage (or cohab-
itation) on the stability of a later cohabitation (or marriage), it is important to allow for the
possibility that individuals with a high risk of separating from a cohabiting partner (accord-
ing to unobserved characteristics) might also have a high risk of marital dissolution. This may
be achieved by jointly modelling the outcomes of marriage and cohabitation and allowing for
residual correlation between the transitions. Using this approach, Steele et al. (2005) found evi-
dence of a positive correlation between the unmeasured woman-specific influences on the risk
of separation from marriage and cohabitation.
760 F. Steele, C. Kallis and H. Joshi
Another example of the endogeneity of previous partnership transitions arises in investiga-
tions of the effect of prior cohabitation on the risk that a subsequent marriage is dissolved.
There has been much debate about the effect of premarital cohabitation on the risk of mar-
ital breakdown. An increased rate of dissolution is often observed among couples who lived
together before marriage, but many commentators argue that this is due to selection rather
than a causal effect. Lillard et al. (1995) tested the selection hypothesis by using a simulta-
neous equations model in which the hazard of marital dissolution was modelled jointly with the
decision to cohabit before marriage; they found that the apparent increased risk of separation
among the cohabitors was due to self-selection of women with a high risk of dissolution into
cohabitation. This selection effect was not captured by covariates and therefore led to a positive
correlation between the woman-specific unobservables that affect the hazard of dissolution and
the propensity to cohabit. Put another way, the decision to cohabit before marriage is endoge-
nous to marital dissolution. No British study of the link between premarital cohabitation and
marital dissolution has allowed for selection on unobservables. Of course cohabitation is not
always a precursor to marriage; cohabitation may also take the form of a short-term convenient
alternative to a non-co-resident sexual relationship, or a longer-term alternative to marriage
(Murphy, 2000). The association between any previous cohabitation experience and marital
dissolution may be subject to the same form of selection as was reported by Lillard et al. (1995),
i.e. that individuals with an above-average propensity to cohabit also have an above-average
propensity to separate if they should marry.
As discussed above, failure to account for correlation between different types of partner-
ship transition may lead to incorrect inferences. There are at least two additional benefits of
adopting a joint modelling approach in the analysis of partnership histories. First, estimates
of the residual correlations across transitions can provide useful insights into partnership pat-
terns. Goldstein et al. (2004), in an analysis of unions that were formed by male members of
the 1958 birth cohort between the ages of 16 and 33 years, found a weak negative residual
correlation between the hazards of entry into and exit from partnerships. A negative correla-
tion may be due to the presence of individuals with a strong latent desire to be in a partner-
ship, who therefore have a tendency towards long partnerships and short unpartnered episodes.
Aassve et al. (2006), however, using data on multiple cohorts from the British Household Panel
Survey, found that the correlation between the hazards of partnership formation and dissolu-
A positive correlation suggests the presence of women who make quick transitions into and out
of partnerships. A second advantage of estimating one joint model for all types of partnership
transition rather than independent models is that tests of the equivalence of covariate effects
across transitions may be carried out. For example, we may be interested in comparing the
effects of previous partnership experience on the dissolution risks of marital and cohabiting
We consider five transitions from three partnership states—single, cohabitation and marriage.
We distinguish between partnership formation, which is defined as a transition into cohab-
itation or marriage from the single state, and partnership outcomes, which include dissolu-
tion (a transition from marriage or cohabitation to the single state) and the conversion of
cohabitation into marriage. Strictly, a married or cohabiting woman could repartner imme-
diately after a separation, leading to a transition to marriage or cohabitation rather than to
Definitions of state transitions and episodes
Formation and Outcomes of Cohabiting and Marital Partnerships 777
tional indicator, current enrolment in full-time education, we find negative effects of enrolment
on the hazard of partnership entry (Table 4) and on the odds of marriage among cohabitors
(Table 5). Our findings agree with those from previous research (Berrington, 2003; Ermisch and
The direction of the effect of attainment on the odds of dissolution is more difficult to pre-
dict. To the extent that more educated women have better employment opportunities, we might
expect that such women would be less likely to remain in an unsatisfactory partnership than
those with little education, leading to a positive effect of education on dissolution. In contrast,
having a low level of education is associated with social disadvantage, which is usually found to
have a destabilizing effect on partnerships. Whereas the empirical evidence from other studies
(Berrington and Diamond, 1999; Steele et al., 2005) supports the latter hypothesis, we find
a non-monotone relationship between education and the risk of dissolution for cohabitation
(Table 5), and no effect of education on the risk of marital separation (Table 6). However, there
finding was reported by Berrington and Diamond (1999) and Ermisch and Francesconi (2000).
Taken together with the negative effect of educational enrolment on partnership formation, the
findings from this and previous studies suggest that full-time education is incompatible with a
Berrington and Diamond, 2000; Steele et al. 2005, 2006) we find weak effects of social class
on partnership formation and dissolution. This is perhaps unsurprising given that the effects
of family background are likely to be mediated by individuals’ own characteristics and experi-
ences, particularly childbearing and education. We find that having a father in a professional,
managerial or skilled occupation is associated with delayed entry into cohabitation. There is
also evidence that women from less advantaged backgrounds (social classes IV and V) are at
decreased risk of experiencing partnership dissolution, whether they are married or cohabiting.
We find that the experience of parental separation during childhood is associated with ear-
lier cohabitation and later marriage, lower odds of moving from cohabitation to marriage and
higher risk of dissolution (Tables 4–6). Similar effects have been found for the 1958 cohort (Ber-
rington and Diamond, 1999; Kiernan and Cherlin, 1999; Steele et al., 2005, 2006). Thus there is
evidence that women who experience marital breakdown during childhood favour cohabitation
lived with both parents throughout childhood. The effects for the category ‘missing’ are also
worthy of mention. Compared with women whose parental status is known, this small group of
women stayed single for longer, were less likely to marry a cohabiting partner and experienced
a higher risk of marital dissolution. Clearly missingness on this particular indicator cannot be
assumed random. As such, the estimated effects of parental separation should be interpreted
There is some evidence of regional variation in partnership transition rates. We find that
single women who were born in the north of England, Scotland or Northern Ireland are the
least likely to cohabit and the most likely to marry (Table 4). (Note that our sample contains
only five respondents who were born in Northern Ireland as those who were born and stayed
there were not followed up.) In their study of first-partnership formation in the 1958 cohort,
Berrington and Diamond (2000) considered the effect of region at age 16 years and found that
men and women from London or South-East England at age 16 years cohabited sooner and
married later than their counterparts from other parts of Britain. We also find that, compared
with women who were born in other parts of Britain, cohabitations that were formed by women
Family background. In common with most previous research (Aassve et al., 2006;
778F. Steele, C. Kallis and H. Joshi
from London and the South-East are shorter in duration and less likely to be converted to mar-
riage (Table 5). There is no regional variation in the risk of marital dissolution in this cohort
(Table 6), nor in the 1958 cohort (Berrington and Diamond, 1999; Steele et al., 2005).
Among cohabiting women, we find that those who have not married before are more likely to
marry than those who have been married previously, after allowing for selection on woman-
specific unobservables. There is, however, no effect of previous partnership breakdown on the
before marriage are no more or less likely to separate than couples who married directly. These
ages 16 and 42 years, although that study did not model partnership outcomes jointly with their
they are at no greater risk of separation than are women in their first partnership. Although
the presence of unobserved heterogeneity in the risks of dissolution of cohabiting and married
women implies that there are women who are more prone to separate than others, it is not the
experience of dissolution that places them at greater risk in their later partnerships. In other
words, partnership dissolution does not appear to have an internally generated dynamic.
eralize to men. For example, Haskey (1999) reported that among divorcees men are more likely
to remarry than women. The analysis is also limited to the early partnership transitions of a
single birth cohort although, as noted above, our findings are broadly consistent with those
for the 1958 cohort to age 42 years (Steele et al., 2005). Finally, we assume that the residual
correlation between the hazards of partnership formation and dissolution is due to their shared
allow for selection on transient unobservables, which would include time-varying attributes of
women (other than age, fertility status and educational enrolment and attainment) and partner
characteristics that vary between different partners of the same woman.
A major difference between this study and others that have used a joint modelling approach
(Aassve et al., 2006; Goldstein et al., 2004) is our treatment of cohabitation and marriage as
separate partnership states. This decision is supported by our finding that the characteristics
are embarking on their second partnership are more likely to cohabit and less likely to marry
riages are much more likely among unpartnered women than among women who are already
cohabiting with the father. The effects on the risk of dissolution of previous partnership expe-
rience and children are similar for cohabitation and marriage. For both forms of partnership,
there is no evidence of an association between a previous partnership breakdown and the risk
of subsequent dissolution, and the presence of a young child appears to have a cementing effect.
Nevertheless, overall, marriages are more stable than cohabitations and there are differences in
the effects of other predictors by type of partnership. For example, pregnancy has a stronger
stabilizing effect for married couples, and the positive association between parental separation
and the respondent’s risk of dissolution is stronger for marriage than for cohabitation.
Formation and Outcomes of Cohabiting and Marital Partnerships779
We thank two referees whose insightful comments on an earlier version of this paper resulted
in substantial improvements to the text. This research was funded by the UK Economic and
Social Research Council under the ‘Research methods programme’ (award H333250044).
in Great Britain. J. R. Statist. Soc. A, to be published.
Berrington, A. (2001) Entry into parenthood and the outcome of cohabiting partnerships in Britain. J. Marr.
Famly, 63, 80–96.
Berrington, A. (2003) Change and continuity in family formation among young adults in Britain. Applications
and Policy Working Paper A03/04. Southampton Statistical Sciences Research Institute, University of South-
Berrington, A. and Diamond, I. (1999) Marital dissolution among the 1958 British birth cohort: the role of
cohabitation. Popln Stud., 53, 19–38.
Berrington, A. and Diamond, I. (2000) Marriage or cohabitation: a competing risks analysis of first-partnership
formation among the 1958 British birth cohort. J. R. Statist. Soc. A, 163, 127–151.
Blossfeld, H.-P., Klijzing, E., Pohl, K. and Rohwer, G. (1999) Why do cohabiting couples marry?: an example of
a causal event history approach to interdependent systems. Qual. Quant., 33, 229–242.
Cambridge University Press.
Bynner, J., Ferri, E. and Shepherd, P. (1997) Twenty-something in the 1990s: Getting On, Getting By, Getting
Nowhere. Aldershot: Ashgate.
Ermisch, J. (2005) The puzzling rise of childbearing outside marriage. In Understanding Social Change (eds A.
Heath, J. Ermisch and D. Gallie), pp. 22–53. Oxford: Oxford University Press.
Ermisch, J. and Francesconi, M. (2000) Cohabitation in Great Britain: not for long, but here to stay. J. R. Statist.
Soc. A, 163, 153–171.
Ferri, E. and Smith, K. (2003) Partnerships and parenthood. In Changing Britain, Changing Lives (eds E. Ferri,
J. Bynner and M. Wadsworth), pp. 105–132. London: Institute of Education.
Goldstein, H., Pan, H. and Bynner, J. (2004) A flexible procedure for analysing longitudinal event histories using
a multilevel model. Understand. Statist., 3, 85–99.
regression analysis. Am. J. Epidem., 142, 1255–1268.
Haskey, J. (1983) Marital status before marriage and age at marriage: their influence on the chance of divorce.
Popln Trends, 32, 21–28.
Haskey, J. (1992) Pre-marital cohabitation and the probability of subsequent divorce: analyses using new data
from the General Household Survey. Popln Trends, 68, 10–19.
Haskey, J. (1996) The proportion of married couples who divorce: past patterns and current prospects. Popln
Trends, 83, 23–36.
Haskey, J. (1999) Divorce and remarriage in England and Wales. Popln Trends, 95, 18–22.
Kiernan, K. E. and Cherlin, A. J. (1999) Parental divorce and partnership dissolution in adulthood: evidence
from a British cohort study. Popln Stud., 53, 39–48.
Lillard, L. A., Brien, M. J. and Waite, L. J. (1995) Premarital cohabitation and subsequent marital dissolution: a
matter of self-selection? Demography, 22, 437–457.
Murphy, M. (1983) Demographic and socio-economic influences on recent British marital breakdown pattern.
Popln Stud., 39, 441–460.
Murphy, M. (2000) Editorial: cohabitation in Britain. J. R. Statist. Soc. A, 163, 123–126.
and Households in the 1990s (ed. S. McRae), pp. 100–138. Oxford: Oxford University Press.
Rasbash, J., Steele, F., Browne, W. J. and Prosser, B. (2004) A User’s Guide to MLwiN, Version 2.0. London:
Institute of Education.
Steele, F., Goldstein, H. and Browne, W. (2004) A general multistate competing risks model for event history
data, with an application to the study of contraceptive use dynamics. Statist. Modllng, 4, 145–159.
Steele, F., Joshi, H., Kallis, C. and Goldstein, H. (2006) Cohabitation and childbearing: changing compatibility
between two cohorts of young British women. Popln Stud., to be published.
Steele, F., Kallis, C., Goldstein, H. and Joshi, H. (2005) The relationship between childbearing and transitions
from marriage and cohabitation in Great Britain. Demography, 42, 647–673.