Children and Marital Instability in the UK

Abstract

There are many good theoretical reasons which suggest that children are a stabilizing factor in marriage. Most empirical research confirms that this is indeed the case. In this paper, we use data from three British surveys to demonstrate that in Britain the effect of children on marital stability has changed from stabilizing to destabilizing in the last two decades. This shift began in the 1980s, and by the 1990s couples with children are at a substantially higher risk of divorce than similar but childless couples. We also explore some concomitants of the change in the children effect, again using multiple data sources. We show that the destabilizing effect of children is most pronounced for low income households.

Full text

Children and Marital Instability in the UK
Tak Wing Chan
Department of Sociology
University of Oxford
Brendan Halpin
Department of Government and Society
University of Limerick
December 2, 2001
Draft only. Comments very welcome. But please don’t cite or quote without permission.
1

Children and Marital Instability in the UK
Abstract
There are many good theoretical reasons which suggest that children are a stabilizing
factor in marriage. Most empirical research confirms that this is indeed the case. In this
paper, we use data from three British surveys to demonstrate that in Britain the effect
of children on marital stability has changed from stabilizing to destabilizing in the last
two decades. This shift began in the 1980s, and by the 1990s couples with children
are at a substantially higher risk of divorce than similar but childless couples. We also
explore some concomitants of the change in the children effect, again using multiple
data sources. We show that the destabilizing effect of children is most pronounced for
low income households.
2

1 Context and Motivation
In a recent paper, Chan and Halpin (2001) report that children destabilize marriage in
the UK. In that paper, they use data from the first eight waves of the British Household
Panel Study (BHPS 1991–98) to examine the hazard of divorce for women in their
first marriage.
1
Their primary interest is to test the so-called independence hypothesis.
Although their main result in relation to this hypothesis seems robust, the finding that
children destabilize marriage is very puzzling. As we shall elaborate in section 2, this
unexpected finding goes against many theoretical arguments and the results obtained
in most empirical research.
We have two goals in this paper. First, we seek to demonstrate that children do
destabilize marriage in the UK. We shall do so by replicating the finding of Chan and
Halpin (2001) with two other British data sets. Our analyses point to a shift over
marriage cohorts in how children affect marital stability. This shift began in the 1980s,
and by the 1990s couples with children are at a substantially higher risk of divorce than
similar but childless couples. Our second goal is to explore the possible reasons for
this shift. We shall consider the trends of social attitude towards divorce, the effect of
children on the chance of repartnering, and the association between home ownership
and divorce. We also explore how the destabilizing effect of children might depend on
their own characteristics (e.g. age, sex) or their parents’ characteristics. We shall show
that household income is a crucial mediating factor.
The rest of this paper is structured as follows. In section 2, we review the the-
oretical and empirical literature on children and marital stability. Section 3 reports
the result of our analysis of the retrospective life history data taken from two British
surveys (the Family and Working Lives Survey 1994–95 and the General Household
Survey 1991–95).
2
In section 4, we draw on data from the British Social Attitudes Sur-
veys, the Family and Working Lives Survey and the British Household Panel Study to
1
In this paper, we use the term divorce to refer to divorce or separation.
2
Technical details of all data sets used in this paper can be found in Appendix A.
3

explore the children’s effect in detail. Finally, in section 5, we summarize our results,
suggest some interpretations, and consider potential policy implications.
2 Theories and Empirical Evidence
Why are children (thought to be) a stabilizing factor in marriage? It is likely that the
observed association between children and marital stability is partly due to selection, in
that couples anticipating divorce will delay or avoid having children. Having said that,
there are good reasons to believe that children actually stabilize marriage. The most
fundamental reason is that children are a marital-specific investment, and they would
become less valuable to the parents should the marriage dissolve (Becker, Landes and
Michael 1977). In a straightforward sense, children ‘belong’ to a couple, not to just
one parent. Should the marriage dissolve, the costs and benefits of having children,
and the related emotional and social implications would change for both parents.
The argument of marital-specific investment can be elaborated and extended in sev-
eral ways. First, couples with children often have a higher level of division of labour
than childless couples.
3
Following the argument of Becker et al. (1977), more spe-
cialization should lead to a higher expected gain from marriage, and therefore greater
marital stability. Morgan, Nye and Condran (1988) remind us that the Durkheimian
notion of ‘organic solidarity’ is quite similar to the argument of Becker et al.: parent-
hood often implies a greater degree of role differentiation between the spouses which
increases conjugal solidarity.
Morgan et al. (1988) also point out that children create a web of obligations and
attachment between parents and children, between the parents themselves, and also
between the nuclear family and the wider kinship group. All these should increase
3
Research into domestic division of labour repeatedly shows that the presence of children increases
the gender gap in housework (South and Spitze 1994, Bianchi et al. 2000). Also, despite the general
increase in female labour force participation, mothers are still less likely to work, and if they do work,
less likely to work full time, than non-mothers. In contrast, marriage and children increase men’s
labour supply (Akerlof 1998), though this association might be modified by men’s gender role attitudes
(Kaufman and Uhlenberg 2000).
4

marital stability.
Children’s effect on marital stability might also work through their anticipated im-
pact on the prospective custodial parent’s (usually the mother’s) chance of repartner-
ing. Divorced women with (especially young) children might have a strong motivation
to find a new partner, partly in order to secure financial support, but also to complete
the family. It is well known that women suffer financially from divorce. Divorcees
with young children face particularly acute difficulties in having to manage childcare
and paid work alone. They might also think that their children need a father. However,
as Koo, Suchindran and Griffith (1984:452–453) point out, ‘the prospect of assuming
the financial and parental responsibilities for a woman’s children may deter prospective
grooms. Children may also interfere with a woman’s ability to engage in social activi-
ties that allow her to search for and develop relationships with prospective spouses, by
limiting her free time and energy and, in some cases, by actively opposing her dating
or re-marrying’.
Most empirical research suggests that the probability of remarriage or cohabitation
is lowerfor divorcees with children (Lampard and Peggs2000).
4
In this sense, children
weaken the custodial parent’s position in the repartnering market. It is likely that most
prospective custodial parents know this, and if we further assume that most of them
would want to find a new partner eventually, children would deter women walking
away from an unhappy marriage.
5
As for the non-custodial parents (usually the fathers), they often lose access to their
children or have access to children restricted after divorce. This in itself is likely to be
a deterrence against divorce. But a further mechanism might operate via their concern
for their children’s welfare. Loss of access makes it difficult for the non-custodial par-
4
Koo et al. (1984) show that in the US this effect is significant only for white women divorcees with
3 or more children.
5
Huber and Spitze (1980) report that having a child under six reduces the probability of both hus-
bands and wives thinking about divorce, although this effect is significant for husbands only. Also, for
couples whose youngest child is between 6 and 11, the thought of divorce increases for both husbands
and wives, but this effect is significant for wives only. On the basis of the latter finding, they suggest
that the presence of children may deter divorce less now than they have in the past.
5

ents to monitor effectively how (much) resources are spent on the children by their ex-
spouse. Weiss and Willis (1985) argue that as children are a public good to the parents,
when effective monitoring of resource allocation is difficult, there is a disincentive for
the non-custodial parents to transfer resources to their children, even though they care
about the children’s welfare. This is an instance of the ‘free-rider’ problem, and it
explains why many divorced fathers fail to pay adequate child support. Partly as a re-
sult of underinvestment, children with divorced parents tend to do worse than children
from intact families over a range of outcome measures, such as educational attainment
(McLanahan and Sandefur 1994). Of course, parents (custodial or non-custodial) need
not know the Weiss and Willis’ argument. As long as they believe that divorce is bad
for their children, and that they care about their children’s welfare, children will be a
deterrence to divorce.
Attitudes towards marriage, family and childbearing have changed drastically in
Western industrial societies, both between birth cohorts and within cohort over time
(Thornton 1985, 1989). However, despite a shift towardsgreateracceptance of divorce,
normative sanction against divorce is still stronger for couples with children than for
childless couples. No doubt this is partly due to concern about the negative effect of
divorce on children. We shall use data from the British Social Attitudes Surveys to
show that this is still the case in the UK.
We have considered several reasons as to why children should be associated with
greater marital stability. In fact, this association has been confirmed by most empirical
research. As readers can see from Table 1, these studies cover a range of industrial
countries, including the UK. They all use nationally representative survey data or, in
some cases, population register data, and appropriate statistical tools (generally speak-
ing hazard models or cognate techniques). There is no reason to doubt the results
reported therein. In short, both theories and empirical evidence weigh against the find-
ing that children destabilize marriage in the UK.
6

Table 1: Empirical research reporting a positive association between children and mar-
ital stability
country reference data used and time coverage
Australia Bracher et al. (1993) 1986 National Survey of the Australian Family
Finland Jalovaara (2001) 1990 Census linked with Divorce Records
Germany Diekmann and Engelhardt (1999) 1988 Family Survey of the German Youth Institute
Sweden Andersson (1997) Swedish Fertility Register (1961–94)
UK Berrington and Diamond (1999) National Child Development Study (1958–91)
US Morgan and Rindfuss (1985) June 1980 Current Population Survey
US Tzeng and Mare (1995) NLS
a
of Youth (1979–87)
NLS of Young Men (1966–81)
NLS of Young Women (1968–85)
US Weiss and Willis (1997) NLS of the High School Class of 1972 (1972–86)
US South (2001) Panel Study of Income Dynamics (1969–93)
b
a
National Longitudinal Survey.
b
The stabilizing effect reported in this study is not statistically significant.
2.1 Counter-evidence
But is the result of Chan and Halpin (2001) really so implausible? Though few and
far between, there are a handful of studies with similar findings. For example, Cherlin
(1977) analyzes data from the first four years of the National Longitudinal Survey of
Women (1967–71), and reports that although preschool children deter separation and
divorce, school children (aged 6 to 17) have no effect on marital stability. This differ-
ence, he argues, exists because much more time, energy, money and other resources are
needed in the care of preschoolers, which dissuades parents with very young children
from divorce.
Similarly, Waite and Lillard (1991) analyze data from the Panel Study of Income
Dynamics (1968–85) and show that first-borns stabilize marriage in their preschool
years, and other children stabilize marriage only when they are very young. Older
children and children born before marriage significantly increases the divorce risk.
More importantly for our purpose, in a recent paper which is also based on the
BHPS (1991–97), B¨oheim and Ermisch (2001) report that the risk of partnership dis-
solution increases with the number of children. But since B¨oheim and Ermisch are
7

primarily interested in the effect of financial surprise on marital stability (Weiss and
Willis 1997), they have not investigated the destabilizing effect of children in detail.
Studies reporting such counter-evidence are in the minority. But they are consis-
tent with research on marital satisfaction which shows that children are a stressor in
marriage. Couples consistently report lower marital satisfaction with the arrival of
children (VanLaningham, Johnson and Amato 2001) and higher marital satisfaction
when the children leave the parental home (White and Edwards 1990). In the extreme
but not very rare case of children with disability, where children-related stress is likely
to be very much greater, there is evidence to suggest that ‘at very age, children in poor
health are more likely to see their parents divorce. This relationship is strongest for
children between six and nine years old’ (Mauldon 1992:356).
6
The results of Cherlin, Waite and Lillard, and B¨oheim and Ermisch also highlight
the need to disentangle the multifaceted ways in which children affect marital stability.
The question is not whether children increase or reduce divorce risks, but under what
conditions would they do so. We have seen that the children effect might depend on
their age and number. Furthermore, the timing of conception and birth (pre-marital vs
marital) might be important. Becker et al. (1977:1151) argue that women who became
pregnant accidentally have an incentive to get married quickly in order to ‘legitimate’
their child. This shortens the search for marital partner, leading to a sub-optimal match,
a lower gain from marriage, and thus greater marital instability. By a similar argument,
children born before marriage might also destabilize marriage.
The sex of the child might also be important. Billed as ‘perhaps the most inter-
esting finding of the decade [of the 1980s]’ (White 1990), Morgan et al. (1988) report
that sons reduce the divorce risk 9% more than daughters. This, they suggest, is be-
cause fathers often take a more active role in raising sons than in raising daughters,
for example, in teaching sons to play and appreciate sports. In contrast, the raising
6
Mauldon (1992:353) notes that, in the 1981 National Health Interview Survey, major long-term
problems that are not usually congenital (e.g. stammer, stutter, or other speech defect) account for 6.1%
of all health problems among children mentioned by the respondent, and major long-term problems that
are mainly congenital (e.g. clubfoot, deformed foot or leg) account for another 4.9%.
8

and socialization of daughters are, to a greater extent, left to mothers. Such a division
in childrearing practice is in effect a second tier of role differentiation, which further
promotes conjugal organic solidarity and reduces the divorce risks.
7
Finally, there is growing evidence that parental investment in children and parent-
child relationship in general vary by family type—intact families of genetic parents
and children on the one hand vs step, adopted or foster families on the other. In the ex-
treme, Daly and Wilson (1988) have documented a higher rate of neglect, abuse, even
homicide of children in step families. Less dramatically, but still of great personal and
social consequences, Biblarz and Raftery (1999) show that, on average, step, adopted
and foster-parents invest less on children than do genetic parents, leading to lower so-
cioeconomic achievement of the child. These findings are amenable to explanations
in terms of evolutionary social psychology. But plausible sociological mechanisms
also come to mind. For example, there might be more conflict within step families, or
perhaps step fathers do not have the ‘full license’ to discipline their children. For our
present purpose, it would be sufficient to note that family dynamics is likely to vary
by family type. Thus, genetic children and non-genetic children may have different
effects on marital stability. Given the growth of non-conventional families, this could
be important in the aggregate.
Given these considerations, we think that a destabilizing effect of children in the
UK is at least conceivable. But why is the result of Chan and Halpin (2001) different
from most previous research? It seems to us that there are three possibilities: (1) faulty
analysis in that paper, (2) deficiency of the BHPS data, and (3) that a real social change
has taken place in the UK. We shall argue that (3) is the case. To do so, we now turn
to analyze two other British data sets.
8
7
But see Andersson and Woldemicael (2000) for negative evidence.
8
We have endeavoured to purge all mistakes from our analysis. But, by the nature of these things,
one can never rule out the possibility of making mistakes. All our program and output files are available
to colleagues for examination on request.
9

3 Change in the Children Effect
3.1 Family and Working Lives Survey, 1994–95
The Family and Working Lives Survey (FWLS) provides retrospective life history data
in the domains of work, education, housing, family formation and dissolution for a
sample of 11 237 individuals living in Britain. In the analysis reported in this subsec-
tion, we select all women who were in their first marriage (N 4 430), and we use
the proportional hazards model (Cox 1972) to examine the divorce risks facing these
women over the course of their first marriage,
λ t λ
0
t exp x t β
where the dependent variable λ t is the hazard of divorce, λ
0
t is the unspecified
baseline hazard rate, x t is the vector of covariates and β the vector of coefficients to
be estimated.
9
Because of the retrospective nature of this survey, the set of covariates available to
us is very limited. Specifically, we consider ‘age at marriage’ and ‘marriage cohort’
which are time-constant covariates, and ‘number of children’ which is time-varying.
We distinguish five marriage cohorts: 1 = 1950s or before, 2 = 1960s, 3 = 1970s, 4 =
1980s, 5 = 1990s.
10
Since the FWLS allows us to distinguish between different types
of children (viz. genetic, step, adopted or fostered), we report two sets of analyses
in Table 2. Panel A uses the full sample, while Panel B excludes those marriages
with step, adopted or fostered children. Because the results of the two panels are very
similar, we will discuss the estimates of Panel A only.
Model 1 shows the expected negative coefficient for age at marriage—women who
got married when they were older have more stable marriage. Also in line with expec-
9
All hazard models in this paper are fitted with TDA, a free software kindly made available by
Rohwer and P¨otter (2001).
10
Further descriptive statistics of the FWLS data are provided in Appendix A.1.
10

Table 2: The hazard of divorce for women in their first marriage, proportional hazards
models as applied to FWLS (1994–95) data.
Panel A: All first marriages model 1 model 2
age at marriage -0.046** (0.008) -0.044** (0.008)
marriage cohort 0.844** (0.034) 0.658** (0.046)
number of children 0.037 (0.026) -0.295** (0.066)
cohort children 0.122** (0.022)
Number of divorce 1139 1139
Log likelihood -8413.85 -8397.41
Panel B: First marriages with step, adopted or fostered children excluded
age at marriage -0.046** (0.008) -0.043** (0.008)
marriage cohort 0.846** (0.034) 0.648** (0.047)
number of children 0.037 (0.026) -0.319** (0.067)
cohort
children 0.132** (0.022)
Number of divorce 1127 1127
Log likelihood -8314.28 -8295.96
Note: Standard errors in parenthesis; ** p 1%.
tation, we see a higher dissolution rate for more recent marriage cohorts. Controlling
for these, children increase the divorce hazard in model 1. This result is consistent
with the BHPS, but the effect is not significant by conventional standards.
Adding an interaction term that allows the children effect to vary by marriage co-
hort (model 2), we see that the main effect of children changes sign. But taking into
account both the main effect and the interaction effect, it is clear that the effect of chil-
dren on marital stability has changed over time: they used to reduce the divorce risks,
but this is no longer true. A small numerical example will make this clear.
1950s :
0 295 1 0 122 0 173
1990s : 0 295 5 0 122 0 315
For those who got married in the 1950s, each additional child was associated with
a 16% (e
0 173
1) reduction in divorce risk. In contrast, for the 1990s marriage
cohort, each additional child is associated with a 37% (e
0 315
1) increase in divorce
risk. This change cannot be explained by the growth of non-conventional families in
11

recent cohorts, as the results in Panel B, where families with step, adopted or fostered
children are dropped from the analysis, are essentially the same.
3.2 Further Test with the General Household Survey, 1991–95
Although the FWLS results in the last subsection are suggestive, our analysis is admit-
tedly rather crude. Family formation behaviour has changed in many ways over time.
For example, while premarital cohabitation was quite a rare occurrence in the past, it is
now the majority practice in Britain and many western societies (see Table 9, also Er-
misch and Francesconi 2000, Murphy 2000). The rise of divorce also means that more
first-time brides would be marrying divorcees. Furthermore, the closing of the gender
gap in educational attainment might also affect the dynamics of family formation and
dissolution. It would be useful to control for these trends in the analysis. With these
considerations in mind, we turn to the General Household Survey (GHS).
The GHS is a continuous multi-purpose survey of people living in private house-
holds in the UK. It began in 1971, with an achieved sample of about nine thousand
households each year. The GHS gives us a set of repeated cross-sectional views of
employment, education, health, leisure and various aspects of social life in Britain.
Since 1991 the GHS has been collecting full retrospective marriage and fertility histo-
ries. We combine the relevant GHS data from 1991 to 1995. This gives us information
of 24,157 women in their first marriage, among whom we observe 6,074 cases of di-
vorce. As before, our main concern is how the children effect on marital stability might
have changed across cohorts. But the main advantage of using the GHS is that it al-
lows us to include three time-constant covariates in the model: whether the respondent
cohabited with her husband before marriage (1 = yes, 0 = no), whether the husband
was a divorcee (1 = yes, 0 = no), and the respondent’s highest educational attainment,
which we distinguish four levels—university, A-levels, O-levels, no qualifications.
11
11
The last of these is the reference category for educational attainment. More information about the
GHS and some descriptive statistics of the covariates can be found in Appendix A.2.
12

Also similar to our analysis of the FWLS, we use the proportional hazards model.
Here we allow both the baseline hazard and the effect of the covariates to vary by
marriage cohorts. This gives the model greater flexibility to fit the data. In practice,
we fit the model to each marriage cohort separately,
λ t λ
c
0
t exp x t β
c
We report our result in Table 3. Panel A refers to all first marriages, and Panel
B excludes those marriages with step, adopted or fostered children. There are many
interesting changes in the parameter estimates. For example, in the early cohorts, it was
the better educated women (those with university degrees, A-levels or O-levels) who
faced higher divorce risks. However, among those who got married since the 1980s,
women with degrees are significantly less likely to divorce, and those with A-levels or
O-levels no longer face higher divorce risk than those with no qualifications.
12
Panel A of Table 3 also shows that for the earliest marriage cohort, premarital co-
habitation was associated with a three-fold increase (e
1
204
) in the divorce risk. The
magnitude of this effect declines monotonically over time, and it is statistically in-
significant for those who got married in the 1990s.
13
A similar trend can be observed
for the estimate of marrying a divorcee. These trends can be understood in terms of
the decreasing selectivity of the two variables. When premarital cohabitation was rare,
those who did cohabit were probably quite a selected group. Perhaps they have un-
conventional views about marriage and the family, which are probably associated with
high divorce risks. As premarital cohabitation becomes a majority practice, it is no
longer a discriminating indicator of underlying attitudes. A similar argument applies
to the covariate of marrying a divorcee.
12
Chan and Halpin (2001) show that once household income and the relative income between the
spouses are controlled for, women’s educational qualifications do not have a significant effect on divorce
in the 1990s.
13
A cautionary note is in order. The relatively small number of divorces observed for the last mar-
riage cohort makes it difficult to detect statistically significant results. But still the downward trend in
magnitude is notable.
13

Table 3: The harzard of divorce for women in their first marriage, proportional hazards
models as applied to the GHS (1991–95) data.
1950s or
Panel A: All first marriages before 1960s 1970s 1980s 1990s
age at marriage -0.177** -0.164** -0.123** -0.107** -0.101**
(0.022) (0.010) (0.008) (0.009) (0.024)
university
a
0.418** 0.385** 0.017 -0.296** -0.829*
(0.155) (0.071) (0.068) (0.094) (0.330)
A-levels
a
0.876** 0.259* 0.148 0.075 -0.337
(0.217) (0.113) (0.081) (0.096) (0.299)
O-levels
a
0.331** 0.109* 0.037 -0.077 -0.270
(0.100) (0.054) (0.051) (0.068) (0.222)
premarital cohabitation 1.204** 0.816** 0.531** 0.530** 0.342
(0.269) (0.106) (0.058) (0.056) (0.186)
husband was divorcee 0.597* 0.446** 0.333** 0.223** 0.254
(0.253) (0.115) (0.078) (0.083) (0.233)
number of children 0.045 -0.076** -0.143** -0.037 0.314**
(0.029) (0.022) (0.024) (0.030) (0.107)
Number of divorce 592 1855 2087 1392 148
Panel B: First marriages with step, adopted or fostered children excluded
age at marriage -0.177** -0.165** -0.122** -0.106** -0.100**
(0.022) (0.010) (0.008) (0.009) (0.024)
university
a
0.415** 0.385** 0.020 -0.300** -0.797*
(0.156) (0.071) (0.068) (0.094) (0.330)
A-levels
a
0.887** 0.262* 0.150 0.081 -0.320
(0.217) (0.113) (0.081) (0.096) (0.300)
O-levels
a
0.326** 0.100 0.038 -0.079 -0.235
(0.101) (0.054) (0.051) (0.068) (0.222)
premarital cohabitation 1.268** 0.826** 0.529** 0.528** 0.345
(0.267) (0.106) (0.058) (0.056) (0.186)
husband was divorcee 0.651** 0.462** 0.337** 0.257** 0.354
(0.251) (0.116) (0.078) (0.084) (0.232)
number of children 0.046 -0.083** -0.148** -0.035 0.336**
(0.029) (0.022) (0.024) (0.030) (0.107)
Number of divorce 590 1847 2081 1388 148
Note: Standard errors in parenthesis; ** p 1%, * p 5%;
a
‘no qualifications’ is the reference category.
14

Coming to the main variable of interest, we note that, with the exception of the
earliest cohort, children used to stabilize marriage. But the children effect began to
shift in the 1980s, and that by the 1990s, each additional child is associated with a
37% (e
0
314
1) increase in the divorce risk. This result is remarkably similar to what
we observe for the FWLS.
As before, we have repeated our analysis of the GHS on a restricted sample in
which women with step, adopted or fostered children are excluded from the analysis
(Panel B). Since the results of the two panels are very similar, we will not discuss
them here. Instead, we show, in Figure 1, the ninety-five percent confidence intervals
of the effect of genetic children on marital stability by marriage cohorts. Note that
the relatively small number of divorces observed for the 1990s cohort implies a much
wider confidence interval. The remarkable thing is that, despite this, the confidence
interval is still well over the line of no effect.
Figure 1: The ninety-five percent confidence interval of the effect of genetic children
on divorce risk, by marriage cohorts, controlling for age at marriage, educational at-
tainment, premarital cohabitation and husband’s previous marital status.
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
1 2 3 4 5
’errorbar.dat’
0
15

4 Why WouldChildrenDestabilize Marriage intheUK?
Having established that the children effect on marital stability has changed over time
in the UK, we now turn to explore why this should have happened. As noted above,
the relevant question is not whether children destabilize marriage, but under what con-
ditions would they do so? We proceed as follows. First, we provide a short time series
of survey evidence to trace how social attitudes towards divorce, especially in relation
to children’s welfare, have changed over time. Second, we examine whether children
weaken a divorced woman’s position in the repartnering market, and how that effect
might be changing. Third, we test if another form of marital-specific capital (namely,
owned home) is losing its stabilizing effect too. Fourth, we examine whether the chil-
dren effect might depend on their age and sex, as some scholars have argued. Finally,
we test if family circumstances, specifically household income, matter.
4.1 Social Attitudes Towards Divorce
‘Should parents stay together for the sake of their children?’ To the best of our knowl-
edge, there are four systematic surveys in the UK which include such a question. Since
the first of these surveys was from 1983, they do not cover the entire period of interest
to us. Also, the four surveys are not exactly comparable. They are based on samples of
different design,
14
and the question wording and the response categories vary between
studies. For these reasons, the evidence presented below must be treated with caution.
The wording of the questions and the response cateotries are as follows:
1983 BSA: ‘Parents with unhappy marriages should stay together for the sake
of their children’—agree strongly, just agree, neither, just disagree, disagree
strongly.
14
We use two data sources here: the British Social Attitudes Survey Series (BSA) and the BHPS.
The population of the BSA are individuals aged 18 and above living in private households in Great
Britain, whereas that for the BHPS are households. We have restricted our analysis in this subsection to
respondents aged 20 to 64. See Appendix A.3 for technical details.
16

1988 BSA: ‘When a marriage is troubled and unhappy do you think it is gener-
ally better for the children if the couple stays together or gets divorced?’—much
better to divorce, better to divorce, worse to divorce, much worse to divorce.
1994 BSA: ‘When there are children in the family, parents should stay together
even if they don’t get along’—strongly agree, agree, neither, disagree, strongly
disagree, can’t choose.
1998 BHPS: As in 1994 BSA but without the ‘can’t choose’ response category.
Table 4: Social attitudes towards divorce(column percentages). Data taken from 1983,
1988, 1994 British Social Attitudes Surveys and 1998 British Household Panel Study.
Panel A: Should parents stay together for their children?
1983 BSA 1988 BSA 1994 BSA 1998 BHPS
divorce 59.1 76.4 65.1 59.6
neither 19.3 — 17.4 26.3
stay together 21.6 23.6 17.4 14.1
N 1231 806 746 7802
Panel B: Does the presence of children matter? (1994 BSA)
couples couples marital
with kids without kids problems
divorce 65.1 88.2 55.8
neither 17.4 7.6 22.2
stay together 17.4 4.2 22.0
N 746 754 738
Panel C: For whom is divorce better? (1988 BSA)
children wife husband
divorce 76.4 86.9 87.0
stay together 23.6 13.1 13.0
N 806 786 785
Note: Respondents were given four or five response categories in the interview (see
text). We have collapsed the responses into two or three, respectively.
Panel A of Table 4 shows that since 1983 less than one quarter of the respondents
say that couples should stay together for their children, and that there is a downward
trend in the level of support for ‘staying together’. When respondents were not given
the ‘neither’ option in 1988, it was the ‘pro-divorce’ group which grew in size.
17

Does the presence of children matter? In 1994, the question we considered in Panel
A was followed by this statement: ‘Even when there are no children, a married couple
should stay together even if they don’t get along’. Only 4% of the respondents agreed
with this statement (see the second column of Panel B; the first column of Panel B
is the same as the third column of Panel A). It is therefore quite clear that normative
sanction against divorce is stronger for couples with children than for childless couples.
While 4% must be seen as a very low level of support for ‘staying together’, we note
that in the 1994 survey, respondents were also asked whether ‘Divorce is usually the
best solution when a couple can’t seem to work out their marriage problems’. The last
column of panel B shows that when divorce is mooted as the best solution to marriage
problems, significantly fewer respondents endorsed the statement.
Three quarters of the respondents of the 1988 survey said that, from the children’s
point of view, divorce is generally better than a troubled and unhappy marriage (see
second column of Panel A). The same survey also asked whether divorce is better or
worse for the wife and the husband. We tabulate these responses in Panel C, which
shows that while a large majority of the respondents thought that under the conditions
specified, divorce is better for all parties concerned, considerably fewer respondents
thought that divorce is better for the children (76% as opposed to 87%).
15
4.2 Children and the Repartnering Market
We have argued that children might contribute to marital stability through an antic-
ipatory mechanism—by weakening the prospective custodial parent’s position in the
repartnering market. If this is true, the shift in the children effect might in part be
due to a parallel shift: that the disadvantage faced by lone mothers in the repartnering
market, as compared to childless divorced women, is declining over time. Just as the
15
There is some recurring structure in the attitudes towards divorce and children. For example, in
all four surveys, women are more likely than men to endorse divorce. This gender difference remains
after controlling for age, educational qualifications, and marital status. Because social attitudes is not
the focus of our research, we shall not discuss these patterns in this paper.
18

increase in the incidence of premarital cohabitation has been accompanied by a weak-
ening of its association with divorce (see Table 3), the impact of children on divorced
women’s chance of repartnering might have weakened as lone motherhood becomes
more common (Harrop and Plewis 1995).
To test this idea, we return to the FWLS data. We select all women whose first mar-
riage has been dissolved through divorce, separation or widowhood. Among this group
of 1,515 women, 584 have repartnered (i.e. remarried or cohabited with a partner) by
the time of the interview. We use the proportional hazards model to examine the effects
on the chance of repartnering of several covariates: ‘respondent’s age at dissolution’,
‘cause of dissolution’ (widowhood vs divorce/separation), ‘length of first marriage’,
‘dissolution cohort’ (1=before 1970, 2=1970s, 3=1980s, 4=1990s), and ‘number of
children at dissolution’. Our main interest is the last of these covariates.
Table 5: The hazard of repartnering for women whose first marriage has been dis-
solved, proportional hazard models as applied to FWLS data.
model 1 model 2
age when first marriage dissolved -0.084** (0.013) -0.085** (0.013)
length of first marriage 0.048** (0.015) 0.049** (0.015)
widow (vs div/sep) -0.671** (0.149) -0.689** (0.150)
dissolution cohort 0.722** (0.082) 0.813** (0.097)
number of children -0.142** (0.035) 0.010 (0.093)
cohort
children -0.065† (0.038)
Number of repartnering 584 584
Log likelihood -3172.37 -3170.90
Note: Standard errors in parenthesis; ** p 1%, †p 10%.
Table 5 shows that women who were older when their first marriage was dissolved,
widows (as opposed to divorcees), those from the early dissolution cohorts, as well as
those with a relatively short first marriage have a lower probability of repartnering. It
also confirms that children weaken a woman’s position in the repartnering market. In
model 2, we add an interaction term which allows the children effect to vary in a linear
fashion by cohort. The deviance between the two models is 2.94 ( 2 3172 37
3170 90 ), which is not significant for 1 degree of freedom. So there is no evidence
19

of any linear change. The sign of the interaction term suggests that, if anything, the
disadvantage faced by divorced women with children has increased across cohorts.
16
On the whole, we see no evidencethat the shift in the children effecton marital stability
can be attributed to the changing position of lone mothers in the repartnering market.
4.3 Home Ownership and Divorce Risks
As we note in section 2, the most fundamental reason for children to be a marriage-
stabilizing factor is that they are a form of marital-specific investment. The shift in the
children effect might then suggest that parents are investing less in their children. It
is also possible that there is a general decline in the stabilizing effect of all forms of
marital-specific capital.
We explore the latter possibility by testing the effect on divorceof home ownership.
A house is obviously not comparable to one’s children, as a divorcing couple can al-
ways sell the house they own and split the proceeds. But compared to other joint assets,
owned home often embodies not just monetary investment, but also lifestyle choice in
terms of neighbourhood and decoration, as well as local social networks which are not
portable. In this sense, owned home can be considered as a form of marital-specific
investment. Of course, as with children, the association between home ownership and
marital stability may in part be due to self-selection, in that couples anticipating mari-
tal difficulties might avoid or delay buying a house. Practically speaking, we examine
the effect of home-ownership because it is a readily available measure.
The analysis for the rest of this paper is based on data taken from British House-
hold Panel Study (BHPS). The BHPS is an annual panel which began in 1991. Its
sample, which is representative of the British population, covers 10,264 respondents
from 5,511 households in 1991. From the first eight waves of the BHPS (i.e. 1991–
98), we have constructed a data set with up to seven wave-on-wave transitions for each
16
We have tested several specifications of the children variable, namely by counting the number of
children under age 2, under age 6 and under age 18 when first marriaged was dissolved. The results are
very similar as those reported here.
20

respondent. This allows us to track their marital status at discrete yearly interval. We
focus on women who were in their first marriage. They remain in the risk set until
their marriage is dissolved (divorce or separation) or our observation is censored. Our
strategy is to use covariates at time t to predict their marital status at t 1, using the
discrete-time logistic regression model:
log
p
t 1
1 p
t 1
x t β
where p
t
1
is the probability of divorce at t
1 given that the respondent was married
at t, x t is a vector of covariates, and β is the vector of parameters to be estimated.
Table 6 provides the basic descriptive statistics of the covariates. The value of
all covariates are updated each year. So in terms of measurement, they are all time-
varying, though some, such as age at marriage, are time-constant by nature. Since
the meaning of most covariates is fairly self-explanatory, we highlight just two derived
variableshere. The variable ‘pay comparison’ compares the monthly wages of husband
and wife (Sørensen and McLanahan 1987), calculated as follows:
pay comparison
wife s pay husband s pay
wife s pay husband s pay
This variable ranges from 1 (indicating a situation where all labour income comes
from the husband, and thus denoting a low degree of economic independence of the
woman) to 1 (where the opposite is true).
17
The variable ‘household income’ refers to
total annual income from all household members, adjusted by household size.
We reproduce the main result of Chan and Halpin (2001) in the first column of
Table 7. Most notably, the estimates for the two children dummies are positive, and
their magnitudes are substantial. For example, the odds of divorcefor couples with one
child is six times higher (e
1 860
) than that facing similar but childless couples. These
effects might seem implausibly large, though as we shall see, the children effect is
17
In cases where a spouse does not work, we impute the value zero for his or her wage.
21

Table 6: Descriptive statistics of the BHPS data
variable name range mean s.d.
duration (years) 1–64 24.69 14.45
duration-squared
a
0.01–40.96 8.19 8.06
year of marriage 30–97 70.08 14.40
age at marriage (years) 16–60 23.08 4.19
household income 0.45–3.89 1.32 0.37
pay comparison -1–1 -0.18 0.61
proportion
home-owners 75.7%
renters
b
24.3%
1 child 14.8%
2+ children 27.0%
none
b
58.3%
degree 7.5%
A-level 27.3%
O-level 30.3%
no qualifications
b
34.9%
Notes: The unit in this tabulation is person-year in the event history data file;
a
duration
duration / 100;
b
reference category.
22

modified significantly by household circumstances.
Adding the covariate of home ownership in model 2, we see that home owners are
about half as less likely to divorce as renters (e
0 568
). To the extent that home owners
have greater investment in their marriage, we see no evidence of a general decline in
the stabilizing effect of marital-specific capital.
Table 7: The hazard of divorce for women in their first marriage, discrete-time
logistic regression model as applied to BHPS data, 1991–98.
(1) (2) (3) (4) (5) (6)
constant -0.907 0.549 0.110 0.779 -1.167 -1.158
(4.892) (4.893) (4.922) (4.904) (5.053) (5.069)
pre-1991 -0.871* -0.858* -0.115 -0.883* -0.826* -0.351
(0.387) (0.385) (0.398) (0.392) (0.392) (0.404)
duration 0.275** 0.293** 0.144* 0.279** 0.263** 0.129*
(0.060) (0.061) (0.062) (0.060) (0.062) (0.063)
duration-squared -0.573** -0.603** -0.309** -0.580** -0.543** -0.250*
(0.106) (0.107) (0.104) (0.106) (0.109) (0.104)
year of marriage 0.102† 0.106* 0.122* 0.104* 0.095† 0.112*
(0.052) (0.052) (0.052) (0.052) (0.053) (0.054)
age at marriage -0.195** -0.188** -0.188** -0.196** -0.194** -0.184**
(0.033) (0.032) (0.032) (0.033) (0.034) (0.034)
degree -0.732 -0.583 -0.932* -0.745† -0.633 -0.787†
(0.453) (0.459) (0.465) (0.448) (0.456) (0.466)
A-level -0.327 -0.197 -0.411 -0.344 -0.324 -0.351
(0.271) (0.277) (0.276) (0.273) (0.280) (0.287)
O-level -0.198 -0.092 -0.234 -0.167 -0.155 -0.178
(0.262) (0.266) (0.265) (0.262) (0.272) (0.277)
pay comparison 0.487** 0.469** 0.426** 0.506** 0.522** 0.488**
(0.150) (0.150) (0.153) (0.150) (0.153) (0.157)
household income -9.993** -9.953** -10.299** -10.046** -7.170** -8.182**
(0.497) (0.497) (0.527) (0.503) (0.757) (0.658)
home owner -0.568**
(0.215)
1 child 1.860** 1.843** 5.115**
(0.336) (0.337) (1.568)
1 child income -3.395*
(1.448)
2+ children 4.069** 4.031** 10.320**
(0.333) (0.333) (1.377)
2+ children income -5.618**
(1.166)
child (0–2) 0.994** 2.148
(0.292) (1.689)
child (0–2) income -1.055
23

Table 7: The hazard of divorce for women in their first marriage, discrete-time
logistic regression model as applied to BHPS data, 1991–98.
(1) (2) (3) (4) (5) (6)
(1.450)
child (3–4) 1.305** 7.021**
(0.281) (1.728)
child (3–4)
income -4.750**
(1.445)
child (5–11) 2.707** 7.183**
(0.265) (1.457)
child (5–11) income -3.592**
(1.167)
child (12–15) 3.005** 4.377*
(0.338) (2.231)
child (12–15) income -1.121
(1.655)
child (16–18) 2.493** 0.074
(0.626) (3.113)
child (16–18) income 1.554
(2.283)
1 boy 1.999**
(0.414)
1 girl 1.603**
(0.410)
2+ boys 3.134**
(0.476)
2+ girls 3.888**
(0.451)
2+ boys and girls 4.347**
(0.351)
Number of divorce 160 160 160 160 160 160
Log likelihood -487.71 -484.34 -474.33 -484.04 -474.61 -457.95
Note: Standard errors in parenthesis; ** p 1%, * p 5%, †p 10%.
4.4 Children’s and Couple’s Characteristics
Does the children effect depend on their age and sex? In model 3, we use separate
dummy variables to capture the effect of having at least one child in the five age brack-
ets of 0–2, 3–4, 5–11, 12–15, and 16–18. Such a specification does not substantially
24

modify other parameter estimates.
18
But it now becomes clear that in the UK even
very young children destabilize marriage. This finding is stronger than those reported
by Cherlin (1977) and Waite and Lillard (1991). However, with the exception of the
last age bracket, we do see an age gradient in the magnitude of the effect: the older the
child, the stronger the destabilizing effect.
We test the hypothesis of gender difference proposed by Morgan et al. (1988) in
model 4. All parameter estimates are positive and statistically significant. At parity 1,
the difference in the effect of boy and girl is only 0.396 (1 999 1 603), which is well
within the range of sampling variation, given the standard errors of both parameters
is about 0.4. The same is true for couples with at least two children. In other words,
there is no evidence of any gender difference.
Figure 2: Histogram of equivalised household income
income
Frequency
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
0 1000 2000 3000 4000 5000
In model 5, we add two interaction terms which allow the children effect to vary by
household income. Here we see that the children effect on marital stability is crucially
18
The exception is the dummy which distinguishes marriages which began during the panel period
for which we have complete information and those which began before 1991.
25

modified by household income. While children destabilize marriage in low income
families, they have no effect or even a stabilizing effect in middle or high income fam-
ilies. Again, let us illustrate this point with a small numerical example. As readers
can see in Figure 2, the distribution of the equivalised household income variable is
bimodal, with a spike just under 1 and another mode at around 1.5. Using the pa-
rameter estimates of model 5, the effect of having one child for three households with
equivalised income at 1, 1.5 and 2 are:
5 115 1 3 395 1 720
5 115 1 5 3 395 0 023
5 115 2 3 395 1 675
In model 6, we test the interaction effects using the various age bracket dummies.
Again, we see evidence of the mediating role of household income, although some
parameters become statistically insignifcant. In particular, children under 2 no longer
have a significant destabilizing effect.
5 Summary, Interpretations and Possible Implications
In this paper, we use data from three British sources to demonstrate that in the UK
the effect of children on marital stability has changed over time. Children used to
be a stabilizing factor in marriage, but they are now associated with greater marital
instability. This shift began in the 1980s, and by the 1990s couples with children
are at a substantially higher risk of divorce than similar but childless couples. This
shift is still evident after we have excluded non-conventional families (those with step,
adopted or fostered children) from the analysis.
Why has this happened? We do not have a good answer to this question. But we
show that since the early 1980s at most one quarter of the respondents think that an
26

unhappy marriage is better than divorce for husband, wife and children. Having said
that, we note that the normative sanction against divorce is still stronger for couples
with children. We also show that children are still a ‘liability’ for divorced women
in the repartnering market. Thus, the shift in the children effect cannot be attributed
to a convergence in the repartnering rate of childless divorcees and divorcees with
children. Also, home owners are still less inclined to divorce than renters. So there is
no evidence of a general decline in the stabilizing effect of marital-specific investment.
There is an age gradient in the children effect—the older the child, the stronger
the destabilizing effect. This gradient is partially consistent with the results of Cherlin
(1977) and Waite and Lillard (1991). But the striking thing is that in the UK even
very young children are associated with marital instability. Contrary to the finding
of Morgan et al. (1988), we detect no gender difference in the association between
children and marital stability.
We see very clear evidence that in the 1990s the children effect are crucially modi-
fied by household income. Perhaps the resources that are available to middle and high
income households make it easier for them to cope with the stress and high costs that
are associated with raising children. If this is the case, one might argue that a high level
of income inequality is bad for marital stability. Income inequality did increase very
rapidly in the UK since the 1980s (Atkinson 1997, Jenkins 1996). Some conservative
political opinion is in favour of both a greater degree of income inequality and stable
families. One interpretation of our results is that, to put it bluntly, you can’t have both.
We hasten to add that our results need to be explored much further. If family re-
sources are key to whether children stabilize or destabilize marriage, one might ask:
what type of resources is relevant? Is it just income? Or perhaps other types of re-
sources such as informal support networks also play a role.
It is also possible that the shift in the children effect can partly be explained by
some selection process into marriage and parenthood. Let us consider a simple hypo-
thetical scenario. Suppose there are two latent classes of people. Those in the first class
27

get married because they want to have children, and those in the second get married for
its own sake. If, for whatever reason, the relative proportion of the two latent classes
changes over time in favour of the second class, one would expect the stabilizing effect
of children to decline.
Still another possibility is that the fertility rates of the various income groups have
changed over time. Since household income is positively associated with marital sta-
bility. If fertility rate declines more rapidly among high income families than among
low income families, there would be a higher proportion of families that are proned to
divorce in the married population.
Our research also suggests the possbility of that parental investment in children
might be declining. There is a need to investigate personal interaction pattern within
household and parent-child relationship in general. The BHPS contains some useful
data in this regard. We plan to examine these data very soon.
A Data Sets
All data sets used in this paper are in the public domain, obtainable from the UK
data archive, based at the University of Essex, Wivenhoe Park, Colchester CO4 3SQ,
United Kingdom (http://www.data-archive.ac.uk/).
A.1 Family and Working Lives Survey (FWLS)
The population of the Family and Working Lives Survey (FWLS) are adults aged 16–
69 years living in Britain. This survey was commissioned by the UK Department for
Education and Employment and was designed to replicate and extend the 1980 Women
and Employment Survey. The main sample of the FWLS was achieved through a two
stage process. First, a stratified random sample of Census Enumeration Districts were
selected with probability proportional to their size. Then within each Census Enu-
meration District, addresses were drawn randomly from the Postcode Address File.
28

The fieldwork was done through face-to-face interview, and was carried out between
July 1994 and May 1995. The FWLS provides retrospective life history data for the
main respondent and his/her partner over a wide range of areas. We use the full sam-
ple in this paper, consisting of the main sample (N 9 139), and a booster sample
for ethnic minorities (N 2 098). The overall response rate of the FWLS is a little
disappointing—only 53.5%. Table 8 reports the descriptivestatistics of the FWLS data
used in this paper.
Table 8: Descriptive statistics—FWLS data.
marriage cohort
Variables 1950s 1960s 1970s 1980s 1990s all
mean and standard deviation
age at marriage (years) 21.18 21.71 21.49 22.59 24.23 22.05
(2.83) (4.27) (4.47) (4.64) (5.97) (4.47)
max number of children 2.36 2.00 1.60 1.26 0.46 1.62
(all) (1.52) (1.37) (1.33) (1.08) (0.86) (1.40)
max number of children 2.33 1.98 1.59 1.25 0.45 1.61
(genetic) (1.52) (1.36) (1.33) (1.07) (0.83) (1.39)
N 952 812 1017 1176 473 4430
21.49% 18.33% 22.96% 26.55% 10.68%
Note: The unit in this tabulation is the individual.
A.2 General Household Surveys (GHS)
The General Household Survey (GHS) is an annual survey which has been running
almost continuously since 1971. It is conducted by the Office for National Statistics
on behalf of a number of government departments. The aim is to provide information
on housing, employment, education, health, and the family for planning and policy
purposes. The population of the GHS are private, non-institutional households in the
UK. The achieved sample of each year, which is nationally representative, is about
9000 households. The response rate of the GHS is around 70%. Fieldwork is done
through face-to-face interview.
29

Table 9: Descriptive statistics—GHS data.
marriage cohort
Variables 1950s 1960s 1970s 1980s 1990s all
proportion
university 7.69 14.14 19.08 20.31 21.77 17.10
A-levels 2.45 4.21 8.35 11.71 15.32 8.05
O-levels 21.81 31.07 37.82 47.58 49.32 37.79
no qualifications
a
68.05 50.58 34.76 20.40 13.60 37.07
premarital cohabitation 1.05 3.30 15.14 41.64 66.31 21.46
husband was divorcee 1.53 3.91 8.66 12.42 16.08 8.21
mean and standard deviation
age at marriage (years) 20.40 21.48 22.10 23.50 25.74 22.41
(1.98) (2.88) (3.80) (4.30) (5.07) (3.94)
max number of children 2.58 2.19 1.89 1.54 0.63 1.85
(all) (1.57) (1.21) (1.17) (1.07) (0.77) (1.27)
max number of children 2.57 2.17 1.88 1.53 0.62 1.84
(genetic) (1.57) (1.22) (1.18) (1.06) (0.76) (1.27)
N 2485 6572 6914 6215 1971 24157
10.29% 27.21% 28.62% 25.73% 8.16%
Note: The unit in this tabulation is the individual;
a
reference category for educational qualifications.
A.3 British Social Attitudes (BSA)
The BSA survey series began in 1983, and has been running almost yearly ever since.
This survey series is conducted by the National Centre for Social Research (formerly
Social and Community Planning Research), and it covers a wide range of issues over
the years, including attitudes towards religion, inequality, work, and the welfare state.
Since 1985, the BSA also includes a module of the International Social Survey Pro-
gramme (ISSP) in its self-completion questionnaire. The 1988 and 1994 data we use
in this paper are part of the ISSP.
The population of the BSA is adults aged 18 or over living in the UK. Until 1991,
the BSA samples were drawn from the Electoral Register. Since 1993, the sampling
frame of the BSA has been the Postcode Address File, which is a list of addresses com-
piled by the Post Office. The BSA sample is nationally representative. The achieved
sample size for each year is between 3,300 and 3,600. But since the BSA has a mod-
30

ular structure, respondents are asked a different set of questions according to which
module they belong to. That is why the Ns of Table 4 are considerably smaller.
A.4 British Household Panel Study (BHPS)
The BHPS is conducted by the Institute for Social and Economic Research, Univer-
sity of Essex. When it began in 1991, its sample contained 5,511 households. These
households were selected through a two-stage clustered probability sampling proce-
dure, using the Postcode Address File. This sampling design is roughly equvalent to
the current sampling method of the GHS.
All adults (aged 16 or over) of the original households were interviewed (N
10 264 in 1991). The same individuals have been re-interviewed in successive waves
and, should they leave their original household, all adult members of their new house-
holds would also be interviewed. Thus, the sample is broadly representative of the
population of Britain as it changes through the 1990s. Additional sub-samples were
added to the BHPS in 1997 and 1999. Data is collected through face-to-face inter-
views. Sample attrition of the BHPS is modest: 87.7% of wave one respondents were
re-interviewed in wave two. Subsequent wave-on-wave recontact rate is at least 90%.
Online information and documentation are available at:
http://www.irc.essex.ac.uk/bhps/index.php
31

References
[1] Akerlof, George A. (1998) ‘Men Without Children’, Economic Journal,
108:287–309.
[2] Andersson, Gunnar (1997) ‘The Impact of Children on Divorce Risks of Swedish
Women’, European Journal of Population, 13:109–145.
[3] Andersson, Gunnar and Gebremariam Woldemicael (2000) ‘Sex Composition
of Children As a Determinant of Marriage Disruption and Marriage Formation:
Evidence from Swedish Register Data’, Max Planck Institute for Demographic
Research Working Paper, WP 2000-007.
[4] Atkinson, Antony B. (1997) ‘Bringing Income Distribution In From The Cold’,
Economic Journal, 107:297–321.
[5] Becker, Gary S., Elizabeth M. Landes and Robert T. Michel (1977) ‘An Eco-
nomic Analysis of Marital Stability’, Journal of Political Economy, 85:1141–
1187.
[6] Berrington, Ann and Ian Diamond (1999) ‘Marital Dissolution Among the 1958
British Birth Cohort: The Role of Cohabitation’, Population Studies, 53:19–38.
[7] Bianchi, Suzane M., Melissa A. Milkie, Liana C. Sayer and John P. Robinson
(2000) ‘Is Anyone Doing the Housework? Trends in the Gender Division of
Household Labor’, Social Forces, 79:191–228.
[8] Biblarz, Timothy J. and Adrian E. Raftery (1999) ‘Family Structure, Educational
Attainment, and Socioeconomic Success: Rethinking the “Pathology of Matri-
archy”’, American Journal of Sociology, 105:321–365.
[9] B¨oheim, Ren´e and John Ermisch (2001) ‘Partnership Dissolution in the UK—
The Role of Economic Circumstances’, Oxford Bulletin of Economics and Statis-
tics, 63:197–208.
32

[10] Bracher, Michael, Gigi Santow, S. Philip Morgan and James Trussell (1993)
‘Marriage Dissolution in Australia: Models and Explanations’, Population Stud-
ies, 403–425.
[11] Chan, Tak Wing and Brendan Halpin (2001) ‘Divorce in the UK’, University of
Oxford, Department of Sociology Working Paper, 2001-01.
[12] Cherlin, Andrew (1977) ‘The Effect of Children on Marital Dissolution’, Demog-
raphy, 14:265–272.
[13] Cox, David R. (1972) ‘Regression Models and Life Tables’, Journal of the Royal
Statistical Society, Series B, 34:187–220.
[14] Daly, Martin and Margo Wilson (1988) ‘Evolutionary Social Psychology and
Family Homicide’, Science, 242:519–524.
[15] Diekmann, Andreas and Henriette Engelhardt (1999) ‘The Social Inheritance of
Divorce: Effects of Parent’s Family Type in Postwar Germany’, American Soci-
ological Review, 64:783–793.
[16] Ermisch, John and Macro Francesconi (2000) ‘Cohabitation in Great Britain: Not
For Long, But Here To Stay’, Journal of the Royal Statistical Society, Series A,
163:153–171.
[17] Harrop, Anne and Ian Plewis (1995) ‘Two Decades of Family Change: Secondary
Analysis of Continuous Government Surveys’, Journal of the Royal Statistical
Society, Series A, 158:91–106.
[18] Huber, Joan and Glenna Spitze (1980) ‘Considering Divorce: An Expansion of
Becker’s Theory of Marital Instability’, American Journal of Sociology, 86:75–
89.
[19] Jalovaara, Marika (2001) ‘Socio-Economic Status and Divorce in First Marriages
in Finland 1991–93’, Population Studies, 55:119–133.
33

[20] Jenkins, Stephen P. (1996) ‘Recent Trends in the UK Income Distribution: What
Happened and Why?’, Oxford Review of Economic Policy, 12:29–46.
[21] Kaufman, Gayle and Peter Uhlenberg (2000) ‘The Influence of Parenthood on
the Work Effort of Married Men and Women’, Social Forces, 78:931–949.
[22] Koo,Helen P., C.M. Suchindran and Janet D. Griffith(1984) ‘The Effectsof Chil-
dren on Divorce and Re-marrige: A Multivariate Analysis of Life Table Proba-
bilities’, Population Studies, 38:451–471.
[23] Lampard, Richard and Kay Peggs (1999) ‘Repartnering: The Relevance of Par-
enthood and Gender to Cohabitation and Remarriage Among the Formerly Mar-
ried’, British Journal of Sociology, 50:443–465.
[24] Mauldon, Jane (1992) ‘Children’s Risks of Experiencing Divorce and Re-
marriage: Do Disabled Children Destabilize Marriages?’, Population Studies,
46:349–362.
[25] McLanahan, Sara and Gary Sandefur (1994) Growing Up With A Single Parent:
What Hurts, What Helps, Cambridge, Mass.: Harvard University Press.
[26] Morgan, S. Philip and Ronald R. Rindfuss (1985) ‘Marital Disruption: Structural
and Temporal Dimensions’, American Journal of Sociology, 90:1055–1077.
[27] Morgan, S. Philip, Diane N. Lye and Gretchen A. Condran (1988) ‘Sons, Daugh-
ters, and the Risk of Marital Disruption’, American Journal of Sociology, 94:110–
129.
[28] Murphy, Michael (2000) ‘The Evolution of Cohabitation in Britain, 1960–95’,
Population Studies, 54:43–56.
[29] Rohwer, G¨otz and Ulrich P¨otter (2001) “TDA User’s Manual”. (Available at
http://steinhaus.stat.ruhr-uni-bochum.de/tda.html)
34

[30] Sørensen, Annemette and Sara McLanahan (1987) ‘Married Women’s Economic
Dependency: 1940–1980’, American Journal of Sociology, 93:659–687.
[31] South, Scott J. (2001) ‘Time-Dependent Effects of Wives’ Employment on Mar-
ital Dissolution’, American Sociological Review, 66:226–245.
[32] South, Scott J. and Glenna Spitze (1994) ‘Housework in Marital and Nonmarital
Households’, American Sociological Review, 59:327–347.
[33] Thornton, Arland (1985) ‘Changing Attitudes toward Separation and Divorce:
Causes and Consequences’, American Journal of Sociology, 90:856–872.
[34] Thornton, Arland (1989) ‘Changing Attitudes toward Family Issues in the United
States’, Journal of Marriage and the Family, 51:873–893.
[35] Tzeng, Jessie M. and Robert D. Mare (1995) ‘Labor Market and Socioeconomic
Effects on Marital Stability’, Social Science Research, 24:329–351.
[36] VanLaningham, Jody, David R. Johnson and Paul Amato (2001) ‘Marital Hap-
piness, Marital Duration, and the U-Shaped Curve: Evidence from a Five-Wave
Panel Study’, Social Forces, 78:1313–1341.
[37] Waite, Linda J. and Lee A. Lillard (1991) ‘Children and Marital Stability’, Amer-
ican Journal of Sociology, 96:930–953.
[38] Weiss, Yoram and Robert J. Willis (1985) ‘Children as Collective Goods and
Divorce Settlements’, Journal of Labor Economics, 3:268–292.
[39] Weiss, Yoram and Robert J. Willis (1997) ‘Match Quality, New Information, and
Marital Dissolution’, Journal of Labor Economics, 15:S293–S329.
[40] White, Lynn (1990) ‘Determinants of Divorce: A Review of Research in the
Eighties’, Journal of Marriage and the Family, 52:904–912.
35

[41] White, Lynn and John H. Edwards (1990) ‘Emptying The Nest and Parental Well-
Being: An Analysis of National Panel Data’, American Sociological Review,
55:235–242.
36